Sex differences in the genetic architecture of clinical quantitative traits in the electronic health record

Sex differences exist in complex diseases, but the genetic architecture of sex differences in quantitative clinical traits is understudied. Here, we performed sex-stratified genome-wide association studies of 508 clinical traits from electronic health records (EHRs) in 64,956 European-ancestry individuals in the Vanderbilt biobank, BioVU, with replication in the UK Biobank and Colorado Center for Personalized Medicine biobank. Most trait distributions differed significantly by sex. We identified a female-specific locus for erythrocyte distribution width at PIEZO1 and sex-differentiated effect magnitudes at APOE (LDL cholesterol) and SLC2A9 (uric acid) that replicated across biobanks, varied with age, and partially mediated associations with heart disease and gout. Opposite-direction sex effects were less replicable. Ascertainment bias could be addressed using inverse-probability weighting. Sex-stratified colocalization with eQTLs nominated additional candidate genes. Differential effect magnitude, not opposing effects, predominate in sex-specific genetic architecture, requiring large sample sizes for discovery that can be achieved using EHR data.