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.
Learning Objective 1:
This work presents a mixture-of-experts machine learning framework to improve fluid and vasopressor administration strategies for sepsis patients in intensive care units using observational data.
We demonstrate how one can use the proposed mixture model approach to adaptively combine nearest-neighbor-based (kernel) and deep reinforcement learning (DRL) experts to achieve better expected outcomes than physician, kernel only, and DRL-only experts.
Xuefeng Peng (Presenter)
Yi Ding, Harvard University
David Wihl, Harvard University
Omer Gottesman, Harvard University
Matthieu Komorowski, Imperial College London
Li-wei Lehman, MIT
Andrew Ross, Harvard University
Aldo Faisal, Imperial College London
Finale Doshi-Velez, Harvard University