Resolving disease causality by integrating polygenic risk into propensity score matching

Background Disentangling causal relationships between diseases from associations driven by shared genetic and environmental risk remains a major challenge. Methods We developed PRIMA (polygenic risk–integrated matching approach), a framework that incorporates exposure-specific polygenic risk scores and clinical covariates into propensity score matching. We evaluated PRIMA using liability-threshold simulations and disease-pair analyses in 333,370 unrelated UK Biobank participants of European ancestry, and compared its estimates with clinical matching, Mendelian randomization and survival analyses. Results In simulations, PRIMA reduced bias due to shared genetic liability and recovered effect estimates closer to the expected causal effects than clinical matching alone, while remaining sensitive to unmeasured confounding. In UK Biobank analyses, integrating polygenic risk improved genetic and clinical balance and materially altered association estimates for a subset of disease pairs. PRIMA identified additional associations not detected by Mendelian randomization and showed greater concordance with survival analyses. Conclusions PRIMA provides a scalable framework for incorporating polygenic liability into observational disease-pair analyses. It complements existing causal inference methods, although causal interpretation remains conditional on appropriate model specification, sufficient overlap and adequate control of confounding.