Assessing the risk of endogeneity bias in health and mortality inequalities research using composite measures of multiple deprivation which include health-related indicators: A case study using the Scottish Index of Multiple Deprivation and population health and mortality data

Abstract

The inclusion of health-related indicators in composite measures of multiple deprivation introduces a risk of endogeneity bias when using the latter in health inequalities research. This bias may ultimately result in the inappropriate allocation of healthcare resources and maintenance of preventable health inequalities. Mitigation strategies to avoid this bias include removing the health-related indicators or using single constituent domains (such as income or employment class) in isolation. These strategies have not been widely validated. This study used population-level health and mortality data with a contemporary composite measure of multiple deprivation (Scottish Index of Multiple Deprivation; SIMD) to assess these mitigation strategies. The differences between deprivation methods (original, health excluded, and income domain) were negligible. The results of quantitative research on health inequalities are unlikely to be affected by endogeneity bias.

Pre-pandemic inequalities in the burden of disease: a Scottish Burden of Disease study

“…Our measure of deprivation is area-based rather than individual measure and is therefore likely to misclassify many individuals into categories that do not reflect their individual experiences [32]. Although the SIMD includes health indicators in its range of domains (thereby raising the theoretical possibility of reverse causality whereby people are ordered by the health outcome rather than socioeconomic deprivation), the employment-income deprivation index (i.e., excluding the health measures) is very highly correlated with the overall SIMD index and so this is unlikely to have changed the results [33,34]. Furthermore, The SII and RII have the advantage that they are based on data about the whole population, rather than just the extremes, and so take into account inequalities across the entire distribution of inequality.…”