Fair Reinforcement Learning for Equitable Health Resource Allocation

Abstract

Inequitable distribution of health resources remains a persistent challenge in many healthcare systems, disproportionately affecting underserved communities, rural populations, and socioeconomically disadvantaged groups. Traditional decision-support tools often optimize for efficiency or overall population benefit, but they rarely account for structural biases that contribute to unequal access. Reinforcement Learning (RL) has emerged as a powerful approach for optimizing dynamic resource allocation; however, conventional RL models risk reinforcing existing disparities when rewards are based solely on aggregate performance metrics.
This study introduces a fairness-aware reinforcement learning framework designed to promote equitable allocation of critical health resources including hospital beds, vaccines, and diagnostic equipment across diverse demographic groups. The proposed approach incorporates explicit fairness constraints, such as counterfactual fairness and group-level equity metrics, directly into the RL reward function and training process. By combining fairness penalties with outcome-based rewards, the model balances efficiency with equity, ensuring that improved system performance does not come at the expense of vulnerable populations.
Simulation results demonstrate that the fairness-aware RL approach significantly reduces disparities in resource allocation while maintaining competitive performance compared to unconstrained RL baselines. The findings contribute to the growing field of equitable AI in healthcare by offering a scalable, adaptable framework that can be applied across various health system contexts. Overall, this work highlights the potential of fairness-constrained RL to support more just, transparent, and inclusive healthcare delivery systems.