Sampling Bias and Its Consequences in Biobank-Based Genetic Studies

Large biobanks linking genetic data with electronic health records have transformed human genetics research. Because most biobanks enroll by convenience rather than probability sampling, recruitment, consent, and participation processes shape whose data are analyzed. We review bias arising from non-probability sampling across the biobank landscape and methods for detecting and reducing it. We then examine its consequences for common genetic analyses: descriptive assessment, population substructure, genetic association, Mendelian randomization, and risk prediction. The impact is not uniform: descriptive quantities and analyses of disease-ascertained samples are especially vulnerable, whereas genetic association tests and risk prediction are often more robust. When the goal is estimating effect sizes rather than testing null hypotheses, bias-reduction tools should be considered. We conclude with recommendations for reporting biobank sampling strategies, defining source and target populations, stating bias detection and reduction approaches, and outline related open challenges for the biobank research community.