Evaluating Reinforcement Learning Algorithms in Observational Health Settings

TL;DR

This paper discusses the challenges and considerations in evaluating reinforcement learning algorithms for healthcare decision-making, emphasizing the importance of careful assessment to ensure reliable and effective treatment policies.

Contribution

It highlights key subtleties in evaluating RL in healthcare, offering guidance to avoid misleading conclusions and improve policy assessment methods.

Findings

Evaluation metrics can be biased by how patient history is summarized.

Statistical estimator variance affects the reliability of policy quality estimates.

Confounders in observational data can lead to misleading policy evaluations.

Abstract

Much attention has been devoted recently to the development of machine learning algorithms with the goal of improving treatment policies in healthcare. Reinforcement learning (RL) is a sub-field within machine learning that is concerned with learning how to make sequences of decisions so as to optimize long-term effects. Already, RL algorithms have been proposed to identify decision-making strategies for mechanical ventilation, sepsis management and treatment of schizophrenia. However, before implementing treatment policies learned by black-box algorithms in high-stakes clinical decision problems, special care must be taken in the evaluation of these policies. In this document, our goal is to expose some of the subtleties associated with evaluating RL algorithms in healthcare. We aim to provide a conceptual starting point for clinical and computational researchers to ask the right questions when designing and evaluating algorithms for new ways of treating patients. In the following, we describe how choices about how to summarize a history, variance of statistical estimators, and confounders in more ad-hoc measures can result in unreliable, even misleading estimates of the quality of a treatment policy. We also provide suggestions for mitigating these effects---for while there is much promise for mining observational health data to uncover better treatment policies, evaluation must be performed thoughtfully.