Current polygenic risk scores are unlikely to exacerbate unfairness in cardiovascular disease risk prediction
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Current cardiovascular disease (CVD) risk prediction models place many individuals in an intermediate risk category where clinical decision-making remains uncertain, highlighting a critical gap in precision prevention. Polygenic risk scores (PRS) represent a promising solution to enhance risk stratification in intermediate-risk groups by identifying individuals with high genetic risk; however, observed differences in performance across ancestry groups may cause health disparities. The emerging field of algorithmic fairness offers a principled frame-work to assess equity in model performance among relevant subgroups, but have rarely been applied to clinical risk tools and PRS.
Objectives
To evaluate the fairness of incorporating PRS in CVD risk prediction, both as a standalone risk factor and as a risk-enhancing factor for individuals at intermediate risk (recommended in current clinical consensus statements).
Methods
Using data from the UK Biobank (N = 327,923), we calculated 10-year CVD risk using QRISK3 (a guideline-endorsed prediction model) and quantified genetic risk using a validated PRS. We assessed fairness among population characteristics relevant for health equity (age, sex, ethnicity, and area-level deprivation) using four algorithmic fairness metrics relevant for prevention (accuracy equality, equal opportunity, conditional use accuracy equality, and treatment equality).
Results
PRS, when used as a stand-alone risk factor, demonstrated fairness levels similar to or better than traditional clinical predictors (age, sex, blood pressure, cholesterol). Some variation in fairness was observed across ethnic groups, especially at extreme risk thresholds. When integrated as a risk-enhancing factor for reclassifying intermediate-risk individuals into high-risk categories, PRS improved sensitivity of CVD risk prediction with minimal impact on fairness metrics across demographic groups.
Conclusions
This study demonstrates that PRS, when incorporated into existing risk prediction frameworks, are unlikely to meaningfully exacerbate disparities in CVD risk stratification. Applying algorithmic fairness metrics provides insight into the equitable implementation of PRS and supports current recommendations for their use in risk-stratification for intermediate-risk individuals.
