Creating Fair Models of Atherosclerotic Cardiovascular Disease Risk

TL;DR

This paper develops a fair ASCVD risk prediction model using adversarial learning on EHR data, aiming to reduce bias across race and gender groups while maintaining predictive accuracy.

Contribution

It introduces an adversarial learning approach to improve fairness in cardiovascular risk models using high-dimensional electronic health records.

Findings

Reduced variability in error rates across groups

Aligned risk prediction distributions across groups

Discussed fairness-performance trade-offs

Abstract

Guidelines for the management of atherosclerotic cardiovascular disease (ASCVD) recommend the use of risk stratification models to identify patients most likely to benefit from cholesterol-lowering and other therapies. These models have differential performance across race and gender groups with inconsistent behavior across studies, potentially resulting in an inequitable distribution of beneficial therapy. In this work, we leverage adversarial learning and a large observational cohort extracted from electronic health records (EHRs) to develop a "fair" ASCVD risk prediction model with reduced variability in error rates across groups. We empirically demonstrate that our approach is capable of aligning the distribution of risk predictions conditioned on the outcome across several groups simultaneously for models built from high-dimensional EHR data. We also discuss the relevance of these results in the context of the empirical trade-off between fairness and model performance.