Reward the Reward Designer: Making Reinforcement Learning Useful for Clinical Decision Making

Reinforcement Learning (RL) provides a promising framework for sequential decision-making in personalized healthcare. However, its clinical adoption hinges not only on algorithms, but also on effective reward design. Unlike games, healthcare lacks natural reward signals; they must be carefully crafted, validated, and interpreted. In this position paper, we argue that reward engineering is the bottleneck for applying RL in clinical time series. Drawing on case studies, we consolidate practical guidelines and propose a cultural shift: reward design should be benchmarked, published, and rewarded as a research contribution in its own right.