Impact of Selection Bias on Polygenic Risk Score Estimates in Healthcare Settings

Abstract

ABSTRACTBackgroundHospital-based biobanks have become an increasingly prominent resource for evaluating the clinical impact of disease-related polygenic risk scores (PRS). However, biobank cohorts typically rely on selection of volunteers who may differ systematically from non-participants.MethodsPRS weights for schizophrenia, bipolar disorder, and depression were derived using summary statistics from the largest available genomic studies. These PRS were then calculated in a sample of 24,153 European ancestry participants in the Mass General Brigham (MGB) Biobank. To correct for selection bias, we fitted a model with inverse probability (IP) weights estimated using 1,839 sociodemographic and clinical features extracted from electronic health records (EHRs) of eligible MGB patients. Finally, we tested the utility of a modular specification of the IP weight model for selection.ResultsCase prevalence of bipolar disorder among participants in the top decile of bipolar disorder PRS was 10.0% (95% CI: 8.8%-11.2%) in the unweighted analysis but only 6.2% (5.0%-7.5%) when selection bias was accounted for using IP weights. Similarly, case prevalence of depression among those in the top decile of depression PRS was reduced from 33.5% (31.7%-35.4%) in the unweighted analysis to 28.9% (25.8%-31.9%) after IP weighting. Modular correction for selection bias in intermediate selection steps did not substantially impact PRS effect estimates.ConclusionsNon-random selection of participants into volunteer biobanks may induce clinically relevant selection bias that could impact implementation of PRS and risk communication in clinical practice. As efforts to integrate PRS in medical practice expand, recognition and mitigation of these biases should be considered.