Background
Research has shown that people living in less deprived areas of high-income countries experience lower mortality and longer life expectancy than those in more deprived areas [
1,
2], and such inequalities are also found for cause-specific health outcomes for various diseases, including many types of cancer [
3‐
7]. In the UK, the NHS (National Health Service) has highlighted the importance of reducing socio-economic health inequalities in its recent long-term plan [
8], which dedicates funding and resources to narrowing inequalities over the next decade.
The majority of research on these inequalities has focused on differentials measured at an aggregated geographical level (largely due to aggregated data being more accessible than individual data), but some studies have also found that inequalities in mortality exist across individual-level socio-economic groups. For example, it has been shown that individuals on higher incomes or with a higher level of qualifications have lower mortality and longer life expectancy [
9‐
11]. In order to successfully target the underlying causes of health inequalities, there is a need to fully understand to what extent differentials are due to area-level factors such as resource distribution, and to what extent they are associated with individual-level factors such as occupational health or personal circumstances affecting ability to access healthcare.
Moreover, there is also a need to explore the potential for area-level and individual-level factors to interact with one another, such that individual-level health inequalities could differ depending on the deprivation context of the area an individual lives in. In this way, data that describes area-level, or group-level, characteristics such as overall level of deprivation or access to services could have differing effects on individual outcomes according to personal characteristics such as education level or type of occupation. These ‘contextual effects’ could be particularly important given recent evidence that the area-level deprivation where an individual lives is not necessarily a good indicator of their individual socio-economic circumstances [
12,
13]. An over-reliance on research that focuses only on area-level patterns of inequalities risks overlooking subsets of individuals who, for example, live in a relatively affluent area but have a low personal income and experience health inequalities differently from low income individuals who live in low income areas. There is also a risk that area-level differentials are interpreted principally at an individual level and the influence of the area itself is not acknowledged.
Contextual effects on all-cause mortality and overall health have been examined in several countries via a combined analysis of area-level and individual-level socio-economic measures [
14‐
16], as well as some studies of contextual effects on cancer-specific mortality [
17‐
20]. However, the effect of deprivation context on cancer survival in the UK remains poorly understood. This is the case despite the potential for such research to inform more effective health policy, especially in the face of evidence that there has been little, if any, improvement in socio-economic inequalities in cancer outcomes in recent years [
4,
7,
21].
Using population-based survival data for patients diagnosed with colorectal, prostate, and breast cancer patients, we aimed (1) to quantify the association between individual-level socio-economic variables, area-based deprivation, and cancer survival, and (2) to investigate whether individual socio-economic survival differentials vary depending on the area-level deprivation context. We discuss the results in terms of gaining a better understanding of the underlying mechanisms of these socio-economic inequalities and the implications for health policy.
Discussion
Our analyses have identified mixed evidence of individual-level socio-economic inequalities in cancer survival, as well as evidence of contextual effect modification for breast and prostate cancers but not for colorectal cancer. Adjusting for area-based deprivation hardly impacted the association between individual socio-economic variables and the EMH for colorectal cancer. Conversely, for prostate and breast cancers, the contextual effects we found widened individual-level inequalities between occupation groups depending on the level of area-based deprivation. We also found that, for prostate and breast cancers, the disadvantageous effect of area-based deprivation was substantially reduced by adjustment for individual-level effects, whereas the opposite was true for colorectal cancer.
We used a relative survival approach for the analysis, which enabled the association between socio-economic circumstances and excess (cancer-specific) mortality to be assessed independently of expected (background) mortality [
31‐
33]. The estimates of background mortality used here were based on the same population cohort (but not restricted to cancer patients), and we have previously shown that, within this cohort, there are wide inequalities in all-cause mortality and adult life expectancy between individual-level socio-economic groups [
11]. Here, by accounting for these differences in background mortality using a relative survival approach, we have focussed our analyses specifically on the association between individual socio-economic status and EMH. Our analysis does not however account for important prognostic factors such as stage of cancer at diagnosis, or treatment undertaken, which are not routinely collected in the ONS-LS dataset, and so these factors could contribute unmeasured confounding effects to our results.
We note that the net survival and EHR estimates from our models have wide confidence intervals for some sex/cancer combinations, as presented in the results. This is likely due to use of a relatively small sample size combined with relatively complex models. Our model selection approach towards examining evidence for contextual effects gives good support for the overall presence and absence of the contextual effects described here, but the wide Wald-type confidence intervals around the EHR estimates for specific socio-economic sub-groups means that we can only describe the observed trends contributing to the contextual effects. On a related note, we would like to point out that we were unable to find evidence of an association between some individual-level socio-economic variables and excess mortality hazard, but it does not mean that such associations do not exist. We also note that all estimates are based on the model with the lowest AIC. The models with an AIC within 2 units of the lowest AIC (Table S
1) could also be supported by the data, and so multi-model inference would be an interesting research avenue for further analysis [
39].
The differentials observed here varied across the different individual-level socio-economic variables included in the analysis. Disparities were most notable according to occupation type for men with prostate cancer and women with colorectal cancer; across income quintiles for women with breast cancer; and across education groups for men with colorectal and prostate cancers. In addition, there was a slight trend for higher EMH in prostate cancer patients associated with higher income. Although unexpected, this trend could be partially explained by the analysis adjusting for individual-level occupation and education, which are likely to be linked to income. Generally, it might be expected to observe individual-level socio-economic inequalities that consistently reflect the well-documented area-level deprivation differentials [
3‐
7]. However, the individual-level variables were based on specific dimensions of socio-economic status, as opposed to the summary deprivation scores used at an area-level, and as such the observed mixed results might suggest that the specific underlying mechanisms of individual-level inequalities can differ between cancer sites.
We found evidence of persistent area-level deprivation inequalities for colorectal cancers in the models that included both individual socio-economic variables and area-level deprivation. Previous literature has comprehensively shown that cancer survival is lower in more deprived areas [
3‐
7], but the increase of area-based deprivation inequalities even after adjustment for individual patient characteristics in this analysis is novel. Conversely, the lack of area-level deprivation inequalities for breast and prostate cancers after adjusting for individual-level effects was unexpected, given area-level differentials documented previously for breast [
4] and prostate cancer [
40], and could suggest that these inequalities are more effectively explained by individual-level variables. Although the variables analysed here might not necessarily explain the inequalities directly, they might act as a more appropriate proxy for the underlying mechanism. For example, inequalities in breast cancer survival could be partially explained by individual characteristics (such as BMI) influencing treatment pathways, and so focussing on individual-level characteristics to estimate inequalities might be more appropriate than the use of area-level scores in certain cases.
The models used here enabled us to simultaneously investigate individual socio-economic status and contextual effects for cancer survival, while properly adjusting for age. To our knowledge this has not been explicitly examined in a UK population setting to date. For prostate cancer, and to some extent for breast cancer, there was evidence that inequalities in cancer survival across individual-level occupation groups were wider in more deprived areas than in less deprived areas. Although some of this effect could be accounted for by a ceiling effect, in which differences between groups will appear increasingly smaller as survival approaches 100%, it is unlikely this phenomenon explains all of the wide differences observed here, particularly for prostate cancer. The observed contextual effect amplifies inequalities, such that the sub-group of individuals in manual/technical occupations who live in the most deprived areas experience a survival disadvantage in addition to that previously estimated in area-level analysis. This differential is not detectable in studies that use only area-based socio-economic metrics. As a result, policies relying exclusively on such studies may overlook these individuals.
In addition to widening inequalities in the most deprived areas, the contextual effect observed for women with breast cancer indicated differences in terms of which occupational groups experienced the highest cancer survival in different area-level deprivation contexts. It is unclear from this analysis why this might occur. It is possible that unmeasured confounding could help to explain this effect, for example, if women in manual/technical occupations living in less deprived areas tend to be in households that are generally less deprived due to the income or occupation of their partner. Further research could consider the underlying reasons for these patterns. Analysis of household income as opposed to individual income may be useful to consider. However in this study, data pertaining to non-ONS-LS members of the relevant households is not widely available. This precluded the estimation of household total income.
For colorectal cancer for both men and women, there was no evidence of any contextual effect modification due to area-level deprivation, whilst there was stronger evidence of area-level inequalities. There was also some indication of individual-level inequalities across education and income groups for men, and across occupation groups for women. These results suggest that individual- and area-level deprivation exert independent effects on cancer survival, and future healthcare policy is likely to benefit from accounting for both these potential sources of inequality.
A major reason for the relative lack of research on individual-level and contextual effects on health outcomes, especially in a UK setting, is that individual-level data is protected and aggregated area-level data are far more accessible. Studies such as the ONS-LS offer rich individual-level data that could be used for follow-up work in this area, and the ONS-LS is particularly useful in this respect due to its representativeness of the overall population [
22]. In spite of this, the cohort is based on only approximately 1% of the whole population, and so when focussed to individual cancer sites, sample sizes are not large enough to consider less common cancers, nor large enough to incorporate more complex effects such as modelling non-proportional hazards for the socio-economic variables. Further work should consider ways to adapt these methods to extend the generalisability and detail of the results. Although it is possible to group cancer sites for analysis, this would need to be done with caution, since we have shown here that contextual effects can differ between cancer sites. These differences are likely to be biologically meaningful, given the disease-specific differences in screening, diagnosis, and treatment, meaning that combining cancer sites for analysis is unlikely to be appropriate.
Contextual effects on cancer outcomes have been explored to some extent in other countries [
15‐
20,
41]. Research in the USA has found evidence of context-specific inequalities and support for multi-level health policy interventions relating to breast, colorectal and prostate cancers [
18‐
20]. These are consistent with our results for prostate and breast cancers, although we find no evidence for contextual effects for colorectal cancer in this population cohort. A study focused on prostate cancer outcomes in a Californian population found individual-level inequalities across education groups consistent with the survival disadvantage we identified here for men with school-level education, and in addition they found contextual effects of these inequalities [
20]. Although our study found evidence of contextual effects across occupational types rather than educational level for men with prostate cancer, education level and occupation type are correlated so further research could explore the underlying causes of these contextual effects in more detail. Furthermore, a systematic review based on USA data highlighted the focus on common cancer sites for such analyses, reinforcing the point that there is a need for studies to be extended to consider contextual effects for less common cancers [
19].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.