Improving Sepsis Treatment Strategies by Combining Deep and Kernel-Based Reinforcement Learning

TL;DR

This paper introduces a personalized sepsis treatment approach using a mixture-of-experts framework that combines kernel-based and deep reinforcement learning methods, leading to improved outcomes over existing strategies.

Contribution

It presents a novel mixture-of-experts model that dynamically combines kernel and deep RL experts for personalized sepsis treatment, outperforming individual methods.

Findings

Outperforms physician and individual RL methods on retrospective data

Demonstrates improved patient-specific treatment recommendations

Validates the effectiveness of combining kernel and deep RL approaches

Abstract

Sepsis is the leading cause of mortality in the ICU. It is challenging to manage because individual patients respond differently to treatment. Thus, tailoring treatment to the individual patient is essential for the best outcomes. In this paper, we take steps toward this goal by applying a mixture-of-experts framework to personalize sepsis treatment. The mixture model selectively alternates between neighbor-based (kernel) and deep reinforcement learning (DRL) experts depending on patient's current history. On a large retrospective cohort, this mixture-based approach outperforms physician, kernel only, and DRL-only experts.