Penalized Fair Regression for Multiple Groups in Chronic Kidney Disease

TL;DR

This paper extends fair regression to handle multiple groups by introducing a penalized framework that imposes group-specific unfairness penalties and reduces to cost-sensitive classification for efficient estimation. It defines true positive rate-based fairness for each group, aggregates unfairness across groups via several synthesis methods, and automatically selects penalty weights through a random search guided by a combined accuracy-fairness score. Through extensive simulations and a CKD ESRD prediction application, the method demonstrates improved fairness across groups with little to no loss in overall predictive performance, while highlighting challenges for very small groups. The work contributes practical methodology, implementation guidance, and empirical insights for deploying multi-group fairness in health-care analytics.

Abstract

Fair regression methods have the potential to mitigate societal bias concerns in health care, but there has been little work on penalized fair regression when multiple groups experience such bias. We propose a general regression framework that addresses this gap with unfairness penalties for multiple groups. Our approach is demonstrated for binary outcomes with true positive rate disparity penalties. It can be efficiently implemented through reduction to a cost-sensitive classification problem. We additionally introduce novel score functions for automatically selecting penalty weights. Our penalized fair regression methods are empirically studied in simulations, where they achieve a fairness-accuracy frontier beyond that of existing comparison methods. Finally, we apply these methods to a national multi-site primary care study of chronic kidney disease to develop a fair classifier for end-stage renal disease. There we find substantial improvements in fairness for multiple race and ethnicity groups who experience societal bias in the health care system without any appreciable loss in overall fit.

Figures (14)

• Figure 1: Simulation Setting 1
• Figure 2: Simulation Setting 2
• Figure 3: Simulation Setting 3
• Figure 5: Accuracy
• Figure 6: Population-Weighted Mean Unfairness