FairFML: A Unified Approach to Algorithmic Fair Federated Learning with Applications to Reducing Gender Disparities in Cardiac Arrest Outcomes

Abstract

Addressing algorithmic bias in healthcare is crucial for ensuring equity in patient outcomes, particularly in cross-institutional collaborations where privacy constraints often limit data sharing. Federated learning (FL) offers a solution by enabling institutions to collaboratively train models without sharing sensitive data, but challenges related to fairness remain. To tackle this, we propose Fair Federated Machine Learning (FairFML), a model-agnostic framework designed to reduce algorithmic disparities while preserving patient privacy. Validated in a real-world study on gender disparities in cardiac arrest outcomes, FairFML improved fairness by up to 65% compared to centralized models, without compromising predictive performance.