Where Polygenic Risk-Based Estimates of Absolute Risk Start To Fail
In this second article of this series, Wayne Delport explores the statistical limits of polygenic risk scores (PRS) and why extreme estimates of absolute disease risk often become unreliable at the far ends of the distribution curve. Focusing on the “tails” of polygenic risk modeling, the article explains how small errors in effect sizes, population assumptions, and prevalence estimates can compound dramatically when predicting very high or very low lifetime disease probabilities.
The piece examines how genomic prediction models are built, why odds ratios do not translate cleanly into absolute risk, and how calibration failures emerge when models are applied outside the populations or conditions they were trained on. It also discusses the implications for precision medicine, consumer genomics, preventive healthcare, and AI-driven risk prediction systems.
By combining statistical analysis with practical examples, the article highlights the importance of transparency, calibration, and clinical context in genetic risk interpretation. It argues that while polygenic risk scores can provide meaningful population-level insights, caution is required when presenting individualized absolute risk estimates — especially when those estimates appear unusually extreme.
This article is relevant to researchers, clinicians, bioinformaticians, and digital health companies working in genomics, predictive medicine, polygenic risk modeling, and AI-based healthcare analytics.
Wayne Delport is a bioinformatics and genomics technology leader with more than two decades of experience building scalable computational biology platforms and advancing precision medicine. As Co-Founder and CTO of Simplify Genomics, he leads the development of AI-driven genomic interpretation and search technologies designed to accelerate clinical decision-making and make whole genome data actionable in healthcare. Wayne has authored widely cited research in evolutionary biology, genomics, and computational methods, with thousands of academic citations across his work. His expertise spans cloud-based bioinformatics systems, genomic data infrastructure, and translating complex biological data into practical clinical applications.