FairFML: fair federated machine learning with a case study on reducing gender disparities in cardiac arrest outcome prediction.

Health equity is a critical concern in clinical research and practice, as biased predictive models can exacerbate disparities in clinical decision-making and patient outcomes. As healthcare systems increasingly rely on data-driven models, ensuring fairness in these systems is essential to prevent perpetuating existing disparities. While large-scale healthcare data exists across multiple institutions, cross-institutional collaborations often face privacy constraints, highlighting the need for privacy-preserving solutions that also promote fairness. We present Fair Federated Machine Learning (FairFML), a model-agnostic solution designed to reduce algorithmic bias in cross-institutional healthcare collaborations while preserving patient privacy. Validated through a real-world case study on reducing gender disparities in cardiac arrest outcome prediction, FairFML improved fairness metrics by up to 90% without compromising predictive performance. FairFML is flexible and compatible with various FL frameworks and models, from traditional statistical methods to deep learning, offering a robust and scalable solution for equitable model development in clinical settings.