Abstracts

Oral communications

S. Accordini¹, A. Corsico², I. Cerveri², L. Cazzoletti¹, E. Duran-Tauleria³, D. Jarvis⁴, C. Janson⁵, J. Martinez-Moratalla⁶, R. Marco¹.¹University of Verona, Unit of Epidemiology & Medical Statistics, Verona, Italy; ²IRCCS Policlinico S. Matteo, University of Pavia, Division of Respiratory Diseases, Pavia, Italy; ³Institut Municipal d’Investigació Mèdica, Barcelona, Spain; ⁴King’s College, Department of Public Health Sciences, London, UK; ⁵University of Uppsala, Dept of Medical Sciences, Respiratory Medicine and Allergol., Uppsala, Sweden; ⁶Hospital General de Albacete, Albacete, Spain

Introduction: According to the Global Initiative for Asthma (GINA) Guidelines, well treated asthmatics are expected to have no limitation of their daily life activities. The aim of this study was to evaluate life impairment and its determinants in adult asthmatics from the general population.

Methods: Life impairment (production losses and leisure time forgone) and hospitalisation (emergency department visits and hospital admissions) were assessed in the frame of the ECRHS-II, a multicentre population based survey performed in 1998–2000 on the adults (28–56 years old) from 13 Western countries who participated in the ECRHS in 1991–1993. For the purpose of the present analysis, the 1291 subjects who reported current asthma confirmed by a doctor at a clinical interview were considered. Asthma severity was classified according to the GINA criteria. Multilevel logistic regression models were used to evaluate the predictors of life impairment and hospitalisation.

Results: 58%, 10.8%, 14.8%, and 16.4% of the patients were intermittent, mild, moderate, and severe persistent asthmatics, respectively. The percentage of subjects who lost at least one working day because of the disease in the previous year was 16.4% (median number of days 7; interquartile range, IQR 3 to 14 days), ranging from 12.1% among intermittent asthmatics to 29.4% among severe persistent patients, whereas 12.4% of the subjects reported at least one impaired day (median number of days 24; IQR 12 to 60 days), ranging from 8.7 to 20.3% according to the level of disease severity. Six per cent of the subjects had at least one hospitalisation in the past 12 months. The risk of life impairment and the risk of hospitalisation were significantly higher for severe persistent asthmatics than for intermittent patients (OR 3.00; 95%CI 1.88 to 4.81; and OR 4.64; 95%CI 2.33 to 9.24, respectively). Sex (female) (OR 1.61; 95%CI 1.12 to 2.32) and low education level (OR 1.49; 95%CI 1.00 to 2.20) were positively associated with the risk of life impairment.

Conclusions: Life impairment is still a substantial problem for adult asthmatics: on average, one patient out of six lost 7 working days and one patient out of eight was impaired for 24 days per year.

E. Agerbo.University of Aarhus, National Centre for Register-Based Research, Denmark

Introduction: This study aimed to estimate the risk of suicide among people whose marital partner has been admitted with a psychiatric disorder or has died by suicide or other causes.

Methods: Danish population based registers were used to identify the 9011 people aged 25–60 years who had committed suicide during 1982–1997, and 18 0220 age–sex time matched controls and their spouses.

Results: The risk of suicide was increased in people whose spouse had been first admitted after December 31 two years earlier (RR 4.97; 95% CI 3.46 to 7.14), whose spouse had committed suicide (RR 22.13; 95% CI 11.30 to 43.35) or died by other causes (RR 7.67; 95% CI 4.98 to 11.81). The suicide rate was unrelated to spousal concordance for psychiatric admission and spousal diagnosis.

Conclusion: The suicide risk is increased in spouses whose partners suffer from a psychiatric disorder or have committed suicide. The mental illness and suicide of a spouse can have a huge impact on the other spouse, and this impact is not necessarily genetically mediated.

W. Ahrens, W. Schill, C. Goddon, J. Weiss, W. Dahlke, B. Marschall.Bremen Institute for Prevention Research and Social Medicine (BIPS), Bremen Volkswagen AG, Corporate Health Division, Wolfsburg, Germany

Introduction: A cluster of testicular tumours in one car manufacturing plant gave rise to a suspected occupational cause. Exposure to electromagnetic fields during resistance welding operations was considered a possible hazard. An incidence study was conducted to describe the occurrence of the disease in different manufacturing plants by department and time period and to find out whether the incidence rate was raised above expected values of a reference population.

Methods: The cohort comprised 167 212 men employed between 1989 and 2000 in six different plants. Over the study period, 160 incident cases were identified through health insurance records. The age–period specific incidence of testicular cancer from the Saarland for the same time period was used as a reference for the calculation of crude and standardised incidence ratios (SIRs, Segi truncated world standard, 15–69). Internal comparisons between departments and plants were carried out using Cox regression. 95% CIs were calculated for SIRs and risk ratios (RR).

Results: The average age of the cohort was 40.7 years. The study confirmed the suspected excess incidence in plant A, which was 66% above the expected rate. In the total cohort, the standardised incidence rate was 9.18 (95% CI 7.46 to 11.30) for the whole period, 6.37 (95% CI 4.29 to 9.45; n = 52 cases) between 1989 and 1995, and 11.42 (95% CI 9.35 to 13.96; n = 108) between 1996 and 2000. For the early period the typical peak of the incidence between 25 and 40 years of age was missing. Compared with the Saarland, the SIR for the early period was below expectation (SIR 0.52; 95% CI 0.39 to 0.68), whereas it was increased in the second half (SIR 1.1; 95% CI 0.89 to 1.30). In plant A the corresponding SIRs were (SIR 1.02; 95% CI 0.56 to 1.84; n = 11) and (SIR 2.18; 95% CI 1.43 to 3.31; n = 22) for the early and late period, respectively. Overall, the excess risk was most pronounced in the older age groups above 50 years, particularly in plant A with an SIR of 24.24 (95% CI 10.47 to 47.77; n = 8). The internal comparison revealed increased risk ratios in the following work areas, where all other areas served as reference: maintenance (RR 2.4; 95% CI 1.05 to 5.41), administration/office work (RR 1.7; 95% CI 1.08 to 2.67), assembly line (RR 1.4; 95% CI 0.94 to 2.05), and machine assembly (RR 1.4; 95% CI 0.94 to 2.07).

Conclusions: The analysis stratified by work areas may give first clues for possible disease risks. However, the classification has to be considered as preliminary. It does not give sufficient information about possible exposures. Moreover, the reduced incidence for the early study period indicates a possible underreporting of cases. Currently an incidence validation study is under way to clarify this issue and to provide data for a nested case control study that should provide more insight into possible explanations for our observations.

J. Allenberg¹, H. Darius², C. Diehm³, R. Haberl⁴, D. Pittrow⁵, B. v. Stritzky⁶, G. Tepohl⁷, H. J. Trampisch⁸.¹University of Heidelberg, Department of Vascular Surgery, Heidelberg, Germany; ²Vivantes Hospital Berlin Neukoelln, I. Med. Clinic, Berlin, Germany; ³Teaching Hospital of the University of Heidelberg, Department of Internal Medicine, Karlsbad Langensteinbach, Germany; ⁴Municipla Hospital Munich Harlaching, Department of Neurology, Munich, Germany; ⁵Technical University Dresden, Department of Clinical Pharmacology, Dresden, Germany; ⁶Sanofi-Synthelabo, Berlin, Germany; ⁷Practice, Internal Medicine/Vascular Medicine, Munich, Germany; ⁸Ruhr-University of Bochum, Department of Medicine, Informatics, Biometry and Epidemiology, Bochum, Germany

Introduction: Previous studies in selected patient samples suggested a high risk for total mortality and cardiovascular morbidity associated not only with symptomatic, but also with asymptomatic peripheral arterial disease (PAD).

Objectives: Our aim was to assess the 1 year risk of death and cardiovascular morbidity associated with PAD in a large, unselected sample of elderly patients in primary care. In addition, the strength of association between low ankle brachial index (ABI, as indicator for PAD) and various accepted risk factors for PAD, and death and outcome was quantified.

Methods: In a prospective cohort study, the German epidemiological trial on Ankle Brachial Index (getABI), 6880 unselected patients 65 years were followed up by 344 primary care physicians. Main outcome measures were 1 year mortality due to all causes, mortality due to cardiovascular and cerebrovascular disease, and severe vascular events (myocardial infarction, coronary revascularisation, stroke, revascularisation at carotids, amputation, or peripheral revascularisation because of PAD).

Results: The baseline characteristics of the patients were as follows: mean age 72.5 years, 58% females, 46% (ever) smoker, 65% hypertension, 25% diabetes mellitus, 52% lipid disorders. The prevalence of PAD in the cohort was 18.0%. At 1 year, all cause mortality was 2.8% in patients with PAD and 0.9% in patients without PAD (odds ratio (OR) adjusted for known risk factors: 2.0; 95% CI 1.3 to 3.3). Mortality due to CHD or CVD was 1.6% v 0.4% (adjusted OR 2.5; 95% CI 1.3 to 4.9). Severe vascular events or death occurred in 8.3% of PAD patients, and in 2.7% of patients without PAD (adjusted OR: 2.2; 95% CI 1.6 to 2.9). Patients with PAD and at least 2 of the following 3 additional risk factors—diabetes, (ever) smoking, history of severe cardiovascular or cerebrovascular events—had a substantially increased risk of dying or suffering from a severe vascular event during follow up in comparison to patients without PAD and at most one additional risk factor: 13.9% v 2.0% (adjusted OR 6.2; 95% CI 4.2 to 9.0).

Conclusions: Patients with PAD have an increased risk for (short term) all cause mortality and severe vascular events. Hence, PAD is closely associated with other manifestations of atherosclerosis ("indicator disease"). In a real life setting the ABI is a useful means of identifying patients with PAD and thus with advanced atherosclerotic disease. A subgroup of patients with a substantially increased risk can be identified when three risk factors, which can be obtained very easily, are taken into account additionally.

M. Almeida¹, G. Alencar¹, L. Rodrigues², I. França³, A. Siqueira³, D. Shoeps¹, M. Novaes⁴.¹Universidade de São Paulo, Epidemiologia/Faculdade de Saúde Pública, São Paulo, Brasil; ²London School of Hygiene and Tropical Medicine, Epidemiology, London, UK; ³Universidade de São Paulo, Saúde Materno Infantil/Faculdade de Saúde Pública, São Paulo, Brasil; ⁴Universidade de São Paulo, Medicina Preventiva, São Paulo, Brasil

Introduction: Perinatal mortality rate in São Paulo City is declining, but fetal mortality has rarely been studied and its risk factors are almost unknown. Social and health inequalities are important in São Paulo and fetal mortality was 932/1000 births in 2002, ranging from 623 to 1034 from the richer to the poorer areas. The south region of the city was selected to carry out a perinatal and fetal mortality study because of its inadequate living and health conditions.

Methods: This is a population based case-control study. Cases and controls were selected from the vital statistics information system from mothers living in the South region of São Paulo city (01/08/2000 to 31/01/2001). Cases (172) included fetal deaths with weight 500 g and over. Controls (313) were obtained by random sampling of the survivors. Home interviews were performed and data were obtained from hospital records. Variables were grouped into blocks according to the conceptual framework, from distal to proximal pathway: socioeconomic, obstetric history and maternal conditions, gestational conditions, and delivery conditions. The data were modelled by hierarchical multiple logistic regression. The variables were adjusted within each block, which were added subsequently.

Results: To classify fetal deaths we used the clinical information of the delivery hospital. Antepartum deaths (156) represented 90.6% of fetal deaths, intra-partum (8) 4.7% and 8 (4.7%) remained unclassified, but these showed the same characteristics as ante-partum deaths, and were grouped (164). Modelling process analyses showed that socioeconomic variables associated were: mother being unmarried (OR 2.98; 1.83 to 4.87), being married less than 1 year (OR 2.79; 1.43 to 5.46) and having had less than 3 years of education (OR 1.91; 1.10 to 3.33). Including maternal history remained significant: mothers having had previous low weight birth (OR 2.20; 1.22 to 3.99). Including variables related to this gestation remained significant: mother being unmarried (OR 2.67; 1.24 to 5.72); being married <1 year (OR 1.87; 1.06 to 3.33); vaginal bleeding (OR 5.19; 1.82 to 14.81), presence of hypertension (OR 2.87; 1.40 to 5.88), inadequate pre-natal care (OR 2.75; 1.75 to 4.32), pregnancy weight gain less than 6.0 kg (OR 2.40; 1.42 to 4.05), diabetes (OR 32.97; 3.72 to 292.21). Adding delivery conditions were significant causes of premature delivery, both those identified (OR 12.04; 4.18 to 34.72) and unidentified (OR 6.09; 2.69 to 13.78); mothers transport to hospital by bus/friends cars (OR 2.41; 1.6 to 5.45), and low birth weight (LBW) (OR 12.39; 6.44 to 23.81). Final model included vaginal bleeding (OR 6; 1.5 to 23.3), hypertension (OR 2.5; 1.02 to 6.3), diabetes (OR 22.3; 1.7 to 290.05), identified/unidentified prematurity (OR 6.1; 2.7 to 13.8; OR 12.0; 4.2 to 34.7), transportation (OR 2.4; 1.1 to 5.5) and LBW (OR 12.4; 6.4 to 23.8).

Conclusions: Ante-partum deaths are the main component of fetal mortality in the south of São Paulo, as in developed countries, but with higher rates. Even though socioeconomic variables did not remain in the final model, mother’s social vulnerability might work through proximal variables such as gestational and delivery conditions (transportation, hypertension, prematurity, and LBW). Adequate health care may have impact in some of these factors, while others can be more difficult to prevent and treat.

Acknowledgement to FAPESP and CNPq

S. Almeida¹, M. Pina², C. Lopes¹.Faculdade de Medicina da Universidade do Porto, Serviço de Higiene e Epidemiologia, Porto, Portugal; ²Instituto de Engenharia Biomédica, Laboratório de Biomateriais, Porto, Portugal

Introduction: In spite of the decrease in cardiovascular mortality in industrialised Western countries, in Portugal, cardiovascular diseases are the leading cause of death and one of the major causes of serious, long term disability among adults.

Objectives: The aim of this study was to analyse the geographical distribution of stroke and coronary heart disease mortality in Portugal, on a municipality level, from 1992 to 2000.

Methods: Mortality data were obtained from the National Mortality Database, aggregated by municipality, with information on number of deaths by cause, sex and age group (10 years), from 1992 to 2000. The data were grouped into 3-year time periods (1992–1994, 1995–1997, and 1998–2000) and the middle period estimated population was used to calculate the indirect age standardised mortality rates of stroke and coronary heart diseases for both sexes. Changes in age standardised death rates were calculated and compared between the three periods. We used a Geographical Information System to identify spatial patterns in the geographical distribution of the mortality rates.

Results: The mortality rates for both diseases present distinct geographical patterns, with the highest mortality rates of stroke in the north of Portugal, and the highest mortality rates of coronary heart in the south of the country. The mortality rates for stroke had decreased from 734.77 per 100 000 in the first period to 638.02 per 100 000 in the last period (13% reduction), while the coronary heart disease had decreased from 283.47 per 100 000 to 262.99 per 100 000 (7% of reduction), respectively. Although the general rates for Portugal have been reduced in the study period, some particular regions had the mortality rates increased. This is the case for Alentejo, in south Portugal, whose mortality rates of stroke had increased about 8% and also the case of the central east of the country, with an increase of about 40% in the mortality rates of coronary heart diseases in the study period.

S. Alves, M. Pina, M. Barbosa.Instituto de Engenharia Biomédica, Laboratório de Biomateriais, Porto, Portugal

Osteoporosis, defined as a skeletal disorder characterised by compromised bone strength predisposing to an increased risk of fracture, affects million of people around the world, with a tendency of increasing incidence rates, due to the aging of the population and due to the increase of risk factors. The most visible and dramatic consequences of osteoporosis are osteoporotic fractures that can occur mainly on three skeleton sites: proximal femur (hip), vertebrae, and distal forearm (wrist). Of those, the hip fractures are considered the most serious osteoporotic fractures, because they almost invariably lead to hospital admission for surgical intervention. The mortality and morbidity rates are both increased in persons who had an osteoporotic hip fracture. Therefore, because hip fractures are almost always treated in public hospitals they are highly documented and easy to count. Some authors consider that hip fractures can be considered a barometer of osteoporosis. Although the osteoporotic fractures vary among different countries, and inside the same country among different areas, suggesting an important paper of environmental and geographical factors, few epidemiological studies have been presented considering the geographical distribution of the cases.

The objective of this study is to identify spatial patterns of femur fractures incidence in Portugal, through the use of Geographical Information Systems—GIS and Spatial Statistical Analysis.

The health data being used in this project are from the National Hospital Inpatients Data Register managed by the Health Informatics and Financial Management Institute (Instituto de Gestão Informática e Financeira da Saúde; IGIF) of the Portuguese Health Ministry. This database contains all the admissions with a diagnosis related to femur fracture that occurred between the years 2000 and 2002 and variables such as sex, age, cause of admissions, and provenience. We selected the registers that occurred in individuals with 50 or more years of age, with diagnosis classified as ICD9-CM codes: 820.x and cause of admission with ICD9-CM codes: E880, E881, E884, E885, E886, E887, and E888.

The demographic and socioeconomic data comes from the 2001 Portuguese Census.

We computed the number of patients with 50 years or more for each municipality according to sex and age (5 year age group), and calculated the age and sex direct standardised rates. The local empirical Bayesian estimator adjusted the incidence rates. The results showed that the incidence of osteoporosis increased with age and was higher in women than in men. Clusters of high rates were identified in the south and northeast of continental Portugal, as well as in the littoral between Lisbon and Aveiro city, we also identified clusters of low rates northwest and central inland.

In conclusion, the incidence of osteoporosis has different spatial patterns within the Portugal mainland territory.

J. Amaral¹, E. Pereira¹, M. Paixão².¹IST, Department of Mathematic, Lisbon; ²INSA, CVEDT, Lisbon

Introduction: A statistical procedure called Back calculation has been a widely used method to estimate the size of the population with HIV/AIDS. Recent discussions suggest that this method should be adapted to take into account the effect of treatment. The aim of this study is to use that method and a worst case scenario, in order to assess the quality of previous projections and obtain new ones.

Methods: The AIDS incidence data were adjusted, taking into account reporting delays, no reporting, and underreporting. A conditional likelihood estimation of the reporting delay distribution was performed using standard algorithms for Poisson regression. Then lower bounds on the size of the AIDS epidemic were obtained. A Weibull and Gamma distribution were considered for the latency period distribution. The EM algorithm was applied to obtain maximum likelihood estimates of the HIV incidence. The density of infection times was parameterised as a step function. The National AIDS/HIV database was used and the methodology was applied to four different transmission categories, injecting drug users (IDU), heterosexual, homo/bisexual and "other", to obtain short term projections (2002–2005) and an estimate of the minimum size of the epidemic.

Results: 2002–2005 projections show an annual decrease of new AIDS cases for the IDU and the homo/bisexual transmission categories. For the year 2003 (in comparison with 2002) a decrease of 7.3% is expected in IDU. In this category the number of new cases is estimated to decrease by 9.5% in 2004 and to suffer a further decrease of 11.8% in the year 2005. However, the heterosexual and the other are estimated to increase. In the first category the increase rate is not very pronounced: in 2003 the incidence is 5.5% higher than estimated for 2002, in 2004 is 3.5% higher when compared to 2003 and 1.3% in 2005 relatively to 2004. In the second, the expected annual increase rate for the number of new AIDS cases is very significant: in 2003 the incidence rises 36% compared to 2002, in 2004 29% higher than in 2003, and in 2005 23.3% higher than in 2004.

Conclusion: This study outlines a methodology for obtaining projections on the size of the AIDS epidemic. It produces a lower bound, because it only estimates the cumulative numbers of individuals that will eventually develop AIDS from those already infected with HIV. The projections indicate for the next few years: an increase in the annual number of AIDS new cases in the heterosexual and other transmission categories, with special incidence on this last one; a slowing down of the annual number of new AIDS cases in the IDU transmission category; an apparent change on the decrease tendency in the homo/bisexual category. These projections indicate that HIV/AIDS will remain a public health problem in Portugal.

O. Amaral¹, C. Pereira², A. Escoval³.¹ISCTE, Lisboa; ²ESEnfV, Viseu; ³ISCTE, Lisboa, Portugal

Background: Overweight and obesity may interfere with social, psychological, and physical activities. This study aimed to analyse the association between obesity and health related quality of life (HRQOL) in adolescents.

Methods: We designed a cross sectional study involving 2144 students (51.5% females) aged 12 to 18 years (14.8; SD 1.7) among seven secondary schools from Viseu, Portugal. Participants completed a self administered questionnaire. Body mass index (kg/m²) was calculated from self reported height and weight and classified into tree groups: normal weight (<24.9), overweight (25.0 to 29.9), and obese (30.0) according to the adolescent’s international body mass index cut off points for overweight and obesity by sex and age. HRQOL was assessed by SF36 (ranging from 0 to 100) assessing nine dimensions: physical functioning, social functioning, role limitations physical, bodily pain, general medical health, mental health, role limitations mental, vitality, and general health perceptions. Proportions were compared by ² test and continuous variables by Kruskal Wallis test.

Results: Overweight prevalence in total sample was 12.6% (13.4 in boys and 11.8 in girls) and obesity was 3.7% (4.1% in boys and 3.4% in girls). Obesity in girls was associated with significantly lower total scores of HRQOL (64.6 (SD 11.3) v 70.6 (SD 12.9) p = 0.02). Overweight was associated with impaired physical well being among girls. In boys, no significant differences were found between overweight and not overweight. Obesity was associated with lower general health perception and poorer physical functioning (girls only) but did not impaire emotional well being.

Conclusion: The presence of obesity is associated with impaired HRQOL only in female sex, with deterioration in physical but not emotional well being. The impairment found in HRQOL in obese girls can be associated with restrictions in daily life.

C. Ancona¹, V. Belleudi¹, C. Marinacci², A. Russo³.¹ASL RME, Epidemiology Department, Rome, Italy; ²Epidemiology Unit, Piedmont Region, Italy; ³ASL Città di Milano, Epidemiology Unit, Milan, Italy

Introduction: Tackling inequalities in health after surgical procedures has just recently begun receiving more emphasis, and findings are controversial.

Objective: To investigate the association between income and 30 day mortality after coronary artery bypass graft (CABG) surgery in Italy.

Methods: We used discharge abstract data to define a cohort of 7163 patients (age 35 years) residents in Rome, Milan, and Turin who underwent isolated CABG surgery (ICD IX code: 36.1) between 1997 and 2000. Vital status 30 days after CABG surgery was obtained from the Municipal Registries. Census tract median income (CTMI) was computed by a record linkage between Tax and Population Registers for the three cities (4.8 million residents). The CTMI distribution was divided into five quintiles ranging from very underprivileged (I quintile) to very well off (V quintile). We evaluated the association between 30 day mortality and the income quintiles using a multiple logistic regression model with empirical risk adjustment (Odds Ratios, OR, and 95% CI). A priori risk factors considered were: age, type of ischemic heart disease, comorbidities, and previous heart surgery. We used two different risk adjustment models for males and females.

Results: In the CABG cohort 82.0 % of patients were males. The mean age was 65.4 years (SD 8.9), patients in the lowest income quintile seem to have CABG earlier in their lives (mean age 64.6 years, SD 8.7 v 66.3 years, SD 8.7 in the V quintile). 21.5% of the cohort was in the I income quintile v 18.4 % in the V quintile. The overall 30 day mortality was 2.2%; 2.6% in the I income quintile and 2.0% in the V income quintile. The crude OR association between 30 day mortality after CABG surgery and income level was 0.76; 95% CI 0.46 to 1.24. After adjusting for the potential confounders no protective effect of the highest income on 30 day mortality was found (for the V level, OR 0.79; 95% CI 0.48 to 1.30). No significant confounding effect of the risk factors was observed.

Conclusions: Income level did not influence early mortality after CABG surgery in Italy in the period under study. Other outcomes such as 6 month mortality and hospital re-admission for coronary revascularisation should also be examined to draw more definitive conclusions about possible inequalities in CABG effectiveness in the Italian NHS.

V. Andreozzi¹, M. Carvalho¹, M. Barreto².¹Oswaldo Cruz Foundation, National School of Public Health, Rio de Janeiro, Brazil; ²Federal University of Bahia, Public Health Institute, Bahia, Brazil

Introduction: The choice of modelling strategies can improve or impair the best designed studies. It is well known the role of recent developed techniques, such as generalised estimating equations and random effects models, on analysing longitudinal data. On the other hand, late updates on survival analysis models allow the inclusion of time dependent covariates, multiple events, and random effects. In this paper we compare the use of these methods in a longitudinal study on the effects of Vitamin A in preventing childhood diarrhoeal disease.

Methods: Data come from a randomised, double blind, placebo controlled clinical trial where over 1000 children aged 6–48 months were assigned to receive either vitamin A or placebo every 4 months over 1 year. Demographic, clinical, nutritional, and socioeconomic characteristics were collected at the baseline. The subjects were visited at home three times a week, to register the number of liquid or semi-liquid motions a day. Three response variables were considered: number of liquid or semi-liquid motions per week, numbers of days with diarrhoea per week, in the longitudinal models; numbers of days without diarrhoea in the survival models.

Results: The survival models are more comparable to the longitudinal models for the number of days with diarrhoea, because both of them model the occurrence of diarrhoea episodes. For the two ways of looking at the problem, no significant effect was found for the vitamin A supplementation. The only significant effect was found when modelling the severity of the disease using the number of liquid or semi-liquid motions per week. The most important covariates were the age of the child, as expected, and the number of domestic appliances (refrigerator, radio, TV).

Conclusion: The comparison of the models showed reasonable consistency; even the differences found can be attributed to different measures and ways of looking at the problem. The major benefit of trying those various models is the ability of detecting small change in point and confidence interval estimates, including independent variable of interest and all potential risk factors. The variety of statistical models is increasing sharply with the development of statistical tools and availability of software. Unfortunately the choice among them is usually more dependent on previous knowledge than on the real question addressed by the study. Quite often data like these are only analysed using simple summary measures in classical models easily adjusted in common statistical packages, causing loss of information so hardly gathered. There is a strong need in the epidemiology environment of incorporation of an entire set of statistic techniques dealing with dependent data, particularly from longitudinal studies.

J. Antunes¹, E. Waldman², C. Borrell³.¹University of Sao Paulo, School of Dentistry, Sao Paulo, Brazil; ²University of Sao Paulo, School of Public Health, Sao Paulo, Brazil; ³Agència de Salut Pública de Barcelona, Barcelona, Spain

Introduction: The use of highly active antiretroviral therapy (HAART) is associated with impressive improvements of AIDS survival after the mid 1990s in affluent countries with a high access to these medications. However, HAART has also been reported as a factor for the increase of socioeconomic inequalities in health, because of unequal access and adherence to treatment. The objective of the present study was to describe trends of AIDS mortality in the city of São Paulo, Brazil, from 1995 to 2002. In particular, we aimed at assessing the association between the annual percentage decrease of AIDS mortality and socioeconomic characteristics of the city’s boroughs, in a context of large scale and cost free distribution of HAART.

Methods: The gathering of population data and information from death certificates allowed the calculation of AIDS yearly death rates for each district of the city, as adjusted by age group and sex distribution. The characterisation of areas assessed income, educational attainment, living conditions, and the human development index. Trend estimation used the auto regression procedure of exact maximum likelihood estimation for time series analysis. The study of association between the annual percentage decrease of AIDS deaths and socioeconomic indices used three schemes of regression analysis: ordinary least squares, simultaneous autoregressive, and conditional autoregressive.

Results: AIDS mortality decreased in São Paulo from 32.1 deaths (per 100 000 inhabitants) in 1995 to 11.2 in 2002. This reduction corresponded to an annual percentage decrease of 16.9%. The decrease of rates was concurrent with a shrinking variation of figures among the districts, as a steeper decline of AIDS deaths was observed for areas with higher prior levels (1995) of mortality. Although the decline of AIDS mortality was not homogeneous in the city, discrepant patterns of decrease in the districts did not associate with figures of social development, with all correlation coefficients corresponding to p values higher than 0.27. Map production indicated that districts with different levels of decrease were similarly distributed in the more affluent central portion of the city, and in its poorer peripheral areas.

Conclusions: The evidence gathered in the present study suggests that the recent reduction of AIDS deaths in São Paulo was not influenced by major estimates of SES indices of the city’s neighbourhoods. The overall reduction of rates and a higher decrease in areas with higher prior levels of deaths indicate that the Brazilian programme of AIDS treatment may have effectively contributed to the reduction of inequalities in the profile of AIDS deaths. Despite its recent decline, mortality from AIDS related conditions remains a major problem of public health, and efforts of health promotion must take account of new challenges posed by the changing epidemiologic pattern of AIDS indices in the country.

M. Arcà, D. Fusco, C. Saitto, C. Cimaglia, P. Colais, C. Sforza, C. A. Perucci.Local Health Authority RME, Department of Epidemiology, Roma, Italy

Introduction: Present health status is generally related to future resource consumption, and a measure of the prospective relationship between health conditions and care costs is needed to cope with the financial uncertainty of providing care to people with specific diseases. In most developed countries, large administrative databases on the health services delivered are currently available. These provide information on individual health status and might represent a cost effective tool in predicting further health expenditures.

Methods: We used hospital discharge abstract databases from two Italian regions (over 10 000 000 inhabitants). The databases include information on every hospital admission, including up to six discharge diagnoses and six treatments, both coded according to the ICD9-CM. Demographic information and an individual identity code is also available. We used data referring to a two year interval (1997–1998) to select a cohort of diabetic patients aged 15 to 80 years to attribute them a score of disease severity and to identify comorbidities. We then followed them up for hospitalisation in the next 3 years (1999, 2000, and 2001) to assess how baseline clinical characteristics could affect subsequent health expenditures and to construct algorithms aimed at predicting costs. The reimbursement attributed to each episode of care by the Italian prospective payment system, based on the Diagnosis Related Groups (DRG) classification, was used as a proxy of health expenditures. We developed a two step predictive model for subsequent expenditures by combining a multivariate logistic regression model, to predict the probability of (at least) one subsequent admission, and a generalised linear model, with logarithmic link function and Gamma distribution, to predict the expected hospital cost of admitted patients. Both models accounted for demographic variables, past hospitalisations, and comorbidities. Health expenditures of each diabetic patient could then be individually predicted by multiplying the probability of being admitted and the predicted cost of admission(s).

Results: Of the 70 519 diabetic patients selected, 57% were classified as having mild diabetes, 34% as moderate, and 9% as severe. 31% of patients were 60–69 years old and 39% were 70–79 years old. 68% of the patients presented at least one chronic comorbidity. During the 3 year follow up, 1.7 admissions per patient were reported, for 16.6 days of average hospital length of stay and a 4700 average hospital cost. After the modelling step, the observed 3 year hospital cost averaged 2944 in the lowest quintile of predicted cost and 7807 in the highest quintile of predicted cost.

Conclusions: Administrative data can be used to collect information on individual health status and health expenditures. Regression models can be developed to reliably predict future resource consumption and health expenditure in subgroups of patients with chronic diseases.

The study was partially funded by a grant from ANIA, the National Association of Insurance Companies.

M. Artama¹, A. Ritvanen², M. Gissler³, J. Isojarvi⁴, A. Auvinen¹.¹University of Tampere, Tampere School of Public Health, Tampere, Finland; ²Stakes, Finnish Register of Congenital Anomalies, Helsinki, Finland; ³Stakes, Finnish Medical Birth Register, Helsinki, Finland; ⁴University of Oulu, Department of Neurology, Oulu, Finland

Introduction: Use of antiepileptic medication during pregnancy is associated with increased risk for congenital malformations in the offspring. Most previous studies on malformations in the offspring of patients with epilepsy have been hospital based with small and selected study populations.

Objective: The aim of this study was to obtain valid and accurate estimates on major congenital malformations in the offspring of patients with epilepsy in a large population based cohort of women with epilepsy.

Methods: All women (n = 6535) who were approved as eligible for full reimbursement for antiepileptic medication for the first time between 1985 and 1994 from the Social Insurance Institution of Finland (KELA) were identified from the KELA database. A reference cohort (n = 14 704) was identified from the Finnish Population Register Center. Information on live births (n = 7548) was obtained from the Medical Birth Register maintained by STAKES—the National Research and Development Center for Welfare and Health. Information on malformations (n = 555) was obtained from the Finnish Register of Congenital Anomalies maintained also by STAKES.

Results: The overall prevalence of malformations and the proportion of offspring with multiple malformations were more than twice as high in the offspring of mothers with epilepsy as in the offspring of mothers without epilepsy. The greatest absolute excess (4–7/1000 births) was observed in cardiovascular malformations, congenital malformations of urinary system, limbs, genital organs, and spina bifida. The risk for spina bifida (OR 11.32; 95% CI 2.34 to 108) and congenital malformations of genital organs (OR 8.39; 95% CI 2.16 to 47.5) was substantially elevated in the offspring of mothers with epilepsy. The risk for congenital malformations varied slightly by maternal age at delivery, but no clear trend was observed. The risk increased with duration of epilepsy (p = 0.02 for trend).

Conclusion: Risk for certain major congenital malformations is elevated in the offspring of patients with epilepsy, especially among those with long duration of maternal epilepsy.

A. Azevedo¹, A. Santos¹, P. Bettencourt², C. Abreu-Lima³, H. Barros¹.¹University of Porto Medical School, Hygiene and Epidemiology, Porto, Portugal; ²Hospital de S. João, Medicina B, Porto, Portugal; ³Hospital de S. João and University of Porto Medical School, Cardiology, Porto, Portugal

Introduction: The importance of clustering and interaction of cardiovascular risk factors, expressed in the definition of the metabolic syndrome, has been increasingly emphasised in the prediction of cardiovascular diseases, particularly coronary heart disease (CHD). The aim of this study was to assess the prevalence of objective structural and functional cardiac abnormalities according to factors that define the metabolic syndrome, either alone or in association.

Methods: Cross sectional study for which community participants aged 45 years were recruited. Data were collected by clinical interview, including Rose’s questionnaire, physical examination, ECG, echocardiogram, and fasting venous blood sample collection. Metabolic syndrome was defined according to NCEP as the presence of at least three of the following: blood pressure (BP) 130/85 mm Hg, triglycerides 150 mg/dl, HDL <40 mg/dl in men, and <50 mg/dl in women, fasting glucose 110 mg/dl, and waist circumference 102 cm in men and 88 cm in women. CHD was considered when there was history of angina or myocardial infarction, either self reported or according to Rose’s questionnaire, or pathological Q waves on ECG. We calculated age, sex, and education adjusted prevalence or means and their 95% CI, by logistic and linear regression, respectively.

Results: As shown in the table, the prevalence of CHD, left ventricular systolic dysfunction (either defined by the subjective impression of the ecographist eyeball or as ejection fraction<45%), left ventricular hypertrophy (LVH), and impaired relaxation was larger in subjects with metabolic syndrome and progressively increased with the number of factors present. There was also a positive and progressive association with the mean value of left ventricular mass index (LVMI), and inverse with the E/A ratio. When analysing each factor separately, the waist circumference and fasting glucose were the ones that were more consistently and strongly associated with the dependent variables; BP was significantly associated with LVMI and LVH, as well as changes in left ventricular filling.

Conclusion: As expected, a significant association was found between the prevalence of CHD and symptomatic or asymptomatic cardiac structural and functional changes and the presence of the metabolic syndrome. However, defining this syndrome by the number of factors disregards the differential contribution factors for cardiac phenotypes.

A. Azevedo¹, R. Rodrigues², F. Friões³, J. Pimenta³, C. Abreu-Lima², P. Bettencourt³, H. Barros¹.¹University of Porto Medical School, Hygiene and Epidemiology, Porto, Portugal; ²Hospital de S. João and University of Porto Medical School, Cardiology, Porto, Portugal; ³Hospital de S. João, Medicine B, Porto, Portugal

Introduction: The diagnosis of heart failure (HF) in routine clinical practice is unreliable, mainly due to the low specificity of symptoms. The goal of this study was to assess how often symptoms suggesting HF in the absence of left ventricular systolic dysfunction are attributable to diastolic HF and how often alternative causes can be identified.

Methods: We designed a cross sectional study with the main goal of determining the prevalence of HF in a community sample of adults living in Porto. Participants were asked about symptoms suggesting HF, and a clinical examination, ECG, spirometry, and echocardiogram were performed. Seven hundred participants have been evaluated in this ongoing study (280 men and 420 women, aged 69.5 (SD 11.8) years). Personal history of coronary heart disease (CHD) was defined as previous myocardial infarction or angina according to self reported information, Rose’s questionnaire, and/or pathological Q waves on ECG. Obesity was defined as body mass index 30 kg/m². Exertion dyspnea was attributed to HF in the presence of left ventricular systolic dysfunction, moderate–severe valvular disease on echocardiogram, or atrial fibrillation. In participants without any of these abnormalities, symptoms were attributed to lung disease if there were moderate to severe obstructive or restrictive changes on spirometry. In the absence of HF and lung disease as defined, symptoms were attributed to obesity or CHD. The characteristics of symptomatic participants in which none of these diseases was identified were compared with those of each of the previously defined groups as well as those of the asymptomatic.

Results: Overall 250 (35.7%) participants declared exertion dyspnea and/or tiredness, leading to mild–moderate functional limitation. In 24 (9.6% of symptomatic subjects) these symptoms were attributable to HF; in 101 (40.4%) to lung disease, in 47 (18.8%) to obesity, and 14 (5.6%) to CHD without systolic dysfunction. Only 64 symptomatic subjects were left and the clinical, analytical, and echocardiographic characteristics (including parameters of transmitral flow to study patterns of left ventricular filling) of these were in general more similar to those of the asymptomatic than any of the other groups.

Conclusion: In most participants who reported symptoms compatible with HF on the questionnaire but who did not have systolic dysfunction, valvular disease or atrial fibrillation, it was possible to identify alternative explanations for those symptoms. Therefore, before attributing them to diastolic HF, it is mandatory to search for these alternatives causes and as far as possible correct them.

A. Azevedo¹, R. Rodrigues², P. Dias², E. Martins², P. Almeida², M. Alvelos³, C. Abreu-Lima², P. Bettencourt³, H. Barros¹.¹University of Porto Medical School, Hygiene and Epidemiology, Porto, Portugal; ²Hospital de S. João and University of Porto Medical School, Cardiology, Porto, Portugal; ³Hospital de S. João and University of Porto Medical School, Medicine B, Porto, Portugal

Introduction: Heart failure (HF) is a clinical syndrome that represents the end stage of many different cardiac diseases and noxious insults. It is responsible for a huge burden of morbidity, mortality, costs with treatment and hospitalisations, and poor quality of life. A quantitative perspective of heart failure in our community would be shortsighted if the prevalence of clinically overt HF alone was measured. The objective of this study was to measure the prevalence of the different stages of cardiac dysfunction, taking into account the progression from high risk, through asymptomatic cardiac abnormalities up to the development of symptoms of HF associated with objective evidence of cardiac disease.

Methods: Population based cross sectional study. Recruitment of participants by random digit dialling for attending a clinical interview at the University of Porto Medical School Department of Epidemiology from January 2001 to December 2003. Individuals aged over 45 years were eligible. Over sampling of individuals aged over 60 years was performed during the last year of the study. Structured clinical interview, ECG, echocardiogram, spirometry, and venous blood sampling collected data. Participants were classified according to the stages defined in the latest guidelines of the American College of Cardiology/American Heart Association (ACC/AHA) for the management of HF: Stage A: high risk of developing HF, with no identified structural or functional cardiac abnormalities and no signs or symptoms of HF (for example, systemic hypertension, coronary heart disease, diabetes mellitus); stage B: structural heart disease that is strongly associated with the development of HF but which has never shown signs or symptoms of HF (for example, left ventricular hypertrophy or fibrosis, left ventricular dilatation or hypocontractility, asymptomatic valvular heart disease, previous myocardial infarction); stage C: symptoms of HF associated with underlying structural heart disease; and stage D: advanced structural heart disease and marked symptoms of HF at rest despite maximal medical therapy. We have studied 700 participants: 420 women (age (mean SD) 68.8 (11.5) years) and 280 men (age 70.6 (12.2) years).

Results: The prevalence (95% CI) of ACC/AHA stages was similar in men and women: low risk: 27.0% (23.7 to 30.4); stage A: 45.4% (41.7 to 49.2); stage B: 23.3% (20.2 to 26.6); stage C: 4.3% (2.9 to 6.1); and stage D: 0% (0 to 0.5). The prevalence of left ventricular systolic dysfunction and valvular heart disease was larger in men than in women, while the prevalence of left ventricular hypertrophy was larger in women. The prevalence of stages B and C increased with age in both sexes.

Conclusion: In this community sample of older adults, there was a large prevalence of cardiac disease, largely asymptomatic. Taking into account the progressive nature of the disease, the proportion of participants at high risk of developing heart failure is worrisome and warrants intervention.

I. Barreto¹, L. Andrade¹, A. Sucupira², A. Santiago³, A. Segall-Correa⁴, J. Lima⁵, A. Santiago³, I. Filgueiras¹, A. Pereira³.¹Universidade Federal do Ceará, Faculdade de Medicina de Sobral, Sobral, Brasil; ²Faculdade de Medicina da Universidade de São Paulo, Departamento de Pediatria, São Paulo, Brasil; ³Prefeitura Municipal de Sobral, Secretaria de Desenvolvimento Social e de Saúde, Sobral, Brasil; ⁴Faculdade de Ciências Médicas da UNICAMP, Departamento de Medicina Preventiva e Social, Campinas, Brasil; ⁵Universidade Estadual do Ceará, Centro de Ciências da Saúde, Fortaleza, Brasil

Introduction: Children between 5 and 9 years of age have not been subjects of population morbidity studies. The available research results focus on specific groups, such as children at school or the ones assisted by the health services, which poses limits on generalising those results. This study aimed to analyse the referred morbidity among children aged 5 to 9 years.

Methods: A cross sectional study was designed to interview mothers, preferably, or another adult at home. A stratified probability sample, by year of birth (from 1990 to 1994) allowed the selection of 3444 children, aged 5 to 9 years, from a dataset of all families with children in this age, living in the urban area of the city. These data were originated from an official form of the city’s Family Health Program (PSF), filled out by the community health workers. Only one child per home was allowed to enter the study randomly. From the total sample, a sub-sample of children was randomly selected for clinic and laboratory exams. Data were collected from November 1999 to October 2000, by trained interviewers, using a pre-tested semi-structured questionnaire. Sickness in the previous 15 days, search for health assistance during those 15 days, hospitalisation, dentistry treatment in the previous 12 months, and the use of medication in the previous 3 days were the requested information on morbidity.

Results: The presented results are those obtained by household interviews. Almost half of the mothers (43.9%) mentioned that their children had some kind of health problem in the previous 15 days. The most frequent complaints concerned upper respiratory illnesses (63.7%), skin problems (7.5%), asthma (4.2%), and recurrent pain (3.6%). Less than half (41.2%) of the interviewed mothers with sick children had looked for medical assistance; 5.4% stated that their children had been hospitalised in the previous 12 months, 27.6% being due to lower respiratory diseases, 20.8% because of accidents, and 18.2% as a consequence of gastrointestinal diseases. 51% of the mothers had taken their children to dental treatments during the previous 12 months, 16.5% of the children had been under medication in the previous 3 days.

Conclusion: The amount of children with health problems in the previous 15 days was extremely high considering their age. Nevertheless most of them were mild diseases that, even though, required health care. Studies of this kind are fundamental for the planning of adequate health actions destined to children of this age group.

O. Basso¹, S. Olsen², M. Frydenberg³, J. Olsen².¹University of Aarhus, Danish Epidemiology Science Centre, Denmark; ²Statens Serum Institut, Danish Epidemiology Science Centre, Denmark; ³University of Aarhus, Department of Biostatistics, Denmark

Introduction: Dichotomising a measure such as birth weight involves loss of information, but it is required when the aim is making decisions about intervention or monitoring of babies. Estimating the expected birth weight for a baby based on the birth weight of its older sibling may offer an interesting alternative approach compared with using gestational age specific birth weight, for identifying high risk babies.

Methods: For 10 577 second babies born to women enrolled in the Danish National Birth Cohort we calculated the expected birth weight based on the method proposed by Skjaerven and colleagues.¹ We then calculated a birth weight ratio (birth weight/birth weight predicted from older sibling). We identified the lowest 10% and compared this with the lowest 10% of the sex and gestational age specific distribution, estimated by using a z score using standards from Norway (SGA). We examined whether selected predictors of birth weight influenced the classification of the baby as small or normal according to the two criteria. On a sub-sample of xxxx babies whose mothers responded to an interview about 6 moths after birth, we examined how these categories predicted motor development.

Results: The two criteria identified 1058 and 1059 babies, respectively, but only 738 overlapped. Babies who were classified as small for gestation but were normal according to the birth weight ratio were more often born to short or thin mothers. The birth weight ratio predicted babies with a delayed motor development as well as the definition of small for gestational age. The group of babies weighing on average 3252 g that was classified as small by the birth weight ratio because babies fell short of their predicted birth weight by about 550 g had an OR of delayed motor development of 1.45 (95% CI 1.06 to 1.96) compared to babies classified as normal by both criteria.

Conclusions: Our results suggest that the birth weight ratio could be a useful addition to the z score in identifying babies at risk. The former method is less likely to classify as small babies based upon their mother’s size. It identifies, on the other hand, babies who grew smaller than their older siblings without being small in absolute terms.

1. Skjærven R, Gjessing HK, Bakketeig LS. New standards for birth weight by gestational age using family data. Am J Obstet Gynecol 2000;183:689–96.[CrossRef][Medline]

F. Bastos, C. Lopes.University of Porto Medical School, Hygiene and Epidemiology, Porto, Portugal

Introduction: Diabetes is a major public health issue. Its frequency is increasing and it is responsible for high mortality and large amounts of potential years of life lost. Diabetics’ education is essential to promote self care activities and adherence to adequate preventive guidelines. The involvement of the partner might play an important role in the diabetic care plan, but few studies addresses the issue under controlled conditions. We designed a randomised controlled trial to determine the effect of partner involvement in the educational program for the diabetic type 2, in to adherence to treatment and metabolic control.

Setting: A primary health care centre.

Methods: Between April and October of 2003, 103 male participants, previously diagnosed with type 2 diabetes, married and without accentuated dependence in self care activities, were approached and accepted to be randomised into two groups. The intervention involved the presence of the partner, and both groups received an educational programme consisting of one month on self care activities as nutrition, physical exercise, foot care, and medication and monitoring of blood glucose, promoting adherence and behaviour changes. The Summary of Diabetes Self-Care Activities Measure – SDSCA scale was used to evaluate adherence, and the Diabetes Knowledge Questionnaire – 24 was used to evaluate knowledge. Both scales were translated and adapted for the Portuguese population. Paired and unpaired differences were compared using non-parametric tests.

Results: Comparing adherence scores at baseline and after 2 months, a significant improvement in both groups for almost every component of self care was observed. The difference was larger in the intervention group though only statistically significant for exercise and medication. The group with wife presence had a significant higher improvement (exercise mean differences: 3.1 CI 95% 2.2 to 3.7 v 1.9 CI 95% 1.1 to 2.2 days a week, p = 0.017, and medication 1.0 CI 95% 0.2 to 1.6 v 0.0 CI 95% -0.5 to 0.5 days a week, p = 0.016. The total score adherence difference was higher and statistically significant in the intervention group (means difference: 1.8/1.1 CI 95% 1.5 to 2.1/0.9 to 1.5 days a week, p<0.001). The basic 2 months programme significantly increased patient knowledge about diabetes, but no further effect was found for the presence of the partner (mean difference 21.0%/22.2% CI 95% 17.9 to 24.2/17.8 to 26.5, p<0.667). Significantly, decreases in diastolic blood pressure and waist circumference and a significant increase in number of blood glucose test performed were observed within but not between groups, despite better results in the intervention group. There were no significant changes in glucose control, although improvements in HbA1c were found.

Conclusions: The partner presence in the diabetic educational programme improved the partners’ adherence to physical exercise and medication intake.

H. Baysson, M. Tirmarche.Institute for Radiation Protection and Nuclear Safety, Radiobiology and Epidemiology Department, France

Introduction: It is well established that exposure to the radioactive gas radon increases risk of lung cancer among uranium miners. To estimate the lung cancer risk linked to indoor radon exposure, a hospital based case control study was carried out in France, with a focus on precise reconstruction of past indoor radon exposure over the 30 years preceding the lung cancer diagnosis.

Methods: The study was conducted from 1992 to 1998 in four radon prone areas of France. During face to face interviews, a standardised questionnaire was used to ascertain demographic characteristics, information on active and passive smoking, occupational exposure, medical history as well as extensive details on residential history. Radon concentrations were measured in the dwellings where subjects had lived at least 1 year during the 5–30 year period before interview. Measurements of radon concentrations were performed during a 6 month period, using two Kodalpha LR 115 detectors, one in the living room and one in the bedroom. The time weighted average (TWA) radon concentration for a subject during the 5–30 year period before interview was based on radon concentrations over all addresses occupied by the subject weighted by the number of years spent at each address. For the time intervals without available measurements, the region specific arithmetic average of radon concentrations for measured addresses of control subjects was used.

Results: A total of 486 cases and 984 controls with at least one dwelling measured were included in the study. After imputation for missing values, the arithmetic mean of the TWA radon concentration experienced by cases during the previous 5–30 years was 146 Bq/m³, while the value for controls was 140 Bq/m³. The relative risks of lung cancer (with 95% CIs in parentheses) in relation to categories of TWA radon concentrations delimited by cut points at 50, 100, 200, and 400 Bq/m³ were: 0.85 (0.59 to 1.22), 1.19 (0.81 to 1.77), 1.04 (0.64 to 1.67), 1.11 (0.59 to 2.09), respectively, with TWA radon concentrations below 50 Bq/m³ used as reference category and with adjustment for age, sex, region, cigarette smoking, and occupational exposure. The relative risk of lung cancer increased by 0.04 (–0.01 to 0.11) per 100 Bq/m³ increase in the TWA radon concentration cumulated over the 5–30 year period before interview. When the analysis was limited to the 257 cases and 593 controls with complete measurements for the 25 years, the corresponding increase was somewhat higher at 0.07 (0.00 to 0.14) per 100 Bq/m³.

Conclusion: Our results indicate a small increase in lung cancer risk, at the borderline of statistical significance, as a result of indoor radon exposure; which is consistent with the findings of previous indoor radon and miners’ studies. Our data are integrated in the pooled analysis of several European case control studies.

F. Bazin, I. Parizot, P. Chauvin.INSERM U444, Research Team on Social Determinants of Healthcare, France

Introduction: The social inequalities in health have endured or even relatively worsened throughout different social groups since the 1990s. In France, universal access to healthcare is theoretically guaranteed through the Social Security system. Nevertheless, in 2002, 11.2% of the French population stated that they had forgone healthcare because they could not afford it during the year preceding the survey. Our objective was to determine which individual psychosocial characteristics (in addition to the socioeconomic ones and health status) were associated with the fact of stating (or not) that healthcare had been forgone because of cost.

Method: 2000 individuals who lived on income welfare in 2000 were enrolled in a cross sectional representative study in January 2003 in France. Statistical associations between such a renouncement and psychosocial characteristics were examined using logistic regression models adjusted on age, sex, income level, health insurance status, and health indicators. Pearson ² and deviance estimate were used to assess the goodness of fit of our model, and bootstrap methods to validate this model.

Results: After adjustment, we observed a higher occurrence of reported forgone healthcare among people who have lost the income welfare at the time of the study, who perceived their financial situation as bad and who have limited knowledge about existing assistance policies. Such a renouncement is also associated with some items of the Rosenberg self esteem (for example, "I am able to do things as well as most other people"; "I feel I do not have much to be proud of").

Conclusions: There results show that other than purely financial hurdles play a role in the non-utilisation of health care services in this population. Health policies, which mainly promote equal financial access to healthcare, have little chance of abating health inequalities.

H. Becher¹, A. Benner².¹University of Heidelberg, Department of Tropical Hygiene and Public Health, Heidelberg, Germany; ²German Cancer Research Centre, Department of Biostatistics, Heidelberg, Germany

A common problem in survival analysis is to estimate the effect of co-variables on the hazard function for which the proportional hazards model is often used. We compare three methods to model a time dependent effect of a binary risk factor on survival in the context of a proportional hazards model. The simplest option is a stepwise constant modelling of the hazard ratio, which, however, yields a step function, which biologically may not be plausible. In addition, the choice of the cut points is arbitrary and may lead to wrong conclusions. The adaptive hazard regression (HARE) methodology circumvents the need to assume proportional hazards and allows time dependent effects of covariates. HARE models these effects through the use of polynomial splines. A third, new method is based on fractional polynomials, which have been used to model the effect of continuous co-variables on the outcome parameter in various regression models. Here, the co-variable enters the model as a time dependent co-variable f(x, b, z) = b_0 + b_1*z_1 (t) + ... b_k*z_k (t), where zj is among the set of functions used for fractional polynomials.

We illustrate the methods using data of 11 312 children from Burkina Faso where we analyse the mortality of twins in comparison to singles. Here, the risk of dying for a twin is strongly increased in comparison to a singleton in the first weeks after birth and gradually decreases. We compare the goodness of the three model fits by use of the integrated Brier score modified to account for the predicted survival of twins and give recommendations on practical procedures.

E. Benito-Garcia¹, F. Wolfe², J. K. Michaud².¹Division of Rheumatology Immunology, and Allergy Brigham and Womens Hospital Boston MA, USA; ²National Data Bank for Rheumatic Diseases, Wichita, Kansas, US

Introduction: Recent European studies of rheumatoid arthritis (RA) status and outcomes have demonstrated substantial differences between countries,¹ suggesting that treatment effects might account for the differences. However, these studies did not collect data on key socioeconomic variables, which are known to influence RA status and outcome. Although Portuguese RA patients have treatment options that are similar to those in other countries, the average education level in Portugal in 2000 was 5.9 years the lowest in Europe. We took advantage of this characteristic to study the effect of socioeconomic status on RA.

Objectives: To determine the extent to which socioeconomic status determines RA status and outcome.

Methods: As part of a longitudinal study of the outcome of RA in Portugal, 94 patients completed a comprehensive assessment of RA in 2003. Questionnaires were administered by trained assessors for patients with inadequate literacy. For comparison, the same questionnaire was administered to 9609 RA patients in the US. In the first stage of comparisons, the US RA cohort was compared to the Portuguese. In the second stage of comparison, a random sample of 94 US patients matched exactly as to educational level was obtained and used as the comparison group. Results of the second stage were adjusted further for age, sex, RA duration, and total household income.

Results: The groups differed strikingly for the key socioeconomic variables (US v Portugal): age (62.4 v 53.8 years), education level (13.7 v 5.4 years), RA duration (17.0 v 11.9 years), and total household income (47 400 v 8400). In the unadjusted analyses of the table, Portuguese patients had much worse scores than US patients. For example, the health assessment questionnaire disability index (HAQ) was 0.51 units greater and quality of life utility scores (EuroQol utility) were 0.29 units lower. In addition depression and the SF-36 mental component scale (MCS) scores were strikingly different. Adjustment for socioeconomic status (SES) resulted in scores that were approximately equal in the two countries with the exception of depression and SF-36 MCS scores.

View this table: [in this window] [in a new window]   Abstract 24

 

Conclusion: Almost all of the clinical differences between Portuguese and US patients can be explained by SES. For the key RA status and outcome variables (HAQ, pain, global, fatigue, utility, and PCS, there are no differences between countries following adjustment for SES. Depression and MCS, however, remained lower in Portugal. SES profoundly influences RA clinic status. A strong case is made for collection of such data in clinical trials and observational studies

1. Antoni C, Smolen J, et al. Results from a crossectional survey (cooperative on quality of life in rheumatic diseases) showed a remarkably high mean HAQ 1.14 despite DMARD therapy in european RA patients. Orlando, Florida: American College of Rheumatology, 2003.

R. Bergantim, A. Azevedo, H. Barros.University of Porto Medical School, Hygiene and Epidemiology, Porto, Portugal

Introduction: Handedness is a unique, uniform and universal characteristic of the human being. Differences between left-handers, right-handers and ambidextrous individuals always raised great interest and have been the centre of multiple discussions within the scientific community. The aim of this study was to assess whether there is a relation between handedness and several cognitive abilities (social sciences, exact sciences, arts, music) and motor abilities (sports) among students of the University of Porto.

Methods: We approached unselected students in 10 faculties of the University of Porto to participate in a cross sectional study on handedness and abilities. Data were collected by an anonymous questionnaire. Handedness was defined according to current hand preference for writing and hand preference for general activities. Participants were considered right-handed if they considered themselves right-handed and reported writing with the right hand; left-handed if they considered themselves left-handed and wrote with the left hand; and ambidextrous if self-percepted as ambidextrous and reported to write indifferently with any hand. Participants whose answers to these questions were contradictory were excluded. Among 4499 students that agreed to participate, only 3966 were included in the present analysis due to missing data in key variables or contradictory answers on handedness. We report prevalences and their 95% CI and used the ² test to compare proportions.

Results: The prevalence of right-handedness was 93.5% (95% CI 92.7 to 94.3); 5.7% (5.0 to 6.5) of participants were left-handers and 0.8% (0.5 to 1.1) ambidextrous. Right-handers had a self reported higher ability for exact sciences (p = 0.008) and ambidextrous individuals for social sciences (p = 0.009). There was no statistically significant association between handedness and arts, music, or sports. Men were more frequently left-handed (6.4% v 5.1% in women), although the difference did not reach statistical significance (p = 0.23). Men had a significantly higher ability for exact sciences and sports, and women had a significantly higher ability for social sciences and arts; there was no significant difference between sexes in ability for music.

Conclusions: There was an association between handedness and exact sciences, for which right-handers have higher ability, and between handedness and social sciences, for which ambidextrous have higher ability. Previous empiric impressions on the association between left-handedness and arts were not confirmed by our study, as well as relationships between handedness and music or sports.

G. Bergman¹, J. Winters¹, K. Groenier¹, J. Pool², B. Jong¹, K. Postema³, G. Heijden⁴.¹University of Groningen, Department of General Practice, The Netherlands; ²Vrije Universiteit Medical Centre, Institute for Research in Extramural Medicine, Amsterdam, The Netherlands; ³University Hospital of Groningen, Centre for Rehabilitation, The Netherlands; ⁴University Medical Centre Utrecht, Julius Centre for Health Sciences and Primary Care, The Netherlands

Introduction: A dysfunction of the cervicothoracic spine and the adjacent ribs (shoulder girdle) is considered to be a predictor for occurrence and poor outcome of shoulder complaints. It can be treated by manipulative therapy, but scientific evidence for its effectiveness is lacking. The objective of the present study is to evaluate the effectiveness of manipulative therapy of the shoulder girdle, additional to usual care, on relief of shoulder complaints.

Methods: In this randomised controlled trial, 150 patients with shoulder complaints and a dysfunction of the shoulder girdle were recruited in 50 general practices in Groningen, The Netherlands. All patients received usual care according to the Dutch Guidelines for General Practitioners and were evenly randomly allocated to manipulative therapy (up to 6 treatment sessions in a 12 week period). Main outcome measures included perceived recovery, severity of main complaint, shoulder pain, shoulder disability, and general health. Data were collected at 6 and 12 weeks after randomisation.

Results: During treatment (6 weeks) no significant differences were found. After completion of treatment (12 weeks), 43% of the patients receiving additional manipulative therapy reported full recovery, and 24% in control group (95%CI 3.7% to 34.5%). At 12 weeks the difference in mean severity of main complaint (d = 1.0; CI 0.0 to 1.8) and shoulder pain (d = 1.9; CI 0.3 to 3.5) favoured manipulative therapy. Also, the difference in mean shoulder disability and general health favoured manipulative therapy, but was not statistically significant.

Conclusions: Manipulative therapy of the shoulder girdle, additional to usual care by the general practitioner, improves recovery and reduces symptoms of shoulder complaints at 12 weeks. Currently, long term data on clinical and cost effectiveness are collected.

M. Biazevic¹, J. Antunes², E. Muller¹, E. Michel-Crosato¹.¹UNOESC, Saúde Colectiva, Joaçaba-SC; ²FOUSP, Odontologia Social, São Paulo, Brazil

Introduction: An improved socioeconomic profile and access to health services are critical issues to a reduced incidence of cervical cancer, early diagnosis, and prolonged survival of patients. The objective of the current study was to describe trends of cervical cancer mortality in the state of Paraná, Brazil, from 1980 to 2000, and to assess socioeconomic conditions associated with a poorer evolution of deaths.

Methods: The Official System for Information on Mortality supplied data related to cervical cancer deaths, as discriminated by age and town of residence. Age adjusted death rates were calculated for 22 regions of the state in each year. The appraisal of trends used the Cochrane-Orcutt procedure of auto regression for time series analysis and estimation of the annual per cent increase of mortality. Population data and socioeconomic information at region level were gathered from censuses performed in 1980, 1991, 1996, and 2000. Comparative analysis assessed socioeconomic indices associated with regions presenting stationary and increasing trend of cervical cancer mortality.

Results: Cervical cancer deaths were on the increase in the state of Paraná, with an annual percent increase of 1.68% (95% CI 1.20% to 2.17%). Despite the overall increase of rates, most regions presented stationary trend. Only three regions presented increasing trend of mortality: Guarapuava, Campo Mourão, and Cornélio Procópio. The comparison of regions presenting increasing and stationary trend indicated poorer socioeconomic indices for the former set, and regions with increasing levels of cervical cancer mortality had significantly higher illiteracy rate (p<0.001), percent of individuals older than 15 years with less than 4 years of study (p = 0.001), and lower per capita income (p = 0.025) and human development index (p = 0.023).

Conclusion: The ecologic assessment of information on cervical cancer suggests that poorer socioeconomic standings of population contributed to a worse evolution of mortality. The quantitative assessment of this association in the Brazilian context may instruct health services aimed at prevention and treatment of the disease and help them target intervention programmes.

M. Bielska-Lasota¹, R. Krynicki², D. Rabczenko³, K. Czerw-Glab⁴, J. Starzewski⁵, A. Chil⁵, J. Hudala-Klecha⁴, A. Swiercz⁴.¹M. Sklodowska-Curie Cancer Centre, Department of Mass Screening Organisation, Warsaw, Poland; ²M. Sklodowska-Curie Cancer Centre, Department of Gynaecological Oncology, Warsaw, Poland; ³National Institute of Hygiene, Department of Medical Statistics, Warsaw, Poland; ⁴Regional Hospital of Oncology, Department of Oncology No 2, Opole, Poland; ⁵Holycross Cancer Centre, Department of Gynaecological Oncology, Kielce, Poland

Introduction: The 5 year survival rate for cervical cancer patients in Poland was evaluated in two cities; Warsaw and Krakow were 48%, which is rather low compared with other European countries. The mortality of cervical cancer varies considerably, mainly due to different access to prevention programmes and treatment. Consequently, differences in survivals should be expected. The aim of the study was to evaluate whether cervical cancer patients living in selected regions of Poland show similar treatment results.

Methods: The cohort was constituted on all the data about cervical cancer incidence in years 1990–96 obtained from the population’s Cancer Registries, collecting data from Kieleckie (K), Opolskie (O), voivodships, and Warsaw city (W). The prognostic factors were made complete by adding information from medical records. The follow up of patients lasted 5 years. SAS was used for the analysis. The 5 year relative survival rates were calculated using life tables method modified by Hakulinen.

Results: The cohort consisted of 1386 women whose average age was 56. The FIGO stage distribution was: I 30%, II 29%, III 23%, IV 9%, and unknown 9%. The squamous cell carcinoma represented 85%, adenocarcinoma 8%. The 5 year relative survival rate for all patients was 52%; according to FIGO: I 84, II 50, III 26, and IV 8. The regions were similar in age group distribution and histological diagnosis, differentiated, however, in FIGO distribution (p<0.0001), which was as follows I: K 41%, O 25%, W 28%; II: K 20%, O 32%, W 33%; III: K-17%, O 18%, W 29%; IV: K 10%, O 10%, W 6%; and unknown: K 11%, O 15%, W 4%. The 5 year survival rates were 61% in Kieleckie and were close to European average in Opolskie 43 and in Warsaw 52. The survival for FIGO I in those regions was comparable, but different for FIGO II and III. Multivariant analysis showed significant risk increase related to stage advancement (p<0.0001) as well as the place of living in Opolskie (p = 0.02) and adenocarcinoma diagnosis (p = 0.05). However, the analysis did not confirm the age of diagnosis as prognostic factor. A relatively high proportion of early stages of the disease in Kieleckie seem to be related to a prevention programme carried out there in 1982–88. Insufficient diagnosis and radiotherapy protocol applied in Opole town hospital at that time occurred in unsatisfactory treatment results in Opolskie.

Conclusions: The cervical cancer treatment results are differentiated in Poland. They are almost satisfactory and close to European average in Kieleckie where prevention was effective, but poor in the regions where prevention was insufficient. In all stages survivals were relatively high and close to clinical series in European hospitals. However, the most crucial discrepancies were in FIGO II showing the local problems in diagnosis and radiotherapy. The poor survivals mainly depend on unsatisfactory proportion of early stages, which should be immediately improved by the implementation of national prevention project.

B. Blondel¹, A. Macfarlane², M. Gissler³, J. Zeitlin¹.¹INSERM U149, Research Unit on Perinatal Health and Women’s Health, Paris, France; ²City University, Bartholomew School of Nursing and Midwifery, London, UK; ³STAKES, Research and Development Centre for Welfare and Health, Helsinki, Finland

Introduction: The rise in multiple birth rates in developed countries over the past 25 years has been described as an epidemic. Increasing maternal age at childbirth and, more importantly, the use of ovarian stimulation and assisted conception explain the rise in multiple birth rates. This trend is of concern because of the high mortality among multiples, due principally to higher preterm delivery rates. Moreover, in some countries, preterm delivery rates have risen among twins, probably because of increasingly active management at the end of pregnancy. We compared the multiple birth rates, the rate of preterm delivery, and the management of the onset of labour in a number of European countries, in order to show the implications of these general trends for perinatal outcome in each country.

Method: We used aggregated data collected by the PERISTAT project, a project to develop health indicators for the European Union. Data were available for Austria, Flanders, and the French community in Belgium, Denmark, Finland, France, Germany (9 Bundesländer), Italy, Netherlands, Portugal, Sweden, and Scotland, and Northern Ireland in the UK. In most countries data were for the year 2000. The main measures were the multiple delivery rates per 1000 deliveries, the proportion of live births before 37 weeks, and the proportion of preterm births after induction or elective caesarean section.

Results: The proportion of multiple deliveries ranged from 11.4 per 1000 in Portugal to 19.4 per 1000 in The Netherlands. There was no clear tendency towards a higher rate of multiple births in the countries in which a high proportion of deliveries are to women aged 30 years or more. Data on ovarian stimulation and assisted conception were very limited and could not be used to explain variations in the multiple delivery rates. In all countries with data, more than 40% of multiple live births occurred before term. The highest proportions were 56% in Belgium and 68% in Austria. The relative risk of preterm birth for multiples compared with singletons varied between 8.1 and 10.5. The percentage of preterm live multiples delivered after induction or by elective caesarean section was known in nine countries and ranged from 26% in Finland to 50% in Germany. The countries that had the highest preterm delivery rates also had the highest proportion of labours without spontaneous onset.

Conclusion: The wide variations in the rate of multiple deliveries and the ways in which these deliveries are managed raise questions about differences in obstetric practice. Better knowledge of the impact of these practices on immediate and long term health of babies from multiple births and their mothers would be useful.

M. Bobak¹, A. Nicholson¹, M. Murphy², M. Marmot¹.¹University College London, Department of Epidemiology and Public Health, London, UK; ²London School of Economics, Department of Social Policy, London, UK

Introduction: Alcohol and binge drinking have been suspected of adverse effects on mortality in Russia but individual level data are sparse. Using modified indirect demographic estimation techniques, we examined the relationships between frequency of alcohol consumption and of binge drinking and adult mortality in Russia.

Methods: A convenience cohort was constructed based on survey respondents’ information about their close relatives. A random sample of 7172 respondents (response rate 61%) selected from the general population of the Russian Federation provided information on 10 475 male and 3129 female relatives, including age, vital status, and frequency of alcohol consumption and binge drinking. These relatives formed the cohort analysed in this report. The relation between all cause mortality and frequency of any drinking and binge drinking was estimated by Cox regression.

Results: There was a strong linear relationship between frequency of drinking and all cause mortality in men; after controlling for smoking and calendar period of birth, the hazard ratio of death in daily drinkers compared with occasional drinkers was 1.52 (95% CI 1.33 to 1.75). Male binge drinkers had slightly higher mortality than drinkers who did not binge (adjusted HR 1.21, 1.12 to 1.31). In women, the increased mortality was confined to a small group of those who binged at least once a month (adjusted hazard ratio 2.68, 1.54 to 4.66).

Conclusions: The results suggest a positive association between alcohol and mortality in Russia. There was no evidence for the protective effect of drinking seen in Western populations. Alcohol appears to have contributed to the high long term mortality rates in Russian men but it is unlikely to be a major cause of female mortality.

E. Brabazon, A. O’Farrell, C. Murray, P. Finnegan.North Eastern Health Board, Public Health, Railway St, Navan, Co. Meath, Ireland

Introduction: The notifiable infectious disease process is essential for prompt public health action and for the monitoring of real time disease incidence in the Irish population. However, anecdotal evidence suggests that the occurrence of notifiable infectious diseases is seriously underestimated. The aim of this study therefore was to address this question by quantifying the burden of notifiable infectious diseases on inpatient hospital admissions from 1997 to 2002 from a health board region in Ireland and comparing the results to the corresponding statutory notifications.

Methods: All hospital inpatient admissions from 1997 to 2002 with a principal diagnosis relating to infectious diseases (ICD codes 001–139) by residents from this health board region were extracted from the Hospital In-Patient Enquiry System (HIPE). All notifiable infectious diseases were identified based on the 1981 Irish Infectious Disease Regulations and the data were analysed in JMP statistical package. These data were compared with the corresponding notification data. This health board services the needs of a population of 344 926 (based on the 2002 census) and has five acute hospitals.

Results: Analysis of the HIPE data from 1997 to 2002 revealed a substantial burden associated with notifiable infectious diseases. There were 2759 hospitalisations by 2455 residents, 17 035 bed days taken up and 31 fatalities. The most common notifiable disease hospitalisations were for gastroenteritis in children less than two (28.3%), viral meningitis (11.1%), and TB (10.4%). Overall, these hospitalisations represented 31.3% of all infectious disease hospitalisations and 0.94% of total hospitalisations. The statutory notifications, however, comprise both GP and hospital clinician notifications. Therefore, only in cases where there are more hospitalisations than notifications can underreporting be accurately identified. This occurred in 10 out of 23 notifiable diseases, for example acute encephalitis (35 hospitalised v 2 notifications); infectious mononucleosis (224 hospitalised v 70 notified); malaria (23 hospitalised v 4 notifications); leptospirosis (9 hospitalised v 3 notifications), and viral meningitis (299 hospitalised v 42 notifications). It is notable that the second most common hospitalisation (viral meningitis) was underreported by 257 cases.

Conclusion: This study has highlighted the extent of underreporting of some notifiable infectious diseases, which is a cause for concern from a surveillance point of view. If this underreporting is similar in other health boards in Ireland, then it would appear that the epidemiology of some notifiable diseases is being overlooked and this negatively impacts on the effectiveness of the notification process as a real time surveillance tool and an early warning system for outbreaks. Furthermore, on 1 January 2004 Ireland introduced new legislation, which includes microbiology laboratories as notifiers, hence studies like this will provide a baseline for comparing the effectiveness of the new reporting system into the future.

J. Braga¹, J. Junior³, M. Silva², W. Ramalho², E. Duarte², R. Paes-Sousa³.¹SVS/MS, Centro de Ref. P. Hélio Fraga, Rio de Janeiro, Brazil; ²SVS/MS, Coordenação de Análise da Situação de Saúde, Brasilia. Brazil; ³PUC/MG, Pontifícia Universidade Católica, Belo Horizonte, Brazil

Introduction: The main objective of epidemiologic surveillance of disease is detecting its time and space variation. However, mapping crude incidence rate is not appropriate to produce accurate risk estimates. A relied identification of the spatial distribution pattern of the events under surveillance can generate ecological causal hypotheses that are useful to improve their control.

Objectives: (1) producing choropleth maps of crude and adjusted morbidity rates of selected diseases, comparing their spatial distribution in the municipalities of the states of Brazilian Northeast Region (BNR) in the period of 1999–2002; (2) detecting spatial autocorrelation in order to identify morbidity risk areas.

Methods: We studied three infectious diseases under surveillance in Brazil: dengue (ICD-10th, A90), tuberculosis (ICD-10th, A15-A19), and visceral leishmaniasis (ICD-10th, B55.0). Morbidity data were obtained from the Brazilian National Morbidity Information System (SINAN) and population data were obtained from the Brazilian Population Census Bureau (IBGE). We calculated rates for the period studied and Freeman-Tukey transformation and empirical Bayes were also performed in order to reduce the data excessive variability and possible spatial outliers’ identification. We investigate spatial autocorrelation using Moran’s I and Geary’s C statistics and local spatial autocorrelation using LISA and Moran scatter plot maps. Spatial weights matrices were constructed using adjacency as neighbourhood criteria.

Results: We found three distinct spatial patterns for the surveillance diseases studied. The evidence of spatial pattern was found only when transformed rate were used. Global and local spatial autocorrelation were found for the incidence data. Dengue, tuberculosis, and visceral leishmaniasis had statistically significant Global Moran’s I values of 0.701, 0.517, 0.644, respectively. LISA maps showed the evidence high risk zones when the three diseases were analysed.

Conclusions: We conclude that dengue, tuberculosis, and visceral leishmaniasis were not spatially randomly distributed in Brazilian Northeast Region in the period studied.

S. Brasche¹, T. Hartmann², E. Heinz³, W. Bischof¹.¹Jena University, Institute of Occupational, Social and Environmental Medicine, Jena, Germany; ²Technical University Dresden, Institute for Thermodynamics, Dresden, Germany; ³Technical University Berlin, IEMB, Berlin, Germany

Introduction: Dampness and mould growth in homes are associated with an increased risk of allergic and respiratory disease. With the aim of obtaining a representative overview of the situation in German homes and analysing the causes and factors affecting the development of signs of dampness and mould growth, a random sample of 5530 homes was investigated.

Methods: The investigation includes an expert rating of building factors following a standardised protocol and a questioning procedure relating to behaviour in the homes and other aspects of the living environment. In addition, the prevalence of asthma, allergic, and respiratory disease of the 12 132 occupants was surveyed. Statistical software SAS, Release 8.2, was used for calculation of prevalence rates and multiple logistic regression analysis.

Results: There were visible signs of dampness (including mould) in 1213 (21.9 %) of the examined dwellings and mould spots in 513 (9.3%) of them. The risk of dampness was significantly greater for increasing moisture emission indoors (OR 1.8; CI 1.4 to 2.4), inadequate user related ventilation (OR 1.7; CI 1.2 to 2.3), edges or corners of outside walls (OR 1.4; CI 1.2 to 1.7), and rented homes in contrast to owner occupied homes. The existence of a user independent exhaust ventilation system (OR 0.5; CI 0.3 to 0.8) or of chimney ventilation (OR 0.7; CI 0.6 to 0.9) and characteristics of modern state of the art buildings such as sealed windows (OR 0.7; CI 0.6 to 0.8) or thermal insulation (OR 0.9; CI 0.7 to 1.0) as well as an increasing rooms per head index (OR 0.9; CI 0.9 to 1.0) are significant protective factors. With regard to mould growth the critical factors are, on the whole, similar. Differences are: the importance of indoor moisture sources decreases (OR 1.3; CI 0.9 to 1.9) while a pet in the home increases the risk of mould significantly (OR 1.4; CI 1.2 to 1.8). Socioeconomic variables become more important (for example rented home OR 2.0; CI 1.6 to 2.6). All types of moisture related damage are significantly associated with an increased self reported prevalence of respiratory and allergic diseases.

Conclusions: The results confirm the unfavourable health effect of dampness and mould known from, for example, Scandinavian literature and emphasise the importance of these problems in Germany. They also enable hints for prevention regarding construction and equipment of buildings and behaviour of occupants.

A. Brito¹, E. Castilho², C. Szwarcwald³.¹Oswaldo Cruz Foundation/Fiocruz, Research Centre Aggeu Magalhães, Recife, PE, Brazil; ²University of São Paulo, DMP/Medical School, São Paulo, SP, Brazil; ³Oswaldo Cruz Foundation/Fiocruz, Scientific and Technologic Information Centre, Rio de Janeiro, RJ, Brazil

Introduction: The purpose of this study is to determine factors for adherence to antiretroviral treatment in adults with AIDS, living in the State of Rio Grande do Norte, Brazil, who received their first prescription of HAART between January 1999 and December 2001.

Methods: It is based on a population based study. The adherence was calculated using data on the number of kept appointments at pharmacies. A patient was considered to have adhered to the treatment if they turned up for at least 80% of programmed visits to the pharmacy; failure to appear at any programmed visit for a period of 6 months subsequent to the date the medication was prescribed was termed refusal.

Results: The study covered 498 patients, about 52.4% of who had already arrived at the specialised treatment centre with some symptom, signal, or disease indicating immunodeficiency. The overall refusal percentage was 10.4% in the first 6 months following prescription and among pregnant women, this percentage rose to 28.6%. The overall percentage for adherence to antiretroviral therapy was 64.1%. On analysis of the association between adherence and socio demographic and clinical variables, no association was found with the following: sex, type of exposure, place of residence, previous identification of HIV infection, the presence of symptoms at the start of antiretroviral therapy, the level of CD4+ and viral charge, nor with the type of antiretroviral combination prescribed. The level of adherence was lower among young adults, aged between 25 and 34, single people, and in the groups with low levels of education. Clinical factors found to be significantly associated with non-adherence were: a history of psychiatric treatment prior to the diagnosis of AIDS, drug use, and the beginning of antiretroviral treatment during a stay in hospital.

Conclusions: After multivariate analysis, significant associations persist between non-adherence and a hospital stay, the use of legal or illegal drugs, a history of psychiatric treatment, a low level of education and an age of between 25 and 34. It is worth pointing out that the analysis of predictive factors for non-adherence did not show any association between adherence and the type of antiretroviral combination prescribed.

C. Cans¹, M. Topp², S. Gliniabaua³, A. Johnson⁴, M. J. Platt⁵.¹CHU Grenoble, SIIM/RHEOP, Grenoble, France; ²NIPH, CP Register, Copenhagen, Denmark; ³Institute of Child Health, NECCPS, Newcastle, UK; ⁴National Perinatal Epidemiology Unit, ORECI, Oxford, UK; ⁵Department of Public Health, Mersey Region Register, Liverpool, UK

Introduction: Because multiple births are at a higher risk of cerebral palsy (CP) than singletons it is of great importance to assess the impact of the increasing multiple birth rates on this long term outcome.

Objectives: Through this work we aimed to answer the following: What are the trends over time of the rate of CP in children issuing from multiple births? and have multiples with CP changed over time with regard to their gestational age (GA), birth weight (BW), and clinical features?

Methods: A collaborative network on CP provided data on CP children and denominators for birth years 1980 to 1996. For this work data from nine centres (6 countries) are presented. Trends over time were analysed using three 5 year periods, that is, children born in 1980–1984, 1985–1989, and 1990–1994.

Results: 1) In the areas concerned by the nine CP registers the rate of multiple live birth increased from 1.8% live births in 1980 up to 3% in 1996. A highly significant trend was observed on pooled data (p<0.001), with no interaction. The proportion of multiple live births below 32 weeks varied between centres and years from 2% to 12%, it doubled from 5.6% in the first period to 8.3% in the last period. The same was observed for the GA group 32–36 weeks (from 19.3% up to 35.0%). 2) Overall 499 children with CP were born from multiple pregnancies in these nine centres between 1980 and 1996. Among all CP cases, the proportion of cases from multiple births increased significantly from 7% in the first period up to 11% in the last period (p<0.001). However, the rate of CP in multiples remained remarkably stable over time, 8.1 per 1000 multiple live births, 8.0 and 8.4, respectively, during the first, second, and third periods (p = 0.95). Among multiples with CP nearly half were very preterm babies. The proportion of children born below 32 weeks increased from 42% up to 55% but not statistically significant (p = 0.17), and a decrease in their mean BW (150 g) as well as in their mean GA (1 week) was observed. These CP children born from multiple presented a predominantly spastic CP type (92%), with 30% of them unable to walk, 25% having a severe intellectual impairment, and 10% a severe visual impairment. This clinical pattern did not change over time.

Conclusions: Despite an increase in the rate of multiple live births, in particular for very preterm multiples, the rate of CP in multiples did not change over time. This probably reflects a balance between improved survival and care management.

M. Carreira¹, J. Vintén², J. Catarino², R. Calado¹.¹National Health Service, DSIA, Lisbon, Portugal

Introduction: In Portugal there is an ongoing survey about adolescent health in school. The main objective of the survey is to monitor the health strategies and health programmes and to know some determinants of health, like attitudes, behaviours and life styles. The first phase of survey was performed in 2003, and included the health evaluation related to prevalence of asthma, tobacco consumption, and physical exercise.

Methods: We selected a clustering sample with 8928 students from all five health regions in continental Portugal. The data were obtained by a structured self responsiveness questionnaire with 96 questions. The questionnaire was composed by five parts: demographics data, parents’ data (professional, occupational situation, level of education, and other data), asthma (wheezing, asthma crisis, nocturne cough, cough and wheezing with exercise, emergency care related to asthma crisis, and drug consumption), physical exercise (in and out of school), and finally about tobacco consumption. The part about asthma was adapted from the ISSAC study questionnaire.

Results: We found a prevalence of 11.3% (95% CI 10.6 to 12.0) for one or more wheezing episode in the past 12 months: males 9.1% (95% CI 9.2 to 10.0) and females 13.1% (95% CI 12.1 to 14.1). By age group the prevalence was 10.7% (95% CI 9.5 to 11.9) at 11–13 years, 10.3% (95% CI 9.3 to 11.3) at 14–16 years and 12.6% (95% CI 12.0 to 13.2) at 17–19 years. The prevalence of daily consumption of tobacco was 11.3% (95% CI 10.6 to 12.0); males 12.0% (95% CI 11.0 to 13.0) and females 10.7 (95% CI 9.8 to 11.6). By age group the prevalence was 1.9% (95% CI 1.4 to 2.4) at 11–13 years, 10.9% (95% CI 18.9 to 20.9) at 14–16 years, and 19.8% (95% CI 18.4 to 21.2) at 17–19 years. 41.2% (95% CI 40.2 to 42.2) of the sample practice physical exercise 1 h or more each day; males 52.4% (95% CI 50.9 to 53.9) and females 30.9% (95% CI 29.6 to 32.2). By age group the practice of 1 h physical exercise or more per day was 41.2% (95% CI 39.3 to 43.1) at 11–13 years, 42.2% (95% CI 40.5 to 43.9) at 14–16 years, and 40.1% (95% CI 38.3 to 41.9) at 17–19 years.

Conclusions: The prevalence of asthma in past 12 months is in the expected rank in comparison with other survey studies in Portugal but higher than the results from the National Health Survey. The one or more wheezing episode in the past 12 months is higher in females than males. The tobacco consumption is very similar in both sex and increases with age. Males practice more intensive physical activity, which is very similar between age groups.

L. Cazzoletti¹, I. Cerveri², A. Corsico², M. Bugiani³, S. Accordini¹, R. Marco¹.¹University of Verona, Division of Epidemiology & Medical Statistics, Verona; ²IRCCS Hospital San Matteo, Clinic of Respiratory Diseases, Pavia; ³CPA-ASL 4 Piemonte, Unit of Pneumology, Torino, Italy

Introduction: During the 1990s guidelines for the treatment of asthma have been developed with a step wise increase in treatment for those with more severe symptoms.

Objectives: The aims of this analysis were to determine asthma severity in a Europe wide sample of physician diagnosed asthmatics and to assess the adequacy of their treatment in relation to the GINA guidelines.

Methods: In the European Community Respiratory Health Survey II, 1291 adults with asthma from 28 centres were identified. Asthma severity was assessed using the GINA criteria, which are based both on clinical features and current treatment. A subject was classified as receiving inadequate treatment if current use of drugs was lower than that suggested by the guidelines for the corresponding severity level. A two level logistic random intercept model identified predictors of "inadequate treatment", with individuals as 1 level units and centres as 2 level units.

Results: Fifty eight per cent of the patients had intermittent asthma, 10.8% mild persistent, 14.7% moderate persistent, and 16.4% severe persistent. Overall, 28.5% (95% CI 26.0 to 31.1) of all the current asthmatics were receiving inadequate treatment, but in those with persistent asthma this percentage increased to 64.3% (95% CI 60.1 to 68.3). Inadequately treated persistent asthmatics had a higher prevalence of exacerbations (66.9% v 52.3%, p = 0.001) than those considered being adequately treated. Subjects owning a peak flow meter (OR 0.55; 95% CI 0.34 to 0.87), compliant with therapy (OR 0.31; 95% CI 0.17 to 0.57) living in Nordic countries (OR 0.50; 95% CI 0.28 to 0.90) and with asthma starting at an earlier age (OR 0.98; 95% CI 0.96 to 0.99) were less likely to be identified as being on "inadequate treatment".

Conclusions: Despite dissemination of guidelines, a substantial percentage of asthmatics in Europe are not taking asthma therapy as recommended. This is particularly marked in those with severe asthma and is associated with increased morbidity.

B. Chaix¹, J. Merlo², P. Chauvin¹.¹National Institute of Health and Medical Research, Research Unit in Epidemiology and Information Sciences, France; ²Malmö University Hospital, Department of Community Medicine, Sweden

Introduction: Past studies that have investigated place effects on health have been conducted with the multilevel approach where variations in a space subdivided into areas are investigated with multilevel models. However, it may be a strong limitation to rely on a space fragmented into areas. Conversely, we suggest using a continuous concept of the space that can be operationalised with a geographical information system, and propose to use spatial modelling techniques and place indicators that continuously consider the space around the individuals’ place of residence. To compare this spatial approach with the more conventional multilevel approach, we investigated geographic variations of healthcare utilisation behaviour in France.

Methods: We used French survey data collected in 1998 and 2000 linked with administrative data on healthcare consumption. The first binary outcome variable indicated whether each individual had a regular primary care physician or not. A second binary outcome indicated whether the individuals had a per cent of 1 year consultations with specialists rather than primary care physicians over 50%. We compared multilevel models with spatial mixed models that define the correlation between individuals as a function of the spatial distance between them. The models were adjusted for several health, demographic and socioeconomic variables. Regarding place indicators, we considered the population density, the per cent of individuals with a low education and indicators of supply of physicians.

Results: From the multilevel models, we found significant variations for both outcomes at the municipality level or at the level of the broad areas (p<0.001). From the covariance parameters in the empty spatial models, we found that the correlation in the risk of not having a regular primary care physician was 0.028 for individuals located at the same place, but only 0.015 for individuals located 10 kilometres away (figures were 0.029 and 0.026 for having a high per cent of consultations with specialists). For both outcome variables, the scaled deviance was markedly lower in the empty spatial model than in the empty multilevel models, indicating a better fit of the spatial covariance structure to the data. The socioeconomic level in the place of residence and the supply of physicians were independently associated with healthcare utilisation behaviour. Associations were stronger when measuring place indicators within a continuous space rather than within the usual administrative areas. These different approaches for measuring place effects also lead to different pictures when identifying places with different levels of exposure to contextual characteristics.

Conclusions: Considering an illustrative example, we found evidence that an approach combining spatial modelling techniques and definition of place indicators in spatial rather than territorial or administrative neighbourhoods was more appropriate to account for the spatial variability of the healthcare utilisation behaviour under study.

B. Chaix¹, P. Chauvin¹, J. Merlo².¹National Institute of Health and Medical Research, Research Unit in Epidemiology and Information Sciences, France; ²Malmö University Hospital, Department of Community Medicine, Sweden

Introduction: Measuring the extent to which health phenomena occur in cluster is highly informative for public health policymakers. Indeed, the magnitude of clustering can be viewed as an indication of the extent to which a contextual dimension must be included in prevention or intervention programmes. For continuous outcome variables, multilevel models provide a convenient measure of clustering in the form of the intraclass correlation coefficient (ICC). However, for binary outcome variables, only approximate definitions exist for the ICC. Therefore, Klaus Larsen and colleagues propose to express the magnitude of clustering in the well known odds ratio scale with an index termed the median odds ratio (MOR) that can be easily computed from multilevel logistic models. On the other hand, alternating logistic regression (ALR) models are now recognised as an interesting alternative to multilevel logistic models. They quantify clustering with a pair wise odds ratio that considers the degree of similarity between individuals residing in the same area. Investigating healthcare utilisation behaviour with Swedish data, we compared these different approaches to measure the magnitude of clustering, and compared their advantages and drawbacks in a user oriented perspective.

Methods: We used Swedish data from the Health Survey 2000 conducted in the county of Skåne in Sweden. The binary outcome variable distinguished the individuals who consulted private healthcare providers from those who only used public providers. Multilevel models and ALR models adjusted for individual level confounders were fitted to the data. We examined whether the per cent of high educated inhabitants in the area of residence had an impact on the propensity to use a private healthcare provider.

Results: The multilevel model indicated that the area level variance in the propensity to use a private provider was highly significant. From the multilevel model, we found that the ICC was equal to 0.08. The MOR was equal to 1.81. The ALR model indicated that the POR was equal to 1.37 (95% CI 1.19 to 1.57). Both the multilevel and ALR models indicated that the amount of clustering was higher among old individuals than among young individuals. This complex pattern of variations was in part explained by the per cent of inhabitants with a high education in the area of residence: the per cent of high educated inhabitants was positively associated with the propensity to use a private provider, and had a stronger impact among old individuals than among younger ones.

Conclusions: Measuring the magnitude of clustering of health related phenomena provides important information for public health policymakers. Since the ICC, the MOR and the POR provide information on clustering under different forms, it is important to compare their statistical consistency and interpretability in a public health user oriented perspective.

J. Chang-Claude¹, C. Lilla¹, S. Kropp¹, H. Bartsch², A. Risch².¹German Cancer Research Centre, Clinical Epidemiology, Heidelberg, Germany; ²German Cancer Research Centre, Toxicology and Cancer Risk Factors, Heidelberg, Germany

Introduction: Sulfotransferase (SULT) 1A1 is involved in the inactivation of oestrogens and in the metabolism of pro-carcinogens such as heterocyclic amines and polycyclic aromatic hydrocarbons, both of which are present in tobacco smoke. Recently, we reported a differential effect of NAT2 genotype on the association between active and passive smoking and breast cancer (Chang Claude et al 2002). The additional investigation of a SULT1A1 genetic variant associated with reduced enzyme activity and stability might therefore provide a deeper insight into the modification of the susceptibility for breast cancer.

Methods: We employed a population based case control study in women up to age 50, which was conducted in Germany. A total of 394 pre-menopausal breast cancer patients and 835 age matched controls with information on genotype and detailed smoking history were included in the analysis. Genotyping was carried out using a fluorescence based melting curve analysis method. We performed multivariate logistic regression analysis to estimate breast cancer risk associated with the SULT1A1 Arg213His polymorphism in relation to smoking and other risk factors.

Results: The overall risk for breast cancer in women who were carriers of at least one variant allele (53.3% of women with breast cancer and 57.7% of controls) was not significantly different from non-carriers of the variant allele (adjusted odds ratio 0.8; 95% CI 0.7 to 1.1). Odds ratios for breast cancer with respect to several smoking variables did not differ substantially for carriers and non-carriers. However, we observed elevated odds ratios associated with passive smoking only (2.2, 95% CI 0.7 to 7.4) and more than 10 pack–years of active smoking (1.8, 95% 0.5 to 6.1) in NAT2 fast acetylators homozygous for the SULT1A1*1 wild type allele but not in NAT2 fast acetylators carrying the SULT1A1*2 variant allele (1.10, 95% 0.4 to 2.7, and 0.7, 95% 0.3 to 2.1, respectively). The risk for breast cancer associated with high endogenous oestrogen exposure was not found to be altered by SULT1A1.

Conclusion: Our data failed to support a possible association between pre-menopausal breast cancer risk and SULT1A1 genotype (Zheng et al 2001; Tang et al 2003) or a significant interaction with active smoking (Saintot et al. 2003). We found no evidence that the SULT1A1 genotype in itself is an important effect modifier of breast cancer risk associated with smoking or endogenous oestrogen exposure. The SULT1A1*1/*1 genotype in combination with NAT2 fast acetylator, however, seemed to increase breast cancer risk in women exposed to tobacco smoke.

S. Conti¹, G. Farchi², E. Buiatti³, S. Arniani³, P. Meli¹, G. Minelli¹, R. Solimini¹, V. Toccaceli¹, M. Vichi¹.¹Italian National Institute of Health, Unit of Statistics, Rome, Italy; ²Italian National Institute of Health, National Centre of Epidemiology, Rome, Italy; ³Region of Tuscany, Regional Agency of Public Health, Florence, Italy

Introduction: There is an increasing need to describe population’s health status as well as that of specific groups of individuals by means of objective tools, which allow comparing data through time and space. This need interests all the institutions in charge with decision making and investments for public health. There are two types of survey that allow to ascertain the health status of a population at a defined time point; the first type requires self administered questionnaires to interview individuals (Health Interview Survey, HIS); the second one is characterised by questionnaires administered by a doctor in a face to face interview as well as by medical investigation that involves physical and biochemical measurements (Health Examination Survey, HES). In Italy, the Italian National Census Bureau (ISTAT), with the title Health Conditions and Health Care Services Use (Condizioni di salute e ricorso ai servizi sanitari), has carried out a HIS. No HES survey has been carried out in Italy yet, while a pilot study Health in Florence took place in Florence between the end of 2000 and the summer 2001. This study was carried out on a part of the general population sample that had previously undergone the investigation during the HIS ruled by ISTAT during the period 1999–2000.

Method: The study Health in Florence concerns the same probability sample drawn by ISTAT in Florence for the HIS, even if only the individuals aged 35-74 years were selected among the HIS examinees, as in this group of age the frequency of pathologies is higher. Data from 343 examined individuals (47% men, 53% women) were analysed. Questions in HIS and questions and measurements and diagnostic criteria in HES, allowed collecting information about some relevant diseases and health problems such as: myocardial infarction, ictus, diabetes, hypertension, hyperglycemia, osteoporosis, as well as physical characteristics (height, weight, body mass index), and lifestyles (smoking habits). Appropriate statistical methods were applied to compare data from the two surveys, such as K statistics, Mc Nemar and Student’s t tests.

Results: From this comparison, it was possible to attribute a higher sensitiveness to the medical investigation in the ascertainment of diseases in comparison with the self administered questionnaires, in particular for hypertension, diabetes, and osteoporosis. Further, in this study, as well as in many other similar investigations, there is a difference between measured height and weight, and those self perceived: men and women perceive themselves taller and slimmer than they are.

Conclusions: This pilot study has firstly shown the feasibility of an HES within the Italian National Health System; further, the comparison between the HIS and the HES shows the usefulness of integrating and sharing information from the two sources.

S. Conti¹, P. Meli¹, G. Minelli¹, R. Solimini¹, V. Toccaceli¹, M. Vichi¹, L. Perini², C. Beltrano².¹Istituto Superiore di Sanità, Unit of Statistics, Rome, Italy; ²UCEA, Italian Central Office for Agriecology, Rome, Italy

Introduction: It is widely recognised that extreme climatic conditions during the summer months can constitute a major public health threat. Persons living in cities have an elevated risk of death when the temperature (and humidity) are high, compared with those living in suburban and rural areas (urban heat island effect). Studies on heat wave related mortality have further demonstrated that the greatest increases in mortality occur in the elderly. Following the unusually hot summer of 2003, together with the dramatic news from neighbouring countries such as France, the Italian Minister of Health requested the Istituto Superiore di Sanità’s Office of Statistics to undertake an epidemiologic study of mortality in Italy during summer 2003, to investigate whether there had been an excess of deaths, with a particular focus on the elderly population.

Methods: Communal offices, which maintain vital statistics, were asked for the number of deaths among resident people, registered during the period 1 June to 31 August 2003 and the same period during 2002, for each of the 21 capitals of the Italian regions. Within the framework of collaboration with the Italian Central Office for Agriecology (UCEA), meteorological data (minimum/maximum temperature and humidity) were obtained for the observed periods 2002 and 2003; this allowed us to calculate discomfort indexes, such as the Humidex.

Results: Compared with 2002, during the three summer months (1 June to 31 August 2003) there was an overall increase in mortality of 3134 (from 20 564 in 2002 to 23 698 in 2003). The greatest increase regarded elderly people: 2876 deaths (92%) occurred among people aged 75 years. The increase in mortality was greatest in the cities belonging to the northwest of the country (31.5%), followed by the south (17.8%), the northeast (16.4%), and the centre (16.3%). With respect to each city, the greatest increase was observed in Turin (45%), where the deaths more than doubled during the first 15 days of August, Milan (30.6%) and Genoa (22.2%). It is noteworthy that some cities in the south experienced their highest increase in mortality during the last period of August; in Bari the overall excess mortality was 33.8%, but in the last part of August it reached 137%. In these cities, the relationship between mortality and climatic indexes (t. max, Humidex) was investigated and a clear correlation was observed.

Conclusion: Lessons learnt and measures to be taken in Italy to prevent the health impact of heat wave: the Italian Minister of Health has started a programme, involving the cities where the highest increase of mortality was observed (such as Milan, Turin, Genoa, and Rome), to identify the elderly at risk, and to provide measures in order to avoid harms to their health during summer.

M. Cordeiro¹, J. Corrente², K. Patricio³.¹School of Medical Sciences/UNICAMP, Preventive Medicine, Campinas, Brazil; ²Institute of Biosciences/UNESP, Statistics, Botucatu, Brazil; ³School of Medical Sciences/UNESP, Public Health, Botucatu, Brazil

Introduction: Acute diarrhoea is common in Brazilian children, however little is known about case aetiology and outbreak dynamics. Epidemiological monitoring and vigilance groups are still coming up against poor laboratory support.

Objective: The object of this work is to analyse the temporal distribution and related variables of diarrhoea in children who attended the emergency department of an important regional hospital in southeast Brazil in 2002 and 2003.

Method: Poisson’s multiple regression model was adjusted, considering serious cases (admission and/or intravenous rehydration therapy (IRT)) as the dependent variable, and the following as independent: age group (less than 1 year old, 1–4 years old, and 5–12 years old), concurrent respiratory symptoms, and time of occurrence.

Results: Case temporal analysis revealed a seasonal pattern with a marked incidence peak in September 2002 and a less intense one in September 2003. The significant variables from the final adjusted model suggested that occurrence time of serious cases was linked to the 5–12 year age group; the curves were similar (p = 0.78). The other groups did not follow the model in serious cases where IRT was given (less than 1 year old p = 0.05, and 1–4 years old p = 0.03). The model showed that the cases associated with respiratory symptoms also had the same distribution pattern as those admitted and/or given IRT.

Conclusion: Even though this was not a population study, case notes in an important regional hospital can function as an early sentinel site, indirectly reflecting what could be happening in the population. This study allowed us to investigate diarrhoea outbreaks in a region where an aetiological study was impossible. Epidemiological vigilance of diarrhoea, particularly in children, assumes differential diagnosis between different aetiologies, which is impossible in regions without laboratory support. The epidemiological vigilance team should take advantage of instruments that allow them to identify outbreaks and raise hypotheses on their aetiology, which would otherwise not be detected.

J. Corrente¹, M. Cordeiro², T. Ruiz³.¹Instituto de Biociências – UNESP, Departamento de Biostatística, Botucatu, São Paulo, Brazil; ²Faculdade de Ciências Médicas – UNICAMP, Departamento de Medicina Preventiva, Campinas, São Paulo, Brazil; ³Faculdade de Medicina de Botucatu – UNESP, Departamento de Saúde Pública, Botucatu, São Paulo, Brazil

Multidimensional instruments have been widely used in studies aimed at measuring heterogeneous aspects of populations. With the advent of electronic data analysis, such instruments have become more and more complex, thus hindering statistical analysis and interpretation. Therefore, a home survey was conducted in a medium sized city in south eastern Brazil using an instrument with 172 variables to learn about the profile of elderly people (welfare, disease prevention, material comfort, social relationships, satisfaction concerning intellectual capabilities and manual skills, and level of physical activity), reported morbidity, mental evaluation, and the population’s opinions on the meaning of quality of life in old age. The variables were evaluated in blocks for each lifestyle category, morbidity, and level of physical activity. The questions were measured on different scales and one question was open. Where answers were on a 1–7 scale, the variables were dichotomised with four established as the cutting point. Where answers were categorised as "yes", "almost always", "sometimes", and "no", they were also grouped into two classes so that "yes" and "almost always" were regarded as one class, and "sometimes" and "no" as another. The open question was categorised according to the most frequent answers and then dichotomised as "yes" and "no" for each elderly person. Thus, the dataset consisted of binary variables in which various cluster analysis methods were tested with various suggestions for the distance matrix using STATISTICA v.5.0 software. There were practically no differences in any clusters; therefore a distance matrix was made using Euclidean distance and the ward clustering method; this is based on analysis of variance, aimed at minimising the sum of squares between two hypothetical clusters in each step. A first cluster was obtained with 11 binary variables from answers to the open question, where three groups of elderly people were observed: i) those who valued aspects related to work, religion, rectitude, and charity as quality of life; ii) those who valued leisure, material possessions, and healthy habits; and iii) those who valued interpersonal relationships, good health, and mental balance. These 11 variables were then reduced according to three clusters and including the other variables from the other domains and observing the formed clusters gradually constructed hierarchical dendrograms. As a final result, cluster analysis revealed four groups and identified the profiles of the elderly individuals.

S. Cortez¹, H. Bettiol¹, M. Barbieri¹, M. Saraiva², M. Goldani³, A. Silva⁴.¹Faculty of Medicine of Ribeirão Preto-USP, Department of Paediatrics, Ribeirão Preto, SP, Brazil; ²School of Dentistry of Ribeirão Preto-USP, Department of Paediatric, Preventive and Social Dentistry, Ribeirão Preto, SP, Brazil; ³Federal University of Rio Grande do Sul, Department of Paediatrics, Porto Alegre, RS, Brazil; ⁴Federal University of Maranhão, Department of Public Health, São Luís, MA, Brazil

Introduction: There has been controversy about the adverse effect of working at early ages on human growth. This debate seems to come from differences in study designs (mostly cross-sectional) and the difficulties in controlling for socioeconomic factors. Moreover, most of the evidence comes from underprivileged communities.

Objective: To assess the effect of childhood work (less than 14 years old) on height of a 23/24 year old population based cohort born in 1978/1979 in Ribeirão Preto, the wealthiest area in a developing country, Brazil.

Methods: From 6728 valid questionnaires obtained from the mothers when they gave birth to a singleton born alive at the initial cohort, 1071 (15%) individuals (518 males and 553 females) were re-examined at 23–24 years old. Childhood work was classified in three categories according to the age at the first job: less than 14; 14–16; 17 years. Those who identified themselves with oriental ascendance were excluded from the study. Height measurements in centimetres were obtained clinically at the follow up examination. Social economic status was measured through education, ethnicity, family income at birth, number of siblings, secondary exposure to tobacco at home, and mother’s age and education. Known determinants of height were also considered: order of birth, birth weight, birth length, and smoking status. The analysis was performed separately for males and females. Covariance analysis followed bivariate and stratified analysis. Variables were selected into the model using a backward stepwise like selection of the variables.

Results: Height average was 175.8 cm (SE 0.3) and 162.7 cm (SE 0.3), respectively for males and females. In the bivariate analysis child work was negatively associated with height for both males (p = 0.002) and females (p = 0.003). However, when adjusted for covariates the strength of association was reduced (males p = 0.229 and females p = 0.142). Trend test was only borderline significant for males (p = 0.086) and no interaction was observed between working at early ages and other covariates in any of the models. For males the most important predictors of height were: mother’s ethnicity (p <0.001), mother’s age (p = 0.008), and birth length (p <0.001). For females the most important covariates associated with height were: mother’s age (p = 0.004), birth length (p = 0.006), and birth order (p = 0.003). Education (p = 0.008) and family income (p = 0.091) were only borderline significant.

Conclusions: Our results seem to not support the hypothesis of working influence on height after considering socioeconomic factors. At least for this population, it seems that socioeconomic factors play the most important role in the determination of height.

Supported by FAPESP, CNPq and FAEPA.

L. Costa^1,2, D. Gal¹, H. Barros¹.¹University of Porto Medical School, Hygiene and Epidemiology, Porto, Portugal; ²Hospital of Alto Minho, Rheumatology Service, Ponte de Lima, Portugal

Introduction: Muscular–skeletal conditions are a fundamental cause of individual disease and of health expenditure in evolved societies. Due to this, the present study was performed to determine the prevalence of some of these pathologies in a representative sample of the adult population of the city of Porto.

Methods: The 1057 participants evaluated (667 women and 390 men; equal to or older than 18 years of age) were selected as part of a population health survey (EPI-Porto), through random telephone digit dialling (participation 70%) of residents in the city of Porto. All responded to a questionnaire, which included questions about lifestyle, behaviours, and personal and family disease history. In particular, they were questioned about whether their physician ever diagnosed them as having rheumatoid arthritis, lupus, ankylosing spondylitis, psoriatic arthritis, hip or knee osteoarthritis, and chronic low back pain.

Results: At least 30.6% of women and 14.6% of men reported one of these pathologies. A prevalence of rheumatoid arthritis in women was 2.2% (95% CI 1.3 to 3.8) and 0% (95% CI 0.3 to 1.4) in men. Globally, the frequency of lupus was 0.3% and 0.7% for ankylosing spondylitis, similar in both sexes. Arthritis in the hip (7.5% v 2.6%), in the knee (14.2% v 6.2%) and lower back pain (17.4% v 7.4%) were significantly more frequent in women and increased with age.

Conclusion: One in every four adults reported at least one diagnosis of a rheumatic pathology revealing the marked public health importance and healthcare needs of individuals with muscular-skeletal conditions.

S. Cotton¹, L. Sharp¹, R. Seth².¹University of Aberdeen, Department of Medicine & Therapeutics, Aberdeen, UK; ²Queens Medical Centre, Department of Histopathology, Nottingham, UK

Introduction and Objectives: Human papilloma virus (HPV) is a sexually transmitted virus. Some high risk types of HPV have been implicated in cervical cancer and pre-cancer. There is current debate as to whether HPV testing could be employed within cervical screening, either as a primary screening tool, or in triage of women to appropriate follow up after an abnormal smear. This debate would be informed by increased understanding of the epidemiology of HPV infection. There is relatively little UK data, and data from other populations may not be generalised due to different patterns of sexual behaviour. In this analysis we describe factors associated with HPV infection in UK women.

Methods: Subjects comprised 5031 women aged 20–59 years, resident in Grampian, Tayside or Nottingham, with a smear taken in the UK Cervical Screening Programmes between 1999 and 2002 and assessed for eligibility for TOMBOLA (Trial Of Management of Borderline and Other Low-grade Abnormal smears). All had a current smear reported as normal, or showing some degree of abnormality, and up to one previous borderline smear. They had an HPV test and completed a lifestyle questionnaire. HPV analysis was by PCR using consensus GP5+/6+ primers and a 14-probe cocktail to detect high risk types. Women were categorised as high risk HPV positive (hrHPV+ve) or negative. Associations between lifestyle factors and HPV status were analysed using logistic regression.

Results: 39.2% of women were hrHPV+ve. The two most important predictors of hrHPV status were age and smear status. Older women were at lower risk of being hrHPV+ve than younger women (odds ratio (OR) 50–59 years v 20–29 = 0.21, 95% CI 0.15 to 0.29). The risk of being hrHPV+ve increased with increasing severity of current smear. Compared with women with a current normal smear, the OR for women with a current borderline smear was 2.8 (95% CI 2.3 to 3.5) and for women with a first mild smear was 6.7 (95% CI 5.4 to 8.5). After adjustment for smear status and age a number of other factors were associated with hrHPV positivity. Having a university/college degree reduced risk (OR = 0.77; 95% CI 0.63 to 0.93), as did having had a child (OR = 0.74, 0.63 to 0.87). Married/co-habiting women were at lower risk than other women. Non-white women were at higher risk than white women (OR = 1.37; 95% CI 1.00 to 1.87). Risk was increased in pill users (OR = 1.23; 95% CI 1.06 to 1.44) and current smokers (OR = 1.19; 95% CI 1.03 to 1.38). Level of physical activity, current condom use, and current use of hormonal contraception (other than oral contraceptives) were not associated with hrHPV infection. Information on number of sexual partners was not available.

Conclusions: In this large series we have found associations between several lifestyle factors and hrHPV status. These factors are highly inter-related, and some may be markers for other risk factors. Epidemiological data of this type will inform the debate on HPV testing.

D. Czeresnia.ENSP/FIOCRUZ, DEMQS, Brazil

An important aspect in the epidemiologic profile of the industrialised Western countries is the tendency of growth of allergic and autoimmune diseases prevalence that has been happening since the middle of the 20th century. The so-called hygiene hypothesis, intensely researched in past years, establishes a connection between this growth and the decrease of infectious diseases in those countries, especially intestinal infections. Recent works formulate that the lymphoid tissue stimulation by infectious agents would be involved in the maturation processes of the mucosal immune system in early childhood and, when inadequate or insufficient, the risk of atopy is accentuated. This new field of problems and discoveries takes new elements into consideration about the epidemiologic transition processes in Western societies. It also allows rethinking some controversial aspects in the history of epidemiologic diseases theories.

The process in which infectious diseases incidence tends to decrease inversely to allergic and autoimmune diseases increase is not just explained by demographic, environmental, and sociocultural changes. Another explanation level in which those diseases would be biologically interconnected is added. Allergic and autoimmune diseases would be a consequence of a disturbance of regulatory mechanisms involving complex molecular interactions in which microorganisms would be involved. Researchers suggest that the protective or harmful effect of bacterial molecules is likely to depend on the dose and "a complex mixture of the timing of exposure during the life cycle, environmental and genetic cofactors". At a molecular level, health and disease would correspond to mechanisms of balance, closer to the idea of predisposition. The objective of the research about the hygiene hypothesis, still cause-centred, is to identify microbes in doses and circumstances capable of preventing atopy without causing disease. However, maybe the scientific development at molecular level is creating the conditions for a new epistemological discontinuity that would make a victory of previously discredited perspectives.

R. Marco¹, L. Cazzoletti¹, I. Cerveri², A. Corsico², S. Accordini¹, M. Bugiani³.¹University of Verona, Division of Epidemiology & Medical Statistics, Verona; ²IRCCS Hospital San Matteo, Clinic of Respiratory Diseases, Pavia; ³CPA-ASL 4 Piemonte, Italy

Introduction: Identifying factors predicting asthma severity might help to prevent the most serious consequences of the disease or at least to better monitor its progression.

Objective: The aim of this study was to identify long term prognostic factors of asthma severity in a cohort of asthmatics, followed from 1992 to 2002.

Methods: Between 1991 and 1993, in the ECRHS I, 1482 adults with asthma from 27 centres was identified in 13 countries. Of these, 980 (66%) were reassessed 9 years later (ECRHS II). At the end of follow up all current asthmatics were classified according to the severity of the disease, using the GINA criteria, which are based both on clinical features and current treatment. Severe asthma was considered present if a subject was classified as moderate/severe asthmatic. A two level logistic random intercept model assessed the association of baseline characteristics with severe asthma, with individuals as 1 level units and centres as 2 level units.

Results: At the end of follow up and according to the GINA criteria, out of these 980 asthmatics, 457 (46.9%) had intermittent asthma, 83 (8.5%) mild persistent, 133 (13.6%) moderate persistent, 141 (14.5%) severe persistent, and 161 (16.5%) were in partial remission (that is, no symptoms, no exacerbations, no asthma medication in the past year). At baseline, severe asthmatics were older, more were women, and a higher percentage reported a familiar history of asthma and a low educational level. They had poorer lung functions (FEV₁, FEV₁% predicted, and FEV₁/FVC) and worse control of symptoms (symptom score) than less severe asthmatics. Furthermore, they had a higher total IgE level, a higher percentage of subjects tested positive to the methacholine test and sensitised to D Pteronyssus, Cat and Cladosporium herbarium. When mutually adjusted in a multilevel logistic model, the baseline factors significantly associated with a subsequent severe asthma were: FEV₁% predicted (OR 0.96; 95% CI 0.96 to 0.98), total IgE [ln] (OR 1.6; 95% CI 1.2 to 2.2), poor control of symptoms (OR 1.14; 95% CI 1.0 to 1.29), a familiar history of asthma (OR 1.8; 95% CI 1.2 to 2.8), and age at onset of asthma (OR 1.02; 95% CI 1.00 to 1.04).

Conclusions: At the end of follow up, a non-negligible proportion of asthmatics were in partial remission (16%). Besides poor lung function and poor control of the disease, a later onset of asthma and a history of familiar asthma are long term predictors of severe asthma.

D. Di Lallo, A. Polo, A. Spinelli, L. Gianrizzo, G. Guasticchi.Agency for Public Health, Servizio Tutela Soggetti Deboli, Rome, Italy

Introduction: Migration to Italy from developing countries is increasing. In 2000 there were about 25 000 babies born to foreign mothers (about 5% of the total national births). Only few data are available on the health conditions of immigrants. The aim of the study was to compare utilisation of prenatal care, modality of obstetric care, and pregnancy outcomes in foreign born (FB) and Italian born (IB) mothers.

Methods: The study was carried out in the Lazio region of Central Italy, which has 5 000 000 million inhabitants. Information was obtained from neonatal administrative hospital discharge records, containing data on maternal and neonatal characteristics. During the years 2001–2002 there were 82 619 live births among IB mothers and 13 015 among FB mothers. Assignment to ethnic groups was done according to the place of birth of the mother: the greatest group was from the Eastern Europe countries (n = 4787), in particular Romania, and the smallest from Central Africa (n = 831).

Results: The utilisation of prenatal care was lower in immigrant mothers. Twelve per cent of FB mothers had the first prenatal visit after 12 weeks of gestation compared with 2.8% of IB mothers. Among women aged above 35 years, the use of amniocentesis was 19.9% in the FB group and 44.4% in the IB group. The prematurity rate (gestational age under 37 weeks) was 8.5% for FB and 6.7% for IB mothers. The caesarean section rate standardised for gestational age was lower in FB than in IB mothers (32.9% v 39.4%). The rate of neonatal transfer to different hospitals for serious medical conditions was 2.1% for FB group and 1.5% for IB group. Low 5 min Apgar score (<7) was more frequent among the newborn of FB mothers: 1.3% compared with 0.8 among those of IB mothers. The inhospital mortality was 4.2/1000 births for those born to FB mothers and 3.2 for those born to IB mothers (p = 0.06). However, there were no differences in the inhospital mortality among babies with gestational age less than 37 weeks (36.1/1000 births for those born to FB mothers and 34.7 for those born to IB mothers) and in the incidence of neonatal respiratory diseases (1.2 v 1.1).

Conclusions: We observed a high disparity in the utilisation of prenatal health services, as well as a higher prematurity rate and inhospital neonatal mortality, among immigrant mothers. These results should be taken into consideration when making policies for the promotion of mother and child health in the immigrant population.

J. Dickute.Kaunas University of Medicine, Department of Preventive Medicine, Kaunas, Lithuania

Introduction: Smoking during pregnancy and unfavourable socioeconomic factors are related to the higher risk of low birth weight.

Aim: To evaluate the risk of low birth weight according to socioeconomic factors among smoking and non-smoking expectant women.

Methods: Case-control study involved 851 newborns with low birth weight (<2500 g) and 851 normal weight newborns. The study started 1 February 2001 and ended 31 October 2002 in six main maternity hospitals in Lithuania. Mothers of newborns were interviewed on the first or second day after delivery using the structured questionnaire. The database was processed by the application of statistical package SPSS for Windows v.10.0.

Results: Prevalence of smoking during pregnancy was higher among mothers with unfavourable socioeconomic factors (19.5–35.2%), if compared to mothers in favourable socioeconomic groups (11.5–16.6%), with the clear tendency to be even higher among cases if compared to controls. Risk of low birth weight was significantly higher for smoking mothers in the univariate stratified analysis. Smoking mothers who were unemployed or had unstable marital status were at the highest risk to deliver low birth weight baby, OR 4.6 and 4.8, respectively. Smoking during pregnancy among mothers with low income increased the risk of low birth weight by 3.5 times, while smoking mothers living in rural area had 2.5-fold higher risk of low birth weight. Logistic regression analysis showed the statistically significant independent influence of low education, low income, unemployment, and smoking during pregnancy on the risk of low birth weight.

Conclusions: Smoking during pregnancy is more prevalent among mothers with unfavourable socioeconomic factors. Smoking during pregnancy in combination with the socioeconomic inequalities is significantly associated with the higher the risk of low birth weight.

H. Dolk, M. Loane.University of Ulster, EUROCAT, UK

Introduction: The EUROCAT project is a European network of population based registries set up in 1979 to carry out epidemiologic surveillance of congenital anomalies. Currently over 1 000 000 births are surveyed per year throughout 36 registries in 18 countries of Europe. Multiple births are known to be at higher risk than singletons for congenital anomaly. The proportion of multiple births is steadily rising due mainly to advances in fertility treatment, as well as increasing maternal age.

Objective: To assess the effect of the rising proportion of multiple births on the prevalence of congenital anomalies in Europe.

Methods: Data from 32 population based registries of congenital malformations (EUROCAT) for the years 1980–2001.

Main Outcome Measure: Proportion of congenital anomaly cases resulting from multiple pregnancies.

Results: There has been a gradual increase in the proportion of multiple births among congenital anomaly cases over time, rising from 3.1% in 1980–84 to 3.5% in 1996–2001. This represents a prevalence of 0.66 per 1000 total births in 1980–84 rising to 0.78 per 1000 total births in 1996–2001. 4.8% of cases from multiple births were from higher order multiple births (triplets and so on), almost doubling from 0.09% in 1980–84 to 0.16% in 1996–2001. Nearly a fifth of twin cases were from twin pairs concordant for malformation. Among non-concordant twins 1996–2001, 5% of cases resulted in termination of pregnancy of the affected twin following prenatal diagnosis compared to 15% of singletons.

Conclusions: One of the implications of the rise in multiple births in Europe associated with fertility treatment is its effect on the number of children born with congenital anomalies. Multiple births with congenital anomalies also present difficult problems on prenatal diagnosis of one affected foetus. Further analysis of EUROCAT data will determine anomaly specific risks, risks among higher order multiple births, and regional variation.

B. Brinkhaus¹, C. Becker-Witt¹, S. Jena¹, D. Selim¹, B. Liecker², S. N. Willich¹.¹Charité-University Medical Centre, Institute of Social Medicine, Epidemiology, Health Economics, Berlin, Germany; ²Insurance company, Techniker KK, Hamburg, Germany

Introduction: Complementary and alternative medicine treatment such as acupuncture is increasingly used by patients with allergic rhinitis (AR). The aim of the study was to investigate the effectiveness of treatment with v without acupuncture in patients with AR in general medical practice.

Methods: In this randomised study, patients (>17 years) with AR were randomised either to an acupuncture group (ACU) or a control group (CON). ACU received up to 15 acupuncture treatments over a period of 3 months after entering the study, whereas CON received no acupuncture in the first 3 months. CON received also up to 15 acupuncture treatments in months 4–6. At any time in the study ACU and CON were free to use routine care for AR. Patients filled in standardised questionnaires including sociodemographic data, rhinitis quality of life questionnaire (RQLQ, score minimum, 0; score maximum, 6) and generic quality of life (SF-36) at baseline and after 3 and 6 months.

Results: Altogether, 981 patients (64% female, 39.4 (SD 11.2) years and 36% male, 40.7 (13.0) years) were randomised. At baseline there was no difference between the groups in RQLQ (ACU and CON 3.1 (1.1)). After 3 months there was a significant difference in the RQLQ score (p<0.001) in favour for ACU (ACU 1.5 (1.2) v CON 2.6 (1.5)). The symptoms of ACU according to the RQLQ score improved again significantly (p<0.001) between month 3 and 6. Similarly, after 3 months we found a significant difference (p<0.001) in favour for ACU v CON on both component scales of the SF-36. After 6 months there was no significant difference for the RQLQ score and both component scales of SF-36 between both groups.

Conclusion: Patients suffering from AR showed a significant and clinically relevant improvement of quality of life after treatment with addiional acupuncture compared to patients in routine care without acupuncture.

J. Prel¹, J. Rosenbauer¹, A. Icks¹, M. Grabert², R. Holl², G. Giani¹.¹Diabetes Research Institute at Heinrich-Heine-University, Department Biometrics and Epidemiology, Duesseldorf, Germany; ²University of Ulm, Division of Applied Information Technology, Ulm, Germany

Introduction: The incidence of type1 diabetes is increasing worldwide. Despite decades of intensive research, the aetiology of type1 diabetes mellitus (T1DM) is still unknown. Socioeconomic conditions have a strong impact on health, particularly on children’s health. Discovering relationships between T1DM in childhood and factors of the social environment could contribute to the aetiological clarification of the disease.

Methods: In an ecological study we investigated the association between the incidence of T1DM in children up to 14 years old and the socioeconomic factors at area level in North Rhine-Westphalia (NRW) between 1996 and 2000. The incidence data were derived from the Diabetes Registry NRW. Regional data for the social status were provided by official statistics. Analytical methods were the Spearman’s rank correlation, the Poisson regression, and a bootstrap method for estimating confidence intervals of the non-normal distributed correlations.

Results: A significant correlation was found between a population based income ratio (quotient of low to high income groups) and the type1 diabetes in children under 5 years (r = 0.57 (95% CI 0.32 to 0.77)) respectively under 15 years (r = 0.63 (95% CI 0.36 to 0.77)). A similar trend was detected for the association between a deprivation index, including household income, education and professional training, and the diabetes incidence. An inverse correlation between population density and type1 diabetes incidence was also found in children under the age of 15 years (r = –0.38 (95% CI –0.62 to –0.10)).

Conclusions: These results indicate an association of a higher diabetes incidence in regions with lower socioeconomic conditions. To avoid an ecological fallacy the correlations should be proven on individual level. As T1DM is a rare disease a well prepared case-control study would fit to verify the hypothesis.

F. Duarte-Ramos, F. Calado, L. C. Pinheiro, J. Cabrita.Faculty of Pharmacy, Lisbon University, Social Pharmacy, Lisbon, Portugal

Introduction: Prevalence of chronic diseases, such as diabetes, treated on a daily basis with drugs that are specific to that pathologic entity can be estimated through the use of both drug sales data and drug consumption pattern. However, the accuracy of this estimate is affected by the proportion of patients treated with combinations of two or more different classes of drugs.

Objective: The aim of this study was to estimate the Portuguese drug treated diabetes type 2 prevalence using drug consumption data, taking into account the proportion of patients that use more than one drug for the same indication.

Methods: The drug treated diabetes prevalence estimate was based on the formula proposed by Sartor and Walckiers (1995), considering: (1) the total amount of oral hypoglyaemic agents (OHAs) sold in Portugal during 2003 based on IMS Health data; (2) the defined daily dose of each OHA, as proposed by the WHO Collaborating Centre for Drug Statistics Methodology; and (3) a weighing factor (w) based on the proportion of individuals taking an association of OHAs, obtained in a national survey. The drug consumption pattern, concerning OHAs, was obtained in a descriptive, cross national survey performed in 2003, in 118 community pharmacies distributed all over the country. Each pharmacy was asked to recruit 15 patients presenting a medical prescription with at least one OHA. Pharmacists using a structured questionnaire collected information regarding sociodemographic and therapeutic characteristics.

Results: Drug use study: we have studied a sample of 1113 type 2 diabetics (575 women, 532 men, 6 sex unknown), with a mean age of 64.3 (SD 11.0) years and a mean duration of the disease of 9.8 (SD 9.0) years. OHAs were prescribed as monotherapy in 46.7% (520/1113) of the patients, with sulphonylureas being the most frequently prescribed class (56.4%). The remaining 593 diabetics (53.3%) received from 2 to 4 OHAs concomitantly. In the 2 agent regimen, Sulphonylurea + Biguanide was the most popular association. Diabetes prevalence: using the methodology proposed by Sartor and Walkiers, the prevalence of drug treated diabetes type 2 was estimated to be 2.52%, which is close to the value assumed by the National Health Authorities, that is, 2.7%.

Conclusions: Attending the high prevalence of polytherapy in our clinical practice, it is imperative to use an estimator based on the proportion of patients that use more than one OHA in order to obtain a valid estimate. The estimate of diabetes prevalence from drug consumption data appears to be a valid approach. The authors considered that this is a cost effective strategy to monitor the disease prevalence from an epidemiological perspective.

1. Sartor Fe, Walkiers D.American Journal of Epidemiology 1995;141(8):782–7.[Abstract/Free Full Text]

D. Dunt¹, G. Elsworth², D. Southern³, C. Harris⁴, M. Potiriadis⁵, D. Young⁶.¹University of Melbourne, Public Health, Melbourne, Australia; ²Royal Melbourne Institute of Technology University, CIRCLE, Melbourne Australia; ³University of Melbourne, General Practice, Melbourne Australia; ⁴Monash University, Centre of Clinical Effectiveness, Melbourne Australia; ⁵University of Melbourne, General Practice, Melbourne Australia; ⁶University of Melbourne, General Practice, Melbourne Australia

Aim and Background: The further integration of primary care within the wider health system is an imperative for reform in all countries. The aim of this paper is to determine the factors associated with GP integration using the recently developed GP integration index.

Methods: A database, derived from an Australia wide mail questionnaire survey of 1256 GPs drawn from a 20% stratified random sample of 123 divisions of general practice (51.7% response rate) was used. The GP Integration Index consists of nine scaled GP integration factors, and their two associated two higher order factors: primary care management (PCM) and community health role (CHR), as well as five GP integration enabling factors. A multivariate multilevel analysis was undertaken. An explanatory model for both PCM and CHR based on the GP integration factors as well as general practice, GP and regional characteristics was proposed.

Results: The principal findings were that CHR and PCM were most strongly associated with GP integration enabling factors (mainly at the individual level) and, for CHR only, with urban rural location (mainly at the area level). The most important single explanatory variable for both PCM and CHR was the GP integration enabling factor, knowledge of local resources. Variables defined by the area unit (The Division of General Practice) explained at most 13% of the overall variance of CHR and almost none for PCM. The important explanatory variables were ones reflecting the way GPs work, rather than their broad "classification" within individual or GP setting groupings. Based on these results, some revision to the proposed model was necessary.

Conclusion: Process factors (as compared to structural factors) were more important in relation to GP integration than previously appreciated. Future policy initiatives to promote GP integration should focus on programmes to improve GPs’ knowledge of local resources.

J. Empana¹, P. Ducimetiere¹, D. Arveiler², J. Ferrieres³, A. Evans⁴, B. Haas², A. Bingham¹, P. Amouyel⁵, J. Dallongeville⁵.¹INSERM U258, France MONICA Project coordinator, Villejuif, France; ²Department of Epidemiology and Pub Health and, Strasbourg MONICA Project, Strasbourg, France; ³INSERM U558, Toulouse MONICA Project, Toulouse, France; ⁴Department of Epidemiology and Public Health, Belfast MONICA Project, Belfast, Northern Ireland; ⁵INSERM U508, Lille MONICA Project, Lille, France

Introduction: Assessment of the absolute risk of coronary heart disease (CHD) is widely used to identify high risk subjects who could benefit from primary prevention. The Second Joint Task Force of European Societies on coronary prevention recommended the use of an algorithm derived from the Framingham risk function although its relevance in the European population is not fully established. Recently, the German PROCAM risk function has been proposed to estimate the absolute CHD risk but its applicability to other populations has never been investigated. We thus assessed whether the Framingham and PROCAM risk functions were applicable to men in Belfast and France.

Methods: We performed an external validation study within the PRIME (Prospective Epidemiological Study of Myocardial Infarction) cohort study. It comprised men recruited in Belfast (2399) and France (7359) who were aged 50 to 59 years, free of CHD at baseline (1991 to 1993), and followed over 5 years for CHD events (coronary death, myocardial infarction, angina pectoris). We compared the relative risks of CHD associated with the classic risk factors in PRIME with those in Framingham (coronary death, myocardial infarction, and angina pectoris), and PROCAM (coronary death or myocardial infarction) cohorts. We then compared the number of predicted and observed 5 year CHD events (calibration). Finally, we estimated the ability of the risk functions to separate high risk from low risk subjects (discrimination).

Results: The relative risk of CHD calculated for the various factors in the PRIME population was not statistically different from those published in the Framingham and PROCAM risk functions. Overall, there was a positive linear relationship between the number of CHD events predicted by the Framingham and PROCAM risk functions and that observed in PRIME. However, the number of predicted CHD events clearly overestimated that observed in Belfast and France. The two risk functions had poor similar ability to separate high risk from low risk subjects in Belfast and France (c-statistic range: 0.61–0.68).

Conclusion: While the use of Framingham and PROCAM risk functions may be suitable for ordering individuals according to their estimated absolute CHD risk, their use seems inappropriate to estimate the absolute CHD risk of healthy middle aged men from low risk (France) and high risk (Belfast) populations because of a clear overestimation. More specific population risk functions are needed. Accordingly, the SCORE system recently provided charts to estimate the 10 year risk of fatal CHD events in individuals from high and low risk European countries, using large dataset of European countries.

M. Engberink¹, J. Geleijnse¹, J. Durga¹, D. Swinkels², W. Kort de³, E. Schouten¹.¹Wageningen University, Human Nutrition, Wageningen, The Netherlands; ²University Medical Centre, Department of Clinical Chemistry, Nijmegen, The Netherlands; ³Sanquin Blood Bank, Nijmegen, The Netherlands

Introduction: Heterozygosity for the Cys282Tyr mutation of the haemochromatosis (HFE) gene and increased serum ferritin levels have both been associated with cardiovascular events. The effect of HFE gene polymorphism on atherosclerosis, however, is still unknown. Therefore, the objective of the present study was to examine the relationship between the Cys282Tyr mutation of the HFE gene, body iron status, and atherosclerosis as indicated by carotid intima-media thickness (IMT).

Methods: We assessed conventional risk factors, serum iron parameters (including non-transferrin bound iron, NTBI), and the Cys282Tyr mutation of the HFE gene in 764 healthy subjects, 50–70 years, recruited from a local blood bank and city registers. Carotid IMT was assessed by B-mode ultrasonography.

Results: Ferritin (25th; 75th percentile) was 87 (49; 138) µg/L in carriers of the Cys282Tyr mutation compared to 65 (34; 125) µg/L in non-carriers (p = 0.06). Similarly, NTBI (SD) was higher in Cys282Tyr carriers than in non-carriers (2.81 (0.95) v 2.38 (0.88) µmol/L, p<0.001). Also, serum iron, total iron binding capacity, and transferrin saturation was significantly higher in Cys282Tyr carriers compared with non-carriers (all p<0.001). Carotid IMT, however, did not significantly differ between subjects with and without the Cys282Tyr mutation (0.82 (0.11) v 0.82 (0.12) mm, p = 0.58). Conventional risk factors, such as age, body mass index, C reactive protein, systolic blood pressure, smoking, and the percentage of current blood donors (54%) did not differ between the two groups.

Conclusion: The Cys282Tyr mutation of the HFE gene is associated with elevated body iron levels. However, this study provides little support for a role of this mutation in the development of atherosclerosis.

H. Englert¹, G. Schaefer², S. Roll¹, K. Beier², S. Willich¹.¹Charité-University Medical Center, Institute of Social Medicine, Epidemiology and Health Econo, Berlin, Germany; ²Charité-University Medical Center, Institute of Sexology and Sexual Medicine, Berlin, Germany

Objectives: To determine the prevalence of erectile dysfunction (ED) in men 40 to 79 years of age in the metropolitan population of Berlin and to investigate the association of the prevalence with different definitions of ED.

Methods: An epidemiological cross sectional study was conducted from May to November 2002. In a population based approach, 6000 men between 40 and 79 years of age were retrieved by the vital statistic office Berlin as a representative sample and received a questionnaire by mail. ED was assessed by five different definitions, of which three are based on the established EF domain criteria: (1) EF "all" (including all men); (2) EF "active" (including only sexually active men); 3) EF "confidence" (including only sexually active men plus inactive men with a low confidence of achieving and maintaining an erection); 4) based on DSM-IV criteria; and 5) based on patients’ self assessment.

Results: A total of 1915 questionnaires were eligible for analysis (response rate 32%). The five definitions yielded the following ED prevalence rates (age-adjusted totals): 1) 48%; 2) 31%; 3) 44%; 4) 18%; and 5) 24%.

Conclusions: ED prevalence rates in a metropolitan population were between 18% and 48% depending upon the definition used. The results indicate the need for standardised approaches in future epidemiological studies on ED.

N. Erazo¹, J. Baumert², R. Engel³, K. Ladwig².¹University Hospital of the Technical University, Psychosomatic Department, Munich, Germany; ²GSF National Research Centre for Environment and Health, Institute of Epidemiology, Munich/Neuherberg, Germany; ³University Hospital of the Ludwig-Maximilians-University, Psychiatric Department, Munich, Germany

Introduction: Railway suicides are considered as a particularly "hard" suicide method, which does not allow any ambivalent behaviour once the suicidal act is initiated. Epidemiological reports, however, consistently found a proportion of survivors on the railway track of approximately 10%. Elucidating circumstances surrounding this 10% survival phenomenon may add further information to improve suicide preventive strategies. Our aim was, therefore, to investigate the factors associated with survival of railway suicides comparing fatal and non-fatal outcome of suicide attempts in an ongoing suicide prevention project of the German Railways AG.

Methods: Cases were derived from the central registry of all passenger accidents on the German railway net satisfying the operational definition of an act of suicidal behaviour during the 6 year observation period of 1997 to 2002. The database included information on outcome, sex, age, date, clock time, and local circumstances of the incidence and contained 5731 fatal and non-fatal suicidal acts during the observation period. Among them, sex was documented for 4003 suicide victims which where considered as study population. Statistical associations between categorised variables and diurnal variations were assessed with ²statistics. To determine which variables were associated with completed railway suicides as opposed to non-completed suicides, logistic regression analysis was used.

Results: In 90.5% (3622 cases) of all cases, outcome was fatal with "death within 30 days" leading to a fatal to non-fatal ratio of 9.5:1 (p<0.0001). In the non-fatal subgroup, 42.3% of the victims were female whereas the female percentage in the fatal subgroup was only 25.4% (p<0.0001). Age did not differ significantly between both subgroups. A bimodal distribution pattern of non-fatal suicidal behaviour with peaks between >9.00 h and 12.00 h and between >15.00 h and 18.00 h (² test for equal proportions; df = 7; p<0.0001) was found whereas fatal suicides peaked between >18.00 h and 21.00 h (² test for equal proportions; df = 7; p<0.0001). Logistic regression analyses identified the following factors for fatal outcome: open track (v station area: adjusted OR 2.96; 95% CI 2.30 to 3.80), main railway line (v local line: adjusted OR 2.26; 95% CI 1.32 to 3.86), and male sex (v female sex: adjusted OR 2.06; 95% CI 1.60 to 2.65) as risk factors. Temporal aspects also reached statistical significance: night (v day: adjusted OR 1.66; 95% CI 1.21 to 2.26) and winter (v summer: adjusted OR 1.35; 95% CI 1.05 to 1.73). The factor age showed no significant impact.

Conclusion: It is widely acknowledged that suicides on the railway track are inappropriate to serve as "gesture" or "cry for help" because no control over the effects of the attempt is available. This is the first study to show for a particular hard suicide method that survivors compared to completers may follow a distinct behavioural pattern with differences in sex, temporal, and local aspects.

N. Erazo¹, J. Baumert², K. Ladwig².¹University Hospital of the Technical University, Psychosomatic Department, Munich, Germany; ²GSF National Research Centre for Environment and Health, Institute of Epidemiology, Munich/Neuherberg, Germany

Introduction: Seasonal and diurnal variations of suicidal behaviour are among the most widely established findings in suicide research. Apart from age and degree of violence of the suicide means, sex was found to influence temporal variations. Evidence on time rhythms of railway suicides as a particularly "hard" method of suicide, however, still is inconsistent. Our aim was, therefore, to examine sex specific time patterns of suicidal behaviour on tracks of the entire German railway net as basis of suicide preventive strategies.

Methods: Cases were derived from the national central registry of all passenger accidents on the German railway net between 1997 and 2002 satisfying the operational definition of suicidal behaviour and including information on sex, age, date, clock time, and outcome of the incidence. Statistical associations between categorised variables were assessed by the ² test for association. Monthly and diurnal variations were carried out with the ² test for equal proportions. Odds ratios were estimated using regression analysis models. Amplitude of the monthly variations in suicides was given as the peak’s percentage above the mean.

Results: During the 6 year observation period 5731 suicidal acts were registered. Among them, sex was documented for 4003 suicide victims. Male to female ratio was 2.70:1 (p<0.001). The female subgroup was significantly older than the male subgroup (p<0.001). The monthly distribution revealed a bimodal pattern, particularly in men younger than 65 years, with a seasonal excess risk in April and September. The significant circannual pattern, however, attenuated in the second half of the observation period. Monday and Tuesday proved to be high risk days for both sexes. For males, the highest incidence was between >18:00 and 21:00 h; in contrast, female suicides occurred significantly more often during daytime and peaked between >9.00 and 12.00 h. A seasonal variation of the diurnal pattern was found with a bimodal distribution in the winter half year (peaks: >6.00–9.00 h and >18.00–21.00 h) and a unimodal distribution in the summer half year (peak: >21.00–24.00 h). Stratification by sex revealed that, at variance with males, females exhibited solely a morning peak (>9.00–12.00 h) in the summer half year. For both sexes, a substantially wider time window of suicide events with a 3 h shift towards earlier morning hours and correspondingly a 3 h shift towards later evening hours was observed in the summer months.

Conclusion: The analysis revealed marked weekly and diurnal peaks of railway suicide intensity with a seasonal modulation of circadian rhythms. Differences between men and women in time patterns indicate sex specific processes underlying their suicidal behaviour. The findings may increase alertness of railway and security personel for particular vulnerable time windows of excess risk for railway suicides.

S. Farchi, F. Chini, P. G. Rossi, L. Camilloni, P. Borgia, G. Guasticchi.Agency for Public Health, Lazio Region, Rome, Italy

Introduction: In the Lazio region, the 3.4% and the 2.4% of children, in 2000, were admitted to the emergency departments (ED) for accidents occurred at home and on the road, respectively.

Objective: The aim of this study is to compare individual and environmental factors associated with hospitalisation consequent to ED admission, among injured children in road traffic and home accidents.

Data source: Pilot road traffic and home surveillance systems, based on the integration between the emergency information system, the hospital information system, and the mortality registry of Lazio region.

Population: All the ED admissions of children, aged 0–18 years.

Methods: Logistic regression has been performed to compare the probability of hospitalisation for different individual and environmental factors. Separate models for four age groups (0 years, 1–5, 6–13, and 14–18) and for place of injury have been performed, to take into account the interaction between these two factors (p<0.0001). Models have been adjusted for the triage code (very urgent, urgent, not urgent, and not appropriate) as a proxy of severity.

Results: The hospitalisation rates are 194 and 219 per 100 000 for home and road, respectively. The highest hospitalisation rate has been found for the teenagers on the road (528/100 000) and for the infants at home (713/100 000). Higher probability of hospitalisation has been found for children living outside of the city of Rome in comparison with the urban children, especially for infants for road traffic accidents (OR 4.27; 95% CI 1.60 to 11.4), and for home accidents (OR 1.78; 95% CI 1.38 to 2.29). Hospitalisation is higher during the warm seasons in comparison to winter, and is less frequent during the day (2pm–7pm v 8pm–6am) (OR 0.80; 95% CI 0.69 to 0.92) for road traffic accidents. Probability of hospitalisation is higher during the weekend (OR 1.16; 95% CI 1.02 to 1.31) and among males (OR 1.34; 95% CI 1.17 to 1.52) only for teenagers involved in road traffic accidents.

Conclusions: This analysis highlights differences between road and home accident hospitalisation risk. The ages at risk are different; road accidents involve teenagers and seem to be more associated to behavioural risk factors, while domestic injuries principally involve infants. This study identifies children living in non-urban area as high risk children, either for road or for home accident hospitalisation.

S. Farchi, P. G. Rossi, L. Camilloni, F. Chini, P. Borgia, G. Guasticchi.Agency for Public Health, Lazio Region, Rome, Italy

Introduction: To date, no commonly accepted definition of home accident (HA) mortality has been proposed.

Objective: Aim of this study is to evaluate the capability to capture the major number of death cases of different criteria.

Methods: HA data come from a pilot surveillance of all the HA occurring in Lazio Region, Italy 2000. The surveillance is integration between the emergency department information system (EIS), the hospital information system (HIS), and the mortality registry (MR). The gold standard definition of HA mortality in this study was obtained through an analysis of the hospitalisation story and of death certificates, using the following criteria: all the deaths, reported in the MR, occurred the same day of the discharge from the emergency department (ED) or hospital; all the deaths for which a trauma/poisoning cause of death (ICD: 800–998.1 and all the E codes, excluding transport, 800–848; and violence, 950–999) present in the MR, for which the cause of death was the same of the diagnosis reported in the EIS-HIS database, and the time span was admissible. Three definitions of HA mortality have been tested: (1) inhospital mortality (IHM); (2) mortality within 30 days from the ED admission or hospital discharge (M30d); and (3) mortality for a trauma/poisoning death cause (ICD: 800–998.1, and all the E codes, excluding transport, 800–848; and violence, 950–999) (TPM). A sensitivity analysis has been proposed to evaluate the mortality outcome, comparing the death cases in our database with the criteria proposed.

Results: The database reports 145 101 episodes of HA (2.7% of the entire population) and an hospitalisation rate equal to (8% of the episodes). The total number of deaths among the subjects present in this surveillance and found in the MR or in the EIS-HIS database was 1321, while the number of HA deaths was 387 and a mortality rate equal to (7.3 per 100 000). Particularly high mortality rate has been found among older people (40.2 age >65 years and 83.7 per 100 000, age >75 years). The three different definitions of mortality were able to capture the following number of cases: IHM, 399 deaths, 75.5% sensitivity and 88.5% specificity; M30d, 563 deaths, 92.8% sensitivity and 78.2% specificity; and TPM, 244 deaths, 58.4% sensitivity and 98.1% specificity.

Discussion: The three definitions of mortality gave different results. M30d is the definition that better individuates the true positives but it classifies as positives also a great proportion of true negatives, while TPM is worse than M30d in identifying the true positive, but it classifies very well the true negatives. Worst results are given by the IHM.

A. Federici¹, F. Bartolozzi², S. Farchi¹, P. Borgia¹, G. Guasticchi¹.¹Agency for Public Health, Rome; ²Campus Biomedico University Hospital, Rome

Introduction: Latium region conducted a set of pilot studies to obtain the information needed to plan a CRCS programme with an evidence-based organisation.

Objectives: To assess the effect of the provider GPs v hospital on the compliance of the population 50–74 year old in returning the FOBT. To analyse the characteristics of the GP, which are associated with high compliance among his beneficiaries.

Methods: We mailed a questionnaire about screening attitudes to 1192 GPs. We asked the GPs to participate in a randomised trial with outcome the compliance to CRCS. We sampled 10 GPs for each 13 districts where the hospital were placed: about 1/10 of the GPs’ beneficiaries 50–75 year old (n = 3657) were invited to pick up and give back the FOBT at the GP surgery and 1/10 (3675) were invited to the gastroenterology centre of the nearest hospital. We tested the association between compliance and GPs’ characteristics performing a logistic regression in which the GP was considered the primary sampling unit. The GPs participating in the trial have been compared with GPs not participating for the characteristics that influenced the level of compliance.

Results: 58.5% of the GPs filled the questionnaire and 22.7% agreed to participate in the trial. The compliance obtained using the GP as providers of the FOBT was 3.4 times higher than the compliance obtained using the hospital as provider (95% CI 3.13 to 3.70), independently from the type of test and the geographical area. There is a strong correlation between the compliance obtained by the GPs in the half practice population assigned to the GP’s surgery arm and the compliance obtained in the half assigned to the hospital arm. We observed a high variability among GPs: GPs with more than 25 patients visited per day and those incorrectly recommending FOBT for CRCS obtained a lower compliance (OR 0.74, 95% CI 0.57 to 0.95 and OR 0.76, 95% CI 0.59 to 0.97, respectively). The GPs who participated in the trial are different from the whole population of the GPs: they are younger, more frequently male, and more likely correctly recommending the FOBT for CRCS (68% v 32%).

Conclusions: The GPs showed scarce interest in the study, probably reflecting a scarce interest to the CRCS. The GPs as providers obtained 3.4 times higher compliance, but this brilliant result was probably not extensible to the whole population of GPs. The involvement of GPs as providers of the FOBT in CRCS can be very effective to enhance the compliance, but the effectiveness is dependent on the willingness of the GP involved.

V. Fonseca¹, R. Sichieri², A. Moura³, D. Hoffman⁴.¹Fundação Oswaldo Cruz, Instituto Fernandes Figueira, Rio de Janeiro, Brasil; ²Universidade do Estado do Rio de Janeiro, Instituto de Medicina Social, Rio de Janeiro, Brasil; ³Universidade do Estado do Rio de Janeiro, Instituto de Biologia Roberto Alcantara Gomes, Rio de Janeiro, Brasil; ⁴Rutgers, the State University of New Jersey, Department of Nutritional Sciences, New Jersey, USA

Introduction: Few studies have evaluated the long term growth of preterm infants associated with iron deficiency.

Objective: This study aimed to address the role of anaemia at birth on the growth pattern of preterm infants.

Methods: We evaluated the association of cord blood haemoglobin level with growth of 96 premature babies, followed during 6 months of corrected age. For longitudinal data analysis, random regression models were used to determine the association of growth, expressed by length, with haemoglobin concentration at birth. The influence of haemoglobin on the anthropometric indexes of growth was analysed based on the temporal changes of anthropometric indexes using the PROC MIXED procedure, SAS, version 8.0 software. The term of interest was the haemoglobin by time interaction, which estimates the change rate of the outcomes. We assumed an exchangeable covariance structure of the models (compound symmetry), and models with random intercept and random slope were tested by the likelihood ratio test.

Results: The rate of growth of those children above median (12.6 cm) of cord haemoglobin was greater than the growth rate among infants below median (10.6 cm). The growth rate was even greater for those infants with values of cord insulin above median (13.5 cm and 11.4 cm, respectively). Interactions with sex were not statistically significant (p = 0.35), indicating that boys and girls showed the same growth pattern. Modelling the growth with a random intercept showed that birth weight was positively associated with length (p = 0.001) and the associations with other confounding variables were not statistically significant in the model. Models with haematocrit instead of haemoglobin showed about the same results.

Conclusion: Even after controlling for gestational age and birth weight, our data indicate that iron status at birth has an important role in the growth pattern of preterm infants, which is independent of the insulin level, but modified by the level of cord insulin.

M. Formiga, P. Ramos.UFRN, Estatística, Natal-RN-Brasil

Objectives: The purpose of this work is to explore the relations between infant mortality, precisely for children below 5 years of age, and socioeconomic factors in the states of the northeast region of Brazil, taking the respective counties as unit of analysis.

Methods: To achieve this goal, a multiple regression model was adjusted, in order to identify the socioeconomic factors (independent variables) that explain the mortality over the referred infant population (response variable) in the Northeast region of Brazil, based on data available on the Human Development Atlas of Brazil (PNUD/IPEA/FJP, 2003), updated by the data from the IBGE 2000 Census. The Average per capita Household Income (RENDPC), the analphabetic rate of the adult population 15 or more years of age (TXALFA), and the proportion of the population living in households with sanitary facilities (PPSAN), were selected as independent or explanatory variables and the mortality rate for children below 5 years of age was taken as dependent or response variable.

Results: These indicators were collected for the 1787 counties of the northeast region, identifying, when convenient, their correspondent states. The adjusted model for the region total reveals, as statistically significant explanatory variables (p <0.05), the variables PPSAN, responsible for 44% of total variation and TXALFA, responsible, after the introduction of PPSAN in the model, for more 3.3% of total variation, completing the 47.4% of explained variation for the model. The variable RENDPC does not give a statistically significant contribution to the model. It is interesting to observe that, in the absence of the variable PPSAN, the resulting adjusted model reveals the variables TXALFA and RENDPC as statistically significant to explain the infant mortality, where TXALFA is responsible for 42.1%, and RENDPC is responsible for the remaining of the 42.8% of the total variation explained by this model, that is, only 0.7%, showing the low explaining power of the income over the infant mortality in presence of education. The adjusted model for each state of the considered region reveals that, only for the states of Rio Grande do Norte and Bahia, the variable PPSAN presented a higher explaining power for the response variable M52000 than the variable TXALFA. For the state of Rio Grande do Norte, the education factor shows a higher power than the income factor, in the absence of the sanitary facilities, and just the inverse for the state of Bahia. For the remaining states, the variable TXALFA appears to be the one with best explaining power for the response variable M52000.

Conclusions: The results giving evidence of the importance of the education factor over the infant survival, pointing out, clearly, that the social area needs to be centred as a focus of investments on this suffered Brazilian region.

D. Fraser¹, A. Levy², R. Dagan³, L. Naggan².¹Ben-Gurion University of the Negev, S. Daniel Abraham Intl. Centre for Health and Nutrition, Beer-Sheva, Israel; ²Ben-Gurion University of the Negev, Epidemiology Department, Beer-Sheva, Israel; ³Soroka University Medical Centre, Paediatric Infectious Disease Unit, Beer-Sheva, Israel

Introduction: Anaemia in the form of iron deficiency anaemia is the most prevalent form of micronutrient malnutrition in the world; however, the causal relationship between anaemia and infection remains unclear due to the complex interactions between morbidity, anaemia, and family and environmental factors.

Objectives: To examine the causal association between anaemia and infection among Bedouin infants.

Methods: Families of 239 infants from a Bedouin Arab township were recruited at birth during the periods of 1989–92 and 1994–97. Infants were followed from birth to age 18 months. The numbers of diarrhoea and respiratory disease episodes as well as total number of days of diarrhoea were ascertained weekly. Haemoglobin levels were obtained at age 6 months. Additional data on feeding practices, environmental, household, and demographic characteristics were obtained throughout the 18 months of follow up.

Results: Diarrhoea before 6 months of age was a risk factor for diarrhoea after that age. When examining the association between anaemia (Hb<11 g/dL) at 6 months and infectious disease morbidity after that age we found that anaemia was an independent risk factor for diarrhoea and respiratory illness which occurred from 7–18 months of age, even after controlling for morbidity up to age 6 months. Furthermore, the independent associations between anaemia and infectious diseases remained significant after controlling, in addition to prior morbidity, for other environmental and socioeconomic factors. In the multivariable logistic regression models, which included the environmental and socioeconomic factors as well as diarrhoea prior to age 6 months, anaemia increased the risk for infection after that age as follows: for diarrhoea the increased risk was 2.9-fold (95% CI 1.6 to 5.3; p<0.01) and for respiratory disease it was twofold (1.1 to 3.6; p = 0.03).

Conclusion: Our findings suggest that anaemia at age 6 months may be an independent risk factor for infectious disease morbidity after that age. This highlights the importance and potential impact of reducing anaemia in infants as a preventive measure to lower disease burden from infectious disease in this and other vulnerable populations.

M. Freitas.S.R.S. Braga, Public Health Department of Braga County, Braga, Portugal

In Portugal the epidemic of AIDS continues to grow at a worrying rate. According to UNAIDS there are 50 000 cases of persons infected with HIV. This study identified the level of knowledge, source of information, opinions, and behaviours referred, of the students from the University of Minho faced with AIDS, and identified differences between the students of health courses (health) and those from different areas of education (other). This study is a descriptive study. A questionnaire was sent to a college population sample of 312 from the 8829 students at the University of Minho. Based on the sample, it is estimated that 30% use condoms (p = 0.05, CI 95%). A random stratified selection was used. There was a 78% response rate and these were divided into the two groups; health and other. The more important information comes from the press and television and the individuals who considered they had "good" knowledge thought that information about AIDS should be provided by health services. There is a higher level of knowledge about AIDS in individuals from health courses compared with those from other courses (p = 0.006). The students have decreased "perception of vulnerability" to AIDS, and this is lower among the students on other courses (p = 0.000). The majority of the individuals answered that they did not feel embarrassed buying condoms; however, this was 80% for males but only 53% for females. 90.7% answered that they would be happy to have an AIDS test, however, barely 45% of them know where they could have the test. 63.4% considered their knowledge about AIDS has an impact on their behaviour and sexual practices; 15.9% assume they have risky sexual behaviours; 17.8% had more than one sexual companion in the past year (of these, 18.4% referred to have occasional sexual meetings). 25.9% referred to have sexual relations under the effect of alcohol; of those, 76.6% said this was without condom use. Apparently it verified that quality on the information about AIDS is better than that transmitted to previous college students.

P. Freitas¹, M. L. Drachler², J. C. C. Leite³, P. Grassi⁴.¹Universidade Federal de Santa Catarina, Mestrado em Ciências Médicas, Florianopolis/Brazil; ²Universidade do Vale do Rio dos Sinos, Mestrado em Saúde Colectiva, São Leopoldo/Brazil; ³Universidade de Caxias do Suluniversidade de Caxias do Sul, Departamento de Psicologia, Caxias do Sul/Brazil; ⁴Governo do Estado do Rio Grande do Sul, Secretaria Estadual da Saúde, Porto Alegre/Brazil

Introduction: Brazil has one of the highest rates of caesarean sections in the world and, in spite of definite regional differences in their magnitude; the increase has been a nationwide phenomenon. The risks and safety of caesarean birth differ from place to place in the world and it has been postulated that as the relative safety of the procedure is an important factor contributing to the rise of caesarean birth, the fact that this has only been achieved in some parts of the world should restrict the indications to perform the operation in the best interests of pregnant women.

Objective: This study was conducted to investigate the effect of social inequalities in caesarean section rates in primiparous women in southern Brazil.

Methods: Annual rates and odds ratios (OR) of caesarean sections among primiparae with single pregnancy and delivering in maternities in the southern Brazilian state of Rio Grande do Sul were estimated for social factors (maternal age and education, skin colour/ethnicity of the newborn, and macro-regions of the health services administration) for post-term delivery, applying unconditional logistic regression on data from the National System of Information on Live Births (SINASC) in 1996, 1998 and 2000.

Results: There were about 180 000 live births per year; 99% in maternities, a third among primiparae. The caesarean section rate was 45%, >37% in all macro-regions. The caesarean section rates increased in most regions in women >30 years, >6 antenatal consultations, brown skin colour, and native Brazilian Indians. Unadjusted and adjusted OR were positively associated with maternal education and age, and white skin colour of the newborn and post-term delivery. These effects were partly mediated by the greater number of antenatal consultations among these women, which have also shown a positive effect on the rates of caesarean sections.

Conclusions: Socioeconomic conditions and differences in use and access to medical technology in Brazil are striking when comparing different regions inside the country. In this context, while distance to medical facilities and lack of means of transport can prevent many women in labour (living in rural settings or in the periphery of metropolitan areas) from access to caesarean section as a life saving procedure, at the same time in urban and central areas of the main cities overuse of caesarean section is the rule. High rates of caesarean sections in southern Brazil are a public health problem beyond the scope of clinical matters. Regional differences in rates of caesarean section are likely to be the consequence of a number of factors operating together, including social, economic, and cultural factors that determine misuse of medical technology during labour and delivery.

P. Freitas¹, T. Marshall².¹Universidade Federal de Santa Catarina, Mestrado em Ciências Médicas, Brazil; ²London School of Hygiene and Tropical Medicine, Maternal and Child Epidemiology Unit, UK

Introduction: Caesarean section is an essential component of good obstetric practice. Used when specific medical indications occur, it can save the lives of mother and baby. In Brazil, rates of caesarean sections are well above those that can be attributed to absolute medical indications. The totality of non-medical issues surrounding birth in Brazil, including personal and institutional factors from women’s and obstetricians’ contexts have been reported to as the "culture of caesarean sections in Brazil". These factors can range from psychological character of the women, their interaction with the healthcare system, and the organisation of obstetric practice. A cross sectional study of 330 recently delivering women, followed by interviews with the obstetrician in charge of the delivery and information from admission and delivery books, was conducted in the main public maternity section in Florianópolis, South Brazil, to investigate how medical staff, women, and the organisation of obstetric practice interact to influence decisions towards caesarean sections.

Methods: Information from interviews with women included factors influencing labour and outcome of delivery from the perspective of these women, their experiences during pregnancy, contact with antenatal services, and other women’s experiences. Also, their experiences during labour and delivery and their interactions with obstetric staff in the maternity ward were addressed as part of the postpartum survey. From the obstetrician perspective, data concerning medical and non-medical factors influencing development of labour was collected. This included decisions concerning type of delivery among these women, timing and use of interventions, labour development, and decisions concerning type of delivery. Methods of multivariate analysis where used to create the hierarchical model of factors influencing delivery by caesarean. The main principle underlying hierarchical regression modelling is that the selection of the exposure variables is not based purely on statistical associations, but also on a conceptual framework describing the logical or theoretical relationships between the potential determinant factors.

Results: The risk factors found to present the largest odds ratios were previous caesarean section (OR = 10.7; p<0.01), previous experience favouring caesarean (OR = 3.6; p<0.01), antenatal care with the on-duty doctor (OR = 4.0; p<0.01), being admitted not in labour (OR = 37.6; p<0.01), and adverse events in labour (OR = 4.0; p<0.01).

Conclusions: Although obstetricians are clearly key players in deciding for caesarean sections, where medical indications are not absolute, the situation is more complex and factors from women’s and obstetricians’ contexts including personal, institutional, and non-medical issues can operate to influence decisions towards type of delivery. The understanding of the high rates of caesarean section in Brazil, from its multifactorial perspective, can be seen as an essential requirement, before potential interventions to reduce the excessive rates can be devised, developed, and evaluated.

A. Gagnon¹, G. Dougherty², R. Platt³, O. Wahoush⁴, D. Stewart⁵, A. George⁶, E. Stanger⁷, J. Oxman-Martinez⁸, J. Saucier⁹.¹McGill University, School of Nursing/Department of Obstetrics and Gynaecology, Montreal, Canada; ²The Montreal Children’s Hospital, Intensive Ambulatory Care Service, Montreal, Canada; ³McGill University Health Centre, Research Institute, Montreal, Canada; ⁴McMaster University, School of Nursing, Hamilton, Canada; ⁵University Health Network, Women’s Health Program, Toronto, Canada; ⁶University of British Columbia, Centre for Community Child Health Research, Vancouver, Canada; ⁷Provincial Health Services Authority, Office for Cross Cultural Care & Diversity, Vancouver, Canada; ⁸McGill University, Centre for Applied Family Studies, Montreal, Canada; ⁹Sainte-Justine Hospital, Department of Psychiatry, Montreal, Canada

Introduction: A steady flow of refugees and asylum seekers arrive in industrialised countries every year. Compared with immigrants who leave their country by choice, these newcomers are at an increased risk of several determinants of ill health including poor nutritional status, reduced social support, and histories of abuse. Asylum seekers may also experience anxiety and stress directly related to their precarious immigration status. Childbearing further increases the vulnerability of female refugees and asylum-seekers, by placing these women in a position of needing additional health services and social support during pregnancy, childbirth, and postpartum. No studies in Canada (and very few in other industrialised countries) have reported the prevalence of harmful health events or outcomes in childbearing refugee and asylum seeking women. To address this gap in knowledge, a Canadian wide study is being conducted to determine the health and social care needs of these women and their infants and whether or not these needs are being met by the current health and social service system.

In the context of this study, a number of scientific challenges on conducting studies in migration were identified. The goal of this paper is to describe these challenges and to discuss solutions in overcoming the complexities that present when studying this vulnerable population.

Methods: 150 women and infant pairs in each of the following groups: refugee, asylum seeker, non-refugee immigrant, and Canadian born women (n = 600), are being recruited across 12 hospital centres in three major Canadian cities (Montreal, Toronto, and Vancouver). Registered nurses make home visits at 1 week post-birth to assess the families (through the use of a standardised assessment protocol and translated questionnaires) for health and psychosocial concerns and to determine care received/planned at that time.

Results: Three significant challenges emerged in conducting this study. Firstly, sample selection (ensuring the study population is representative of the "target" poulation): (a) defining eligibility criteria and "exposure" categories (that is the independent variable, migration history); (b) determining recruitment sites; (c) applying eligibility criteria (while ensuring ethical protection of subjects and given budget limitations); and (d) confirming complete ascertainment of study sample. Secondly, unbiased measurement in several languages: (a) eliciting migration history and other information; and (b) defining outcome criteria and operationalising assessment. And thirdly, unbiased analyses: (a) loss of subjects over the course of the study; (b) recruitment rate differences by migration history and city.

Conclusion: To overcome these challenges the following is required: a high level of commitment to answering the research questions; patience with your own learning curve; and teaching reviewers and others about this field; "thinking" time, money, and strong support staff.

Recognising the challenges and options for resolving them leads to: stronger study designs; greater efficiency (obtaining a valid answer at reduced costs); and ultimately optimising the health of newcomers.

A. Galas, A. Penar, W. Jedrychowski.Jagiellonian University, Medical College, Chair of Epidemiology and Preventive Medicine, Krakow, Poland

Introduction: Diet is assessed as one of the most important preventable causes of cancers. There is no clear relationship, but what kind of micronutrients play an important protective role in cancer development? The association between nutrient intake and the risk of stomach cancer was assessed in a hospital based case-control study in Poland.

Methods: Material covers 190 histologically confirmed cancer cases (mean: 57.1 years) and 548 controls (57.4 years). Information about frequency and quantity of consumption of 174 alimentary items was gathered by questionnaire (1 year dietary recall, 5 years prior to diagnosis). Comparisons of daily intake have been performed using Mann-Whitney test and logistic regression model.

Results: In the cancer group a lower average daily intake (in comparison with controls) of carotene (1.21 mg v 1.32 mg; p = 0.004), vitamin E (4.24 mg v 4.44 mg; p = 0.005), vitamin C (39.87 mg v 42.97 mg; p = 0.032), phosphorus (0.75 mg v 0.78 mg; p = 0.024), iron (6.31 mg v 6.61 mg; p = 0.000), fibre (2.0 g v 2.1 g; p = 0.008), and linoleic acid (4.2 mg v 4.4 mg; p = 0.004) was observed, and higher intake of alcohol (2.1 g v 1.5 g; p = 0.006) and cholesterol (198.3 mg v 191.4 mg; p = 0.028). The subjects of the study were subsequently categorised by daily consumption of nutrient items by quartiles. Low quartile of daily intake in the control group was defined as the low level of consumption. The lower stomach cancer risk was found in cases with low daily intake of energy (OR = 0.63; 95% CI 0.40 to 0.98). After adjustment for energy, age, sex, martial status, and place of birth the higher risk was found in the group of lower intake of carotene (OR = 1.56; 95% CI 1.07 to 2.27), vitamin C (OR = 1.46; 95% CI 1.01 to 2.12), fibre (OR = 1.51; 95% CI 1.03 to 2.20), and iron (OR = 1.99; 95% CI 1.39 to 2.85). Lower intake of alcohol decreases the risk of stomach cancer in our study (OR = 0.62; 95% CI 0.40 to 0.98).

Conclusions: The results of the present study indicated that vitamins (especially carotene and vitamin C) and fibre might have protective effect against stomach cancer, moreover, low consumption of alcohol may decrease the risk of cancer.

A. Gbska-Kuczerowska, M. Wysocki, P. Goryski, J. Car.National Institute of Hygiene, Warsaw, Poland

Introduction: Approximately 50% of world’s injury related deaths occur in young people aged 15–44. Despite that, the injury related mortality rates, in all regions all over the world, increase respective to age.

Objective: The objective is to identify the main reasons for hospitalisation caused by injuries and poisonings in the group of people aged 65+ years living in Poland.

Material and Methods: The data on all hospitalised cases (100%) in 2001, in four regions of Poland were analysed. The hospitalisation rates by causes were evaluated in the group of 1 470 664 cases. 29% of that number constituted elderly patients (65+). The problem of injuries and poisonings that caused hospitalisation of persons aged 65 is presented.

Results: In four selected regions of Poland, 132 646 patients were hospitalised due to injury or poisoning in 2001. The cases of injuries and poisonings constituted 9% of all hospitalisations. In all hospitalised cases of injuries and poisonings, there was 16% of elderly people and 20% people aged 20–34. The higher hospitalisation rate caused by injuries and poisonings was observed in the group of elderly patients, living in cities 150/100 000 v 119/100 000 in general population and 149/100 000 in the group of people 15–19 years old. Younger men living in cities were admitted to hospitals due to injuries and poisonings more often than men 65 years old (rate: 161/100 000 v 141/100 000). For men (65+) living in rural area, the hospitalisation rate caused by injuries and poisonings was higher than for men living in urban areas (151/100 000 v 134/100 000). The hospitalisation rate for elderly women was higher than in general population (156/100 000 v 150/100 000) and especially high for women living in rural areas (192/100 000). The main causes of hospitalisation due to injuries and poisonings in the group of elderly patients were following fracture of the femur (43/100 000; five times higher rate than in general population) and complications of orthopaedic procedures (1.8/100 000). For elderly patients fractures of the thigh (43/100 000), lower leg and knee (12/100 000), forearm (9/100 000), and lumbar vertebrae with hip (7/100 000) were the most typical.

Summary: Elderly women living in rural areas are exposed to a high risk of injuries and poisonings. Injuries are predominant in that large group of injuries and poisonings. The most common injury suffered by elderly patients was fractures of lower extremities.

J. Geleijnse, E. Schouten, D. Kromhout.Wageningen University, Division of Human Nutrition, The Netherlands

Introduction: Whether dietary omega-3 polyunsaturated fatty acids are causally related to cardiovascular disease is a major, unresolved question in preventive cardiology. Essential omega-3 fatty acids are eicosapentaenoic acid (EPA; C20: 5,n-3) and docosahexaenoic acid (DHA; C22: 6,n-3), and their parent compound alpha-linolenic acid (ALA; C18:3,n-3). Evidence is accumulating that EPA-DHA reduce the risk of cardiac arrhythmia and sudden cardiac death in humans.

Methods: The Alpha Omega Trial is a multicentre placebo controlled, double blind intervention study in post myocardial infarction patients to examine the effect of low doses of omega-3 fatty acids on coronary heart disease mortality. The trial includes 4000 men and women aged 60–80 years who had a clinical diagnosis of myocardial infarction in the past 10 years. The trial has a 2x2 factorial design with random assignment of equal numbers of patients to one of four interventions: 1) 400 mg/day of EPA-DHA; 2) 2 g/day of ALA; 3) both EPA-DHA and ALA; or 4) placebo. Omega-3 fatty acids are supplied via enriched margarine for a period of 3 years. The trial started in May 2002. Funding has been obtained from the Netherlands Heart Foundation (topdown programme 2000T401).

Results: A pilot study was conducted in the first 400 participants (25% women, mean age of 69 (SD 6) years) who were re-examined after 3 months of intervention. Fatty acid composition of plasma cholesteryl esters indicated good compliance, with 3 month increases in ALA in two groups (by +35% and +29%; p<0.001) and 3 month increases in EPA in two groups (by +40% and +33%; p = 0.004). Changes in subjective health, blood glucose, blood pressure, heart rate, and body weight were similar in all groups. There was no significant treatment effect on serum total cholesterol (p = 0.50), HDL-cholesterol (p = 0.71), LDL-cholesterol (p = 0.19), or triglycerides (p = 0.38). Seven patients dropped out and 2 died during the first 3 months of intervention.

Conclusion: The Alpha Omega Trial is safe and feasible. Final results are expected in the year 2008.

P. Rossi¹, F. Bartolozzi², S. Farchi¹, P. Borgia¹, G. Guasticchi¹.¹Agency For Public Health, Latium Region, Rome, Italy; ²Campus Biomedico, University Hospital, Rome, Italy

Introduction: Latium region conducted a randomised trial to obtain the information needed to plan a colorectal cancer screening (CRCS) programme with an evidence based organisation.

Aim: To assess the effect of the type of faecal occult blood test (FOBT), Guaiaco v immunochemical, on the compliance to screening of the 50–74 year old population. Other logistical and organisational aspects of the two screening test have also been analysed.

Methods: We sampled 130 GPs who gave the consent to participate to a randomised trial with outcome the compliance to CRCS. We randomised one half of the GPs to the Guaiaco arm and half to the immunochemical arm. We sampled 2/10 of the GPs’ beneficiaries 50–75 year old (n 7332), we randomise this population in two halves, one invited to pick up and give back the FOBT at the GP surgery and the second invited to the gastroenterology centre of the nearest hospital. We considered the percentage of returned tests as the principal outcome, the interval confidence of the relative risk takes into account the cluster randomisation. We also compared the prevalence and the variability among zones of positive and inadequate tests in the two arms.

Results: The immunochemical test had a compliance of 36.5% and the Guaiaco of 30.6%, RR of 1.22 (95% CI 1.02 to 1.46). The difference is completely due to a higher probability of returning the test and not to the picking up: 93.8% and 88.6% of returning the test given the picking up for immunochemical and Guaiaco, respectively (RR 1.06, 95% CI 1.02 to 1.10). The Guaiaco test has a higher prevalence of positives (10.3% v 6.3%, RR 0.603; 95% CI 0.433 to 0.837). The difference is mostly due to the results of three centres very high prevalence: 32%, 27%, and 26%. There is no difference in the mean prevalence between the two providers, GPs and the gastroenterology centre. There is a higher variability in the results obtained with the Guaiaco test compared with the immunochemical (F(1,12) = 16.25; p = 0.0017). The Guaiaco test had a higher proportion of inadequate samples, compared with immunochemical: 2.1% v 1.1%, this difference is not significant, taking into account the variability among GPs and districts (OR 1.94; 95% CI 0.80 to 4.71).

Conclusions: The immunochemical test enhances the compliance to CRCS comparing to Guaiaco, independently from the provider. The prevalence of positive tests with Guaiaco, in our setting show a higher variability. In the implementation of a screening programme it important to consider a period of standardisation of the test reading in order to avoid unexpected work overload to the second level colonscopy services.

L. Granja, M. Severo, C. Lopes.University of Porto Medical School, Hygiene and Epidemiology, Porto, Portugal

Introduction: Based on the evidence that consumption of fruits and vegetables plays an important role in the prevention of chronic disease, the international recommendations include the goal of increasing consumption to five or more servings per day. Methods used to assess intake are often laborious and unattractive, particularly in children and adolescents, to whom the evaluation is especially important for promoting consumption. This study aimed to develop interactive software to evaluate adolescents’ fruit and vegetable consumption and compare it with a traditional food frequency questionnaire.

Methods: The program was developed in Visual Basic 6.0 and the answers were saved in a Microsoft Access database. The program was designed to estimate fruit and vegetables intake and also to supply nutrition information according to the "5 a day" healthy eating message, so that healthy individual choices can be promoted. Among 70 adolescents, aged 14–17 years, consumption of fruits and vegetables was evaluated using firstly a self-administered food frequency questionnaire, comprising questions assessing fruit (fruit, juice fruit) and vegetables (salads and cooked vegetables, vegetables soup), which estimates dietary intake over the previous week. Half an hour after they answered, again using interactive software that estimates dietary intake in the same period of time, we compared the prevalence of adolescents consuming five or more servings per day of fruits and vegetables and the separate prevalence of fruits (3/day) and vegetables (2/day) obtained from both methods using ² test or exact Fisher test. Kappa coefficients were calculated to evaluate the agreement between methods.

Results: The prevalence of five or more fruits and vegetables servings consumption was 14.3% (95% CI 7.1 to 24.7) using the questionnaire and 12.9% (95% CI 6.1 to 23.0) using the software. Also, the prevalence of fruits and vegetables separately was respectively 24.3% (95% CI 14.8 to 36.0) and 8.6% (95% CI 3.2 to 17.7) using the questionnaire; 22.9% (95% CI 13.7 to 34.4) and 11.4% (95% CI 5.1 to 21.3) using the software. For total consumption of fruits and vegetables and for them separately the kappa coefficients were 0.09, 0.33, and 0.21, respectively.

Conclusions: The crude prevalence of consumption of fruits and vegetables was low but similar according to both methods. However, there was poor agreement between rating approaches that need further explanation.

A. Grjibovski¹, L. Bygren¹, A. Yngve¹, M. Sjöström¹, Y. Tedder².¹Karolinska Institute, Unit for Preventive Nutrition, Department of Biosciences, Stockholm, Sweden; ²Northern State Medical University, Institute for Hygiene and Medical Ecology, Archangelsk, Russia

Introduction: The importance of adequate antenatal development on future health is widely recognised. Birth weight (BW) is a strong predictor of infant mortality and morbidity in childhood. Moreover, compromised fetal growth is associated with increased risks of cardiovascular diseases and diabetes in adults. Socioeconomic disparities in BW exist in various settings. The rapid transition to a market economy considerably increased financial and social inequalities in Russia. This study quantifies the effects of sociodemographic and lifestyle factors on BW in a town of Severodvinsk.

Methods: Altogether, 1599 pregnant women (>99% of all pregnant women in the town in 1999) were enrolled in a cohort from 1 January to 31 December 1999. They were followed through delivery. Sociodemographic (maternal age, education, marital status, occupation, parity, weight, timing of prenatal care initiation, infant’s sex, and BW) and lifestyle (smoking and drinking habits for both partners, housing conditions, type of housing, family structure, and psychosocial stress) characteristics were obtained from the medical files and a questionnaire. After excluding women who had abortions (n = 77), twins (n = 11), stillbirths (n = 5), and women lost during follow up (n = 67) after registration, the analysis was based on 1399 women and their live infants. Multiple linear regression was applied to estimate independent effect of the studied factors on the BW. Adjustment for gestational age was made to reflect fetal growth rather than duration of gestation.

Results: When only sociodemographic characteristics were included in the model, a clear gradient of BW in relation to maternal education (ME) was found. Babies of mothers with basic, secondary, and vocational education were 207 g (95% CI 55 to 358), 172 g (95% CI 91 to 253), and 83 g (95% CI 9 to 163) lighter than the babies of mothers with university education after adjustment for other sociodemographic factors.

When all the factors were incorporated in the model (R² = 0.42), maternal smoking, living in shared apartments, living in crowded housing situations, and psychosocial stress were significantly associated with BW loss on 126 g (95% CI 54 to 198), 89 g (25 to 153), 82 g (28 to 136), and 61 g (7 to 116), respectively, compared with the reference groups. Moreover, the associations between BW and ME remained significant in this model.

Conclusions: Observed social variations in BW (by ME) are larger than those previously reported from other European countries. In the light of the fetal origins hypothesis, future variations in adult health in Russia might also be at stake. Maternal smoking, stress, and poor housing conditions are important determinants of fetal growth in transitional Russia. Social variations in pregnancy outcomes should be monitored to ensure that all parts of the society benefit from on going economic and social reforms.

P. Guallar-Castillón¹, P. Peralta², J. Banegas¹, E. López-García³, F. Rodríguez-Artalejo¹.¹Universidad Autónoma de Madrid, Department of Preventive Medicine and Public Health, Spain; ²Ministerio de Sanidad y Consumo, Centro Nacional de Epidemiología. Instituto de Salud Carlos, Spain; ³Harvard, School of Public Health, Boston, Department of Nutrition, EE.UU. USA

Background and Objective: This study examined the relationship between leisure time physical activity (LTPA) and health related quality of life (HRQL) in the older adult population of Spain.

Methods: Household cross sectional survey on 3066 subjects representative of the non-institutionalised Spanish population aged 60 years and over. Data on LTPA were obtained with a structured questionnaire and HRQL was measured with the SF-36 instrument. Analyses were done through linear regression, where the dependent variable was each of the eight scales of the SF-36 and the main independent variable was LPTA. Analyses were adjusted for sociodemographic and social network variables, health habits, health services use, and chronic diseases.

Results: A total of 42.7% subjects had a sedentary activity, 54.2% light LTPA, and 3% moderate/intense LTPA. As compared with sedentary activity, light LTPA was associated with a higher score in all SF-36 scales, except for physical role and emotional role, among men and women. For subjects with light LTPA the increase in score was over 3 points in most SF scales, which is usually considered as a clinically relevant change in HRQL. Results did not vary materially by age, level of education, obesity, or chronic disease. The higher LTPA the better HRQL (p for linear trend <0.05 in most scales of the SF-36 questionnaire).

Conclusions: Light LTPA is associated with better HRQL than sedentary activity. Because this association did not change with age, level of education, obesity, or chronic disease, it is suggested that older adults could improve their HRQL with, at least, light LTPA.

L. Haiek, D. Gauthier, D. Brosseau, L. Rocheleau.Montérégie Regional Health Authority, Public Health Department, Longueuil, Quebec, Canada

Introduction: Lack of uniformity in defining breastfeeding and the recognition of the dose-response phenomenon related to the benefits it confers has emphasised the importance of using internationally recognised definitions and classifications and of reporting separately breastfeeding categories and patterns over time. The study objective was to measure breastfeeding rates and patterns during the first 6 months of life in the Monteregie region using the WHO international definitions.

Methods: Telephone survey with a representative sample of women with 6 month old infants residing in the Monteregie, the second largest socio-sanitary region in the Province of Quebec, Canada.

Results: Among 632 participants, 80% initiated breastfeeding but only 68% exclusively breastfed (breast milk only) their babies during the first 24 h after birth. Regarding breastfeeding patterns, at 1 month, less than two thirds of breastfeeding women were exclusively breastfeeding since birth; by 4 months, only one third of breastfeeding women were exclusively breastfeeding since birth and by 6 months, it was almost non-existent. The proportion of women predominantly breastfeeding (breast milk plus water, water based liquids, or juice) since birth is very small at all periods (between 2 and 6%), whereas the proportion of women giving complementary feeds (breast milk plus non-human milk or solids) since birth increases as the proportion of exclusive breastfeeding diminishes.

Conclusions: The rapid drop in exclusive breastfeeding indicates a need for improvement in order to comply with the current recommendations of feeding babies only breast milk for the first 6 months of life. The most likely types of food to break breast milk exclusivity in the first months of life are primarily non-human milk or solids, resulting in a shift from exclusive breastfeeding to complementary breastfeeding without passing through predominant breastfeeding. This pattern is different from the one described in certain European studies were predominant breastfeeding is a more frequently observed feeding behaviour. The earlier introduction of non-human milk in breastfed infants may represent culturally based differences in infant feeding practices between North America and Europe. Because the introduction of non-human milk in the first months of life is usually intended to replace breast milk, it may also represent a marker of unresolved difficulties and premature weaning, consistent with the lower breastfeeding rates during the first year of life documented in North American studies. A high rate of complementary breastfeeding, combined with a lack of knowledge and a non compliance with most international, national and provincial breastfeeding recommendations, will require important efforts and multiple strategies in order to increase the initiation and the duration of exclusive breastfeeding in the region and assure that Monteregie babies receive optimal nutrition during the first 6 months of life.

L. Haiek, D. Gauthier, D. Brosseau, L. Rocheleau.Montérégie Regional Health Authority, Public Health Department, Longueuil, Quebec, Canada

Introduction: The launching of the WHO/UNICEF Baby Friendly Hospital Initiative (BFHI) in 1991 has encouraged the adoption of evidence-based policy and practices by maternity care units around the world. The Monteregie Breastfeeding Regional Program has chosen the BFHI as its main strategy to increase breastfeeding (BF) initiation, duration and exclusivity in the region. In order to diagnose in the planning stage baseline organisational behaviour regarding the initiative’s proposed policies and practices, a first assessment of the BFHI implementation level in each of the Monteregie hospitals was performed in 2001. However, diffusion of the results to hospital administrators and staff was slowed down by the usual time consuming tasks of data analysis and preparation of personalised reports for each participating institution. With programme ongoing activities being carried out at both the regional and local level, an evaluative monitoring of the BFHI implementation level is planned every 3 years, the first follow up study scheduled for March to April 2004. The objective of this study is therefore to perform a second measure of the BFHI implementation level in each of the Monteregie hospitals using an informatised tool that assures timely feedback of results to hospital administrators and staff.

Methods: The study examines the implementation level of policies and practices, recommended by the BFHI’s Ten steps to successful BF and International code of marketing of breastmilk substitutes, in all nine hospitals offering obstetrical care in the region. Using as data sources for each hospital the perspective of mothers (n = 25), professionals (n = 10), and external observers (n = 2), 3 to 14 indicators were measured for each step and some of the code’s articles. In order to provide hospitals with a summary measure, two synthetic indexes were constructed: the ten steps implementation score ranging between 0 and 10 and the code implementation score ranging between 0 and 5.

Results: A demonstration of the use of the informatised tool will be done. Preliminary results of the 2004 study will be presented and compared with the data collected in 2001.

Conclusions: The use of an informatised tool to assess organisational behaviour allows one not only to adequately measure policies and practices but also to promptly diffuse the results to key players. In the case of maternity care, this timely diffusion will allow study results to act as a catalyst to bring about organisational changes required for the Monteregie hospitals to progress towards achieving the international standards required for Baby Friendly certification.

M. Hakobyan¹, L. Yepiskoposyan², T. Avagyan¹.¹Health Information and Statistical Centre, Health Information Analysis Unit, Yerevan, Armenia; ²Institute of Man, Laboratory of Human Genetics, Yerevan, Armenia

Introduction: The International Development Targets, and the successor, the Millennium Development Goals (MDGs), as poverty reduction strategy explicitly adopt a range of social goals, including reductions of two thirds in infant and under 5 years mortality by 2015. Infant mortality is considered among basic indicators of population health and social economic development of countries and society. During the past decade economic and social situation in Armenia as well as in several countries of the CIS can be characterised as critical. International experience shows that such a situation negatively affects population health, especially infants as the most vulnerable group of population.

Objective: The main goal of this study is to investigate trends in infant mortality during the past decade and define whether Armenia meets MDGs in this concern.

Methods: Calculation of infant mortality rates (IMR) is based on the primary data on live births and infant deaths presented by the National Statistical Service and the Ministry of Health, Republic of Armenia. Statistical analysis of data included calculation of infant mortality rate, its standard deviation and confidence intervals (CI).

Results: Overall, statistically significant (p<0.01) decline (of more than 17%) in infant mortality rate in Armenia has been registered while comparing IMR in 1992 and 2001. The results of more detailed analysis show that noted (p<0.50% (SD 0.01)) decrease in IMR took place only in 1992–1995 (from 18.3 (0.53%)); after that the rate reveals a definite trend to rise—up to 14.03 (0.67%) in 2001 (SD 15.15).

Conclusions: The results obtained witness that the current state of IMR’s reduction in Armenia cannot be considered as consistent with the Millennium Development Goals and respective initiatives must be developed and implemented in the country’s health care sector in full accordance with the WHO recommendations in this area.

G. Hammer¹, B. Kouyaté², H. Becher¹.¹University of Heidelberg, Department of Tropical Hygiene and Public Health, Germany; ²Centre de Recherche en Santé de Nouna, Nouna, Burkina Faso

Objective: We evaluated the accuracy of data from a Demographic and Health Survey (DHS+) with regard to risk factors for childhood mortality, using information from a Demographic Surveillance System (DSS) as the gold standard.

Methods: We performed an analysis of childhood mortality based on birth histories available from the DHS+ survey in Burkina Faso, West Africa 1998–99,¹ which included 4812 households. These surveys are believed to be representative for the whole country. Of all live births from the period 1994–98 (n = 5953), 3544 were taken into account. Similar information (on 7371 children) is available from the Demographic Surveillance System based in Nouna, western Burkina Faso, with a total population of about 60 000 inhabitants. We used all-cause childhood mortality as outcome variable (877 and 559 deaths from DHS, DSS, respectively) Additionally, we investigated the magnitude of underestimation of childhood mortality when using maternal histories to assess childhood deaths (the method used in DHS+ surveys).

Results: A simultaneous estimation of hazard rate ratios by a Cox regression model yielded roughly similar estimates for the DHS+ and DSS data, in line with previous findings² (RRs of 1.05 and 1.07, respectively, for males, 3.0 and 2.4, respectively, for "twin birth", but no agreement for "religion"). Of the 7371 children in the DSS data, 118 (95 alive and 23 deceased) would have been missed in maternal histories because the mother was deceased.

Conclusions: These findings demonstrate that, despite some limitations, DHS+ surveys are broadly comparable to the more precise DSS data, and are therefore a valuable tool for assessing the importance of risk factors for childhood mortality in sub-Saharan Africa. This will help us in the investigation of new methods of estimating childhood mortality for a whole country, based on DSS data and knowledge about the distribution of the main risk factors.

1. Institut National de la Statistique et de la Démographie and Macro International Inc. Enquête Démographique et de Santé Burkina Faso 1998–1999. Calverton, Maryland, USA: Macro International Inc, 2000. 2. Becher H, Müller O, Jahn A, et al. Risk factors of infant and child mortality in rural Burkina Faso. Bull WHO, (in press), 2004.

J. Haukka¹, K. Wahlbeck¹, T. Klaukka², J. Lönnqvist³, J. Tiihonen⁴.¹STAKES, Finnish Information Centre for Register Research, Helsinki; ²Social Insurance Institution, Research and Development Unit, Helsinki; ³National Public Health Institute, Department of Mental Health and Alcohol Research, Helsinki; ⁴University of Kuopio, Department of Forensic Psychiatry, Kuopio, Finland

Introduction: The dynamics of drug usage are of paramount interest when the different medications are compared with naturalistic study approach. The aim of this study was to model the dynamics of antipsychotic drug usage outside the hospital, and hospitalisations.

Methods: 3232 individuals were followed up after the first episode of hospitalisation due to schizophrenia. After the first episode, the hospitalisations due to schizophrenia, the usage of antipsychotic drugs, and death were recorded for each individual. All data were gathered by register linkage of the Finnish Hospital Discharge Register and the Drug Imbursement Register of National Health Insurance Institute. Follow up period was divided in 30 intervals and the usage of antipsychotic drugs, hospitalisation, or death was determined on each interval. Using these data we composed a state space of 14 states (12 out-hospital drug usage, hospitalisation, and death). 152 411 transitions between states were observed. The dynamics of this state space were modelled with Markov assumption using graphical model approach. Current and previous state of medication, age, sex, length of the first hospitalisation, and the follow up period were used in modelling. Akaike Information Criterion (AIC) was applied in model selection. The results are presented as conditional probabilities of current state given the previous state.

Results: The fitted models showed high connectivity between all of the variables. The probability of continuing in the same state was highest (over 80%) for clozapine and olanzapine, and lowest (about 60%) for levomepromazine and perphenazine depot injections. The probability of hospitalisation was the highest for no medication and haloperidol (oral) groups, and lowest in perfenazine group.

Conclusions: The probability of continuing medication was highest for clozapine and olanzapine. However, the probability of hospitalisation was the lowest for perphenazine users.

A. Hosseinpoor¹, M. Forouzanfar¹, M. Yunesian², D. Farhood³.¹School of Public Health, Tehran University of Medical Sciences, Epidemiology and Biostatics, Tehran, Iran; ²School of Public Health, Tehran University of Medical Sciences, Environmental Health, Tehran, Iran; ³School of Public Health, Tehran University of Medical Sciences, Human Genetics, Tehran, Iran

Introduction: Health effects of air pollution have been studied in many different parts of the world. However, few studies have explored the cardiovascular impacts of air pollution. For the first time we studied the association between air pollutants and admission due to two cardiovascular disorders, angina pectoris and cor pulmonale, in Iran.

Methods and Materials: This is a retrospective time series study. The variables of the study include the level of five air pollutants—NO₂, CO, O₃, SO₂, and PM₁₀—as independent variables; daily hospitalisation due to angina pectoris and cor pulmonale in 25 academic hospitals in Tehran as dependent variable; and mean daily temperature and relative humidity, seasonality, time trend and day of week as potential confounders. All variables were measured during a 5 year period from 21 March 1996 to 20 March 2001. The data of the 24 h average levels for SO₂, NO₂, CO, PM₁₀ and 8 h maximum levels for ozone were collected from one of the stations of Tehran’s Air Quality Control Corporation. Data were analysed using Poisson regression models. Relative risks (RR) of hospital admissions for angina pectoris and cor pulmonale were calculated for an increase of 1 mg/m³ for CO and 10 µg/m³ for the other pollutants.

Results: Daily admission due to angina pectoris was significantly related to CO level, after controlling for confounder effects. Each unit increase in CO level causes a 1.009 increase in the number of admissions (95% CI 1.004 to 1.015). This association was verified with a lag of one day. There was no significant association between the other air pollutants and number of daily admissions due to angina pectoris.

There was no significant association between any of air pollutants and daily admissions for cor pulmonale.

Conclusion: We found that with increasing level of CO pollutant, the number of admissions due to angina pectoris rises. Ischaemic heart disease is one of the leading causes of death in Iran. Air pollution control will reduce the number of this preventable disease and resulting death.

A. Jahn, J. Petersen, O. Razum.University of Heidelberg, Department of Tropical Hygiene and Public Health, Heidelberg, Germany

Introduction: Antenatal care is well established in Germany, with an almost complete coverage and 10 antenatal consultations per participating woman. Antenatal screening procedures include laboratory tests, physical examination, measurement of blood pressure and weight, and three routine ultrasound scans around week 10, 20, and 30. While specific single screening procedures have been tested in clinical trials, little is known about the combined effect of the whole set of antenatal screening activities in routine antenatal care. Therefore, we investigated the effect of antenatal screening in terms of suspicious or abnormal findings as reported by the mother and/or recorded in the antenatal card, and assessed the predictive properties for selected outcomes.

Methods: Cohort study comprising 360 women enrolled in antenatal classes with follow up from first attendance of antenatal classes (around week 26) until after delivery. The women were interviewed with structured questionnaires with regard to their previous antenatal consultations and related test results and findings; 273 women had a second interview before delivery and 340 women were interviewed postnatally. The postnatal interview included also questions on the pregnancy outcome. In addition, the antenatal cards of the interviewees were reviewed and analysed.

Results: In the first round of interviews 36% of women reported suspicious or abnormal findings; this figure increased to 43% till the end of pregnancy. In total, 495 such findings were reported (some women reported more than one finding). Almost half of these findings (45%) were based on ultrasound scans, 15% on physical examination, 11% on CTG, 10% on laboratory results, 8% on body weight and blood pressure, and 11% on other investigations. Among the ultrasound based findings, there were 16 suspected malformations, of which three (19%) were confirmed postnatally. Nine other malformations were not detected antenatally (sensitivity 25%). Predictive properties for other outcomes were also poor.

Conclusions: Routine antenatal care in Germany confronts a large proportion of participating women with suspicious findings and test results, most of which turn out to be unrelated to adverse outcomes but may cause considerable distress. The screening quality needs to be improved and women should receive a pre-test counselling on the scope and limitations of antenatal screening procedures.

S. Jena¹, C. Becker-Witt¹, S. Roll¹, K. Wegscheider², S. Löbel¹, D. Selim¹, B. Brinkhaus¹, S. Willich¹.¹Charité University Hospital, Institute for Soc. Medicine, Epidemiology & Health Economics, Berlin, Germany; ²University of Hamburg, Institute of Statistics and Econometrics, Hamburg, Germany

Introduction: The aim was to evaluate whether different imputing strategies for missing values affect the results of a study on low back function after acupuncture or control. Relevant differences were expected, since missing values were unbalanced and thought to be due to withdrawals because patients were not randomised to their preferred treatment.

Methods: The analyses were based on data of 2841 patients with chronic low back pain participating in a randomised interventional study of 15 sessions of acupuncture (ACU) v control (CON) without acupuncture. We imputed new values to missing data of the primary outcome (back function according to a German questionnaire (FFbH-R)) and of quality of life measured by the SF-36 filled in by patients at baseline and after 3 months. We compared the following imputing strategies for the main outcome: a) last value carried forward (LVCF); b) three different hot deck methods (missings replaced out of all cases (MRA), missings replaced from the other randomised group (MRO), missings replaced from the same randomised group (MRS). In addition, we performed a complete case analysis.

Results: Missing data were found in 11% of the patients (ACU 9.3% v CON 13.2%). The analysis for complete cases showed significant differences between ACU and CON in back function after 3 months (FFbH-R: ACU 74.0 (SD 20.5), CON 65.4 (21.7), p<0.001). Different imputing strategies showed similar and significant p values. Comparing the complete cases analysis with the most conservative imputing strategy (MRO), no relevant changes in means of the outcome were found (FFbH-R: ACU 73.1 (20.8), CON 66.7 (21.8)). The most common strategy of LVCF showed also similar results (FFbH-R: ACU 73.9 (20.5), CON 65.4 (21.7)). For quality of life (subscales of the SF-36) no relevant differences were found as well using the different imputing strategies.

Conclusion: In a randomised study with differential frequencies of missing values, unexpectedly the choice of the imputation strategy had no impact on the results.

E. João¹, M. Cruz¹, J. Menezes¹, H. Matos², G. Calvet¹, M. D’Ippolito³, L. Salgado¹, S. Silva¹, R. Braga².¹Hospital dos Servidores do Estado, Infectious Diseases, Rio de Janeiro, Brazil; ²Hospital dos Servidores do Estado, Epidemiology, Rio de Janeiro, Brazil; ³Hospital dos Servidores do Estado, Obstetrics, Rio de Janeiro, Brazil

Introduction: Worldwide, millions of children are estimated to have been infected with HIV through mother to child transmission (MTCT). However, in some settings, antenatal screening for HIV infection, proper management with antiretroviral drugs, obstetrical interventions, and avoidance of breastfeeding have reduced the risk of MTCT to less than 2%.¹ In Brazil, pregnant women have been offered HIV testing since 1996. Those found to be HIV seropositive have been offered perinatal transmission prophylaxis with zidovudine according to the ACTG 076 protocol.² Subsequently in 2000, Brazilian guidelines regarding the management of HIV infected women have included recommendations regarding administration of other antiretroviral drugs (ARVs) during pregnancy. Triple ARV therapy (HAART) has become standard of care for patients with CD4 counts <350 cells/mm³.

Objectives: To describe a cohort of HIV infected pregnant women in Rio de Janeiro, Brazil, and evaluate risk factors for vertical transmission.

Subjects and Methods: A cohort of HIV infected pregnant women was followed from January 1996 through December 2001. Data regarding women and their children were collected prospectively as part of routine care using the same clinic form throughout the 6 year period and included: demographic data, HIV related history, clinical, obstetric and laboratory history and assessment; infant birth weight, and gestational age. Viral load and CD4 counts of the mothers were obtained at baseline and near delivery. Antiretrovirals (ARVs) were offered for prophylaxis and/or treatment according to Brazilian guidelines. Mothers were counselled not to breastfeed. Zidovudine was given to infants during the first 6 weeks of life and artificial formula was provided. Children were followed up to definition of HIV status. Bivariate analysis and multivariate logistic regression were performed to analyse risk factors.

Results: 297 out of 346 HIV+ women referred to our hospital were included in the analysis cohort. Overall transmission rate was 3.57% (95%CI 1.82 to 5.91) and remained constant over time. Low birth weight was independently associated with a higher risk of vertical transmission (p = 0.01), while a longer duration of receipt of ARVs during pregnancy was independently associated with a lower risk of transmission (p = 0.04).

Conclusion: Despite availability of free drugs for HIV management in Brazil, further decreases in vertical transmission should be pursued with efforts towards earlier diagnosis and increase in the length of prophylaxis.

K. C. Johnson.¹Health Canada, Centre for Chronic Disease Prevention and Control, Ottawa, Canada

Objectives: To estimate the risks of breast cancer associated with passive and active smoking, and to explore the heterogeneity in individual study results.

Methods: Examination of the 16 published epidemiologic studies of breast cancer risk related to exposure to secondhand tobacco smoke. Analysis of the risk of breast cancer in relation to passive smoking among never active smokers and examination of the risk of breast cancer associated with active smoking after control for passive smoking in the never smoker referent group.

Main Outcome Measures: Summary relative risk of breast cancer among: 1) life long non-smokers with regular passive exposure to tobacco smoke; and 2) women who had smoked compared with women never regularly exposed to tobacco smoke.

Results: The summary breast cancer risk estimate for exposure to passive smoking among women who had never smoked was 1.41 (95% CI 1.17 to 1.70). The individual study risk estimates were heterogeneous: for the 10 studies that had not collected quantitative information on the three major sources of passive smoke exposure (childhood, adult residential, and occupational) the summary risk estimate was 1.15 (95% CI 0.98 to 1.35); where these three major sources of passive exposure were collected, the summary risk estimate was 1.85 (95% CI 1.56 to 2.20). For passively exposed premenopausal women who never smoked the summary risk estimate was 1.99 (95% CI 1.50 to 2.64). For women who had smoked the summary breast cancer risk was 1.68 (95% CI 1.22 to 2.32) compared with women never regularly exposed to tobacco smoke. Individual risk estimates were again heterogeneous: with incomplete passive smoking assessment, the active smoking summary risk was 1.17 (95% CI 0.85 to1.60); with more complete passive exposure assessment, the active smoking summary risk estimate was 2.14 (95% CI 1.56 to 2.92).

Conclusion: There is a growing body of evidence that implicates passive and active smoking as risk factors for breast cancer. More studies are needed with thorough passive smoking assessment and that explore the biological mechanisms that might explain the unexpected similarity of the passive and active risk.

T. Kabakian-Khasholian.American University of Beirut, Faculty of Health Sciences, Beirut, Lebanon

The Arab region is characterised by a wide variation in population development indicators, fertility rates, and income levels. The region is undergoing a rapid shift from home to hospital births, with an increased uptake of medical technology used in maternity services.

The findings of a coordinated research programme on childbirth issues in the Arab region, namely in Lebanon, Egypt, Palestine, and Syria, provide evidence on the variation in the process of care in these countries, as well as on the discrepancy between routinely followed practices and best practices identified by the literature. Childbirth is highly medicated with a substantial number of unnecessary and even harmful practices such as routine perineal shaving, labour induction, enemas and episiotomies, infrequent use of rooming in, minimal provision of breastfeeding advice and support, and lack of family planning advice being the rule rather than the exception. Women complain about the quality of care in hospitals, especially lack of privacy, bad treatment, and inability to have a companion during labour. They prefer professional delivery in hospitals. They are usually trustful of the physician’s decisions of procedures, although they are discontent with many procedures performed. In general, women are not involved in the decision making process for the provision of care during childbirth and are not much vocal about their complaints.

The constellation of the studies conducted in this programme pinpoint problems in the quality of maternity services provided in the region and the lack of women’s involvement in the overall process of care. Currently, a number of intervention studies targeting behavioural change among providers and/or women as well as studies evaluating the effectiveness of practices with unknown outcomes are being conducted in the region.

This programme, working through a consolidated network of researchers in the Arab region, provides elucidation of one of the most important rites of passages in women’s lives in a region with high fertility rates. The research is also important, as childbirth in the region is salient concerning the implications of shifts from traditional to western biomedical health care systems for an event still highly embedded in traditional understandings and practices. Moreover, the variation in terms of practices and health care systems in our region offers researchers the opportunity to increase knowledge about a variety of intervention practices, with the potential of extrapolating findings to similar situations beyond the region.

M. Kelaher¹, S. Paul², H. Lambert³, A. Waqar³, S. Fenton⁴, G. Smith³.¹University of Melbourne, School of Population Health, Australia; ²Centre for Research in Primary Care, UK; ³University of Bristol, Department of Social Medicine, UK; ⁴University of Bristol, Department of Sociology, UK

Introduction: There is little doubt that exposure to trauma can have negative health sequels but there has been relatively little research on the impact of discrimination on health and its interaction with socioeconomic status. In this study we examine the relationship between education, discrimination, and health among white (n = 227), African Caribbean (n = 213) and Indian and Pakistani (n = 233) adults aged between 18 and 59 years living in Leeds as measured in a stratified population survey.

Analysis: Measures of discrimination included any physical attack, verbal abuse, and unfair treatment on the basis of sex and race. A combined variable, any discrimination, was also developed. Education (above secondary level and secondary level or below), anxious or depressed (quite a lot and a little/not at all), and health status (fair/poor and good/excellent) were coded dichotomously. Analyses were conducted examining the relationship between education and discrimination, discrimination and health, and discrimination and health controlling for education.

Results: People educated above secondary level were more likely than people educated at secondary or below to report being physically attacked (p = 0.001), verbally abused (p = 0.001), and unfairly treated (p = 0.00) on the basis of sex. There were no differences between people educated above secondary level compared with people educated at secondary level or below in reports of being physically attacked (p = 0.6), or verbally abused (p = 0.2) on the basis of race. The more educated group were more likely to report unfair treatment on the basis of race (p = 0.05). People educated above secondary level were more likely that people educated at secondary or below to report more discrimination of any kind (p = 0.00). Any discrimination (p = 0.06), physical attack (p = 0.07), and verbal abuse (p = 0.04) were associated with being very anxious or depressed. Unfair treatment due to sex (p = 0.8) or race (p = 0.1). These results remained the same when education was taken into account. There was no association between having been exposed to any kind of discrimination and having fair or poor health before education was taken into account. Once education was taken into account, having been exposed to any discrimination (p = 0.04) and verbal abuse on the basis of race (p = 0.03) were associated with poorer health status.

Conclusion: The relationship between education and discrimination in this study is counterintuitive, with people with higher education reporting greater levels of discrimination. The results also suggest that there is a relationship between exposure to discrimination and mental and physical health. However, the effect on physical health is not apparent unless education is taken into account.

L. R. Kerr-Pontes¹, C. N. Evangelista², M. L. Barreto³, M. Gomide¹, H. Feldmeier⁴.¹Federal University of Ceará, Community Health Department, Fortaleza, Brazil; ²Health Office of the Ceará State, Department of Epidemiology, Fortaleza, Brazil; ³Federal University of Bahia, Public Health Institute, Salvador, Brazil; ⁴Freie Universitat Berlin, Fachbereich Humanmedizin, Berlin, Germany

Introduction: Brazil report almost 80% of leprosy cases in America. In Ceará, one of the poorest states in North east Brazil, the distribution of leprosy shows a remarkable heterogeneity and a tendency to increase over the past decade. We proposed that analytic epidemiology might provide a clue as to why control efforts have not met the goals set by WHO. We decide to study the risk factors related to leprosy in four areas highly endemic for Mycobacterium leprae in Ceará.

Methods: A case-control study was carried out in those areas. Cases: diagnosis of leprosy by clinical characteristics and at least one positive skin smear; patients were registered in the leprosy register not longer than 2 years ago; no case of leprosy reported in the household or in the near neighbourhood; 20 years. Controls: an individual 20 years with no signs of skin disease, who lived in the same municipality as the leprosy case; no leprosy case in the family or in the near neighbourhood and who procured medical advice in the same health unit unrelated to skin problems in which the leprosy patient was monitored. Controls were matched for age and sex, and for each case four controls were studied. A semi-structured questionnaire was used to collect demographic, socioeconomic, environmental, and behavioural data from March to August 2002. Variables were first analysed in a bivariate manner to identify those variables useful for later logistic regression. In a second step a multivariate hierarchical analysis was performed using a model especially developed for this study.

Results: 200 cases and 800 controls were examined. A low education level, previous food shortage, frequent contacts with natural water bodies as well as a low frequency of changing bed linen or hammock were all significantly associated with leprosy. A BCG vaccination was found to be a highly significant protective factor.

Conclusions: In our study all variables that remained significant after logistic regression are—in one way or another—linked to poverty. However, they may act on different levels on the transmission of M. leprae and/or the progress from infection to disease. Interestingly, variables reflecting high risk factors for person-to-person spread such as a crowding or sharing the bed or hammock with other household members did not show a significant association with leprosy. These puzzling observations have generated the hypothesis that beside person-to-person spread M. leprae might be transmitted by indirect means and that other reservoirs should exist outside the human body. Water has been considered a putative source of infection with M. leprae already in the early days of leprology. In conclusion, the study shows that certain environmental risk factors exist that favour the occurrence of leprosy in an endemic area and which could be targeted in control measures.

K. Khunti¹, M. Stone¹, A. Burden², M. Burden², R. Baker¹, A. Dunkley¹, N. Raymond³.¹University of Leicester, Dept of Health Sciences, Leicester, UK; ²Heart of Birmingham, Diabetes Care, Birmingham, UK; ³Warwick Medical School, Division of Health in the Community, Coventry, UK

Introduction: In the UK, general practitioners have an increasing role in the care of people with type 2 diabetes and intensive management has been shown to be effective and safe. The aim of this study was to evaluate the use of a near patient test (NPT) for glycated haemoglobin (HbA1c) in the management of patients with type 2 diabetes, in primary care.

Methods: The study design was a randomised controlled trial, set in eight general practices in Leicestershire, UK. Patients were randomised individually, to a control group receiving "usual care" (venepuncture) for laboratory measurement of HbA1c or an intervention group tested and managed using the Bayer DCA2000 NPT system. The study follow up period was 12 months. Proportions and 95% CIs of patients with good glycaemic control, defined as HbA1c 7% were estimated and compared in intervention and control groups at baseline and end of study, using laboratory results in all cases. Multiple logistic regressions were used to adjust for potential confounding factors and odds ratios (95% CI) were estimated.

Results: 638 patients (319 interventions, 319 controls) were randomised and included in the intention to treat analysis. Intervention and control groups were similar at baseline with respect to sex, age, duration of diabetes, and treatment. The proportions (95% CI) achieving good metabolic control (HbA1c < = 7.0%) were NPT 0.44 (0.39 to 0.49) v controls 0.40 (0.35 to 0.46) at baseline and NPT 0.40 (0.35 to 0.46) v controls 0.37 (0.32 to 0.42) at follow up. These differences were not statistically significant at the 5% level at either time point. The unadjusted OR (95% CI) for NPT patients (compared to controls) achieving "good" control at follow up was 0.96 (0.70 to 1.31). The final model resulting from multiple logistic regression modelling included sex, duration of diabetes, treatment at baseline, deprivation score, and baseline HbA1c status. The OR (95% CI) for the NPT v control group was 0.84 (0.58 to 1.22), representing an 16%, statistically non-significant reduction in achievement of good control after the intervention and one year of follow up compared to controls.

Conclusions: Proportions of patients achieving good control were very similar in both groups throughout this study. The use of a NPT to measure HbA1c in patients with type 2 diabetes in primary care is unlikely on its own to lead to improvements in metabolic control. Perhaps 1 year of follow up is insufficient to detect improvements. Rapid feedback of test results to patients is an appealing concept, and further evaluation, perhaps using an algorithm to guide prescribing to achieve specified targets would be useful.

A. Kieltyka¹, K. Szafraniec¹, W. Jedrychowski¹, W. Laszewicz².¹Collegium Medicum, Jagiellonian University, Chair of Epidemiology and Preventive Medicine, Kraków, Poland; ²Medical University, Department of Gastroenterology and Internal Diseases, Bialystok, Poland

Introduction: Helicobacter pylori infection is predominantly acquired during childhood. Despite its high prevalence and serious clinical results, there have been few large population based studies examining epidemiology of its infection. The purpose of the study was to assess the distribution and determinants of H. pylori infection in Polish children and youth.

Methods: The cross sectional study on prevalence of H. pylori infection was administered to 2382 children aged 1 to 18 years in six university centres in Poland. H. pylori status was determined by level of specific IgG antibodies (>24 IU/ml). Data on demographic factors, socioeconomic status, living conditions, and hygienic habits were collected by interview (with children over 14 and with parent of younger ones).

Results: The overall prevalence of H. pylori infection was 32.0% (95% CI 30.4 to 33.6). Prevalence increased with age from 26.4 in 1–3 years old to 36.5 in 15–18 years old. Infection is not associated with sex (34.0% in boys and 31.7% in girls). It was more common in rural areas (38.1%), than in small (<100 000 dwellers) and big cities (respectively 34.2% and 25.5%). Probabilities of having infection were statistically higher in children from families greater than four persons (36.0% v 29.2%), with more than two children (39.5% v 30.5%), rather poor (public assistance help) (45.4% v. 31.0%), living in flats with less then 10m² per person (40.7% v 30.8%), with wells as the source of water (42.0% v 34.2%), and without sewage system (37.6% v 30.7%). Prevalence of infection was higher if a family of child owned a piece of land (39.4% v 29.7%), farm animals (39.4% v 29.9%), and children helped in farm works (39.3% v 29.9%). Drinking non-boiled water increased infection prevalence (39.4% v 28.5%), like eating not washed fruits (39.8% v 31.6%) and having a bath no more than once a week (40.9% v 29.7% for at least once a day). Following variables remained statistically significant after adjusting for sex, age, and place of living and were taken into the final multivariate logistic model: number of children in a family OR = 1.13 (95% CI 1.03 to 1.25), housing density <10 m²/person OR = 1.53 (95% CI 1.12 to 2.09), public assistance OR = 1.40 (95% CI 1.02 to 1.92), help in soil cultivation OR = 1.35 (95% CI 1.08 to 1.68), and drinking non-boiled water OR = 1.40 (95% CI 1.13 to 1.74).

Conclusions: The importance of household living conditions and poverty in acquisition of H. pylori infection in childhood was confirmed. Household crowding, living with others children, and necessity of public assistance were identified as risk factors for infection. Additionally work in contact with soil and drinking non-boiled water could increase the probability of infection.

R. Kregzdyte, D. Baranauskiene, R. Naginiene, S. Ryselis.Kaunas University of Medicine, Institute for Biomedical Research, Kaunas, Lithuania

Introduction: Biological monitoring is based on the repeated measures data analysis. But the contingent often varies over the whole period of longitudinal study. Therefore, the problem of data analysis arises in estimating loses of subjects. Missing data may reduce the precision of calculated statistics. Because lead is known as toxic heavy metal, the accuracy of its effect estimation is necessary. The objective of this study was to find an appropriate method for estimation of missing values in the assessment of changes of one of the most sensitive biomarkers of lead toxicity, delta-aminolevulinic acid dehydratase (delta-ALAD) activity, and lead concentration in blood (PbB).

Methods: The biological monitoring of workers, occupationally exposed to lead, has been conducted over 6 years (1998–2003). Blood lead concentration was determined by atomic absorption spectrophotometer with Zeeman Effect. Delta-ALAD activity in blood was measured according to the European standardised spectrophotometric method. Four methods of handling with missing values were considered: listwise deletion, adding/subtracting mean of differences between adjacent measurements, expectation maximisation, and regression. Repeated measures analysis of variance was applied for evaluation of changes in delta-ALAD activity and PbB over all period.

Results: The mean of PbB without filling missing data in every year of study was 6.28, 8.65, 4.73, 4.77, 4.44, and 3.62 µg/dl respectively. The mean of delta-ALAD activity without filling missing data in every year of study was 0.646, 0.440, 0.439, 0.352, 0.568, and 0.474 µmol/s*l, respectively. All methods of missing values estimation produced similar results: the long term exposure to lead was related to negative changes in delta-ALAD activity even after cessation of lead exposure and might cause disorders in haem biosynthesis.

Conclusions: Listwise deletion is easy but inefficient method for handling missing data. Filling missing data with plausible values is more efficient, but it requires care to avoid data distortion. Regardless of the imputation method merits, imputed values are only estimates of the unknown true values.

A. Kroke¹, S. Hahn¹, A. Liese², A. Buyken¹.¹Research Institute of Child Nutrition, Nutrition and Health, Dortmund, Germany; ²University of South Carolina, Epidemiology and Biostatistics, Columbia, SC, USA

Introduction: Prevalence of overweight and adiposity among children and adolescents have increased dramatically. Research focuses on critical periods over the life course, such as the time of adiposity rebound (AR). Recent data showed that age at AR is associated with higher body mass later in life. AR occurs during childhood (between ages 3–8) when the child starts to gain more weight relative to the growth of length. To study its effects or determinants, AR needs to be determined. To date, two approaches have been applied: the visual inspection (VI) of individual body mass index (BMI) curves, and the use of a polynomial equation (PE) of varying degrees. Different approaches might lead to different results and thereby impair conclusions. So far, no formal evaluation was published comparing these approaches. We used one dataset to compare different approaches to determine AR.

Methods: Children were selected from the DONALD (DOrtmund Nutritional and Anthropometric Longitudinally Designed) study, an ongoing study assessing detailed anthropometry between infancy and adulthood. Weight and height are measured annually and BMI was calculated as weight (kg)/height (m)². Children with data gaps >1.5 years between age 2 and 10 were excluded, as were children with abnormal growth curves. Age at AR was determined both with VI of individual BMI curves according to preset criteria, and with PE of 2nd, 3rd and 4th order. In 119 children AR was determinable with all methods.

Results: Mean age at AR ranged from 5.27 (years) to 5.67 in boys and from 5.44 to 5.96 in girls, depending on method. In both sexes, the VI yielded the highest, the 3rd degree PE the lowest estimate. Mean differences in age at AR between VI and the different PE are ranged from 0.30–0.40 in boys, and 0.39–0.52 in girls. We categorised age at AR estimates from both methods (early, medium, late) and cross classified them. Compared with VI, the 2nd order PE had the highest concordance (58% same category, no opposing classification). Within the PE approach, the 3rd and 4th order PE had the highest concordance (77% same category, no opposite classification). Goodness of fit measures of the PE indicated that the 2nd order polynomial had the best fit.

Conclusion: In a dataset with highly complete measurements and no undeterminable AR, the use of PE did not yield substantial advantages over the VI. In a more realistic study setting including unusual growth curves, a visual inspection of all growth curves in which PE cannot determine AR would be necessary. If identical approaches to determination of AR are favoured, VI for all curves is required. We conclude that the VI is a practical method with satisfactory results. However, in the face of missing data, an issue not studied here, PE approaches may have advantages.

M. Kulig¹, M. Nocon¹, D. Jaspersen², J. Labenz³, T. Lind⁴, W. Meyer-Sabellek⁵, M. Stolte⁶, P. Malfertheiner⁷, S. Willich¹.¹Charité University Medical Center, Institute for Social Medicine, Epidemiology and Health Econo, Berlin, Germany; ²Klinikum Fulda, Fulda, Germany; ³Ev. Jung-Stilling-Krankenhaus, Siegen, Germany; ⁴AstraZeneca R&D, Mölndal, Sweden; ⁵AstraZeneca, Wedel, Germany; ⁶Institute of Pathology, Klinikum Bayreuth, Bayreuth, Germany; ⁷Otto-von-Guericke University, Department of Gastroenterology, Magdeburg, Germany

Objective: Gastro-oesophageal reflux disease (GERD) is a prevalent condition with symptom frequency up to 50% in the general population. The disease is characterised by frequent relapses and requires long term management with considerable medical and socioeconomic implications. We aimed to assess disease related costs in relation to patients and disease characteristics including quality of life (QoL) during two years of routine care (RC).

Methods: ProGERD is a multicentre cohort study of 6215 outpatients with GERD predominantly from Germany (mean age 54 (SD 14), 47% female). Patients were endoscoped and received initial standardised treatment with a proton pump inhibitor for 2 to 8 weeks. During the following observational period, patients received RC at the discretion of their primary care physician. After 1 and 2 years of RC, medication, and other patient and disease variables were assessed by patient questionnaires (response 90% and 86%), including the Reflux Disease Questionnaire for reflux symptoms (RDQ, min: 0, max: 36), and the Quality of Life in Reflux and Dyspepsia for disease specific QoL (QOLRAD, min: 1, max: 7). Disease related direct cost data (medication, physician visits, and hospital admissions) were calculated by multiplying disease related medical resource units with cost factors by unit. Indirect costs were not included in this analysis. Factors associated with those direct costs were analysed by multiple regression analysis.

Results: Disease related direct costs amounted to a mean of 361 (SD 911) per patient per year. Significantly higher amounts were observed in patients with severe GERD (erosive GERD: mean 438 per patient per year, or Barrett oesophagus: mean 566 per patient per year) v non-erosive disease (282), in patients with long-standing GERD > = 5 years (438) v duration <5 years (331), and in patients with concomitant diseases (394) v no concomitant diseases (246). Direct disease related costs were increasing with decreasing disease specific QoL (highest quartile of QOLRAD score: 308 v lowest quartile: 496). Medication costs were the main proportion of the total direct costs (71%), followed by hospital care (22%) and physician visits (7%). According to multiple regression analysis, the amount of direct costs was significantly associated with concomitant diseases and with GERD related characteristics including frequency and severity of reflux symptoms (RDQ score), GERD classification, and disease specific QoL (p<0.05). Disease related direct costs did not differ significantly with respect to socioeconomic factors such as age, sex, and education.

Conclusion: In RC, direct costs for GERD varied considerably, mainly depending on disease specific characteristics and not on socioeconomic factors. Medication costs contributed markedly to the disease related costs and were similar to those in patients with statin therapy for hyperlipidaemia1.

Supported by a grant from AstraZeneca GmbH, Germany

1. Schwabe U, Paffrath D. Berlin: Arzneiverordnungsreport Springer Verlag 2003.

I. Kulmala¹, J. McLaughlin², M. Pakkanen³, K. Lassila³, L. Hölmich⁴, L. Lipworth², J. Boice², J. Raitanen¹, R. Luoto¹.¹Tampere University, Tampere School of Public Health, Tampere, Finland; ²International Epidemiology Institute, Rockville, Maryland, USA; ³Hospital Siluetti, Helsinki, Finland; ⁴Danish Cancer Society, Institute of Cancer Epidemiology, Copenhagen, Denmark

Introduction: Use of silicone breast implants in cosmetic breast surgery has increased rapidly and many questions have been raised regarding the safety of silicone implants. Recently, the occurrence of local complications has been set as the primary concern in research related to cosmetic breast implant surgery. The main aim of this study was to find out the frequency rates of local complications in a population of cosmetic breast implant patients in Finland.

Aim: To also collect information on postoperative personal opinions and quality of life.

Methods: Patient records of 685 women were identified to collect information on implant characteristics, complications, and treatment procedures. Information on personal characteristics, medical and reproductive history, and postoperative quality of life were obtained through structured questionnaires mailed to 470 women of the same cohort.

Results: 36% of the women had at least one complication diagnosed in patient records. Capsular contracture was the most common complication, diagnosed in 17% of the women. Other complications were found to be quite rare. 26% of the women had undergone at least one postoperative treatment procedure. Capsular contracture was the main indication for additional surgery. Most of the women (89%) were satisfied with their decision to undergo the implantation operation. However, only 40% of the women were satisfied with the preoperative information on possible health effects related to breast surgery. Only 9% of the women reported increase in their postoperative health status, whereas majority reported positive effects on their self confidence (79%) and appearance (89%).

Conclusions: Epidemiological information on postoperative complications as well as quality of life questions and personal satisfaction related to cosmetic breast implant surgery has been insufficient and sparse. Our data confirm previous results of capsular contracture as a most significant and common complication in cosmetic breast implant surgery and of quite low frequency of other complications and represent interesting perspectives on personal satisfaction and postoperative quality of life related to breast surgery.

J. Kyncl¹, B. Prochazka², M. Havlickova¹, M. Otavova. J. Castkova¹, B. Kriz¹.¹National Institute of Public Health, Centre of Epidemiology and Microbiology, Prague, Czech Republic; ²National Institute of Public Health, Department of Biostatistics, Prague, Czech Republic

Introduction: Annual influenza epidemics differ in duration and magnitude. Influenza infection is underestimated, being easily mistaken for one of acute respiratory diseases (ARD) since it has similar clinical symptoms. The aims of this paper are to find correlation between mortality and influenza morbidity and to model mortality in different weeks of the year outside the influenza epidemic.

Methods: Data on daily deaths from all causes and deaths from diseases of the circulatory system in the Czech Republic were available for 1982–2000 (altogether 2 323 424 and 1 293 786 deaths reported, respectively). Data on the incidence of influenza and other ARD were taken from the surveillance programme. The weeks in which ARD morbidity exceeded the epidemic threshold and at the same time, circulation of influenza virus among the population was reported by the NRL for influenza were considered as influenza epidemic weeks. Analysis was based on the assumption that outside the epidemic periods, deaths are distributed according to the Poisson distribution with a linear trend depending on time and with periodic behaviour during the year. The morbidity rate is only expected to increase in the epidemic compared with non-epidemic period.

Results: When comparing the weekly morbidity from acute respiratory illnesses and weekly mortality for all causes of death, the peaks of these two parameters almost overlap. In the epidemic period (178 weeks) 49.4% of findings were above the unilateral 95% tolerance limit of the model, compared with the non-epidemic period (813 weeks) with only 6.8% of findings above this limit. The mean estimated excess of annual deaths from all causes was 2764 (min –630; max 7049). The median of deviations of the estimated number of deaths from the actual number of deaths is negligible, that is, 0.8 (95% CI –5.3 to 7.0), for the non-epidemic period, being equal to 204.7 (95% CI 175.4 to 249.6) for the epidemic period. Similar results were found for deaths from diseases of the circulatory system accounting for 55.7% of all deaths in the study period. The median of deviations of the estimated number of deaths due to diseases of the circulatory system from the actual number of deaths is 1.0 (95% CI –3.2 to 6.5) for the non-epidemic period, being equal to 133.2 (95% CI 116.4 to 166.3) for the epidemic period.

Conclusions: The presented results confirm clearly and unambiguously excess in death rates during the influenza epidemic periods, depending on the duration and magnitude of the epidemic. The mean annual excess rate for the Czech Republic is 2.24% population, the major part of this rate being attributable to influenza. Vaccination against influenza proved both effective and cost-effective and therefore is to be recommended as the most important preventive measure.

C. Kyobutungi, U. Ronellenfitsch, O. Razum, H. Becher.University of Heidelberg, Tropical Hygiene & Public Health, Heidelberg, Germany

Introduction: Since the late 1980s, almost 3 million ethnic German "Aussiedler" have migrated from Eastern Europe to Germany. Their health status could be influenced by the "mortality crisis" in Eastern Europe, where cardiovascular diseases and external causes cause an extraordinarily high number of premature deaths, and by their comparatively low socioeconomic status in Germany. We assessed their mortality by age, sex, and cause and compared it to that of the German population.

Methods: We established a cohort of 34 394 selected from 281 356 Aussiedler aged 15 years and above who arrived in North Rhine, Westphalia, Germany’s largest federal state, between 1990 and 2001 from the former Soviet Union. We established vital status at 31 December 2002 through local registries for the first episode of residence. For deceased cases, we obtained ICD coded cause of death from the state’s statistical office. We calculated person years for 5 year age groups, sex, and calendar period. For comparison, we used German mortality data in three different time periods (1990–93, 1994–97, 1998–2002) to account for secular trends in mortality.

Results: The cohort contributed about 214 000 person years with a mean follow up of 6.25 years. 1498 cohort members (4.36%) had died and cause of death information was available for 98%. The Aussiedler had a lower mortality than the general German population. SMR adjusted for age, sex, and calendar period were significantly low for all causes at 0.8 (95% CI 0.7 to 0.9), cardiovascular at 0.7 (0.6 to 0.8), and external causes 0.8 (0.7 to 1.0). Age specific SMR for all causes and CVD were however slightly raised among the younger (15–34) age groups at 1.2 (0.9 to 1.6) and 1.5 (0.8 to 2.4) but significantly lower among the older (65+) age groups at 0.8 (0.7 to 0.9) and 0.7 (0.6 to 0.8), respectively.

Conclusions: Describing the health status of this large minority group in Germany may justify developing appropriate intervention programmes for them. The slightly higher mortality in the younger age groups may be explained by the postulated causes of the "mortality crisis" in Eastern Europe while selection effects and better health care may explain the lower mortality in the older age groups.

K. Ladwig¹, B. Marten-Mittag², J. Baumert¹, A. Döring¹, H. Löwel¹.¹GSF-National Research Centre for Environment and Health, Institute of Epidemiology, Munich/Neuherberg, Germany; ²University Hospital of the Technical University, Psychosomatic Department, Munich, Germany

Introduction: 0besity is associated with an increased risk of multiple severe disease conditions and substantially reduces life expectancy. Additionally, clinical studies have found a direct relationship between obesity and depression. The association was not entirely confirmed or even rejected in epidemiological studies pointing to obesity as related to low levels of depression which led researchers to surmise that the fat were more "jolly". The aim of the present study was to ascertain whether being jolly or depressed exerted an opposed long term total and cardiovascular mortality risk in obese subjects. We also aimed to assess whether gender effects modify the proposed interaction of depression and obesity.

Methods: Data were derived from three subsequent population based surveys (1984–1995) and from an outcome analysis in 1998 of the prospective population based MONICA-KORA Augsburg Cohort Study. We assessed the crude and adjusted hazard ratios of obesity and depressive mood in 3.472 middle aged men and in 2.932 women for the prediction of total mortality. A total of n = 391 male and n = 163 female fatal events were recorded. Multivariate analysis was adjusted for age, survey, cigarette smoking, hypertension, diabetes mellitus, and hypercholesterolaemia Body mass index (BMI) was calculated as weight in kilograms divided by height in square metres. Trained medical staff following a standardised protocol determined body height and body weight. Obesity was defined as BMI 30 kg/m². Subjects with a BMI of <18.5 were excluded from the analysis. Depressive symptomatology was assessed using an eight items subscale from the von Zerssen affective symptom check list ranging from 0 to 3, leading to a Likert-like scoring range of 0–24.

Results: In comparison to a subgroup of non-depressed and non-obese subjects, men with a BMI >30 (n = 786) but without signs of depression yielded a crude Cox proportional hazard ratio (HR) to predict total mortality of 1.08 (95% CI 0.76 to 1.54) and an adjusted HR of 1.05 (95% CI 0.74 to1.49) which was not significant. However, obese male subjects in the highest stratum of depression yielded a crude age adjusted HR to predict total mortality of 2.31 (95% CI 1.55 to 3.44; p0.0001) and an adjusted HR of 1.94 (95% CI 1.29 to 2.93; p = 0.002). For women, the HR to predict total mortality in the high-risk subgroup was 1.77 (95% CI 0.99 to 3.15; p = 0.054) for the crude model and was 1.49 (95% CI 0.81 to 2.73; p = 0.149) for the adjusted model.

Conclusions: This is the first prospective population based study to show that depression and obesity interact significantly and yield markedly different mortality risks. The effect in men is more pronounced than in women.

B. Larroque¹, A. Burguet², S. Marret³, M. Kaminski¹.¹INSERM, U149, Villejuif, France; ²Hopital Charles Nicolle, Service de Néonatologie, Rouen, France; ³Hopital Jean Bernard, Service de Néonatologie, Poitiers, France

Objective: To assess developmental outcome of very preterm children at 1 year in a large population based cohort by questions to parents in a postal questionnaire and to examine the influence of gestational age, gender, intrauterine growth retardation, and cerebral lesions at neonatal ultrasound on infant development.

Methods: All very preterm births in nine regions of France were included; 2276 parents of infants discharged home agreed to a follow up (96%) and 1855 (81%) answered the postal questionnaire before the child was 20 months old. Non-responders did not differ from responders for neonatal morbidity of the child, but they were more often socially disadvantaged. Questions on development were grouped in four scores adding succeeded items: motor, coordination, language, and socialisation. A ratio was calculated by dividing the score of the child by the mean score of children of the same age in months. A delay was defined as a ratio lower than the 20th percentile.

Results: The incidence of a delay increased with decreasing gestational age for motor score: from 32% of delay at 24–25 weeks to 15% at 32 weeks. The same pattern was seen for coordination and socialisation scores. An intrauterine growth retardation was significantly associated with a delay in motor score (odds ratio (OR) adjusted for gestational age 2.1; confidence interval (CI) 1.1 to 3.0). Boys had more delay in language (1.5; 1.1 to 2.0) and socialisation scores (1.5; 1.2 to 1.9) than girls. Cerebral lesions at neonatal ultrasound were related with a delay in the four areas of development, with an increased frequency of delay with increasing severity of ultrasound lesions: 63% motor delay for major white matter disease, 27% for echo densities, ventricular dilatation, or intraventricular haemorrhage of grade 3, 22% for intraventricular haemorrhage of grade 1 or 2, and 15% for no ultrasound lesions.

Conclusion: A postal questionnaire with simple questions on developmental skills to parents can be a useful method to obtain basic information on child development in a survey. Developmental delay was related to lower gestational ages and severity of cerebral lesions.

B. Larroque, K. Supernant, L. Marchand.INSERM, U49, Villejuif, France

Objective: To determine mortality and development in the first year based on gestation type (multiple or single). A cohort of very preterm births was examined.

Methods: The sample was composed of 3673 live births or stillborn (28% multiple gestation) which occurred before 33 weeks births in nine regions of France in a 1 year period. Of these, 2459 were discharged home. For 1877 infants (79%), parents answered a postal questionnaire sent at 1 year of corrected age. Questions on development skills were grouped into four scores: motor, coordination, language, and socialisation. Non-responders did not differ from responders for neonatal morbidity of the child, but they were more often socially disadvantaged.

Results: A multiple pregnancy was associated with a lower rate of stillbirth (14% versus 24%, p = 0.001) than singleton. Among live born infants, a trend to a lower survival was seen in infants of multiple than single gestation (83% versus 86% p = 0.05). After taking into account gestational age, antenatal steroids, gender and intrauterine growth retardation, this difference was significant (odds ratio (OR) 0.7; confidence interval (CI) 0.5 to 0.9, p = 0.01). Infants born from multiple pregnancy had a higher percentage of motor delay (23% versus 18%, p = 0.01), and socialisation score (21% versus 17%, p = 0.04) which was still significant after controlling for gestational age, gender, bronchodysplasia, cerebral lesions, intrauterine growth retardation. For language delay (16% versus 13%, p = 0.08) and coordination delay (17% versus 14%, p = 0.17), the percentages were slightly higher among multiple than singleton births, but the difference was not significant.

Conclusion: Very preterm infants from multiple gestations have a slight increased risk for mortality or delay of development in the first year compared with infants from single pregnancy.

B. Larroque, M. Delobel.INSERM, U49, Villejuif, France

Objective: To determine behavioural outcome at 3 years age in very premature infants in a regionally defined, prospective cohort study.

Methods: The Epipage study includes all live born infants of <33 weeks’ gestational age, born in 1997 in nine regions of France and a control group of term infants; 2276 (96%) parents of premature infants discharged home and 557 parents (84%) of term infants agreed to a follow up. A postal questionnaire was answered at 3 years old for 1880 (83%) very preterm survivors and 453 (81%) term infants. Non-responders did not differ from responders for neonatal morbidity of the child, but they were more often socially disadvantaged. Behaviour was assessed with the Strengths and Difficulties Behaviours Questionnaire. We include in this analysis 1228 singleton premature infants without severe neurodevelopment disabilities and 447 term infants.

Results: On the behaviour questionnaire, the very premature infants showed significantly higher difficulties than term children: 20% versus 9% for total difficulties score, 20% versus 11% for hyperactivity score, 16% versus 10% for conduct problems score, 15% versus 10% for emotional symptoms score, 14% versus 7% for peer problems score, 15% versus 11% for pro social behaviour score. Boys had higher total difficulty and hyperactivity scores. A lower socioeconomic status was related to a higher frequency of behavioural problems, which was significant for total difficulty score, conduct problems score and pro social behaviour score. After controlling for gender, mother’s age, and socioeconomic and marital status, most differences between very preterm and term children were still significant: for total difficulties (odds ratio (OR) 2.3; 95% confidence interval (CI) 1.6 to 3.3; p = 0.001), hyperactivity (OR 2.0; CI 1.4 to 2.8; p = 0.001); conduct problems (OR 1.7; CI 1.2 to 2.4, p = 0.006), emotional symptoms (OR 1.6; CI 1.1 to 2.4; p = 0.007); and peer problems (OR 1.9; CI 1.3 to 2.9; p = 0.002). These differences in behaviour persisted when additional controls were added for health or development of the child.

Conclusion: The prevalence of behavioural problems assessed by parents at 3 years of very premature infants was higher than in term infants.

R. Laurenti, M. Jorge, S. Gotlieb.Faculdade de Saúde Pública USP, Epidemiologia, São Paulo, Brazil

Background: During pregnancy or within a period of up to a year afterwards (the so-called expanded pregnancy-puerperal cycle), the woman may die due to known obstetric causes and there is a higher risk of death from non-obstetric causes. The American Centers for Disease Control and the American College of Obstetricians and Gynaecologists use the terminology "pregnancy associated death" to identify such events.

Objective: To compare the distributions of deaths of women of fertile age (10–49 years) according to the underlying causes and their actual conditions (if they were in the expanded pregnancy puerperal cycle or not).

Methods: The investigation of the mortality of women between 10 and 49 years old, following the Reproductive Age Mortality Survey methodology, in the 26 Brazilian capitals of state and the federal district (Brasilia), made it possible to evaluate the actual causes of death. The population of study was taken from death certificates of the residents of these areas, deaths that occurred in the first semester of 2002. For each case, a household interview was performed with the family of the deceased, and one with physicians that handled the case. Medical records and autopsy reports were consulted and data were copied. Using all the additional information, a new death certificate was filled for each case.

Results: There were a total of 7332 female deaths, of which 6869 had not been pregnant or given birth within a year before the death (Group 1). Among the women that were in the expanded pregnancy puerperal cycle, 239 died of obstetric, and 224 of non-obstetric causes (Group 2). It was noted that 23.2% and 63.8% of these deaths, respectively, were of women aged 30 years. The distribution of cases according to the underlying cause (chapters of the IDC-10) showed that external causes were responsible for 33.5% of the deaths in the Group 1, and 15.5% in Group 2. Deaths due to infectious disease and respiratory system problems occurred more frequently in Group 1 than Group 2 (16.5% and 12.17% versus 7.1% and 4%, respectively). In relation to external causes, it was found that homicide was responsible for 19.2% of the deaths of pregnant women, and 5.8% of the non-pregnant. Suicide was present in 7.1% and 2.6%, respectively, making it possible to evaluate the association of depression with these cases.

Conclusions: The important findings of this study open new perspectives for the pursuit of researches and actions in the area of the woman’s health, mainly in relation to the cases of non-obstetric deaths, in the expanded pregnancy puerperal cycle.

M. Leal, S. Gama, C. Cunha.National School of Public Health-Oswaldo Cruz Foundation, Department of Epidemiology, Rio de Janeiro-Brazil

Objective: To analyse sociodemographic and prenatal and childbirth care inequalities and their repercussions on birth weight.

Methods: This study was based on a sample of 10 072 postpartum women treated in public maternity hospitals, hospitals contracted out by the Unified National Health System, and private hospitals in the municipality of Rio de Janeiro in 1999–2001. To verify the association between maternal characteristics and birth weight, multiple linear regressions were performed, stratifying postpartum women by level of schooling. The bootstrap method was used to obtain non-biased estimates of the model’s performance, and accurate confidence intervals for the estimates of effects.

Results: For both groups of schooling, birth weight was associated directly with the modified Kotelchuck index and gestational age and inversely with history of prematurity. In the sub-group with less schooling, birth weight was inversely associated with smoking and skin colour (children of white mothers had the highest mean birth weight, while those of black mothers had the lowest). In the sub-group with more schooling, birth weight was inversely associated with the number of hospitals visited before succeeding in being admitted for labour and childbirth. The extreme ranges of maternal age and parity displayed different behaviours from the central range of data, and thus quadratic terms were used.

Conclusions: The variables that explained birth weight in the newborns of mothers with more schooling in RJ were eminently biological, as opposed to the strong presence of social determinants among women with less schooling. In addition, prenatal care exerted a protective role, and smoking had a negative effect, regardless of the mother’s level of schooling.

M. Leal, S. Gama, C. Cunha.National School of Public Health-Oswaldo Cruz Foundation, Epidemiology, Brazil

Objective: To examine prenatal care in the city of Rio de Janeiro, Brazil, in a sample of 9920 single birth mothers. A modified Kotelchuck index (KI) was used.

Methods: Multivariate ordinal logistic regression and multivariate linear regression were used to estimate the importance of selected sociodemographic and obstetric factors for prediction of prenatal services use and its effects on birth weight. A modification of the KI included as the inadequate category both pregnant women initiating prenatal care after the fourth month, having attended more than 50% of the consultations, and women initiating before that period but attending less than half of the consultations.

Results: Only 38.5% of mothers were classified as having adequate and intensive use levels. Educational level (odds ratio (OR) 2.0), living with the newborn’s father (OR 2.0), attempted abortion (OR 2.0), diabetes (OR 1.9), satisfaction with pregnancy (OR 1.8), race (OR 1.8), parity (OR 1.7), age of the mother (OR 1.4), and place of residence (OR 1.3) were maintained as predictors of modified KI controlled for other variables. For RLM, age of the mother, living with the newborn’s father, smoking, gestational age, parity, previous loss of a child, diabetes, and modified KI were associated with birth weight.

Conclusion: Adequate use of prenatal care in the city of Rio de Janeiro was associated with birth weight, worse social and educational conditions, family support, and the obstetrics risk of the mothers.

A. Leclerc¹, J. Chastang¹, G. Menvielle¹, D. Luce¹, B. Geoffroy-Perez², E. Imbernon², E. Jougla³, I. Robert-Bobée⁴, M. Saurel-Cubizolles⁵.¹INSERM, U88, St-Maurice, France; ²InVS, DST, St-Maurice, France; ³INSERM, CépiDc, Le Vésinet, France; ⁴INSEE, Paris, France; ⁵INSERM, U149, Villejuif, France

Introduction: An increase in social inequalities in mortality rates during the past decades has been reported for several European countries. In France these inequalities are larger than in other European countries; however, the changes over time have been only partly documented.

Objective: To compare the magnitude of socioeconomic inequalities among adults in France between four different periods, from 1968 to 1997.

Methods: The results are issued from a longitudinal national dataset from INSEE (in charge of censuses) linked with causes of deaths from CépiDc, INSERM (in charge of mortality data). Four sub-cohorts comprising about 1% of the French population aged 30–64 years were defined at a census, in 1968, 1975, 1982, and 1990, and the deaths in a 7 year follow up period were recorded. The relative gap in death rates between upper and lower socioeconomic groups in a census was quantified for each of the four periods with a RII (relative index of inequality) using a Cox model.

Results: In analyses restricted to active men, the magnitude of socioeconomic inequalities remained rather stable over the period. However, considering the whole male population, an increase in socioeconomic inequalities from 1968 to 1997 was observed. The contribution of unoccupied and early retired to social inequalities explained this change over time. Around 1970, the percentage of the population in these categories was about 10%, with a high rate of mortality. In 1990 more than 15% of the population belonged to these categories, and the mortality in these subgroups of the population has not decreased in the last 30 years. The magnitude of socioeconomic inequalities was smaller for women than for men. In a context of increasing unemployment, two mechanisms can explain the observed changes: an increase in health selection in various activity sectors, and long term effects of unemployment.

Conclusions: Socioeconomic inequalities remain larger in France than in other European countries, especially for men. In addition to inequalities within the active population, inequalities associated with employment status are increasing. The results highlight the importance of methodological aspects in comparisons between periods.

A. H. Leyland.University of Glasgow, MRC Social and Public Health Sciences Unit, Glasgow, UK

Introduction: Cardiovascular disease (CVD) is known to have a pattern according to both individual social class and area deprivation. Moreover, smoking is a known risk factor.

Objective: To explore the geographical patterning of CVD in Scotland and its association with deprivation, and in particular questions whether the association with area deprivation reflects the geographical distribution of individual social circumstances (measured by social class) or of individual risk factors (smoking). A further objective was to examine the effects of social class and smoking when aggregated to the area level.

Methods: The data relate to 8804 adult interviewees aged 18–74 years clustered in 312 small areas, with complete data for smoking and CVD, taken from the 1998 Scottish Health Survey. The response considered was the presence of a cardiovascular condition, which relied on the self reporting of doctor diagnosed conditions (including heart attack, angina, and stroke). Occupation was classified into professional, intermediate, and manual or missing. Smoking was grouped into never smoked, ex-smokers, light, moderate, and heavy. Area deprivation was measured using the Carstairs score (an external measure derived from the 1991 Census). The data were analysed using multilevel logistic regression, and all analyses adjusted for the respondent’s age and sex.

Results: The expected gradient was seen for social class, with higher CVD prevalence among manual workers (odds ratio (OR) 1.19, 95% confidence interval (CI) 1.04 to 1.36) relative to professionals. Only ex-smokers showed significantly increased odds of CVD relative to non-smokers (OR 1.18, 95% CI 1.03 to 1.33). The effect of area deprivation was substantial, with a third of the population in the most deprived areas having an OR of 1.32 (95% CI 1.16 to 1.49) compared with the least deprived. The social class gradient disappeared when area deprivation was included, but both area deprivation and individual smoking remained significantly and independently associated with CVD. When individual social class and smoking were aggregated to the area level they showed strong correlations with the area deprivation measure (0.69 and 0.70), and CVD prevalence was strongly associated with both measures.

Conclusions: The lack of a relationship with smoking, apart from among the ex-smokers, is not surprising in a cross sectional study examining prevalence rather than incidence. Despite the lack of an individual relationship, individuals who lived in areas with higher rates of smoking were more likely to have CVD. Given the strong associations with CVD and the Carstairs score, aggregated individual characteristics may provide adequate substitutes for external deprivation measures where these are unavailable. Although there was a strong relationship between CVD prevalence and area deprivation, considerable unexplained geographical variation remained. Neither individual social class nor individual risk factors could explain the relationship between CVD and area deprivation.

J. Lindert, M. Blettner.University, IMBEI, Mainz, Germany

Introduction: The number of people living as refugees has grown over the past several decades. Manmade disasters have a severe impact on health across men and women and across cultures. The impact of violence on human health is an important area of epidemiological research. There is conflicting evidence about gender differences of the prevalence of post-traumatic stress disorder symptoms (PTSD). Female preponderance in prevalence, incidence and morbidity risk for posttraumatic stress disorder is under review.

Objectives: To assess the prevalence of morbidity of Kosovan refugees living in Germany and to analyse gender differences in symptom levels; to unmask differences in morbidity between men and women by assessing carefully symptoms.

Methods: A cross sectional cluster sample survey was conducted among refugees living in Germany. Main outcome measures were PTSD symptoms, depression, and anxiety. Using the Harvard Trauma Questionnaire and the Hopkins Symptom Checklist 25, 99 refugees were interviewed face to face. Main biographical data were collected.

Results: Men and women were exposed to high levels of adversity; most commonly reported events were forced expulsion and deprivation of water, food, and shelter. Values higher than the threshold level were reported by 41.9% of the sample. No differences in symptom levels between men and women were found.

Conclusion: High levels of impairment are likely to occur in populations affected by mass violence. PTSD seems to be as likely in men as in women after exposure to manmade disasters. Modelling PTSD as a unidimensional construct seems to mask symptom differences between men and women. Further research is needed to get to know better the impact of manmade disasters on health of men and women.

C. Lopes¹, A. Gaio², A. Santos¹, H. Barros¹.¹University of Porto Medical School, Hygiene and Epidemiology, Porto, Portugal; ²Faculty of Sciences, University of Porto, Pure Mathematics, Porto, Portugal

Introduction: Epidemiological studies on diet and coronary heart disease mainly focused on the effect of single nutrients and foods, but the effect of overall dietary patterns on the disease remains to be elucidated. South European countries, where specific dietary patterns are well recognised, may add useful information for understanding the risk relationship of diet to heart disease.

Objective: To evaluate the associations between overall dietary patterns and acute myocardial infarction (AMI), using a population based case–control study

Methods: Cases were 511 consecutive male patients, older than 39 years, with a first non-fatal myocardial infarction, admitted to the coronary intensive care unit of a major teaching hospital. Community controls were 599 individuals from the same catchment area, with no evidence of infarction, selected by random digit dialling (participation proportion 70%). Trained interviewers using a validated 86 item semi-quantitative food frequency questionnaire obtained dietary intake. Five dietary patterns were identified using independent component analysis on the caloric contribution of 25 chosen food groups: (1) higher consumption of vegetables, fish, beans, and soup; (2) higher consumption of alcoholic beverages, especially beer and spirits, and lower consumption of vegetables and fruits; (3) low caloric intake; (4) higher consumption of wine and red meat; and (5) higher consumption of milk. Odds ratios were calculated using unconditional logistic regression.

Results: Cases were significantly younger and less educated than controls, and significantly more often reported a family history of AMI (29.9% versus 19.4%), hypertension (40.7% versus 30.7%), diabetes (14.1% versus 7.0%), and angina (5.1% versus 2.5%). Cases were also significantly more often current smokers (56.1% versus 30.7%) and less frequently engaged in leisure time physical activities (20.4% versus 41.6%). In crude analysis and considering pattern 1 as reference, we found an increased risk of AMI for every other dietary pattern, the odds ratio (95% confidence interval) being 4.70 (2.21 to 9.98), 2.04 (1.11 to 3.76), 3.65 (1.98 to 6.76), and 1.95 (1.01 to 3.75), respectively. After adjusting for age, education, leisure time physical activity, smoking, and family history of MI, the increased risk remained, with slight modification of the odd ratios: 2.72 (1.19 to 6.23), 2.67 (1.37 to 5.21), 3.21 (1.64 to 6.27), and 2.28 (1.11 to 4.67) respectively for patterns 2–5.

Conclusions: This study identified a dietary pattern clearly associated with a lower risk of AMI, characterised by higher consumption of vegetables, fish, beans, and soup. The other four patterns identified according to dietary choices showed no different effect on AMI risk.

C. Lopes¹, E. Faerstein¹, D. Chor², G. Werneck¹.¹State University of Rio de Janeiro, Department of Epidemiology, Social Medicine Institute, Rio de Janeiro, Brazil; ²Oswaldo Cruz Foundation, Department of Epidemiology, National School of Public Health, Rio de Janeiro, Brazil

Introduction: Violence exposure has been identified as an important stressful life event associated with mental disorders. In large Brazilian cities such as Rio de Janeiro, violence is reaching epidemic levels, affecting society as a whole; however, this issue is largely unexplored in relation to mental disorders.

Objective: To investigate the role of victimisation (hold up/theft and physical aggression) on the incidence and maintenance of common mental disorders (CMD), after two years.

Methods: A prospective cohort study (Pró-Saúde Study) was conducted, in which socioeconomic and demographic factors, as well as being a victim of physical aggression and/or hold up/theft were measured at baseline (phase 1–1999), and CMD were measured at phases 1 and 2 (2001). Participants were civil servants of a university in Rio de Janeiro, aged 20–75 years at baseline, of whom 55% were women and 45% men. The overall response rate for the study at follow up was 81%, comprising 3252 subjects. A standardised self administered questionnaire assessed the occurrence of victimisation in the previous 12 months, and the socioeconomic and demographic characteristics of the population. Common mental disorders were assessed using the General Health Questionnaire (GHQ-12), a self assessed measure of psychiatric morbidity. Logistic regression was used to adjust for potential confounding variables and to examine the independence of the associations.

Results: The risk of CMD after 2 years of follow up was higher for those who had suffered a physical aggression (odds ratio (OR) 2.2; 95% CI 1.3 to 3.7) or had been a victim of hold-up/theft (OR 1.5; 95% CI 1.1 to 2.1). Having suffered physical aggression was also associated with a higher risk of maintenance of CMD (OR 1.7; 95% CI 0.9 to 3.2). The analysis performed separately for men and women, showed that the association between physical aggression and incidence of CMD was stronger for women (OR 3.1; 95% CI 1.5 to 6.4) than for men (OR 1.4; 95% CI 0.6 to 3.4). For maintenance, the pattern was the same, and physical aggression was only associated with CMD in women (OR 2.3; 95% CI 1.0 to 5.5). Being a victim of hold up/theft was associated with a higher risk of onset of CMD for men (OR 1.6; 95% CI 1.0 to 2.7) when compared with women (OR 1.4; 95% CI 0.9 to 2.1); but no association was found for both men and women for upholding CMD.

Conclusions: Being a victim of physical aggression was strongly associated with both onset and maintenance of episodes of CMD in women only. Being a victim of hold up/theft increased the risk of onset of CMD but not their maintenance; this effect is stronger for men than for women. Differences in the impact of type of victimisation in the risk of CMD for women and men need to be better understood, as this study does not evaluate in which context victimisation occurred.

B. Lumbreras-Lacarra¹, I. Jarrín-Vera¹, I. Ferreros², I. Hurtado², S. Pérez-Hoyos², M. García-De La Hera¹, J. Del Amo¹, I. Hernández-Aguado¹.¹Medicin School of University of Miguel Hernández, Public Health, Alicante, Spain; ²EVES, Valencia, Spain

Objective: To describe the changes in the mortality rates and the distribution of the causes of death between 1987 and 2001 among a cohort of intravenous drug users (IDUs).

Methods: Cohort study of 6178 IDUs recruited between January 1990 and December 1996 in three AIDS prevention and information centres for voluntary counselling and testing in Valencia (Spain) and followed up until November 2001. Sociodemographics, drug using behaviours and HIV status were obtained at study entry. Causes of death were ascertained from death registries and classified as AIDS related deaths, deaths due to liver disease, drug related deaths, deaths due to cardiovascular disease, violence related deaths, and tumours. Cause specific rates were calculated for periods before and after 1997 (date of introduction of highly active antiretroviral therapy (HAART) in Spain) and were adjusted through Poisson regression assuming independence. Rate ratios (RR) were adjusted by sex, age at first visit, and duration of drug use. The seroconvert group was excluded of the statistics analysis because of its sample size.

Results: Median age at study entry was 26 years (interquartile range (IQR) 23 to 30); median duration of addiction was 6 years. IDUs comprised 77.4% men and 22.6% women; 3320 (53.7%) were HIV negative, 2614 (42.3%) HIV positive, and 244 (3.9%) had seroconverted (IQR 2.3 to 10.0). Median follow up time was 8.7 years (IQR 6.7 to 10.7). There were 52 099.5 total person years of follow up time and 1.091 deaths. Overall death rate was 20.94/1000 person years (19.73 to 22.22), with a lower risk of death after 1997 (RR 0.85; 95% CI 0.75 to 0.95). Mortality rate was 10.45/1000 person years (9.34 to 11.71) in HIV negative cases and 35.45/1000 person years (32.99 to 38.10) in HIV positive. After 1997 the risk of death decreased in HIV positve cases (RR 0.73; 95% CI 0.63 to 0.85); but increased in HIV negative (RR 1.35; 95% CI 1.07 to 1.70). AIDS related deaths declined after 1997 (RR 0.55; 95% CI 0.47 to 0.67); however, deaths due to cardiovascular disease (RR 2.39; 95% CI 1.28 to 4.46) and tumours (RR 2.08; 95% CI 1.14 to 3.77) showed the opposite trend. Death rates due to other causes (drug use, liver disease, and violence related deaths) did not change over the study period.

Discussion: The overall death rate has decreased since the introduction of HAART in1997 in HIV positive patients because of the reduction in AIDS related deaths. Nevertheless, the risks of other causes of death, mainly cardiovascular disease and tumours, have increased. It is essential to continue to monitor all causes of mortality to identify changes.

N. Lunet, C. Lopes, H. Barros.University of Porto Medical School, Hygiene and Epidemiology Department, Porto, Portugal

Introduction: The decline in gastric cancer incidence in industrialised countries, the variation in incidence rates across geographical areas, and the effect of migration on risk within two generations point towards the aetiological role of environmental factors. In Portugal, the stomach remains a major cancer location and the decline in gastric cancer is slower than expected.

Objective: We performed the first epidemiological investigation on nutrition and gastric cancer in a Portuguese population.

Methods: Gastric cancer cases were identified in two major hospitals from the northern region of Portugal (Hospital de S. João, Porto, and Instituto Português de Oncologia Francisco Gentil, Porto) between January 2001 and December 2003. Population controls were recruited by random digit dialling using households as the sampling frame, followed by simple random sampling to select one eligible person among permanent residents in each household. Participants were questioned by trained interviewers and completed a structured questionnaire including sociodemographic, dietary and other lifestyle characteristics. A validated food frequency questionnaire was used to evaluate food intake pertaining to the previous year or in the year before onset of symptoms when applicable. Preference for salty foods and access to refrigeration was assessed using specific questions. Participants with cancer history, cognitive impairment (after the Mini Mental State Examination), or who changed their dietary habits were excluded. Data from 287 cases with histological confirmation and 1096 controls were included in the analysis. Odds ratios (OR) and 95% confidence intervals (CI) adjusted for age, sex, education, smoking status, and total caloric intake were computed by unconditional logistic regression, using Stata®.

Results: Using the lowest quartile of dietary intake as the reference category, the OR for the highest quartiles was 0.5 (95% CI 0.33 to 0.80) for raw vegetables, 0.5 (95% CI 0.35 to 0.79) for fruits, and 2.0 (95% CI 1.37 to 2.95) for pulses. Compared with current non-drinkers OR for the consumption of two or more alcoholic drinks per day was 1.7 (95% CI 1.17 to 2.54). Participants in the highest quartile of preference for salty foods had an increased risk of gastric cancer (OR 2.7; 95% CI 1.46 to 4.85), as did those who lived in a house without a refrigerator at the age of 20 years (OR 1.8; 95% CI 1.03 to 3.26).

Conclusion: Gastric cancer risk was lower in consumers of higher amounts of fruits and raw vegetables while preference for salty foods and lack of access to refrigeration was associated with an increased risk. Consumption of pulses and alcoholic beverages was more frequent in cancer patients. These findings favour an important role for social determinants in gastric cancer epidemiology.

B. Manktelow, K. Abrams, E. Draper.University of Leicester, Department of Health Sciences, Leicester, UK

Introduction: The performance of healthcare providers is often summarised using the ratio of the number of observed events to the expected number. The expected number of events is usually estimated by using the coefficient estimates from a logistic regression model using a large reference dataset. A variety of methods have been suggested for estimating confidence intervals for such ratios and it is unclear which method is the most appropriate, particularly for small datasets where the assumptions made may not hold.

Methods: The performance of six proposed methods for estimating confidence intervals was investigated and compared using simulated data. For two of these methods (normal approximation and bootstrap) only uncertainty from the observed number of deaths was allowed for. The four other methods additionally included uncertainty from the estimated coefficient estimates used to calculate the expected number of events: bootstrap, a Bayesian approach using MCMC, and two proposed extensions to the Normal approximation proposed by Hosmer & Lemeshow (1995) and Zhou & Romano (1997). For each simulation, two datasets were created: a small dataset to represent the healthcare provider of interest and a large dataset to represent the reference data. Morbidity scores were included, with the same empirical distribution for each dataset, and observations were then sampled using a logistic model with a known linear predictor. These observations were used to estimate the outcome ratio and confidence intervals. Simulations were made of 27 different scenarios, with 1000 repetitions in each case, varying: (a) the size of the dataset of interest; (b) the size of the reference dataset; (c) the underlying probability of death within the dataset of interest. For each scenario the type I error rates and coverage were calculated and compared. Each method was also illustrated using mortality data from 16 UK neonatal intensive care units.

Results: The methods based on the Normal approximation tended to have inflated error rates when the true ratio was greater than unity and reduced rates when the true rate was less than unity. Of these methods, that proposed by Hosmer and Lemeshow performed best, although the error tended to be one sided, especially when the observed number of events was small. Both of the bootstrap methods and the Bayesian model produced more consistent type I error rates.

Conclusions: This study provides evidence that the commonly used method based on the normal approximation performs poorly, particularly when the true standardised ratio differs from unity. The methods based on the bootstrap were straightforward to implement and produced more appropriate type I error rates. The Bayesian approach also performed well and has a number of potential advantages, including allowance for a wider range of model uncertainty and the ability to be extended to more complex scenarios.

C. Marinacci², M. Mangia³, P. Rossi¹, P. Borgia¹.¹Agency for Public Health, Lazio Region, Rome, Italy; ²Epidemiology Unit, Piedmont Region, Grugliasco, Italy; ³Local Health Unit, Roma B, Rome, Italy

Objective: To discover the proportions of compliance to colorectal cancer screening (CRCS) with two types of primary screening test: faecal occult blood test (FOBT) and flexosigmoidoscopy (FS); and to evaluate the role of sociodemographic factors in determining participation.

Methods: This randomised two arm trial was conducted in two health districts of Rome. The units of randomisation were 20 the practice with an average target population of 150 beneficiaries, 50–74 years old. All the individuals were classified according to the participation to the proposed screening as follows. Contacted: subjects who actually received the letter, not contacted: subjects who were unknown at the given address; participants: subjects who either booked an FS or picked up an FOBT kit; performers: subjects who actually performed the screening test, non-performers: subjects who booked an FS or picked up an FOBT kit but did not do the test. The analysis was intention to treat. The effect of individual sociodemographic variables (age, sex, socioeconomic level, length of time of being registered with a doctor) and organisational variables (provider of test, posting of follow up, distance between residence and centre of reference) on compliance was analysed.

Results: There were 1447 individuals referred for FOBT and 1461 for FS. The percentage of non-contacted was similar (6.1% versus 6.9%). FOBT obtained a higher level of participation (21.3% versus 8.2%; RR 2.37; 95% CI 1.92 to 2.93). The probability of performing the test, given the participation, was not statistically different (80.8% with FOBT and 90% with FS). The probability of performance decreased as the distance from the centre where the test is provided increased, independently of the type of test offered. Sex, socioeconomic level, and age bracket did not appear to be associated with performance to a statistically significant degree. The socioeconomic status was an effect modifier of the type of test: in lower classes the difference in compliance was smaller.

Conclusions: Our figures confirm the low level of participation and clearly demonstrate the need for active measures to increase compliance. The type of test is a determinant of participation in the screening, which is independent of sociodemographic variables. FOBT can be considered the test ‘most likely to succeed’, in terms of implementation of the screening. Efforts must be performed to bring the screening test as close as possible to the healthy target population.

C. Marinacci¹, P. Schifano², G. Cesaroni², N. Caranci³, A. Russo⁴.¹Piedmont Region, Epidemiology Unit, Grugliasco, Italy; ²Local Health Unit RME, Department of Epidemiology, Rome, Italy; ³Emilia Romagna Region, Regional Health Agency, Bologna, Italy; ⁴Local Health Unit, Epidemiology Unit, Milan, Italy

Introduction: While socioeconomic inequalities in mortality and morbidity measures have been documented, few European studies have evaluated the impact of income in generating a socioeconomic gradient in hospitalisation, acting through a differential incidence of diseases and access to health services. Moreover, little is known on how increasing amounts of income, measured at small area level, affect hospitalisation risk.

Objective: To evaluate the association between small area income and general acute hospitalisation and to analyse the shape of this relationship, with data from four Italian cities.

Methods: Census tract (16 371 areas, median 260 residents) median per capita family income (CTMI) was computed through record linkage between 1998 national tax and local population registries for the cities of Rome, Turin, Milan, and Bologna (total 5.6 million residents). CTMI was assigned to 2 885 541 general acute hospital admissions occurring in the period 1997–2000 among residents in the cities, on the basis of patient’s residence. Within each stratum, defined by gender and broad age group (0–64, 65+ years), standardised hospitalisation rates (SHR) x 1000 residents were computed for quintiles and for 50 quantiles of CTMI distribution. After assigning the CTMI midpoint value to each quantile, a segmented non-linear model was fitted to the 1998 SHR as a function of CTMI quantiles, with three different connected linear parts. Knots and parameters of the three different linear segments were estimated through the Gauss-Newton method.

Results: Hospitalisation rates were showed to strongly decrease with CTMI, with no difference between cities or across the 4 years of the study period (RR 1.575, 95% CI 1.567 to 1.582 and 1.470, 95% CI 1.461 to 1.479 in the lowest versus highest quintile respectively among the 0–64 and 65+ year age groups). Within the younger group, a negligible linear gradient was found between the 1998 SHR and CTMI in the first income range (up to 6700 income per year for males and 7550 for females). Within the second segment (up to 13 840 for both genders), significant decreases in the hospitalisation rates were found with increasing income; the corresponding slope estimates (expressing increasing SHRx1000 for each 100 increase in income) were –0.85 (95% CI –0.97 to –0.74) for males and –0.91 (95% CI –1.01 to –0.81) for females. In the highest income range, no significant relationship was found. A similar pattern was obtained among older men and women, with a steeper gradient within the intermediate income range, compared with the younger age group, and no gradient among extreme ranges.

Conclusions: Small area income seemed to have a strong impact on hospitalisation risk; however, among the poorest and richest, preliminary results showed that increasing amounts of income corresponded to very slight improvement in health, with small differences by gender and age. If confirmed in future analyses, the negligible health impact of highest incomes could be an additional issue supporting redistribution policies.

L. Marin-Leon, M. Barros.UNICAMP, Preventive Medicine, Campinas, Brazil

Introduction: Traffic accidents (TA) are mainly caused by inadequate driving behaviour, especially drunken driving, and frequently involve young people. As accident causes are preventable and as the population at risk is mostly young, it is an important task to monitor the mortality from this cause.

Objective: To analyse if there had been any changes in the mortality from TA in relation to the Traffic Legislation changes in 1998.

Methods: Age standardised mortality rates from municipal districts with 40 000 or more inhabitants in the state of São Paulo were calculated for each district and divided by sex. Deaths coded V01–V89 using ICD-10 were downloaded from the Mortality Information System. The rate difference between 1999–2000 and 1996–1997 was calculated. The correlation of 1999–2000 rates with some sociodemographic variables was assessed.

Results: Although in both periods male rates were higher in all districts (mean M:F rate ratio 4.9 in 2000) the mean reduction in both sexes was 29%. In almost every district with 400 000 inhabitants or more the male rate reduction was bigger than the mean reduction of the districts as a whole. Four districts presented an increment of male rates between 8.7% and 60%. Masculine rates correlated inversely with the number of inhabitants by car and female rates presented positive correlation with illiteracy. Male rates are quite higher than European rates and also are higher than US and Latin American countries.

Conclusions: The difference in gender of the correlation pattern may point to differences in the type of accidents that kill men and women. In some districts, the rates did not decrease as much as the state means and in some they even increased. The reduction in TA mortality in the majority of districts is probably consequent to several measures implemented after the legislation changed.

C. Marino, C. Saitto, D. Fusco, M. Arcà, C. A. Perucci.Local Health Authority RME, Department of Epidemiology, Roma, Italy

Introduction: Few studies have examined the effectiveness of rehabilitation and most considered only particular rehabilitation treatments for specific conditions.

Objective: To investigate the association between rehabilitation treatments and functional gain by diagnostic class and severity of impairment.

Methods: Information on the characteristics of patients and the rehabilitative treatments was collected using an Italian version of the Minimum Datase Post Acute Care. The questionnaire was prepared by the Centers for Medicare and Medicaid Services. For each patient the questionnaire addressed clinical and functional status, diagnosis, rehabilitation, and other treatments received. It was completed at regular intervals throughout the hospital stay. The study included 1918 patients admitted from July 2001 to July 2002 in five rehabilitation facilities. We used factorial analysis to summarise each questionnaire section in a single variable. The functional gain was calculated as the difference between functional status at the beginning and at the end of the hospitalisation. A multiple linear regression analysis was performed to evaluate the association between rehabilitation treatments and functional gain, after adjusting for patient characteristics and severity at admission. As the result of the regression model, we estimated: (1) the effectiveness of the stay—that is, the functional gain predicted by the regression equation; (2) the effectiveness of the stay in the absence of rehabilitation treatments—that is, the functional gain predicted by the regression equation when the treatment variables are set to null; and (3) the effectiveness of the rehabilitation treatments, defined as the difference between 1 and 2. The homogeneity of treatment effectiveness across diagnostic classes and quintiles of severity was analysed.

Results: The highest effect of treatment was observed in multiple trauma, where the median of treatment effectiveness was 0.82 (95% CI 0.65 to 1.23) or 57% of total effectiveness. No effect of treatment is observed in Cardiac disease (median –0.01). Treatments were effective in 81% of the cases, while in 76% the positive effect of treatments combined with the positive effect of the stay in the absence of rehabilitation treatments. The positive effect of treatments opposed spontaneous deterioration of patient functional status in 11% of the cases. Overall, hospital stay was effective in 90% of the cases. The functional gain associated with treatments differed across diagnostic classes and was directly associated with severity of functional status at admission.

Conclusions: It is possible to evaluate the effectiveness of treatment in a population of unselected rehabilitative cases. The functional gain associated with the treatments can be combined in different ways with the effectiveness of the stay in the absence of rehabilitation treatments to determine the overall effectiveness of the hospital stay in post acute rehabilitative care.

Acknowledgements: The study was partially funded by the Italian National Health Service - Progetto "Sviluppo di una classificazione isorisorse per la remunerazione prospettica della riabilitazione" (Ministero della Sanità - Dipartimento della Programmazione "Programmi speciali" - Art.12, comma 2, lett. b), del d. lgs.502/92).

P. Marques-Vidal¹, C. M. Dias².¹Instituto de Medicina Molecular, Unidade de Nutrição e Metabolismo, Lisboa, Portugal; ²Instituto Nacional de Saúde Dr. Ricardo Jorge, Departamento de Epidemiologia e Bioestatística, ONSA, Lisboa, Portugal

Objective: To assess the trends in the prevalence of overweight and obesity in the Portuguese population for period 1995–1999.

Methods: Data from the National Health Surveys, conducted in 1995–6 and 1998–9, amounting to 36 682 men and 61 702 women aged 18 years, were collected. Weight and height were self reported. Obesity was defined as a body mass index (BMI) 30 kg/m²; overweight was defined as 25BMI<30 kg/m².

Results: In men, obesity increased from 10.1 to 11.3% and overweight from 39.4 to 41.9% (Mantel-Haenszel test adjusting for age group: p<0.001) whereas mean (SD) BMI increased from 25.3 (0.1) to 25.6 (0.1) kg/m² (general linear model: p<0.001). In women, overweight remained relatively stable from 31.8 to 31.9% whereas obesity increased from 12.5 to 14.0% (Mantel-Haenszel test adjusting for age group: p<0.001), and mean (SD) BMI increased from 24.8 (0.1) to 25.1 (0.1) kg/m² (general linear model: p<0.001). In men, the increase in the prevalence of overweight was stronger in the Centro region (40 to 46%), while a slight decrease was found in the Algarve (40 to 37%), whereas prevalence of obesity increased in all regions. In addition, the increase in overweight and obesity was comparable across age group, educational level, or employment status. In women, prevalence of overweight decreased in age group 25–45 years but increased afterwards, whereas prevalence of obesity increased or remained stable in all age groups. Finally, no differences in the increase of overweight or obesity were found between regions, educational level, or employment status.

Conclusions: Prevalence of overweight and obesity increased in Portugal during the period 1995–9, and this increase concerned virtually all the subgroups analysed. The health and economic burden of this pathology is currently being analysed.

P. Marques-Vidal¹, C. M. Dias².¹Instituto de Medicina Molecular, Unidade de Nutrição e Metabolismo, Lisboa, Portugal; ²Instituto Nacional de Saúde Dr. Ricardo Jorge, Departamento de Epidemiologia e Bioestatística, ONSA, Lisboa, Portugal

Objective: To assess the trends in the prevalence of diagnosed hypertension in the Portuguese population for period 1987–1999.

Methods: Data from the three National Health Surveys, conducted in 1987–8, 1995–6, and 1998–9, amounting to 55 323 men and 61 702 women aged 15 years, were collected.

Results: In men, prevalence of diagnosed hypertension decreased slightly from 16.9% in 1987 to 15.5% in 1999, respectively (Mantel-Haenszel adjustment for age group: p<0.0001). Mean age at diagnosis was 40 years and remained constant throughout the study period. The strongest decrease was found in the Algarve region (20 to 14%), whereas a slight increase was found in the Northern region (16 to 17%). Differential evolutions were found according to educational levels: decrease among subjects with less than 6 years of education, increase in subjects with 7–12 years, and stability among subjects with more than 12 years of education. The decrease was also more pronounced among employed (14 to 11%) than non-employed subjects (23 to 22%). In women, prevalence of diagnosed hypertension also decreased slightly from 25.8% in 1987 to 23.9% in 1999 (p<0.001) and concerned all age groups. Mean age at diagnosis was 39 years and remained constant throughout the study period. As for men, a slight increase in hypertension prevalence was found in the Northern region (24 to 25%), whereas the strongest decrease was found in the Algarve (30 to 23%). In addition, the same differential trends of hypertension prevalence according to educational level or employment status were found.

Conclusions: Compared with other countries, the prevalence of diagnosed hypertension is relatively low in Portugal, and a slight decrease has been noted. Whether this trend reflects a true decrease in hypertension prevalence or is due to decreased screening remains to be assessed.

J. Marrugat¹, R. Elosua¹, L. Molina², V. Valle³, M. Fiol⁴, M. Fito¹, M. Covas¹, G. Sanz⁵, F. Arós⁶.¹Institut Municipal d’Investigació Mèdica, Unitat de Lipids i Epidemiologia Cardiovascular, Barcelona, Spain; ²Hospital del Mar, Cardiologia, Barcelona, Spain; ³Hospital Germans Trias i Pujol, Cardiologia, Badalona, Spain; ⁴Hospital Son Dureta, Cardiologia, Palma de Mallorca, Spain; ⁵Hospital Clinic, Institut de Malaties Cardiovasculars, Barcelona, Spain; ⁶Hospital Txagorritxu, Cardiologia, Vitoria, Spain

Introduction: Some patients with myocardial infarction (MI) have no classical cardiovascular risk factor (CRF), such as smoking, hypercholesterolaemia, hypertension, ir diabetes. Emerging risk factors (ERF) such as fibrinogen, lipoprotein (Lp) (a), C reactive protein (CRP), antibodies against Chlamydia pneumoniae and homocystein (Hcy) may play a role in patients without CRF.

Objective: To analyse whether ERF may explain the occurrence of MI in these patients.

Methods: The study comprised 25–74 year old MI patients free from CRF admitted within 24 hours after symptom onset in 23 Spanish hospitals and a control group of healthy subjects free from CRF taken from a random representative population sample. Laboratory measurements were centralised and samples taken within 24 hours after symptom onset. CRF were meticulously studied.

Results: There were 104 male MI patients and 155 men in the random population sample who were free from CRF. Characteristics of healthy controls and MI patients were: age 44.2 (14.0) and 61.8 (9.6) years (p<0.001), body mass index 25.45 and 26.67 (p = 0.004), HDL cholesterol 53 (13) and 44 (11) (p<0.001), LDL cholesterol 125 (22) and 120 (31) (p = 0.226), Lp (a)* 19 (8 to 66) and 19 (11 to 39) (p = 0.886), fibrinogen* 280 (239 to 349) and 337 (225 to 396) (p = 0.372), Hcy* 13 (10 to 16) and 11 (8 to 14) p = 0.099), C Reactive protein* 0.60 (0.60 to 0.60) and 0.99 (0.60 to 1.95) (p<0.001), and prevalence of antibodies against C. pneumoniae 70.1% and 87.3% (p = 0.007), respectively (*median (25th to 75th centiles, all other values mean (SD)).

Conclusion: Patients with MI and no CRF have higher levels of ERF compared with general population without CRF. This finding suggests that ERF may have a role in the pathophysiology of atherosclerosis and MI in men.

C. Fernández, I. Vera, J. Baró, A. Cantero, J. Valero, I. Aguado.UMH, Salud Publica, Historia de la Ciencia y Ginecologia, Alicante, Spain

Introduction: Violence is an emerging problem, which is being addressed by the World Health Organization from a public health perspective.

Objective: To measure prevalence and risk factors of school based hostility in a city in Spain from a gender perspective in order to design appropriate preventive strategies.

Methods: This was a multicentre study involving 14 high schools, public and private, selected by maximum diversity, geographical criteria, and willingness to participate. An ad hoc designed questionnaire was administered to all students from 11–18 years from December 2001 to May 2002. Sociodemographic characteristics, and individual and family behaviour was requested. Hostility was measured through an adaptation of previously validated scales and modelled in both continuous and categorical scales. Internal consistency analyses were carried out with Cronbach’s alpha coefficient. A multivariate model was fitted through logistic regression adjusting for school clustering with robust 95% confidence intervals (95% CI) controlling for confounders and checking for interaction.

Results: There were 1924 children, 49.9% boys and 50.1% girls. Median hostility score was higher in boys (9; range 0–52) compared with girls (5; range 0–37), (odds ratio (OR) 0.28; 95% CI 0.21 to 0.37). Age was not associated with hostility in boys (p = 0.665) but was in girls (p = 0.018). There was a bimodal age pattern in girls’ hostility in unvaried analyses, being highest in those aged 13–14 (OR 4.9; 95% CI 0.69 to 36.81) and 15–17 years (OR 2.87 95% CI 0.38 to 21.48) compared with those aged 11–12 years (baseline). Because variables associated with hostility differed by sex, different multivariate models were performed. In boys, having a father without education was associated with a threefold increase in hostility compared with those with a university degree (OR 3.47 95% CI 1.12 to 10.75), as was owning a videogame (OR 2.25 95% CI 1.33 to 3.28). In addition, boys who were consulted about important decisions that affected them were less hostile than otherwise (OR 0.45 95% CI 0.37 to 0.56). For girls, only age and being consulted about important decisions that affected them had statistically significant associations with hostility (OR 0.56 95% CI 0.35 to 0.89).

Conclusions: Hostility is markedly lower in girls compared with boys, and variables associated with school based hostility are also different by gender. Owning a videogame is associated with hostility only in boys, as was having a father with no education. For both boys and girls being consulted at the time of taking a decision was associated with a lower level of hostility. This implies that preventive strategies should be gender differentiated.

C. Fernández¹, F. del Alamo², A. Cantero¹, I. Aguado¹.¹UMH, Salud Publica, Historia de la Ciencia y Ginecologia, Alicante, Spain; ²CIMOP, Madrid, Spain

Introduction: Violence is a complex issue with multiple dimensions, the understanding of which requires multidisciplinary methodologies. The study of violence has been lately analysed within the social epidemiology framework as its components are more influenced by groups than by individuals.

Objective: To explore headteachers’ perception of school based hostility in a city in Spain in order to design appropriate preventive strategies.

Methods: This was a multicentre study involving 14 high schools, public and private, selected by maximum diversity, geographical criteria, and willingness to participate. An ad hoc designed questionnaire was administered to all students from 11–18 years from December 2001 to May 2002. Both headteachers and their students aged 11–18 years were included in the study. The qualitative part of the study involved 14 in depth semi-structured interviews with the high schools’ headteachers. The topics addressed were: (a) perception of the magnitude of the school based hostility in their centre; (b) opinion on the causes of school based hostility; (c) evaluation of the positive and negative aspects of their violence prevention school policies; and (d) recommendations to develop violence prevention policies. Discourse analyses were performed by textual data from interview transcriptions.

Results: All headteachers agreed that the level of violence is low, occasional, and among well characterised individuals. The type of violence is mainly "verbal" with insults and threats (bullying) and very occasionally fights among well characterised individuals. It is noteworthy that headteachers label school absenteeism, lack of interest, and unpunctuality (thus characterised as behavioural disorders) as hostile to other students and teachers. Regarding causes, the commonest cause mentioned is age and is felt as temporary as the violence disappears as they grow older. Poor language skills are also highlighted: "to compensate for their poor language and communication skills, they fall more easily into violent behaviours". Boys are labelled as more violent than girls; however, hostility is intra-gender, with boys bullying boys and girls being hostile to girls. Hostility is also related to social violence disclosed in TV and video games and to poor parenting. They all acknowledge carrying out violence prevention policies and are content with them. Their recommendations for future actions include the demand for resources and materials to develop them.

Conclusions: Headteachers perceive that school based violence in this setting is low and thus prefer the better term of hostility. They also identify behavioural disorders as a form of hostility. Most of the identified causes for hostility are external to the schools and highlight different age and gender patterns; poor language skills are recognised as an important cause.

M. Á. Martínez-González¹, Á. Alonso¹, A. Sánchez-Villegas², C. Fuente¹, M. Bes-Rastrollo¹, J. Irala-Estévez¹.¹University of Navarra, Department of Epidemiology and Public Health, Pamplona, Spain; ²University of Las Palmas, Division of Preventive Medicine and Public Health, Las Palmas de Gran Canaria, Spain

Introduction: Many components of Mediterranean diet have been independently associated with a lower risk of hypertension. However, whether an overall Mediterranean dietary pattern, with a high amount of energy from fat, is associated with a lower risk of hypertension has not been assessed.

Objective: To evaluate the relationship between an a priori defined Mediterranean dietary pattern and the risk of hypertension in a Spanish cohort study.

Methods: The SUN (Seguimiento Universidad de Navarra) Study is an open enrolment cohort currently including 13 500 university graduates, recruited and followed up through biennial postal questionnaires. Diet was evaluated at baseline with a semi-quantitative food frequency questionnaire, previously validated in Spain. In the first follow up questionnaire, information about a new physician made diagnosis of hypertension was requested. To build a diet score, eight energy adjusted items were categorised in quintiles and the value of the quintile (+1 for the first quintile to +5 for the fifth) was directly imputed for olive oil, fruits, vegetables, fibre, alcohol, fish/shellfish, legumes, and low fat dairy products. Another two items (meat/meat products and glycaemic load) were categorised in energy-adjusted quintiles and inversely weighted (+1 for the fifth quintile to +5 for the first). The 10 items were summed, yielding a composite score ranging from 10 to 50 points. Associations between adhesion to this score and the risk of hypertension were assessed using a Cox proportional hazards model, considering the score as a continuous variable or as a categorical variable (<25, 25–29, 30–34, >34), and adjusting for known risk factors for hypertension.

Results: The follow up rate for a median of 28.5 months was 88.4% after five mailings. We identified 144 new cases of hypertension among 4825 participants included in this analysis, accounting for a total study base of 11 003 person years. Compared with those with <25 points in the Mediterranean score, the hazard ratio (HR) (95% confidence interval (CI)) of incident hypertension for those scoring 25–29 points was 0.73 (0.44–1.22), for those scoring 30–34 was 0.62 (0.37 to 1.03), and for those scoring >34 points was 0.74 (0.44 to 1.25). When considering the Mediterranean score as a quantitative variable, the HR for each additional point was 0.98 (0.95 to 1.01; p = 0.18); the inclusion of a quadratic term significantly improved the model. The HR (95% CI) for the linear term was 0.76 (0.63 to 0.91; p = 0.003) and 1.004 (1.001 to 1.007; p = 0.005) for the quadratic term, representing a flattening of the dose–response curve for the highest scores.

Conclusions: An a priori Mediterranean dietary pattern was associated with a lower risk of hypertension in a Mediterranean population after adjustment for main risk factors for hypertension with an apparent threshold effect.

H. Matos¹, N. Duppre², E. Sarno², C. Struchiner².¹UERJ, Informática Médica, Rio de Janeiro, Brazil; ²FIOCRUZ, Instituto Oswaldo Cruz, Rio de Janeiro, Brazil

Introduction: At the turn of the century, leprosy is still a public health problem worldwide, particularly in tropical zones. The leprosy elimination programme sponsored by the World Health Organisation (WHO), which is aimed at the detection and treatment of newly detected cases, has attained a relative success in the reduction of global prevalence of the disease. The WHO’s objective is to reduce the prevalence rate below 1 case/10 000 inhabitants in a worldwide scale until year 2000. Nevertheless, detection rate of new cases continues to grow in some areas, including Brazil, where the prevalence rate is still high (4.33/10 000), as is the detection rate (25.86/100 000). Contact surveillance is a valuable strategy for leprosy control. At the Ambulatory Souza Araújo da Fiocruz, Rio de Janeiro, a dynamic cohort study for leprosy contactants was initiated in 1987, partially financed by WHO. Up to now about 3000 contacts have been registered in the study.

Objectives: To: (a) estimate incidence rates in the cohort; (b) identify major risk factors for infection stage of the disease; and (c) characterise risk factors for disease among followed contacts.

Methods: Statistical models used in epidemiology were utilised to achieve these objectives: these were (a) logistic regression models; and (b) non-parametric, semi-parametric, and parametric survival models for studying incidence.

Results: The incidence rate of leprosy among contacts was estimated as 0.01694 persons/year in the first 5 years of follow up. The starting time for follow up was the moment of diagnosis of the primary case. The factors associated with becoming diseased were: (a) non-vaccination with BCG; (b) a negative Mitsuda reaction; and (c) a multibacillary clinical form of leprosy. In particular, contacts whose index cases had a high bacilloscopic index at the end of treatment (>1) were at risk. Factors associated with infection, evaluated as seropositive reactions for anti-PGL1 IgM, were: (a) young age (<20 years); (b) low Mitsuda reaction score (below 5 mm); and (c) high bacilloscopic index of the primary case.

Conclusions: The main conclusions of this study point to additional strategies for leprosy control in areas with a high incidence of the disease, including vaccination with BCG, detection of anti-PGL1 IgM, and follow up of newly treated patients, especially characterisation of their bacilloscopic index.

S. Medina¹, L. Artazcoz², E. Boldo¹, E. Saeger³, B. Forsberg⁴, A. Paldy⁵.¹National Institute for Public Health Surveillance InVS, Environmental Health, Saint Maurice, France; ²Public Health Agency, Environmental Health, Barcelona, Spain; ³European Commission, Joint Research Centre, Ispra, Italy; ⁴Ümea University, Public Health and Clinical Medecine, Ümea, Sweden; ⁵National Public Health Centre, Environmental Health, Budapest, Hungary

Introduction: The Apheis programme aims to provide European decision makers, environmental health professionals and the general public with comprehensive, up to date and easy to use information on air pollution and public health, so they can make informed decisions about the political, professional, and personal issues they face in this area. For this purpose, Apheis delivers standardised, periodic reports based on health impact assessments in 26 cities in 12 European countries.

Methods: Apheis centres have been created in all cities participating in the programme. Apheis adopted WHO guidelines for environmental health risk assessment, and we developed our own guidelines for gathering and analysing data, and communicating findings on air pollution and its impact on public health. We estimated the acute impact of PM₁₀ and black smoke (BS) on premature mortality using the most recent European exposure–response functions (ERFs). For the chronic impact of PM₁₀ and PM_2.5 on premature mortality we used Kunzli’s 2001¹ and Pope’s 2002² ERFs respectively. Different scenarios on the health benefits of reducing particulate pollution levels are proposed.

Results: The total population covered in this health impact assessment includes nearly 39 million inhabitants of Western and Eastern Europe. BS measurements were provided by 15 cities, Athens showing the highest mean BS levels (77 µg/m³). PM₁₀ measurements were provided by 23 cities; in most of them, mean values were below 50 µg/m³. PM_2.5 measurements were provided by 12 cities, with mean values below 20 µg/m³. Lower levels would improve public health; for example, in the city of Budapest, all other things being equal, reducing annual mean PM_2.5 levels (21 µg/m³) to 15 µg/m³ would prevent around 1700 cases of all cause mortality, 1300 cardiopulmonary, and 222 lung cancer mortality cases in a population of 1.8 million inhabitants.

Conclusions: Apheis has created an active public health and environmental information network on air pollution related diseases in Europe using a standardised methodology. Our findings show that even very small and achievable reductions in particulate pollution levels have a beneficial impact on public health, and thus justify taking preventive action in all cities, no matter how low their levels of air pollution. Our final goal is to deliver the information we provide taking into account the specific information needs of our target audiences.

Acknowledgements: The Apheis programme is supported by the European Commission DG SANCO programme of Community action on pollution elated diseases (contract Nos. SI2.131174 (99CVF2-604)/SI2.297300 (2000CVG2-607])). E Boldo was supported by a grant from the Regional Ministry of Health, Madrid Regional Government, Spain (Orden 566/2001).

1. Künzli N, Kaiser R, Medina S, et al. Public-health impact of outdoor and traffic-related air pollution: a European assessment. Lancet 2000;356:795–801.[CrossRef][Medline]
2. Pope A, Burnett R, Thun M, et al. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA 2002;287:1132–41.[Abstract/Free Full Text]

R. Medronho, I. Valencia, M. Passos, C. Casali, F. Bruno, M. Pereira.Universidade Federal do Rio de Janeiro, Núcleo de estudos de Saúde Colectiva, Rio de Janeiro, Brazil

Objective: To evaluate clinical and epidemiological differences among the serotypes of dengue in Rio de Janeiro’s 2001–2002 outbreak of the disease.

Methods: Of 362 cases that had viral isolation samples, notified by the Information System for Notification Diseases from January 2001 to June 2002, 62 were caused by serotype 1, 62 by serotype 2, and 238 by serotype 3.

Results: Compared with serotype 2, an individual infected by serotype 3 had a 6.07 times higher chance (odds ratio (OR) 6.07; CI 1.10 to 43.97) of presenting with shock and a 3.55 times higher chance (OR 3.55; CI 1.28 to 9.97) of having exanthema. Compared with serotype 1, serotype 3 had a 3.06 times higher chance (OR 3.06; CI 0.99 to 9.66) of causing abdominal pain and a 3.61 times higher chance of exanthema (OR 3.61; CI 1.16 to 11.51).

Conclusions: It was found that individuals infected by the serotype 3 of the virus had signs indicating a more severe disease.

R. A. Medronho¹, Luis I. Valencia¹, M. R. Campos².¹Universidade Federal do Rio de Janeiro/UFRJ, Núcleo de Estudos Colectiva, Rio de Janeiro; ²³Fundação Oswaldo Cruz/FIOCRUZ, Escola Nacional de Saúde Pública, Rio de Janeiro, Brazil

Introduction: Ascariasis is an important problem of public health, especially in developing countries. The environment plays an important role in the transmission of this infection. The high prevalence of Ascaris lumbricoides is associated with precarious sanitary conditions, constituting an important health indicator.

Objective: To estimate risk areas for A. lumbricoides parasitic overload, using geoprocessing and geostatistic methods of analysis.

Methods: A coproparasitological and domiciliary survey was conducted in 19 selected census districts of the state of Rio de Janeiro, Brazil. A sample of 1664 children aged 1–9 years was selected and plotted in their own home centre. Geostatistic techniques allowed spatial exploratory analysis, variographic study, and ordinary kriging. Student’s t test, odds ratio, and confidence intervals were used in the statistical analysis.

Results: A prevalence of 27.5% was found for A. lumbricoides. Household income, the mother’s education level, and the domiciliary conditions were identified as significantly associated factors for occurrence of ascariasis. An isotropic spherical semivariogram model with 150 m reach, contribution of 0.45, and nugget effect of 0.55 was employed in ordinary kriging.

Conclusions: Peridomiciliary impact on ascariasis is confirmed by a spatial continuity of 150 m. Disease occurrence could be estimated in the study area and a risk map elaborated using ordinary kriging.

R. de Mello¹, M. Dutra², J. Ramos¹, P. Daltro³, M. Boechat³, J. Lopes¹.¹Instituto Fernandes Figueira–Fundação Oswaldo Cruz, Departamento de Neonatologia, Rio de Janeiro, Brazil; ²Instituto Fernandes Figueira–Fundação Oswaldo Cruz, Unidade de Epidemiologia, Rio de Janeiro, Brazil; ³Instituto Fernandes Figueira–Fundação Oswaldo Cruz, Departamento de Radiologia, Rio de Janeiro, Brazil

Introduction: The enhancement of neonatal physiology knowledge and medical technology advances has improved the prognosis for critically ill babies. Nevertheless, this improvement may be accompanied by higher morbidity due to prematurity and/or intensive care sequelae. The morbidity among many survivors is still a public health problem, leading to numerous cases of hospitalisation for respiratory diseases in the first year of life.

Objective: To evaluate the association between neonatal risk factors and respiratory morbidity during the first year of corrected age of pre-terms.

Methods: The present study included in the cohort newborn infants of adequate development for their gestational age with birth weights of less than 1500 g and gestational ages of less than 34 weeks admitted to the nursery of a Brazilian hospital from 1998 to 2000. Of 179 babies admitted, 11.2% died during hospitalisation period and in four cases parents refused participation. A total of 58 babies were excluded: 23% were small for gestational age, 4% had genetic syndromes, 4% had congenital malformations, and 1.3% had congenital infections. Before discharge, lung functional tests and high resolution chest tomography (HRCT) were performed. During the first year of life the children received monthly medical follow up and were evaluated for the presence of respiratory morbidity/obstructive airway syndromes, pneumonia, or hospital admission due to respiratory conditions. The association of neonatal risk factors and outcome variables (respiratory morbidity) was independently assessed by relative risk (RR), which was also adjusted through logistic regression.

Results: The cohort constituted 97 premature infants with mean (SD) gestational age of 28.5 (2.3) weeks and birth weight of 1113 (233) g; 28% of these children presented birth weight lower than 1000 g and 42% gestational age lower than 28 weeks. The mean time of oxygen admission was of 24 days and the median was 6 days. Mean (SD) use of mechanical ventilation was 12 (16) days and length of hospital admission was of 58.3 (26) days. The results for lung compliance and lung resistance were altered in 40% and 58.8% of the babies, respectively. Tomography alterations appeared in 62 children (72%). There was no loss during 1 year follow up, and respiratory morbidity was found in 52 children (53%). The incidence rates of obstructive airway syndrome, pneumonia, and hospital admission were 28%, 36%, and 25% respectively.

Conclusions: In this cohort, a high percentage of alterations in lung mechanics and in HRCT was found among asymptomatic premature infants. More than 50% of the children presented respiratory morbidity in the first year of life. The statistically significant risk factors for the outcomes were: simultaneous alterations of lung compliance and thorax tomography (RR 2.7; CI 1.7 to 10), neonatal pneumonia (RR 5.2; CI 1.3 to 20) and the use of mechanical ventilation (RR 3.3; CI 1.3 to 8.2).

S. Meneghel¹, C. Victora², N. Faria², L. Carvalho³, J. Falk⁴.¹UNISINOS, PPG Health Sciences, São Leopoldo/Brazil; ²UFPEL, PPG Epidemiology, Pelotas/Brazil; ³UERGS, Health, Porto Alegre/Brazil; ⁴UFRGS, Medical School, Porto Alegre, Brazil

Objective: To profile the issue of suicide in the state of Rio Grande do Sul.

Methods: Suicide deaths were put together as case histories based on data from the Ministry of Health’s mortality notification system. Suicides from 1980 to 1999 were grouped according to patterns from the OMS global population and after analysis using standard demographic variables.

Results: Rio Grande do Sul evidenced the highest rates for suicide in Brazil during the period under study (using proportional mortality and coefficients). The standardised coefficients went from levels around 9/100 000 in the 1980s to 11/100 000 in 1999. This increase in mortality is due mainly to the male population, where the figures went from 14/100 000 to the current 20/100,000. The male:female ratio increased from 2.5:1 to 4.4:1. The highest ratios corresponded to the elderly, although the ratio has been increasing in young adults as well. Widowed persons and farmers exhibit high mortality rates.

Conclusion: The study points to suicide as a collective health problem in Rio Grande do Sul and reveals characteristics that could contribute to preventative action. The monitoring of violent events in the community is one of the strategies in health promotion and allows the identification of potential factors for early intervention.

G. Menvielle, D. Luce, J. Chastang, A. Leclerc.INSERM, Unité 88, Saint-Maurice, France

Introduction: European studies have shown that social inequalities in mortality are larger in France than in other countries. However, very few studies have been conducted in France, and all studies conducted on mortality by specific causes suffered from methodological problems due to the absence of direct linkage between socioeconomic data and causes of death.

Objective: To investigate social inequalities in cancer mortality among men and women in France during the period 1975–90, using data including a direct linkage with the French national death index.

Methods: We used a dataset including around 1% of the French population followed since 1968 at each French census. All men and women aged 35–59 years and who answered the 1975 census were included in this study. Subjects were followed until 31 December 1990. We considered both educational level and occupational class to characterise the socioeconomic status in 1975. The occupational class was coded according to the social class scheme of Erikson, Golthorpe, and Portecarero (EGP). Analysis was conducted using a Cox proportional hazards model.

Results: Large social inequalities in mortality were observed. For educational level, inequalities among men were more pronounced for pharyngeal (RR 9.2 for no diploma versus higher education), laryngeal (RR 6.2), oral cavity (RR 2.7), lung (RR 3.5), oesophagus (RR 3.1), stomach (RR 2.5), and rectal (RR 3.4) cancer mortality. No association between educational level and cancer mortality was observed for colon (RR 1.0) and lymphatic and haematopoietic tissue cancer (RR 1.0). Among women, social inequalities in cancer mortality were less pronounced but nevertheless observed for uterine (RR 1.9), stomach (RR 4.1), and lung (RR 1.6) cancer. No association was shown for breast (RR 0.8) and ovarian (RR 1.0) cancer mortality. Similar results were found for occupational class.

Conclusion: The analysis showed large inequalities in France in cancer mortality, and strong differences according to cancer site. The social differences in mortality are particularly pronounced for pharyngeal and laryngeal cancers. Further studies are needed to investigate the underlying mechanisms.

E. Mierzejewska, K. Szamotulska, Z. Brzezinski.Institute of Mother and Child, Department of Epidemiology, Warsaw, Poland

Background: Neural tube defects (NTDs) have been associated with biochemical factors involved in the conversion of homocysteine to methionine and with methylenetetrahydrofolate reductase (MTHFR) polymorphisms.

Methods: The prevalence of the C667T and A1298C MTHFR polymorphisms and their correlation with biochemical phenotype (serum folate, cobalamine, and total homocysteine (tHcy) and red blood cell folate) among spina bifida cases, their parents, and healthy controls were investigated. The sub-sample of 51 mothers of NTD offspring and 87 control women non-pregnant at the time of the interview who were not supplemented with B vitamins was selected to determine differences in Hcy metabolism in mothers compared with women who had not given birth to a child with NTD.

Results: The distribution of the polymorphisms did not differ significantly between mothers and controls (p = 0.4 for both C667T and A1298C). Genetic analysis revealed homozygosity for the C667T polymorphism in 6% of mothers and 13% of controls and for the A1298C polymorphism in 6% of mothers and 3% of controls. Mean plasma folate levels were higher in mothers than in controls (9.7 versus 7.8 ng/ml, p = 0.03) and plasma cobalamine level did not differ significantly between groups. In mothers red blood cell folate was higher than in controls (421 versus 350 ng/ml, p = 0.2). The mean plasma tHcy level was significantly higher in mothers than in controls (10.1 versus 8.8 µmol/L, p = 0.001) but after stratification by tercile of plasma folate of the control distribution, the mean tHcy remained significantly different between mothers and controls only in the two lower terciles. In the logistic regression model having a tHcy level above the 75th centile of the control distribution was associated with an odds ratio (OR) of 3.3 (95% CI 1.6 to 6.8) for being a mother of NTD offspring. However, after estimation of separated models for three terciles of plasma folate concentration, the risk of being the mother of NTD offspring remained significant for the two lower terciles of folate level groups. In a subgroup with plasma folate concentration in the lowest tercile, having a tHcy concentration a >75th centile conferred an OR for being an NTD mother of 6.9 (95% CI 1.6 to 30.9), while in the middle tercile of plasma folate OR was 4.8 (95% CI 1.1 to 19.9).

Conclusions: In the sample of Polish population studied, no association between MTHFR polymorphisms and greater risk of having an offspring with NTD was found. It seems that it is not folate insufficiency but rather the possible barriers to proper utilisation of folate and the teratogenic effect of homocysteine that are responsible for NTDs. That effect could be compensated by additional folic acid supplementation in mothers in the periconceptional period. A further research on genetic risk factors disturbing folate utilisation is needed.

M. Morales-Suarez-Varela¹, J. Olsen², P. Johansen³, L. Karlev², P. Guenel⁴, P. Arveux⁵, G. Wingren⁶, L. Hardell⁷, W. Ahrens⁸.¹University of Valencia and Hospital Dr Peset, Unit of Public Health and Unit of Clinical Research, Valencia, Spain; ²University of Aarhus, the Danish Epidemiology Science Centre, Copenhagen, Denmark; ³Aalborg Hospital, Reberbansgade, Institute of Pathology, Aalborg, Denmark; ⁴Unite 170, INSERM, Villejuif, France; ⁵Registre de Cancers, Registre des cancers du Doubs, France; ⁶Department of Molecular and Clinical Medicine, Faculty Health Sciences, Division Occupational and Environmental Medicine, Sweden; ⁷Orebro Medical Centre, Department of Oncology, Sweden; ⁸Division of Biometry and Data Processing, Bremen Institute for Prevention Research-Social Medicine, Bremen, Germany

Background: Mycosis fungoides (MF) is a T cell cutaneous lymphoma that starts off as plaques on the skin, which continue to the lymphatic nodes, affecting the spleen and other visceral organs. Its predominance in men, its start point after the age of 50–60 years, and the fact that it occurs quite rarely before this age, that it is not a family linked factor, and the frequent reports of patients previously having had dermatitis, led to us consider the role that occupational exposure plays in its development.

Material and methods: A European multicentre case–control study was conducted from 1995 to 1997, and included seven rare cancers, one of which was MF. Patients between 35 and 69 years of age who had been diagnosed with MF (n = 140) were recruited, and a reference pathologist, who classified 83 cases as definite, 35 cases as possible, and 22 cases as unaccepted, checked the diagnoses. Among the 118 accepted cases, 104 cases were interviewed, of which 76 were definite cases. We selected the control group from population controls and colon cancer controls to serve all seven case groups. Altogether 833 colon cancer controls and 2071 population control subjects were interviewed. The overall response rate was 91.5% for cases and 66.6% for controls. Odds ratios (OR), which were adjusted for matching variables (age, place of residence), were estimated by unconditional logistic regression. We performed a occupational matrix for Cr(VI) and its salts, and for aromatic, polycyclic, and/or halogenated hydrocarbons (APH)

Results: We observed that there is a greater risk of MF among workers exposed to the APH, in both men and women.

Conclusions: We suggest that chronic exposure to APH might be responsible for a chronic antigen immune stimulation, resulting in malignant lymphocyte clonality and giving rise to MF.

E. Moreira¹, D. Santos², M. Souza³.¹Oswaldo Cruz Foundation, Brazilian Ministry of Health, Department of Epidemiology and Biostatistics, Salvador, Brazil; ²Fundação Irmã Dulce, Núcleo de Apoio a Pesquisa, Salvador, Brazil; ³Laboratorios Pfizer, Divisão Médica, São Paulo, Brazil

Introduction: Chronic pain is a major health problem and an important cause of morbidity, causing considerable suffering and disability. Our understanding of the epidemiology of chronic pain remains limited as most studies available are based on clinic samples.

Objective: To quantify and describe the prevalence and distribution of chronic pain in Brazil using a population based study.

Methods: A random sample of 6579 individuals, aged 40 years and over, was drawn from two major cities in Brazil (São Paulo and Salvador) and surveyed by a face to face interview (response rate 76%). The standardised questionnaire included case screening questions, a brief pain inventory, a chronic pain grade questionnaire, a visual analogue scale (VAS) for pain intensity, and sociodemographic questions. Chronic pain was defined as continuous or intermittent pain for 3 months in the 6 months prior to interview. Migraine cases were excluded.

Results: The mean age of the 5000 participants (response rate 76%) was 54.5 years, and 56.6% were women. An overall chronic pain prevalence of 6.6% was found; 9.0% for women and 3.4% for men. Mean pain intensity on the VAS scale was 83 and 74 among women and men, respectively. Of the affected subjects, 39% reported to have chronic pain continuously and 25% very often. Backward stepwise logistic regression modelling identified age, sex, higher educational attainment, poor self rated health, and presence of chronic medical conditions as significant predictors of the presence of chronic pain. Medical treatment had been sought by 41% of the patients with chronic pain had sought medical treatment; 9% of them experienced no improvement and 13% had a complete response. One third of the patients with chronic pain reported they had lost 2 weeks or more of work in the previous 6 months, and 19% of men and 25% of women with chronic pain reported some degree of interference with daily activities caused by their pain.

Conclusions: Chronic pain is a major problem in the community and certain groups within the population are more likely to have chronic pain. A detailed understanding of the epidemiology of chronic pain is essential for efficient management of chronic pain in primary care.

M. Movahedi¹, D. T. Bishop¹, J. Barrett¹, G. Law².¹Cancer Research UK, St James’s Hospital, Genetic Epidemiology Unit, Leeds, UK; ²Paediatric Epidemiology Group, Leeds University, LRF Unit, Leeds, UK

Introduction: The global incidence of many cancers varies according to socioeconomic status. However, the direction and strength of these associations may be different between colon and rectal cancer.

Objective: To describe the variation in the incidence of colorectal cancer by anatomical subsites across the Yorkshire and Northern cancer registry and information service (NYCRIS) region and examine association with community level deprivation.

Methods: Incidence data were obtained from NYCRIS for the period 1976–2000. Using the anatomical site of the tumour (ICD-10) around 89 000 patients with colorectal cancer were grouped into proximal, distal, and rectal sites. Small areas were characterised by their levels of affluence/deprivation by deriving a Townsend score for each Enumeration District (ED) from the 1991 census. EDs were grouped into quintiles of the population by their Townsend scores. Age standardised incidence rates were calculated for quintiles, and Poisson and related regression models were used to investigate the association between affluence/deprivation and cancer.

Results: In men the age standardised colorectal cancer incidence rate ranged from 43.2/100 000 (most affluent) to 49.1/100 000 (most deprived). The age standardised rectal cancer incidence rate also increased with deprivation, ranging from 18.3/100 000 (most affluent) to 22.3/100 000 (most deprived), while the trend for proximal cancer was reversed (9.4 versus 8.8/100 000). Poisson and related models showed a weak significant positive association between deprivation level and colorectal cancer in both genders, particularly in men, with a rate ratio of 1.12 (95% CI 1.08 to 1.15) for the most deprived compared with most affluent quintile. Rectal cancer had a strong positive significant association with deprivation level in men with a rate ratio of 1.25 (95% CI 1.19 to 1.32) for the most deprived compared with the most affluent quintile. Cancer of unknown sites of colon also showed a significant positive association with deprivation in both sexes. In contrast there was an inverse association between deprivation level and proximal colon cancer in both genders, with rate ratios of 0.92 (95% CI 0.86 to 0.98) for the most deprived quintile in men and 0.90 (95% CI 0.84 to 0.95) in women compared with the most affluent quintile.

Conclusions: Overall, we found a positive association between colorectal cancer and deprivation. We also found the direction of association between proximal colon cancer and socioeconomic status to be the opposite of that found for rectal cancer. However, the opposite direction of proximal colon cancer with deprivation might be affected by some other factors like the unknown sites of colon cancer. It is likely that multiple lifestyle factors correlated with deprivation influence disease risk. These findings suggest that lifestyle risk factors for colon and rectal cancer differ, therefore the different sites of colorectal cancer should not be combined in aetiological studies.

J. Müller-Nordhorn¹, H. Englert¹, F. Sonntag², H. Völler³, E. Windler⁴, W. Meyer-Sabellek⁵, K. Wegscheider⁶, H. Katus⁷, S. Willich¹.¹Charité University Medical Centre, Institute of Social Medicine, Epidemiology and Health Econom, Berlin, Germany; ²Cardiology Practice, Henstedt-Ulzburg, Germany; ³Rehabilitation Centre for Cardiovascular Diseases, Rüdersdorf, Germany; ⁴University of Hamburg, Department of Internal Medicine, Hamburg, Germany; ⁵AstraZeneca, Wedel, Germany; ⁶University of Hamburg, Department of Statistics and Econometrics, Hamburg, Germany; ⁷University of Heidelberg, Department of Cardiology, Heidelberg, Germany

Introduction: Hypercholesterolaemia (HC) is a major risk factor for cardiovascular diseases.

Objective: To describe health related quality of life (HRQoL) in patients with HC and to determine factors associated with HRQoL.

Methods: The present analysis is part of the ongoing ORBITAL study, a randomised controlled trial evaluating the long term cost effectiveness of a compliance enhancing programme in patients with HC requiring statin therapy according to Joint European Guidelines. At baseline, patients were asked about sociodemographic factors, lifestyle, cardiovascular risk factors, co-morbidity, resource utilisation, employment status prior to enrolment, and HRQoL. HRQoL was assessed with the generic Short Form (SF)-12 health status instrument and the Physical and Mental Component Summary (PCS-12, MCS-12) measures were calculated. Mean PCS-12 and MCS-12 scores of the German norm population are given for comparison. One way analysis of covariance models were used to determine factors associated with PCS-12 and MCS-12 scores.

Results: A total of 7598 patients (44% female, mean (SD) age 61 (11) years), were included in the study. Of these, 32% were employed at the time of inclusion, 18% had a history of myocardial infarction, 8% a history of stroke, 62% had hypertension, and 31% diabetes. The mean (SD) PCS-12 score was 43 (6) for all patients (German norm population 49 (9)) and the mean MCS-12 score 47 (11) (German norm population 52 (8)). A variety of factors were significantly associated with higher PCS-12 scores including male gender, higher socioeconomic status, employment status (yes versus no), a healthy lifestyle, a body mass index <25 kg/m², and the absence of diseases such as angina pectoris and cardiac arrhythmias. Factors significantly associated with higher MCS-12 scores were age 65 years, type of employment (blue versus white collar worker), intake of fruits and vegetables at least daily, and the absence of cardiac arrhythmias.

Conclusions: In comparison with the German norm population, HRQoL is impaired in patients with HC. The long term effectiveness of healthcare programmes needs to be assessed in patients with this disorder, not only regarding clinical outcome and costs but also patient-based outcomes such as HRQoL.

H. K. Neuhauser, U. Ellert, B. Kurth.Robert Koch Institute, Department of Epidemiology and Health Reporting, Berlin, Germany

Introduction: Estimation of absolute risk of coronary or cardiovascular events with prediction models based on the Framingham heart study is widely used in Europe. Recently, a new risk estimation model based on data from 12 European cohort studies (SCORE project) has been used for definition of high risk patients in the new European guidelines on cardiovascular risk prevention issued by the Third Joint Task Force of European and other societies on cardiovascular disease prevention.

Objective: To determine how well Framingham and SCORE based risk estimations agree in a representative sample of the German population with respect to discrimination of high risk patients and to magnitude of predicted risk.

Methods: We compared 10 year risk for fatal cardiovascular events in participants of the 1998 German National Health Interview and Examination Survey based on a Framingham versus SCORE model. SCORE based risks were calculated using both coefficients for high risk countries (SCORE-High, which is recommended for Germany) and for low risk countries (SCORE-Low). Inclusion criteria account for the recommendations for use of the models: age 30–74 years, and absence of cardiovascular disease, absence of markedly raised levels of single risk factors (total cholesterol 320 mg/dl, LDL cholesterol 240 mg/dl, blood pressure 180/110 mmHg) and absence of diabetes with microalbuminuria. A total of 3918 participants (1882 men and 2036 women) were eligible for analysis. High risk was defined as 10 year risk of fatal CVD 5% now or if extrapolated to 60 years of age according to the new European guidelines.

Results: Framingham based estimates yielded the same risk group allocation as SCORE-High in 92.4% of women and 71.2% of men (kappa 0.61 and 0.44). Framingham and SCORE-Low allocated 96.6% of women and 77.5% of men to the same risk group (kappa 0.73 and 0.54). Modifying the risk group definition by leaving out the extrapolation to 60 years improved also in men agreement of risk group allocation between Framingham and both SCORE variants to over 90%. SCORE-High predicted 109 fatal events in our sample, Framingham 75, and SCORE-Low 54. In most age groups and for men and women mean SCORE-High risk was highest, followed by Framingham, and by SCORE-Low.

Conclusions: Framingham and SCORE based cardiovascular risk estimations show good agreement in women but only moderate individual agreement in men with regard to risk group allocation. Especially in men, this agreement is sensitive to extrapolation of risk to 60 years of age. Mean predicted risks differ considerably according to the risk function used, the SCORE variant recommended for Germany (SCORE-High) yielding the highest estimates. As a recent study showed that the Framingham risk function, which we used, overestimated coronary heart disease risk in two German cohorts (PROCAM and MONICA Augsburg), our results suggest that SCORE-High may overestimate absolute risk at the population level in Germany.

A. M. N. Andersen¹, K. Hansen², P. K. Andersen², G. D. Smith³.¹University of Copenhagen, Department of Social Medicine, Denmark; ²University of Copenhagen, Department of Biostatistics, Denmark; ³University of Bristol, Department of Social Medicine, UK

Introduction: Age at reproduction has increased dramatically in Europe during the last decades. Fetal mortality has usually been studied according to the characteristics of the mother, and many studies have shown a significant impact of the biological characteristics of the mother, most importantly maternal age. Studies of paternal age effects on fetal loss are rare, and face several methodological problems, such as selective fertility, a strong correlation between maternal and paternal age, and the fact that routine registrations on fetal loss seldom include information on paternal age. However, paternal ageing may increase the risk of point mutations in the sperm and thus influence fetal survival.

Methods: The study is a prospective cohort study of 23 821 pregnancies recruited consecutively to the Danish National Birth Cohort. The women were interviewed about obstetric history, exposures in pregnancy, and sociodemographic background factors, and the pregnancies were followed up regarding outcome of pregnancy. Paternal age was obtained by record linkage with the Central Person Registry, which contains information on identity (and age) of fathers of all live born children. Furthermore, the participating pregnant women’s husbands or cohabitants at time of conception were traced to identify the fathers of pregnancies ending in fetal loss. The paternal age related overall risk of fetal death and early and late fetal loss were estimated by using Cox regression, taking maternal age, obstetric history, and lifestyle during pregnancy into account. The relationship between maternal and paternal age was examined, identifying possible problems with co-linearity in the data material. Possible residual confounding by maternal age was adjusted for in 5 year intervals, 1 year intervals, and by including the shape of the effect of maternal age as a variable in the model.

Results: Pregnancies fathered by a man aged 50 years or more at the time of conception have almost twice the risk of ending in a fetal loss compared with pregnancies with younger fathers (hazard ratio (HR) 1.88, 95% CI 0.93–3.82). Various approaches to adjustment for potential residual confounding of the relation by maternal age did not attenuate the relative risk estimates substantially. The paternal age related risk of late fetal death was higher than the risk of early fetal death (HR 4.56, 95% CI 1.39 to 14.94). Results should, however, be interpreted cautiously because of the restricted number of late fetal deaths.

Conclusions: High paternal age seem to contribute to the risk of fetal death.

K. O’ Rourke¹, F. Quinn¹, S. Munn², M. Browne¹, J. Sheehan³, K. Iftikhar⁴, S. Cusack¹, M. Molloy¹.¹Cork University Hospital, Department of Accident and Emergency, Cork, Ireland; ²Cork University Hospital, Department of Paediatrics, Cork, Ireland; ³University College Cork, Department of Epidemiology and Public Health, Cork, Ireland; ⁴Waterford Regional Hospital, Department of Accident and Emergency, Waterford, Ireland

Background: In the Irish sporting arena, participation by children in soccer, rugby, and Gaelic football is increasing in popularity. As a result, sport related injuries (SRIs) are common. A comparison of the demographics of injuries in these three physical contact ball sports in Irish children has not previously been adequately performed, as reflected by the paucity of publications in the medical literature.

Objectives: To provide up to date demographic information on the nature of these SRIs using a cross sectional study

Materials and method: Data were collected on all children (<17 years of age) injured in these three sports, presenting to an emergency department of a major teaching hospital over a 6 month period. The data, which included sport, age, sex, cause, type, site, and management, were entered into a specially designed database.

Results: Retrospective analysis was performed on 23 000 charts, and 409 SRIs were identified over a 6 month period. Soccer injuries presented most frequently (56%), followed by Gaelic football (24%) and rugby (20%) injuries. Overall, injured males presented more frequently (82–93%), and were 0.5 years older than their female counterparts in each of the three sports. In the 5–8 and 9–12 year age groups, soccer and Gaelic football injuries predominated, but in the 13–16 year age groups, rugby injuries predominated. The predominant mechanism of injury was different in each sport: in soccer these were falls (38%), in Gaelic football, collisions with football (37%), and in rugby, collisions with persons (55%). Upper limb injuries predominated in all three sports, but were the commonest injury in Gaelic football (66%). Lower limb injuries were more common in soccer (35%), and head (14%) and trunk (9%) injuries were more common in rugby. Fractures were the commonest type of injury in soccer (50%) and Gaelic football (42%), and soft tissue injuries were the commonest type of injury in rugby (44%). Plain x rays were performed in 91% of cases, with 48% revealing fractures. Analgesia was required by 50% of cases, 8% required admission, and 82% required follow up. No documented use of protective gear or preventative advice was recorded for any of the SRIs seen.

Conclusions: This is the first reported study to compare injury in these three physical contact ball sports. The data provided from this study may raise awareness of the nature of SRIs affecting children in each of these sports, and should be helpful in formulating much needed injury prevention strategies.

A. O’Farrell, E. Brabazon, J. Downey, E. Flanaghan, D. Bedford.NEHB, Public Health, Navan, Ireland

Introduction: Alcohol consumption has increased significantly in Ireland during the last 5 years. Excessive alcohol consumption increases the risk of assaults and violence. Studies have shown that drunkenness is not only a risk factor among perpetrators of violence but is also a risk factor among victims of violence.

Objective: To see if alcohol intoxication was recorded among victims of violent incidents who required emergency hospital inpatient admission.

Method: All emergency admissions related to "homicide and injury purposely inflicted by other persons" (ICD-9 codes E960–E969 inclusive) by residents in a health board region were extracted from the Hospital Inpatient Enquiry (HIPE) database for the years 1997–2001. The HIPE database includes data on all hospital inpatient admissions. The data were analysed in the JMP statistical package. Age standardised rates were calculated using StatsDirect.

Results: There was a total of 1451 emergency hospital admissions due to violence, with more admissions occurring at the weekend (76.2% versus 23.8%, p<0.001). Over 13.5% of all the admissions were drunk on admission (had a diagnosis of acute alcohol intoxication, ICD Codes 303.0 and 305.0 also recorded on their HIPE record). There was no significant difference in the number of men and women presenting drunk (14.0% versus 0.3%, p<0.14). The age adjusted hospital admission rate resulting from patients presenting with violence related injuries whilst drunk was significantly higher in 2001 than in 1997 (12.6/100 000 population versus 5.8/100 000 population, p<0.001). The men presenting were significantly younger than the women (28.5 versus 30.3 years, p<0.04), and the most common age group 20–24 years.

Conclusion: This study shows that the increase in drink related violence hospital admissions mirrors the increase in alcohol consumption over the same time period. This study reinforces the need to tackle the increasing positive association between alcohol and violence in Irish society.

M. I. Parahyba¹, D. Melzer².¹Instituto Brasileiro de Geografia e Estatística, Departamento de População e Indicadores Sociais, Rio de Janeiro, Brazil; ²University of Cambridge, Department of Public Health and Primary Care, Cambridge, UK

Introduction: Socioeconomic differences in disability prevalence in old age have been reported from the US, Europe, Japan, and more recently, also from Brazil. Brazil is exceptional in having extreme income inequalities, and provides an opportunity to identify whether there are threshold levels of income above which gains in disability reduction are relatively small.

Objective: To analyse income inequalities in disability among the elderly in Brazil, using the family income per capita as a marker, weighted through an equivalent scale (OECD modified).

Subjects and methods: Data used in this study came from the 1998 National Household Survey (PNAD/1998), conducted by the Brazilian Institute of Geography and Statistics. The study involved a nationally representative sample of 28 943 people aged 60 years or over. Difficulty in walking more than 100 metres, a measure of moderate disability, was taken in this study as a proven marker with prognostic value for disability progression in older people. The data were analysed in SPSS v.10 and the statistical procedure used was multivariate analysis, with the proportion of elderly people with difficulty walking more than 100 metres as the dependent variable, controlling by gender, age, and region of residence, education, and urban/rural residence.

Results: The prevalence of difficulty in walking more than 100 metres for elderly people below the median of the family income per capita was 23.2 for men (95% CI 21.0 to 25.4) and 34.5 for women (95% CI 32.7 to 36.3), decreasing in the richest elderly group (90th centile) for 9.9 (95% CI 4.7 to 15.1) and 18.1 (95% CI 13.6 to 22.6), respectively. Even for those elderly in the highest income group (95th centile), disability prevalence rates decreased when the value of family income increased.

Conclusions: In developed countries, there is evidence of a gradual reduction in the disability prevalence rates among the elderly over recent decades. For the 16 million of elderly people in Brazil, functional impairment occurs in a context of poverty and extreme social inequalities. Reducing the burden of disability in the older population of Brazil is likely to require reduction in income inequalities and improvements in institutional support.

T. Pekmezovic¹, J. Drulovic², M. Jarebinski¹, N. Stojsavljevic², D. Kisic¹, S. Mesaros².¹School of Medicine, Institute of Epidemiology, Belgrade, Serbia and Montenegro; ²Clinical Centre of Serbia, Institute of Neurology, Belgrade, Serbia and Montenegro

Introduction: The putative role of risk factors for multiple sclerosis (MS), such as tobacco, alcohol, coffee, and tea consumption, and physical activity has been evaluated only through several epidemiological studies so far, but with contradictory results.

Objective: To analyse the role of some habits (cigarette smoking, alcohol, coffee and tea consumption, and physical activities) as environmental risk factors for MS in the Belgrade population using a case–control study.

Methods: The study group consisted of 110 cases with clinically and/or laboratory supported MS according to the Poser’s criteria, in whom onset symptoms occurred up to 2 years prior to the interview. The same number of controls, individually matched by sex, age, and area of residence, was recruited from patients with various non-autoimmune neurological disorders. Statistical analysis included conditional univariate and multivariate logistic regression models.

Results: Subjects with MS were smokers significantly more frequently than controls (odds ratio (OR) 1.81, 95% CI 1.06 to 3.21, p = 0.031). For smokers who had smoked for 21 years (OR 2.38, ² test for trend = 6.291, p = 0.012), and for those who smoked 11 cigarettes/day (OR 2.13, ² test for trend = 6.236, p = 0.012), the risk increased significantly over twofold. Coffee consumption was significantly more frequent in the study group than in the control group (OR 2.28, 95% CI 1.07 to 5.17, p = 0.032), and the risk of MS significantly increased with the length of the period of coffee consumption (² test for trend = 3.885, p = 0.048). The analysis of alcohol drinking showed a significant association between spirit consumption and MS compared with drinking of wine and beer (OR 3.78, 95% CI 1.06 to 13.45, p = 0.040). According to the multivariate analysis the consumption of spirits (OR 6.86, 95% CI 1.42 to 33.0, p = 0.016) and the practice of sports (OR 2.61, 95% CI 1.09 to 6.23, p = 0.031) were significantly related to MS, and an inverse association was found for tea consumption (OR 0.18, 95% CI 0.03 to 0.99, p = 0.049).

Conclusions: The findings obtained in this study support the a possible role of certain habits such as cigarette smoking, alcohol and coffee consumption, and physical activities, in addition to the other factors, in the occurrence of MS in this environment.

C. Pereira, R. Campos, A. Rijo, C. Costa, C. Santos, L. Vasconcelos, R. Antunes, O. Amaral, N. Veiga.ESEnfV, Viseu, Portugal

Introduction: The pattern and duration of breastfeeding are influenced by many social factors and constitute important determinants of health.

Objective: To quantify the prevalence of breastfeeding and to identify the main social determinants of initiation and duration from birth until 4 months.

Methods: Mothers consecutively admitted from June to September 2003 in obstetric departments of S. Teotónio Hospital of Viseu constituted the sample. On the third day after delivery, 661 mothers completed a self administered questionnaire with questions about breastfeeding situation and social variables. At the fourth month a telephone contact was carried out with the objective of measuring the prevalence and duration of breastfeeding. Only 621 mothers could be contacted by telephone; 20 were not available. The ² test was conducted to compare the proportions. Crude odds ratio (OR) with a 95% confidence interval was used to measure the strength of the associations for weaning at the second month.

Results: The prevalence of exclusive breastfeeding on the third day of life was 94.1% (95% CI 92.1 to 95.7), after 1 month it was 82.2% (95% CI 79.1 to 85.0), after 2 months 71.8% (95% CI 68.2 to 75.4), after 3 months 59.4% (95% CI 55.6 to 63.2) and after 4 months 34.9% (95% CI 31.3 to 38.7). Weaning at the second month was associated with the mother’s age (<20 years OR 2.1; 95% CI 1.2 to 13.6), mother’s occupation (manual OR 2.2; 95% CI 1.3 to 3.9), multiparity (OR 1.6; 95% CI 1.1 to 2.3), marital status (singled/divorced OR 1.9; 95% CI 1.0 to 4.3) and mother’s professional situation (employed OR 1.5; 95% CI 1.0 to 3.1).

Conclusion: We found a higher prevalence of exclusive breastfeeding at the third day after delivery. Duration of breastfeeding was shorter in children of younger women, mothers with manual occupation, and in multiparous, singled/divorced, and employed women.

C. Pereira¹, O. Amaral², N. Veiga³, A. Escoval².¹ESEnfV, Viseu; ²ISCTE, Lisboa; ³CRB-UCP, Viseu, Portugal

Background: Eating disorders are common in adolescents and are associated with important psychiatric morbidity.

Objective: To quantify the prevalence of abnormal eating behaviours in a Portuguese community sample of adolescents.

Participants and methods: In a cross sectional study we evaluated the eating attitudes and the prevalence of eating disorders in a sample of 2144 students (1105 females) aged 12–18 years (mean (SD) 14.8 (1.70)), among seven secondary schools from Viseu. Participants were surveyed according to the Portuguese Validated Eating Disorder Inventory (EDI2) and questions about socioeconomic and demographic situation. We considered a score of 4 points as the cutoff for bulimia.

Results: Lifetime prevalence of any eating disorder was 8.3%, significantly higher in females (13.2 versus 3.2, p<0.01). The prevalence of full syndrome anorexia nervosa in the total sample was 1.0%, and partial syndrome was 7.6%, significantly higher in females (12.8 versus 2.5, p<0.01). Prevalence of bulimia in the past year was 0.8%, and of binge eating disorder was 2.9%. Point prevalence of body dissatisfaction was 10.9%, significantly higher in females (17.6 versus 3.7, p<0.01). Current dieting was more prevalent in girls (28.2%) than in boys (8.9%). Laxative or diuretic use was reported by 1.0% and self induced vomiting by 0.9% of total sample.

Conclusion: This study show higher prevalence of abnormal eating behaviours in adolescents. Educational programmes among children and adolescents will be considered, aimed at preventing abnormal eating behaviours and related morbidity.

F. Perlman, M. Bobak, M. Marmot.University College London, Epidemiology and Public Health, London, UK

Introduction: Since the fall of communism in Russia in 1991, mortality has fluctuated dramatically. Mortality has increased most in middle aged people, men, and those with less education, and poor self reported health is also commoner in people with lower education and income. Education and income are less closely correlated than in Western societies.

Objective: To study prospectively the associations of different socioeconomic measures with mortality using one of the very few longitudinal studies in Russia.

Methods: The Russia Longitudinal Monitoring Survey is a panel survey of households and individuals from centres across Russia. We analysed data from seven rounds between 1994 and 2002. We included subjects aged 18 years and over with individual and household data available for the round of entry and at least one subsequent round. Household members reported deaths. We explored the associations between several socioeconomic measures and mortality using Cox regression. These variables were education, whether goods were sold for money, and household and individual income quintiles. Incomes were adjusted to the value of the 1992 rouble. We then explored the influence of health behaviours (smoking and alcohol), self reported health and other socioeconomic factors on these associations.

Results: In the final analysis, 13 455 participants (5950 males and 7505 females) were included. Participants spent a mean of 4.06 years in the study. There were 845 deaths (488 males and 357 females) reported. Health was described by 17% of respondents as poor or very poor at entry. Of the participants, 47% had higher education, 3% had sold goods for money. After adjusting for age, mortality was significantly lower in men and women with greater than secondary education (0.75 (0.58 to 0.97) men, 0.42 (0.29 to 0.62) women) and higher in men with less than secondary education (1.80 (1.40 to 1.31)). Mortality was significantly higher in men (1.66 (1.08 to 2.51)) and women (1.91(1.14 to 3.21)) who had sold goods for money, and significantly lower in both sexes in the highest individual income quintile and in men in the second highest quintile compared with the middle individual income quintile. Household income was only weakly related to mortality. The relationships with education, income, and selling goods to mortality persisted after adjusting for smoking, alcohol intake, and initial self reported health (poor self reported health, smoking, and frequent alcohol consumption were each independently significantly associated with increased mortality). After adjusting for other socioeconomic and behavioural factors, only the relationship with education remained significant.

Conclusions: Overall education was most strongly associated with mortality, a relationship not fully explained by initial health, socioeconomic factors, and health behaviours. This is consistent with the findings of others.