Different socioeconomic indicators detect survival differentials of varying magnitude and definition. For all cancers combined, the four indicators show similar effects. For individual cancers there are differences between indicators. But where there is an association, all indicators show poorer survival with lower socioeconomic status.
Many studies have demonstrated associations between socioeconomic indicators and mortality or survival following a cancer diagnosis, although others have inconclusive results.
Kogevinas et al, using the LS, found wide survival differentials between housing tenure groups. A later LS study found rather weak effects of better survival among owner-occupiers, compared to tenants, for younger subjects suffering from breast, ovarian or prostate cancer.
Using data from the South Thames Cancer Registry, Schrijvers et al investigated relative survival from ten common cancers. The Carstairs index was used as a proxy of socioeconomic disadvantage. They found poorer survival with increasing deprivation level of area for 7 out of the 10 cancers. Stomach, pancreas and ovarian cancer showed directionally similar gradients but these were weaker and insignificant in controlled models.
The largest and most detailed study of cancer survival in the UK to date is the CST study. Using the Carstairs index, it showed improved survival among those living in less-deprived areas. However, this was only confirmed statistically for 21 cancers. Despite its size and comprehensive scope the CST study did not employ multivariate methods, indeed only some results were age-standardised. The Scottish study was more modest in scope but did test survival by age and sex controlled Cox models. This also used the Carstairs score as a socio-economic indicator and gradients were found for many cancers.
McDavid et al conducted one of the few studies to date using controlled multivariate models of relative excess mortality to investigate cancer survival. Survival from cancers of the breast, prostate, lung and colorectum in Kentucky were all strongly associated with type of health insurance held – itself related to socioeconomic status.
Greenwald et al suggested socioeconomic influence on survival may have separate mechanisms of action – survival differentials for highly lethal cancers may be the result of income differentials buying better care (in the US), whereas for cancers of better prognosis education differentials may affect disease progression. Supporting this, an investigation into cancer survival in Turin, showed a strong positive relation between survival and higher levels of education for some less lethal cancers only.
Level of education is perhaps the missing socioeconomic indicator in this study. The British census has made various attempts to capture educational level in censuses and the questions asked in the 2001 census are very promising. However questions in the 1981 and 1991 censuses asked only about education after age 18. For the older cohorts in this study, especially women, this is even less useful than social class.
To our knowledge this is only the second application of multivariate models of relative excess mortality (the complement of relative survival) to British cancer data: previous studies using less sophisticated case fatality rates, Cox regression models or even simpler statistics. In addition our models are not only controlled for age and sex but also for period of diagnosis and geographic zone.
The fact that lifetables are not easily available by socioeconomic status and therefore national tables are used here for estimating relative survival could theoretically over-emphasise socioeconomic differentials. This effect, though recognized, is likely to be small and is largely obviated by socioeconomic controls in survival models. Overall relative survival estimates obtained were very similar to those presented by the CST study, which did use specially constructed lifetables reflecting socioeconomic gradients.
The finding that combined cancers show strong socioeconomic gradients for survival is not quite as clear-cut as it appears because it is confounded by the fact that many of the cancers with higher incidence among the disadvantaged are those that are rapidly lethal (lung, oesophagus). Those that show slightly higher incidence in the more advantaged (breast, prostate) have longer survival periods. In other words the "case mix" of combined cancers varies with socioeconomic status and this affects survival.
All indicators suggest a socioeconomic gradient for lung cancer. This gradient is small but well defined and is perhaps an example of improved power of detection of such small effects for a common cancer. For other individual cancers there are differences between the indicators. Survival gradients by Social Class are generally not well defined, probably due to the known problems of unclassified individuals, especially older women.
The import of demonstrating survival differentials between groups with rate ratios in the order of 1.04 (lung) is questionable but is related to the speed of lethality of the cancer and may well reflect only lead time differences. For sites such as bladder and cervix/uterus, for which survival times can be lengthy, rate ratios approaching 2 between groups are substantial, important, and less likely to be merely lead time bias.
The Carstairs score picks out bladder and breast cancer as having a socioeconomic gradient for survival but only for bladder cancer do the car access and tenure indicators endorse this effect, showing a much stronger association than the ecological indicator. The effect concurs with recent findings for bladder cancer but may well have been missed had only the Carstairs score been used. Car access and tenure detect an association for colorectal cancer. This is not particularly well defined but is not detected by the Carstairs measure at all. This is unexpected since the CST study found a noticeable gradient with Carstairs quintile, although their deprivation results were not age-standardised. However not all studies have found socioeconomic differentials in survival from this cancer and this association may well vary by indicator used.
For prostate and cervix/uterus a gradient by Carstairs' score of the local area is suggested in uncontrolled models (not shown) but not confirmed in controlled models. This suggests that at least some of the gradients identified in the CST study might not have persisted after age/sex adjustment. However the Scottish trends study does suggest a socioeconomic gradient (measured using Carstairs score) for age adjusted Cox models of prostate and cervix (but not uterus).
Survival gradients in combined leukaemias are difficult to demonstrate, probably because they are not a homogenous group of diseases. The Scottish study does not confirm a survival gradient by Carstairs quintile for the combined group but the larger CST study demonstrates a large gradient for chronic lymphoid leukaemia but very much smaller effects in other leukaemias.
The indicator with an interesting profile of results is car access: an indicator that is widely used in social science studies. It shows a socioeconomic differential for colorectal cancer and quite strong effects for cervix/uterus and ovary that are not shown by Carstairs quintile or particularly well by tenure. The effect for ovarian cancer is particularly interesting as it seems to be a strong effect. Although the Scottish and CST studies detect a significant gradient by Carstairs quintile it is not strong. These differences suggest that some indicators are particularly sensitive to differences in survival for particular cancers, possibly the more lethal women's cancers, and may be more useful in these cases.
Although there is undoubted correlation between the indicators used here it may well be that the utility of different indicators is their ability to select rather different social groupings. For instance no car access (in Britain) may be selecting a relatively small, particularly deprived, and possibly rural biased, group as well as the elderly.
It is tempting to explain socioeconomic differentials in survival by the simple explanation that patients of lower social class, or education, ignore symptoms and present later for diagnosis. Some studies have found less compliance with screening and more co-morbidity among certain groups, including the socially disadvantaged [24–29]. However in two studies using comprehensively controlled multivariate methods, adjustment for stage at diagnosis did nothing to change socioeconomic differentials[6, 15]. Other studies have also failed to show definitive links between deprivation and cancer stage or biology or tumour size [23, 26, 27, 30–32]. Therefore, although delayed presentation must play a part, it is by no means a complete explanation.
The fact that socioeconomic association with survival appears to be specific for certain cancers does not endorse any simplistic biological theory linking disease progression in general with poverty or disadvantage. In populations with largely equitable healthcare access this leaves differential treatment, health status and co-morbidity, coping and support strategies (embracing level of income), and understanding of, and ability to influence, disease progression (embracing educational level) as explanatory mechanisms for differential survival by socioeconomic status.
Ways of measuring socioeconomic status are manifold and their utility for older people is especially problematic. Nevertheless it is important to consider different measures, especially when it is possible that the measures most widely used in cancer studies – the ecological measures – may not be the most sensitive.