Empirical design of population health strategies accounting for the distribution of population health risks

Abstract

Recent discussions in epidemiology have emphasised the need to estimate the heterogeneous effects of risk factors across the distribution of health outcomes for better aetiological understanding of the determinants of population health. We propose using quantile regression-based decomposition to expand the empirical discussion on population health intervention strategies for health equity by incorporating population homogeneity/heterogeneity in the risk–outcome association. We theorised that the ‘proportionate universalism’ approach presumes population homogeneity in the risk–outcome association with varying risk intensities, which decomposition analysis shows as the ‘covariates part’ between groups. Conversely, the ‘targeted approach’ assumes population heterogeneity in the risk–outcome association across the outcome range, which the analysis identifies as the ‘coefficients part’. Our demonstration, using a case of education-related disparity in dietary behaviours, exemplified that differences between education groups were mainly explained by the coefficients part. This finding suggests heterogeneity in their risk profiles, necessitating a ‘targeted approach’ across outcome quantiles to close the gap. The ‘proportionate universalism’ strategy could be partially applied to specific quantile segments where the covariates part remained significant as a supplementary intervention. However, simply increasing the magnitude of certain risk factors (e.g., income) showed conflicting directions between covariates and coefficients parts. Structural modifications of risk–outcome associations would therefore be more equitable. We also discuss the potential strengths and limitations of the analysis, suggesting that it may be complemented by data-driven methods using machine learning to identify discriminating risk factors for population health equity.