Deprivation and Cancer Survival in Scotland
Session type: Proffered paper sessions
Theme: Epidemiology and prevention
Risk of death from cancer is influenced by numerous factors. This analysis concentrated on the effect of deprivation (Scottish Index of Multiple Deprivation-SIMD).
Net survival was calculated for all Scotland by deprivation quintile for the most common twenty cancers. Eight cancers (lung, breast, prostate, colorectal, head and neck, malignant melanoma of skin, oesophageal, and liver) demonstrated statistically significant (p<0.05) differences between the most and least deprived quintiles and/or there was a linear trendline across all deprivation quintiles.
Differences in the hazard of death (HR) and excess mortality for these cancers were compared across deprivation groups to those in the least deprived group (at baseline). Six cancers demonstrated significant differences in excess mortality for the most deprived quintile: breast (1.89), colorectal (1.45), head and neck (1.61), liver (1.28), lung (1.08), and prostate (1.98).
Next, multivariate techniques were used to determine the HR whilst considering other factors potentially driving the variation – separately and in combination. The additional factors varied by cancer, but typically included information on the patient (e.g. existing co-morbidities), the tumour (e.g. tumour grade), and treatments (e.g. use of surgery).
There were two outcomes across these six cancers: the differences in survival between the most and least deprived was accounted for by the factors modeled (liver, lung), or the final model was not able to explain all the variation in survival by deprivation (breast, colorectal, head and neck, and prostate).
When modeled factors explain deprivation's impact, improving these may help reduce inequalities. For example, if stage at diagnosis explains much of the variation, then interventions such as early detection awareness may improve detecting cancer at an earlier stage, reducing differences in survival. Unexplained variation is likely due to factors not in the model (e.g. smoking), measurement error, or other issues, such as differing expectations of and access to health services.