Augmenting Missing Individual Social Determinants of Health with Area-Level Social Deprivation Index for Gene-Environment Interaction: Application to Opioid Use Disorder Prediction.

Detecting gene-environment (G×E) interactions is challenging when individual-level social determinants of health survey responses (iSR) are highly missing. We hypothesize that an area-level Social Deprivation Index (aSDI), derived from ZIP3 codes with <1% missingness, can augment or substitute for iSR variables, which exhibit ∼62% missingness in the All of Us Research Program. We developed a generalized linear mixed-effects model framework incorporating ZIP3-level random effects, assessed concordance between iSR and aSDI through correlation and quintile analyses, and evaluated G×E interactions between polygenic risk scores and aSDI. We applied this framework to prediction of opioid use disorder (OUD) in All of Us (N = 123,325 opioid-exposed adults). Composite quintile concordance between iSR and aSDI was moderate (r = 0.52). Models using aSDI alone (AUC = 0.822) performed comparably to iSR-only models (AUC = 0.825). Leveraging aSDI to address missing iSR expanded the analytic sample from ∼43,000 to ∼123,000 and enabled detection of significant gene-poverty and gene-income interactions. The full model integrating clinical, genetic, individual, and area-level social features achieved AUC = 0.835.