A0887
Title: Polygenic risk score methods to improve disease screening
Authors: John Witte - Stanford University (United States) [presenting]
Abstract: Polygenic risk scores (PRS) combine information across large numbers of genetic variants to give an individualized genetic susceptibility profile that may be useful for disease prediction. PRS can also be leveraged to improve biomarker and screening accuracy. For example, genetically adjusting prostate-specific antigen (PSA) with a PRS for this biomarker improves prostate cancer screening. Specifically, diagnostic decisions based on PSA values adjusted using a PRS would have avoided 31\% of negative prostate biopsies but also resulted in 12\% fewer biopsies in prostate cancer cases, mostly in patients with Gleason score <7 tumours. Important outstanding questions surrounding the use of PRS will be considered, including the most appropriate PRS modelling approach, especially to maximize PRS transferability across populations. PRS developed in populations of European genetic ancestries may have poor predictive performance in individuals of other ancestries; thus, their use can potentially increase health disparities. Recent advances are considered in PRS modelling, including Bayesian shrinkage of the risk score weights and SuperLearner ensemble methods across multiple PRS approaches. Such innovative developments can help to generate PRS that benefit diverse populations.