Background: Pharmacogenetics is a cornerstone of precision medicine, with several gene–drug pairs already embedded in clinical guidelines. Previous pharmacogenetic studies were typically limited to small cohorts, restricting discoveries to common variants with large effects. In contrast, our study leverages large-scale prescribing records from the UK Biobank, enabling comprehensive detection of both common and rare variants, including those with more modest effect sizes.

Analysis: We analyzed dose and treatment-duration associations for 37 drugs in cardiology, neurology, and psychiatry using PLINK2 in 140,000 participants. Prescribed doses and treatment durations were algorithmically extracted from primary care prescription data, accounting for intervals and significant dose changes over time. Genetic predictors included 362 standardized HLA variants and pharmacogene activity scores for six cytochrome P450 (CYP) genes. We robustly replicated known associations for warfarin dose (CYP2C9 and CYP4F2, BH-adjusted p = 2e-80 and 8e-5) and amitriptyline (CYP2D6, p = 8.8e-3).

Findings: Novel findings include associations of diazepam dose and gabapentin treatment duration with CYP pharmacogenes, and bisoprolol dose with HLA variants, potentially reflecting differences in tolerability, clinical response, or prescribing practices. This work highlights the UK Biobank’s unique value — combining deep phenotypic, genetic, and longitudinal primary care data — as a powerful resource for large-scale pharmacogenetic discovery.