Pervasive interactions between exposures and polygenic risk can inform more effective clinical and behavioral...

The generalizability of polygenic scores (PGS) remains a major hurdle in the pursuit of equitable genomic medicine. Differences in disease prevalence across groups, potentially including social strata, influence the relationship between PGS and risk. Here we quantified the magnitude of PGS-by-context (PGS×C) interactions for seven human diseases and pairs of 75 contexts in the UK Biobank (n = 408,801). Across 24,198 PGS×C models, 746 (3.1%) had significant interactions by two criteria, up to fourfold more than expected by chance, improving predictive accuracy. The predominant mechanism for PGS×C is the amplification of genetic effects in adverse contexts, such as low polyunsaturated fatty acids or social determinants of ill health. We introduce the notion of the proportion needed to benefit as a metric that quantifies the expected effectiveness of interventions as a function of polygenic risk. Our results highlight the need for more comprehensive sampling across groups experiencing adverse exposures.