Article Text
PDF
Abstract
Background Previous studies have shown that cancer incidence is related to a number of individual factors, including socioeconomic status. The aim of this study was to refine the current knowledge about indicators associated with cancer incidence by evaluating the influence of neighbourhood characteristics on breast, prostate and lung cancer incidence in Denmark.
Methods All women aged 30–83 years were followed for breast cancer between 2004 and 2008, men between 50 and 83 years were followed for prostate cancer and both sexes between ages 50 and 83 were followed for lung cancer. Registry data obtained from Statistics Denmark included age, sex, availability of breast cancer screening, marital status, education, disposable income and occupational socioeconomic status on the individual level and population density and neighbourhood socioeconomic status (the proportion of unemployed) on the parish level. Frailty modelling with individuals on the first level and parishes on the second level was conducted.
Results A significantly lower HR of breast cancer was found in areas with low population density (HR=0.93; CI 0.88 to 0.99), while neighbourhood unemployment had no effect. Inhabitants of lower unemployment areas had a higher risk of prostate cancer (HR=1.14; CI 1.08 to 1.21) compared with those in higher unemployment areas, whereas population density had no effect. Risk of lung cancer was lower in areas with lowest population density (HR=0.80; CI 0.74 to 0.85) and lowest in areas with lowest unemployment (HR=0.88; CI 0.84 to 0.92).
Conclusions In addition to individual-level factors, characteristics on the neighbourhood level also have an influence on breast, prostate and lung cancer incidence.
- Small-area analysis
- multilevel
- risk factors
- residence characteristics
- neoplasms
- cancer
- cancer: breast
- cancer: cervix
Statistics from Altmetric.com
- Small-area analysis
- multilevel
- risk factors
- residence characteristics
- neoplasms
- cancer
- cancer: breast
- cancer: cervix
Introduction
Breast, lung and prostate cancer are more common cancer types in Denmark and accounted for 40% of all cancer diagnoses in 2008.1 Major risk factors include alcohol, heredity, high age at first birth or no birth, physical inactivity, oestrogen treatment and obesity after menopause for breast cancer; animal fat, vasectomy, family disposition for prostate cancer and tobacco, passive smoking and work environment for lung cancer.2
Studies based on individual-level data show that high individual socioeconomic status (SES) is associated with increased risk of breast cancer3–6 and reduced risk of lung cancer,7 ,8 whereas results differ for prostate cancer risk.9–11 Increased breast cancer risk in high SES women has been linked with older age at first childbirth, fewer children, frequent use of oral contraceptives and hormone replacement therapy as well as high alcohol intake.12 Excess risk of prostate cancer in higher SES men has been explained partially by more frequent prostate-specific antigen (PSA) testing in those without prior clinical symptoms13 and by dietary and lifestyle factors.9 The association between low individual SES and risk of lung cancer has been linked to higher smoking prevalence7 ,14 and occupational exposures.8
Few studies have combined individual- and area-level data to investigate whether area characteristics are independently associated with cancer incidence. Two multilevel studies have found that inhabitants in urban5 and high SES areas5 ,6 were at higher risk of breast cancer when accounting for individual demographic and socioeconomic variables. A study which adjusted for similar factors found that inhabitants in low SES neighbourhoods were at higher risk of prostate cancer.10 To our knowledge, no multilevel studies have explored the influence of neighbourhoods on lung cancer incidence. The aim of this study is to identify possible area characteristics associated with breast, lung and prostate cancer incidence using register data on the entire Danish population. It investigates whether population density and area-level socioeconomic status (ALSES) are associated with the three cancer sites while controlling for individual characteristics.
The literature suggests that community norms and values affect individual lifestyle.15 In American studies, areas with high SES have been linked with higher incidence of breast cancer because of greater availability of mammograms.5 ,6 This probably does not apply to Denmark since high individual education has been associated with lower screening participation,16 possibly because the concerns about the diagnostic effectiveness of mammography are disseminated primarily to higher educated women. Instead we hypothesise that higher alcohol consumption in affluent areas17 could increase individual consumption leading to higher breast cancer incidence.
The rate of PSA testing is higher in affluent neighbourhoods,13 and it is the hypothesis in this paper that this encourages individual testing and therefore leads to more prostate cancer diagnoses. We expect a link between residence in a deprived area and lung cancer risk because of higher availability of tobacco and because the increased level of smoking in low SES groups could encourage other inhabitants to smoke.
Lower fertility and later age at first birth in cities18 ,19 could increase breast cancer incidence in areas with high population density. Elevated exposure to traffic-related air pollution20 or to light at night21 have also been suggested as possible links between breast cancer and urbanicity. Increased alcohol consumption in urban areas17 also could increase breast cancer risk. There is evidence of an association between agricultural pesticides and prostate cancer. Population density can serve as a proxy for an urban–rural divide with pesticide exposure higher in farming areas.22 Finally, population density could be associated with lung cancer incidence because of increased exposure to traffic-related air pollution23 and because smoking is more prevalent in urban communities.24
Materials and methods
Design
All persons with residence in Denmark as of 1 January 2004 were followed through 2008 for cancer incidence, death or emigration, whichever came first. Individual information on all residents came from registers in Statistics Denmark and follow-up for cancer was made through linkage to the Danish Cancer Register using a personal identification number. Permission was obtained from the Danish Data Protection Agency. Populations were limited to individuals younger than 84 years because of missing data on education for older age groups. The breast cancer study population was restricted to those 30 years or more to ensure that most had completed their educations. For prostate and lung cancer, individuals younger than 50 years were excluded since few are diagnosed before this age.
Since a cancer diagnosis can result in major life events (eg, job loss), individual characteristics were measured 2 years prior to diagnosis. Information on those censored was measured in 2004. Characteristics of neighbourhoods where individuals lived in 1995 were used to allow for a latency period between neighbourhood exposure and diagnosis. The year 1995 was chosen because a 10-year incubation period between risk factor and diagnosis is often used in cohort studies. In 2006, 1.4% and 4.4% of all Danes between 50 and 83 years moved across and within municipality boarders, respectively.19 Individuals not present in Denmark when individual- and area-level factors were measured were excluded.
Individual-level factors
Individual-level factors included age, sex, marital status, education, family disposable income and occupation-based SES. Age was divided into 5-year age groups. Education, measured as the highest obtained education, was recoded according to the International Standard Classification of Education25 and collapsed into basic, non-tertiary and tertiary education. Disposable income was measured as annual family income after taxation and interest deflated to 2000 price levels divided by the number of household members elevated by a factor of 0.6. This is routinely done by official economic institutions in Denmark to account for the economic advantages of sharing costs in households.26 Marital status was dichotomised into cohabiting/married or not. Occupation-based SES was based on the six-class European Socio-Economic Classification27: Salariat, intermediate employee, small employers and self-employed, lower white collar, skilled manual and semi- or unskilled. Persons not employed were grouped into four additional categories: students, unemployed, disability pensioners and pensioners. Pensioners were classified according to last position if they had had any after 1990. Individuals under age 60 who had been unemployed for more than 2 years were categorised as unemployed; individuals who had been unemployed under 2 years were classified according to last position.
Previous investigations in Denmark have found that the three cancers are associated with individual education, disposable income and social class when mutually adjusted.11 ,12 ,28 All three factors were therefore included in the analysis.
Individual information on invitation (compliance 70%) to organised breast cancer screening was included in analyses of breast cancer incidence. Mammography is offered every second year to all women aged 50–69 years. It was introduced in different years across counties beginning in Copenhagen city in 1991, but for 75% of the country, the programme was initiated in 2007–2008. Due to higher incidence in the initial 2-year screening wave,29 the factor was coded ‘never invited to screening’, ‘screening introduced within two years’ and ‘screening introduced two or more years ago’. For inhabitants living in counties where screening was introduced during follow-up risk time was distributed to screening categories according to screening start.
Neighbourhood factors
ALSES was calculated for all parishes (n=2121) as the proportion of unemployed between 20 and 60 years. Population density was calculated as the number of persons (all ages) per square kilometre. Quartiles were 0–28, 29–49, 50–164 and 165+ persons per square kilometre. A parish is an area sharing a church and is the smallest geographical unit in Danish registers. They have no political or administrative purpose, and in 1995, they differed in size from 0.1 to 156.0 km2 (median=16 km2). Parish populations ranged from 8 to 20 442 persons (median=1107). Unemployment and population density were chosen because these factors appear to represent significant determinations of all-cause mortality.30
Statistical analyses
Hazard rates were estimated using shared frailty models, which are multilevel random effect models for survival data accounting for a latent multiplicative effect on the hazard function, the ‘frailty’. A multilevel framework is useful because it assumes that inhabitants living in the same parish have risk in common, that is, neighbourhood environment. Individuals in a parish are therefore not treated as being independent; they are assumed to share the same frailty. The term frailty refers to the random components of the variance in cancer incidence in the parish. The shared frailty model allows individuals to be nested within parishes and the intercept to vary between parishes.
The streg procedure in Stata V.11.1 MP with the two-parameter Weibull survival distribution and a γ frailty distribution was used to estimate the frailty variance and effect of individual-level and parish-level factors on cancer-specific incidence. Variance was estimated with the θ value provided in output of the streg procedure.31
First a model without explanatory factors was estimated with a random intercept for each parish. Then the association between cancer and each of the explanatory factors was estimated controlling for age and screening programme for breast cancer, age for prostate cancer and age and sex for lung cancer (table 2). In table 3, all significant individual factors from table 2 were then mutually adjusted. Model 1 in table 3 included all remaining significant individual factors. These were subsequently examined together with each of the parish-level factors. In model 2 (table 3), all significant individual-level and parish-level factors were included.
Results
The background populations consisted of 1 539 162 women for breast cancer, 799 839 men for prostate cancer and 1 662 228 women and men for lung cancer analyses. The number of cancer diagnoses for the three cancers were 17 854, 14 612 and 17 361 persons, respectively. Table 1 also shows that persons with a cancer diagnosis had an older age profile, were lower educated and had a lower disposable income compared with the total population. There was an overweight of breast cancer among women with no access to organised breast cancer screening and women who were unmarried and with lower occupations. For prostate cancer, there was a higher proportion with a diagnosis among men who were married or cohabiting and those with higher occupations. For lung cancer, there was an overweight among men, persons married or cohabiting and of lower occupations. Table 1 also shows a higher proportion of breast and prostate cancer cases among inhabitants in areas with lower proportions of unemployed, whereas the opposite was true for lung cancer. There were no major differences for population density except for highest density where lung cancer was more prevalent.
Numbers and percentages of total populations and diagnosed cases of breast, prostate and lung cancer (2004–2008) in Denmark by individual-level (2002–2006) and area-level (1995) factors
Frailty model HRs and 95% CIs of association between breast, prostate and lung cancer incidence in Denmark between 2004 and 2008 and individual- and parish-level factors in models only adjusting for age, sex and breast cancer screening programme
HR and 95% CI of significant individual factors (model 1) plus significant parish-level factors (model 2) on breast, prostate and lung cancer incidence in Denmark between 2004 and 2008, estimated in a frailty model mutually adjusted for all factors in the model
The effect of each individual-level factor adjusted for age and sex (and for breast cancer also for invitation to screening) was significantly associated with the three cancers (table 2). Only individual education was not associated with prostate cancer after controlling for all other individual-level variables (results not shown).
Table 2 shows that parish-level SES was not associated with breast cancer and that population density was not associated with prostate cancer. Both area-level factors were associated with lung cancer.
Models 1 of table 3 show that older age, higher income and higher occupation-based SES, being married/cohabiting or being unemployed all increased the risk of breast and prostate cancer. For breast cancer, higher education and access to screening were also indicators of increased risk. Lung cancer incidence was associated with being male, older, single, having lower education, lower income and receiving disability pension.
In model 2 of table 3, breast cancer incidence was highest among women in parishes with highest population density (ref) compared with high (HR=0.95; CI 0.91 to 0.99), low (HR=0.93; CI 0.88 to 0.99) and lowest (HR=0.94; CI 0.87 to 1.01) population density. For prostate cancer, the lowest risk was found among men in parishes with highest proportion of unemployed (ref) increasing through high (HR=1.05; CI 1.00 to 1.11), low (HR=1.08; CI 1.02 to 1.14) and lowest proportions (HR=1.14; CI 1.08 to 1.21). For lung cancer, incidence risks were higher in areas with highest population density (ref) decreasing to lowest density (HR=0.80; CI 0.74 to 0.85) and in areas with highest unemployment (ref) decreasing to lowest unemployment (HR=0.88; CI 0.84 to 0.92).
Compared with the null model, individual factors explained 82%, 36% and 66% of the variation in breast, prostate and lung cancer incidence, respectively. Adding neighbourhood factors increased explained variation to 84%, 39% and 85% indicating that area characteristics explain relatively little variation in breast and prostate cancer, while they are of greater importance for lung cancer.
Rerunning model 2 for breast cancer without including ‘breast cancer screening’ had no impact on estimates for high and low population density but had a small influence on the estimate for lowest density (HR=0.92; CI 0.86 to 0.99). Thus, the effects of population density appear to only to have a minor relationship with individual access to the organised breast cancer screening.
In sensitivity analyses, ALSES was added to model 2 for breast cancer and population density was added to model 2 for prostate cancer. Both factors remained insignificant and the inclusions altered no estimates. We found no correlation between population density and ALSES, r=0.01 (p=0.57).
The variation explained by using the proportion of unemployed was compared with the explanatory power of other ALSES measures. A composite deprivation index is not available for Denmark, and therefore alternative measures were considered by availability in Danish registers and by reviewing previous studies.32 They were proportion of disability pensioners, proportion of manual workers, average education and average income. Proportion of manual workers explained slightly more variation than proportion of unemployed for breast cancer but was insignificant when evaluated together with population density. For both prostate and lung cancer, proportion of unemployed remained best at explaining variation, perhaps because unemployment is also related to low incomes and low education.
Discussion
Consistent with other studies, it was found that higher individual SES was associated with increased risk of breast cancer,3–5 ,12 prostate cancer9 ,11 and reduced risk of lung cancer.28 ,33 However, our results contradict a previous study reporting a negative association between SES and prostate cancer incidence.10
Densely populated areas were associated with increased risk of breast cancer, but no effect was found for ALSES after adjusting for individual factors. This contrasts with findings in a previous study where both ALSES and urbanisation were associated with breast cancer risk5 and another where higher ALSES, but not urbanisation, was linked to higher risk.6 The contradicting results could be due to the adjustment for three SES factors in this study or the free access to health care in Denmark. The associations found in the previous studies could therefore be attributable to residual confounding. The association with population density found in this study could similarly be due to insufficient control for individual-level factors other than SES, for example, age at first child for breast cancer.18
The urban environment could also promote certain life styles that affect breast cancer risk. A Danish study has shown that heavy and harmful drinking is significantly higher in the capital region after controlling for individual demographic and socioeconomic factors.17 A pathway could be that there are higher densities of alcohol outlets in urban environments making alcohol more available and therefore consumption is higher. The literature on this remains inconclusive.34 ,35 Areas with high population density are also subject to more air pollution and light at night, which have been associated with breast cancer.20 ,21 Our results also showed that screening availability did not reduce the effect of population density. This agrees with previous studies finding that effects from ALSES and urbanicity are unaffected by mammography attendance.5 ,6 Future studies could investigate the role of timing and frequency of screening for breast cancer incidence.
Results showed that risk of prostate cancer was higher in affluent/low unemployment areas, contradicting a previous multilevel study.10 This association could be related to individual lifestyle shaped by the physical and social environments, but we suggest that PSA testing could be a central explanation.13 A common assumption is that individuals with high SES possess more knowledge about how to live healthfully. Inhabitants in high SES areas would therefore be more likely to exchange knowledge about the possibility of being tested. General practitioners in high SES areas might also be more likely to offer PSA tests because of increased awareness. There was no association between population density and risk of prostate cancer. It could be that our proxy for a farming environment was too crude, and farming practices could differ in Denmark from countries where previous research was conducted.
We found increased risks of lung cancer in densely populated areas and in areas with low SES. Two multilevel studies have found associations between higher risk of lung cancer mortality in areas of low SES36 or high population density,37 whereas a third found no association with ALSES.38 Recent studies have also linked lung cancer with traffic-related air pollution39 that would explain the elevated lung cancer risk in densely populated areas. Similarly, a study has demonstrated that smoking in several European countries, including Denmark, is higher in urban communities.24 A number of studies have also found associations between lower SES areas and higher levels of smoking.40 ,41 It has been hypothesised that social influences between inhabitants in neighbourhoods can account for some of the variation in smoking between neighbourhoods.41 It has also been suggested that higher density and shorter distances to convenience stores are associated with smoking.40 These factors might indicate why higher risks of lung cancer are found in areas with low SES. Additionally, a study has found that low SES areas are more exposed to air pollution.42
This study has particular limitations and strengths. First, we did not have access to data on a number of important individual-level confounders such as, for breast cancer, use of mammography, menopause, age at first child and family history of breast cancer, and for prostate cancer, the use of PSA testing. Thus, there is a risk that we overestimated area-level effects because we have not sufficiently controlled for individual risk factors. On the other hand, we have controlled for SES on the individual level, which could have captured some of this information. Second, it would have been helpful if we had had direct measures of area-level features, such as availability of alcohol and tobacco outlets. A particular strength is the large study population. Using nearly the entire Danish population significantly reduces selection bias. Moreover, using registry data increases reliability.
The present paper broadens the usual focus on individual-level risk factors by including neighbourhood characteristics when investigating the determinants of cancer incidence. Based on our findings, we encourage future research to investigate mediating conditions between breast cancer and population density as well as between ALSES and prostate cancer risk.
What is already known on this subject
Previous studies have shown that breast, prostate and lung cancer incidence is associated with factors on the individual and the area levels. Only few studies have examined both levels simultaneously.
What this study adds
In addition to individual socioeconomic status, higher breast cancer incidence was found in areas with highest population density. Increased prostate cancer incidence was found in areas with higher socioeconomic status. Higher lung cancer incidence was associated with both high population density and low socioeconomic status on the area level.
Acknowledgments
We would like to thank Anne Mette T Kejs, Danish Cancer Society, for assisting in the analysis.
References
Footnotes
-
Funding This work was supported by the Danish Council for Independent Research, Medical Sciences grant number 271-06-0549 and the Danish Health Insurance Foundation grant numbers 2007B048 and 2009B077.
-
Competing interests None declared.
-
Provenance and peer review Not commissioned; externally peer reviewed.