Cancer is one of the leading causes of mortality worldwide and the second in the developed countries. It is thought to be responsible for around 13% of the total number of deaths, approximately 7.6 million persons dying from cancer in 2008. While cancer survival continues to improve essentially thanks to progress in treating patients and to screening, the observations concerning incidence are much less encouraging. Social deprivation can be singled out as responsible for part of this cancer incidence and the struggle against social inequalities in cancer constitutes a priority for international organizations 1 .

Public action to reduce this gradient must rely in part on the proper assessment of the burden of cancer associated with social environment and on the knowledge of the mechanisms underlying such inequalities.

Studies of this type have initially focused on mortality data 2 3 . But it is important to differentiate between social disparities in incidence of cancer and social disparities in survival as it was the case in the literature of the recent years. The relationship between cancer incidence and socioeconomic status is dynamic and needs to be continuously monitored.

The mechanisms by which the social environment influences the risk of cancer are many and varied. None of these mechanisms are exclusive and all interact. Based on the work of previous authors, these mechanisms are organized in behavioral models focusing on individual determinants 4 5 (alcohol, tobacco, diet, physical exercise, practice prevention, etc.), or contextual models focusing on complexity determinants 6 7 (occupational exposure, general exposure, access to health system, etc.). This complexity suggests that a proper evaluation of the social environment should not be limited to any particular indicator such as financial resources, education or profession, but should appreciate the social environment in its entire individual and collective dimension. Geographical approaches are thus particularly relevant for studying the link between social environment and cancer incidence. Moreover, from a public health point of view, the measure of the human cost of these inequalities at an aggregated level is particularly relevant for potential further actions.

The objective of the study was to quantify the part of the cancer burden related to social deprivation. We firstly identified the cancer sites linked to the socioeconomic status of the living area and secondly measured for each one the proportion of cases of cancer associated with social deprivation.

For the whole study period, 68 967 cases of cancer were recorded in Calvados and Manche, 40 080 men and 28 887 women.

The most frequent sites in decreasing order were prostate, breast, lung, colon-rectum and lips-mouth-pharynx (Table 1).

Concerning the continuous deprivation index EDI, the minimum was -3.77 for the most affluent IRIS and the maximum was 8.98 for the most deprived IRIS, the median being -0.45. Quintiles being defined at a national level, 20% of the population was situated at the first quintile, 22% at the second, 23% at the third, 23% at the fourth and 12% at the fifth.

Tables 2 and 3 present the results of modelling using the continuous version of EDI.

^aPositive for an over-incidence in deprived areas, negative otherwise.

^bSignificant CIs are in bold type.

^aPositive for an over-incidence in deprived areas, negative otherwise.

^bSignificant CIs are in bold type.

The Potthoff-Whittinghill test and the Moran test were significant for a majority of sites.

The link between incidence and social deprivation was not significant for a majority of cancer sites in both genders, was positive for 9 sites in males and 4 sites in females and was negative for two in males and none in females. For lung, lips-mouth-pharynx and unknown primary sites, the link was positive in both genders. We obtained similar betas for both genders but the sites concerned were more frequent in males so the impact in terms of number of cases was greater in males. The link was positive in males only for bladder, liver, esophagus, larynx, central nervous system and gall-bladder and in females only for cervix uteri. The highest relative risks concerned lips-mouth-pharynx in both genders, larynx and gall-bladder in males and cervix uteri in females.

Tables 4 and 5 present the relative risks calculated using the quintile version of EDI and the results of the calculation of the PAF.

^aPAF calculated with quintile 1 as reference except for prostate cancer and melanoma.

^aPAF calculated with quintile 1 as reference.

Using the calculation of PAF, the greatest part of the incidence associated with deprivation was found for lips-mouth-pharynx cancer, esophageal cancer, laryngeal cancer and gall-bladder in males, respectively 26.0%, 19.6%, 23.2% and 30.1%. In females, the greatest part of the incidence associated with deprivation was found for unknown primary sites (18.0%) and lips-mouth-pharynx (12.7%). For prostate cancer and melanoma in males, the sites where incidence increased with affluence, the part associated with affluence was respectively 9.6% and 14.0%. The excess cases due to social deprivation are represented in Figures 1 and 2. The highest number of cases attributable to social deprivation concerned lips-mouth-pharynx cancer in males (n = 820) (Figure 1) and unknown primary sites (n = 120) (Figure 2) in females and for prostate cancer, 1115 cases can be considered as excess cases due to affluence and for melanoma in males, 90 cases can be considered as excess cases due to affluence. By adding excess cases associated with deprivation, we find 2287 excess cases in men (5.7% of the total number of cancers in men) and 353 in females (1.2% of the total number of cancer in females).

This study provides evidence of social disparities in the incidence of cancers. Most of these disparities consist in an over-incidence for the most deprived, especially for lips-mouth-pharynx, lung, unknown primary sites, bladder and larynx cancers. Both genders are concerned, but the impact is greater in men, considering the huge frequency of these cancer sites for them. These inequalities in incidence are all the more serious and the cancer burden is all the greater in that the cancer sites concerned are those associated with very low survival. For the period 1997–2009, analysis with the PAF showed that the social gradient generated 2287 (5.7%) excess cases in men and 353 (1.2%) in females. By analyzing site by site, the social gradient generated up to 30.1% (gall-bladder) and 18.0% (unknown primary sites) extra cases in men and women respectively.

The sites identified as linked with socioeconomic status are not surprising and consistent with previous papers. Thus, the highest incidence for lung, lips-mouth-pharynx, esophagus, larynx, bladder and liver cancer in low socioeconomic status can be explained by a higher consumption of alcohol and tobacco in the most disadvantaged 5 18 19 . Similarly, the trend in over-incidence of cervical cancer in deprived women can be explained by sexual behaviors and/or lower participation in pap smear screening 20 . The highest incidence of cancers with unknown primary sites in males and females with a low socioeconomic status can be explained by the fact that the group of “unknown primary sites” mainly comprised subjects with metastatic cancers where the primary site could not be identified, a situation more frequent in people with a low socioeconomic status 21 . Results in the literature concerning the relation between incidence of central nervous system cancer socioeconomic status are contradictory. The etiology of cerebral tumors remains unclear 22 23 . The results concerning gall-bladder are consistent with previous papers. People with a low socioeconomic status may have a diet and a feeding behavior which contribute to a development of the disease 24 . The trend in over-incidence of prostate cancer may come from the higher participation of high socioeconomic classes in screening activities and since PSA screening is associated with over diagnosis 25 . The higher participation of high socioeconomic classes in screening activities can also explain the higher incidence for affluent patients for melanoma in males and this higher incidence can also be explained by holidays abroad and exposure to natural UV 17 26 . Conversely, the absence of a social gradient in the incidence of breast seems surprising, since it is targeted by screening associated with social inequalities in participation, and because well-established risk factors such as late age at first birth or hormone replacement are more prevalent in high socioeconomic groups 6 . The spatial nature of the data and its specificities (spatial autocorrelation and non spatially structured heterogeneity) was accounted in our modelling thanks to the Bayesian approach ensuring a good consistency of the statistical analysis. Such a methodology was not integrated in previous studies treating cancer incidence and social disparities, preferring a classical Poisson regression, and thus risking to underestimate the standard error and to wrongly conclude at a significant effect of deprivation on cancer incidence 27 .

Our study has several limits. By using the PAF and in absence of individual data, we sought to quantify social inequalities in incidence of cancer, rather than understand the underlying mechanisms. Using a neighborhood-based index instead of a set of individual indicators has the advantage of incorporating both individual and collective determinants that jointly mediate the social environment, but this inevitably introduces an ecological bias for appropriate measurement of individual socioeconomic status. Moreover, it considerably limits the search for causative factors explaining the links between social environment and occurrence of cancer, individual measures of socioeconomic status and behavioral risk factors being the best means to explore in more depth the mechanisms responsible for the influence of social environment on cancer risk. In addition, the social environment was measured only at the time of diagnosis, using the current address of patients but ignoring their history of mobility, which could be geographical and across social classes. Furthermore, we focused on the consequences of previous social inequalities owing to the delay between exposure and diagnosis. Despite the large number of cases analyzed from cancer registries that have a high level of case ascertainment, consistency and representativeness, a lack of power cannot be excluded for the less frequent cancer sites.

Extrapolation of the PAFs needs further investigations in order to ascertain their variability due to gradient in relative risks, or to distribution across social quintiles. Errors in interpretation can appear, as highlighted in the article by Rockhill, et al. 16 with the use of the PAF. Firstly, Rockhill et al. point out many errors possible when analyzing multiple risk factors which is not the case of our study. The second point is the overuse of the word “explain” in the interpretation of the PAF. Rather than explain, it measures the extent of the phenomenon of deprivation on cancer incidence. The PAF should be considered as the population resultant of the overall excess of cases in deprived compared with privileged people. The socioeconomic environment is not a causal factor of cancer in the biological sense of the term. However, since much of the proximal risk factor is more prevalent in the deprived, the socioeconomic environment can be considered as the "cause of the cause", a distal determinant, pathways from deprivation to health including different types of mediators such as behavioral, community, social, educational, work-related, cultural and political factors 28 . Such quantification of social disparities at a community level points to the need to jointly take actions in a universal approach and also in approaches targeting deprived people, rather than global population actions only that fail to reduce social gradients because they generally benefit the more affluent. The PAF makes it possible to estimate the collective gain that could be obtained by public actions aiming to reduce the social gradient of incidence by measuring the extent of the population for which it is necessary to lead effective cancer prevention.

This study proposes an estimation of the proportion of cancers associated with social deprivation and show how by decreasing socioeconomic variation in incidence with policies aiming to reduce social inequalities, an important impact could be made on the burden of cancer.

The authors declare that they have no competing interests.

JB, OD and GL worked on the conception and design. OD, VB, AVG, XT, FGS, SB, CP and LL participated in the acquisition of data. JB performed the analysis and interpreted the data with OD, VB, MC, LG and GL. JB, OD, VB, MC, CP, LG and GL revised the manuscript and all authors read and approved the final manuscript.