Article Text

Download PDFPDF

Socioeconomic determinants for compliance to colorectal cancer screening. A multilevel analysis
  1. C Pornet,
  2. O Dejardin,
  3. F Morlais,
  4. V Bouvier,
  5. G Launoy
  1. ‘Cancers & Populations’, ERI 3 INSERM, CHU Caen, Faculty of Medicine, Caen Cedex, France
  1. Correspondence to Professor Guy Launoy, ‘Cancers & Populations’, ERI 3 INSERM, CHU Caen, Faculty of Medicine, Avenue Côte de Nacre 14032 Caen Cedex, France; guy.launoy{at}unicaen.fr

Abstract

Background Compliance in cancer screening among socially disadvantaged persons is known to be lower than among more socially advantaged persons. However, most of the studies regarding compliance proceed via a questionnaire and are thus limited by self-reported measures of participation and by participation bias. This study aimed at investigating the influence of socioeconomic characteristics on compliance to an organised colorectal cancer screening programme on an unbiased sample based on data from the entire target population within a French geographical department, Calvados (n=180 045).

Methods Individual data of participation and aggregate socioeconomic data, from the structure responsible for organising screening and the French census, respectively, were analysed simultaneously by a multilevel model.

Results Uptake was significantly higher in women than in men (OR=1.33; 95% CI 1.21 to 1.45), and significantly lower in the youngest (50–59 years) and in the oldest (70–74 years) persons, compared with intermediate ages (60–69 years), with OR=0.70 (95% CI 0.63 to 0.77) and OR=0.82 (95% CI 0.72 to 0.93), respectively. Uptake fell with increasing level of deprivation. There was a significant difference of uptake probability between the least deprived and the most deprived areas (OR=0.68; 95% CI 0.59 to 0.79). No significant influence of the general practitioners density was found.

Conclusion Multilevel analysis allowed to detect areas of weak uptake linked to areas of strong deprivation. These results suggest that targeting populations with a risk of low compliance, as identified both socially and geographically in our study, could be adopted to minimise inequalities in screening.

  • Mass screening
  • socioeconomic factors
  • multilevel analysis
  • colorectal neoplasms
  • cancer screening
  • colorectal
  • multilevel modelling
  • socioeconomic
  • colorectal neoplasms

Statistics from Altmetric.com

Colorectal cancer (CRC) is the second leading cause of death from cancer in France,1 the United Kingdom2 and the USA.3 Randomised controlled trials have demonstrated that CRC mortality can be reduced by screening using the faecal occult blood test.4–6

In the UK, a pilot campaign of organised screening for CRC was set up in 2000, followed by a national screening programme in 2006.2 In France, a pilot campaign was launched in 2002, throughout 23 of 96 geographical departments.7 Local screening management structures invited the target population by post to consult the general practitioner (GP) of their choice, to obtain tests that were to be conducted at home. Individuals forwarded the tests to the analysis centre that, in turn, informed the management structure of test results. A few months later, individuals who had failed to respond to the initial invitation directly received tests by post.

Among countries having established an organised CRC screening programme, France occupies an intermediate position in terms of participation rate (42%)8 between Japan (17%)9 and England (60.6% for the first round of screening).10 Compliance to screening is known to be favoured by a high individual socioeconomic level.11–17 However, most of the studies regarding compliance proceed via a questionnaire and are thus limited by self-reported measures of participation and by participation bias. Some studies regarding social inequalities in cancer have used data aggregated into a geographical level10; however, although this method avoids selection bias, no study has investigated the social determinants of screening by combining individual participation data and aggregate socioeconomic data, within the entire target population, and at a sufficiently accurate geographical level.

This study aimed at investigating the influence of socioeconomic characteristics on compliance to an organised colorectal cancer screening programme on an unbiased sample based on data from the entire target population within a French geographical department (Calvados).

Materials and methods

Study population

The first campaign of CRC screening took place from June 2004 to June 2006 in the department of the Calvados. The Association Mathilde was in charge of organising screening in Calvados. This association was therefore given the target population data from the various health insurance systems, concerning particularly identity and address. The study's target population, aged from 50 to 74 years and located within the department, included 180 045 individuals (figure 1).

The availability of exact addresses enabled a geographical unit to be assigned to each participant (geocoding). The geographical units used were Ilôts Regroupés pour l'Information Statistique (IRIS, or regrouped statistical information block), as defined by the National Institute for Statistics and Economic Studies (INSEE), and are the smallest geographical census units available in France. The regional capital and other major towns are divided into several IRISs, and small towns form one IRIS. Our study zone, the department of Calvados, has a total of 829 IRISs (INSEE website, http://www.insee.fr). Because geocoding procedures were not completely automated, it would appear unreasonable to consider that the study concerns the entire population. Sampling of 10 000 individuals was conducted, by simply drawing lots, without handing over, and according to the law of uniform distribution. The final study population comprised 8758 individuals after geocoding (135 not geocoded), 1107 individuals having been excluded (figure 1). Among these 1107 excluded individuals, 762 individuals were excluded for medical reasons. The medical exclusions were collected by the Association Mathilde after the medical visit. The GP checked the following medical exclusions:

  • Patients with a recent digestive symptomatology;

  • Patients with a normal complete colonoscopy within the last 5 years;

  • Patients with a high risk or very high of CRC, as those having a history of CRC.

Final sample representativeness was checked in terms of participation rate, age and sex (table 1). However, there was a significant difference of insurance coverage between the final sample and the target population, the civil servant system being overrepresented in the final sample.

Table 1

Comparison between the final sample (n=8758) and the target population (n=159 014*)

Measures

Participant or non-participant status was known for each individual (table 1). Participants were defined as having undergone a screening test within the duration of the study.

Two types of variables were used: individual and aggregate.

Level 1: individual variables

Sex, age and insurance coverage were obtained from exhaustive registries held by the Association Mathilde and via all insurers within Calvados (table 1). In France, medical insurance coverage is virtually universal. Homeless people were excluded from our population study (as well as people living in mobile homes) because of the lack of address that was crucial for sending them the invitation to the screening. Unemployed and retired persons were included in our population study because had an available address. Different systems of insurance coverage exist in France: the most common is the general medical insurance scheme, the Mutuelle Sociale Agricole covering agricultural occupations, other coverage including civil servant systems, schemes for the self-employed or other specific systems. The special coverage systems concerns persons belonging to major French enterprises such as Electricité De France and Société Nationale des Chemins de Fer.

Level 2: aggregate variables

Each individual was attributed the socioeconomic characteristics of his/her appropriate neighbourhood (IRIS level) using French census data provided by the INSEE (1999). The INSEE produces a large number of socioeconomic indicators. Consequently, because deprivation is multifactorial,18 the selection of relevant variables for the determination of the socioeconomic status of each IRIS proved difficult. We therefore used a composite index of deprivation, less sensitive to measurement bias than individual variables considered independently.19 20 We chose the Townsend index,18 which is widely used and acknowledged in Anglo-Saxon countries.20 This index is based on the unweighted sum of four centred and reduced socioeconomic variables, after log transformation of the first three: percentage of overcrowned households, percentage of households without a car, percentage of unemployed individuals of an economically active age and the square root of the percentage of non-homeowner households. To allow international comparisons, a standardised division into quintiles of the distribution of the Townsend index score for each IRIS was carried out: quintile 1 represented the most privileged IRIS and, conversely, quintile 5, the most deprived.

GP density per 100 000 inhabitants for each IRIS was used as a proxy variable determining access to healthcare.

Statistical analysis

Multilevel statistical models provide a technically robust framework with which to analyse the correlated nature of the outcome variable and are pertinent when predictor variables are measured simultaneously at different levels.21

The influence of aggregate data and individual data on CRC screening compliance, dependent variable (participation vs non-participation), was analysed using two-level (multilevel) logistic regression models with individuals (level 1) residing within the IRIS (level 2).

We adopted a modelling strategy that consists in increasing model complexity at each step, using a random intercept model.22

First, the empty model, without explanatory variables, enabled us to test the influence of context on participation, that is, heterogeneity between IRISs. At this first step, we tested the null hypothesis according to which the variance of the random intercept is zero to check the existence of an inter-IRISs heterogeneity (test of random intercept). This test is a likelihood ratio test, or deviance difference, classically used to compare two fitted models, noted: deviance (D0)−deviance (D1), where D0 is the deviance of the model without random effects and without parameters; and D1 is the deviance of the random intercept multilevel model with one parameter, the random intercept. Under the null hypothesis, the distribution of this statistic of the deviance difference can be approximated by the χ2 distribution with 1 df.23

Second, individual data significantly associated with the dependent variable were added (model 1). Next, significantly associated aggregate variables were added; that is, socioeconomic environment as represented by the Townsend Index (model 2). Finally, we added GP density, considered as a potentially confusing factor (model 3).

The contribution to the variance by the stepwise introduction of the different variables in the models was determined by the proportional change in variance at different levels: ((V1−V2)/V1)∗100, where V1 is the level 2 variance of the multilevel model with M1 variables and V2 is the level 2 variance of the multilevel-adjusted model M2 with M1+1 variables.24 Statistical analyses were conducted using SAS software, V.9.1, the NLMIXED procedure being used for multilevel models.

Results

Participation at the CRC screening and characteristics of the population study are summarised in table 2.

Table 2

Participation at the CRC screening and characteristics of the population study (n=8691)

Women participation was 1.20 compared with men participation (reference group=1). Participants were significantly older (mean age=61.2 years) than non-participants (60.5 years), and their distribution in age groups was significantly different (p<0.01). Uptake was significantly lower in the youngest (50–59 years) and in the oldest (70–74 years) persons, compared with intermediate ages 60–69 years, with OR=0.70 (95% CI 0.63 to 0.77) and OR=0.82 (95% CI 0.72 to 0.93), respectively. The distribution of health insurance coverage was significantly different between participants and non-participants (p<0.01), participants having more often special coverage systems (OR=1.67; 95% CI 1.30 to 2.16).

The CRC screening participation varied significantly according to IRIS (p<0.01) (table 3, empty model). Adjustment for level 1 variables increased differences between IRISs by 42.5%, signifying that disparities in inter-IRIS participation were not explained by these individual factors (table 3, model 1). Addition of the Townsend index to the model (table 3, model 2) reduced the inter-IRIS variance by 45.6%, signifying that socioeconomic status explained almost half of the disparities between IRISs. The participation in least deprived IRISs (IRISs belonging to the quintile 1 of Townsend index; reference category) was higher than in the most deprived IRISs (IRISs belonging to the quintile 5 of Townsend index; OR=0.68 (95% CI 0.59 to 0.79)).

Table 3

Contextual socioeconomic factors of participation in the first organised CRC screening campaign in a French department, Calvados (2004–2006)—multilevel analysis (n=8691)

Quintiles 2–4 of Townsend index representing a socioeconomic “average” category had a negative influence on the participation compared with the first quintile, but this association was borderline and without marked linear decreasing tendency from a quintile to the other one. On the other hand, a sharply significant negative association existed between the fifth quintile of Townsend index representing the most deprived socioeconomically category and participation.

In the last model (model 3, data not shown), GP density was not significantly associated with participation (OR=1.05; 95% CI 0.94 to 1.16).

We compared the socioeconomic level of the 8758 individuals comprising our study population to those of 762 individuals excluded for medical reasons, for whom other care or screening strategies were required. The latter significantly belonged more often to the most privileged IRISs (16.1%) and less often to the most deprived IRISs (41.3%) than the study population (12.7% and 43.0%, respectively; p=0.04), suggesting that individuals within the most privileged socioeconomic categories are more likely to be subjected to an “independent” colonoscopy. We also compared the demographic characteristics of the 283 individuals not living in the address indicated to those of the individuals included in the study. The individuals not living in the indicated address were not different from the inclusive individuals regarding sex (χ2=1.64; p=0.20), but they were more often of extreme age (74.20%) than the inclusive individuals (67.88%; χ2=5.04; p=0.025).

Discussion

This study, based on a representative sample of a target population, suggests that participation in organised CRC screening is strongly associated with contextual socioeconomic factors. After adjustment for individual factors, multilevel analysis emphasises that participation was weakest in IRISs where deprivation was most prominent. No significant influence of GP density was observed.

Our study presents some relevant methodological aspects. Using aggregate socioeconomic data protects the study from the unavoidable selection bias observed in studies using self-questionnaires. Moreover, our randomly composed sample is representative of the population concerned by organised screening. The use of data on participation or non-participation obtained from a screening organisation provided us with information on actual attitudes, hence protecting analysis from the risk of measurement bias, more common when using auto-questionnaire.11

This study suffers from certain methodological limits. Our study population only comprised individuals at an average risk of CRC, the 762 high-risk individuals or those having undergone “independent” colonoscopy in the previous 5 years being excluded from mass screening. Because these individuals were from less deprived IRISs than those included in our study, their inclusion would have reinforced our conclusions. The only putative bias regarding the representativeness of our sample derives from persons not living at the indicated address. However, these last persons were more often of extreme age, age more at risk of not participating, than those included in our study. Their inclusion would reduce the participation rate of the final sample. From there, the final sample would be even more representative of the total target population. Our study suffers from the lack of information on previous IRIS socioeconomic status. Indeed, aggregate data are derived from the last exhaustive French census led in 1999 that prevents us from taking into account the potential changes in IRIS social characteristics over the study period. In this study, we used a deprivation index that refers to a population not to individuals. Consequently, we can suspect a classification bias of individuals in IRIS. This bias could not be controlled in our analyses. A last methodological limitation was the fact the multilevel analysis does not take into account potential autocorrelations between close geographical units.25

Because no deprivation index was available in France, the four socioeconomic variables in our study were those included into the Townsend index. As distinguished by Townsend, these variables enable investigation of both the social and the material concepts defining deprivation.18 However, in the field of cancer screening, the Townsend index, similarly to the Index of Multiple Deprivation,26 lacks an important notion appreciating a social concept of deprivation: social instability. Indeed, social instability, synonymous to a break in social relationships, could play a key role in the decision to participate or not. Social instability can be appreciated in various ways, particularly by the social isolation it can generate. Some studies revealed the negative influence of living alone on CRC screening participation.11–14 Another way of appreciating social instability is to consider the rate of migration. A Swedish study on breast cancer screening 27 showed that the rate of migration was the main contextual factor influencing participation, after the rate of unemployment. This association is based on the hypothesis that high levels of residential instability, in terms of migration, tend to weaken social networks and reliable relationships within districts. Consequently, it would appear necessary to consider social instability, as represented by marital status or by the rate of migration, in future studies on cancer screening participation.

Our results are in accordance with self-report studies 11–17 and with a recent study assessing the relationship between aggregate socioeconomic factors and CRC screening behaviours.10 In this last study, a positive association was observed between living in a poverty-stricken area and not being screened for CRC. Multivariate analysis of this study reported that the participation was higher in least deprived areas (areas belonging to the first quintile of Index of Multiple Deprivation; reference category) than in the most deprived areas (areas belonging to the fifth quintile of Index of Multiple Deprivation; OR=0.41 95% CI 0.39 to 0.43), whereas our multilevel analysis revealed an OR equal to 0.68 with a lesser precision (95% CI 0.59 to 0.79). This difference of probability between the two studies can be due to difference in index of deprivation used or/and in range of socioeconomic deprivation in study population. The difference of precision can be due to a difference in the number of analysed individuals and perhaps to the lack of accounting of hierarchical structure of the data in the previous study, which can lead to an underestimation of the variance of the contextual factors.28

To our knowledge, our study, within the limits described above, is the first multilevel examination of CRC screening compliance in a representative sample of a target population.

In our study, access to healthcare, represented by GP density, was not associated with participation. Nevertheless, the spatial accessibility to primary care can only be measured correctly by combining the dimensions of availability and accessibility, such as distance to GP.29 Because this last dimension was not available for our own analysis, we could not precisely estimate the effect of access to healthcare on participation. Moreover, as recently demonstrated by a French study on hepatitis C (VHC) screening, GPs are the cornerstones of any screening programme. Using multilevel analysis, this study showed that the main factor limiting the detection of VHC was the proximity of a GP to a greater extent than the socioeconomic context.30 Because the GP does not represent the sole and unique “access” to VHC screening (patients can also be detected in anonymous and free screening units or in hospitals), the distance to the GP would therefore appear even more relevant in the case of CRC screening as compared to VHC screening, as the GP provides the only access to screening for the former.

Over and above this concept of spatial accessibility to primary care, the GP's implication is also crucial in increasing overall participation, through his/her encouragement, as highlighted by numerous studies on CRC screening.11 12 14 15 17 31 Unfortunately, no information regarding GPs' attitudes towards screening was available.

Even if our results concern the target population of Calvados, this strategy of analysis could also be applicable to other French departments, and even in other countries where there is already, or there is going to be, an organised CRC screening programme implemented. Besides, at first and for international comparisons, the Townsend index could be used because of the simplicity of its calculation from four easily available variables in spite of its lack of exhaustiveness of the socioeconomic factors associated with participation to CRC screening.

Multilevel analysis on an unbiased sample highlights infra-town areas of weak participation linked to areas of strong deprivation. Despite the controversial association between socioeconomic position and stage at diagnosis for patients with CRC,32 33 this important finding could partially explain late diagnosis for certain patients, as previously emphasised in research in France.34 These results may generate proposals for new forms of screening organisation aiming at improving participation within deprived areas, hence limiting the social and geographical disparities involved in participation.

What is already known on this topic

  • Because successful mass population screening for colorectal cancer depends on participation, it appears crucial to identify the reasons for no participation. Among these, a low individual socioeconomic level is known to influence participation negatively.

  • However, research has often been limited by self-reported measures of participation and most of the studies have been restricted to survey respondents, and so fail to take account of the socioeconomic circumstances of non-responders.

  • To our knowledge, there are few socioeconomic data of the entire target population of an organised colorectal cancer screening programme and no multilevel analysis taking into account the hierarchical structure of data.

What this study adds

  • This present study, which combines individual data of participation and aggregated socioeconomic data that came from the structure responsible for organising screening and French census, respectively, is protected both from selection bias and measurement bias.

  • Moreover, it is the first multilevel examination of colorectal cancer screening compliance, at an accurate geographical level.

  • In a representative sample of the target population of organised colorectal cancer screening, participation was strongly and significantly determined by contextual socioeconomic factors, whereas no significant association was found between healthcare access and participation.

Acknowledgments

This study received financial support from the Fondation de France. Calvados census data were provided thanks to the Centre Maurice Halbwachs (http://www.cmh.ens.fr). We also thank the Association Mathilde for having provided their data set.

References

Footnotes

  • Funding This study received financial support from the Fondation de France (Paris, France).

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.