Practical Considerations for using Social Determinants of Health for Disease Prediction in All of Us

Abstract Growing recognition that social determinants of health (SDoH) strongly influence health outcomes has expanded their inclusion in biomedical research, underscoring the need to evaluate how best to incorporate them into disease prediction models. To this end, we applied the Healthy People 2030 framework to transform rich individual-level SDoH survey data from the All of Us Research Program into theory-driven composite scores. We then compared these composite scores with area-level indices, and evaluated their associations with nine common chronic conditions. We found that diseases have distinct “social architectures,” differing in the strength and direction of associations across individual- and area-level measures. We then developed disease-specific polysocial risk scores (PsRS). Income and education generally captured the majority of disease-related signal from more complex individual-level data. Many PsRS improved when both individual- and area-level SDoH were included. Our findings underscore the value and complexity of utilizing diverse SDoH measures in disease risk modelling.