FairGRPO: Fair Reinforcement Learning for Equitable Clinical Reasoning

2025-10-22

AI Summary

Medical AI exhibits significant performance disparities across demographic groups, exacerbating health inequities. To address this, we propose FairMed, a fairness-aware hierarchical reinforcement learning framework that dynamically integrates representation quality, task difficulty, and data source reliability via an adaptive importance-weighting mechanism. In the absence of explicit demographic labels, FairMed leverages unsupervised clustering to discover latent population structures and enables continuous in-training fairness optimization. The method synergistically combines group-relative policy optimization (GRPO) with multimodal alignment, supporting five clinical imaging modalities: X-ray, CT, dermoscopy, mammography, and ultrasound. Evaluated on seven benchmark datasets, FairMed reduces average prediction disparity by 27.2% and improves F1-score by 12.49%. We further release FairMedGemma-4B, a fairness-enhanced clinical large language model, which substantially narrows cross-group performance gaps.

Technology Category

Machine Learning: Ethics, Bias, and FairnessPhilosophy and Ethics of AI: Bias, Fairness & EquityComputer Vision: Bias, Fairness & Privacy

Application Category

User Modeling, Personalization and Recommendation: Fairness-aware retrieval and rankingSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingEconomics, Online Markets and Human Computation: Fairness and ethical considerations in crowd work and in human-in-the-loop AI systems

Abstract

Medical artificial intelligence systems have achieved remarkable diagnostic capabilities, yet they consistently exhibit performance disparities across demographic groups, causing real-world harm to underrepresented populations. While recent multimodal reasoning foundation models have advanced clinical diagnosis through integrated analysis of diverse medical data, reasoning trainings via reinforcement learning inherit and often amplify biases present in training datasets dominated by majority populations. We introduce Fairness-aware Group Relative Policy Optimization (FairGRPO), a hierarchical reinforcement learning approach that promotes equitable learning across heterogeneous clinical populations. FairGRPO employs adaptive importance weighting of advantages based on representation, task difficulty, and data source. To address the common issue of missing demographic labels in the clinical domain, we further employ unsupervised clustering, which automatically discovers latent demographic groups when labels are unavailable. Through comprehensive experiments across 7 clinical diagnostic datasets spanning 5 clinical modalities across X-ray, CT scan, dermoscropy, mammography and ultrasound, we demonstrate that FairGRPO reduces predictive parity by 27.2% against all vanilla and bias mitigated RL baselines, while improving F1 score by 12.49%. Furthermore, training dynamics analysis reveals that FairGRPO progressively improves fairness throughout optimization, while baseline RL methods exhibit deteriorating fairness as training progresses. Based on FairGRPO, we release FairMedGemma-4B, a fairness-aware clinical VLLM that achieves state-of-the-art performance while demonstrating significantly reduced disparities across demographic groups.

Problem

• Addressing performance disparities in medical AI across demographic groups
• Mitigating bias amplification in clinical reinforcement learning systems
• Developing fair AI without requiring demographic labels through unsupervised clustering

Innovation

• Hierarchical reinforcement learning for equitable clinical reasoning
• Adaptive importance weighting based on representation and difficulty
• Unsupervised clustering for demographic groups without labels