EHR-MPC: Inference-Time Control for Sepsis Treatment with Generative Patient Digital Twins

2026-07-07

AI Summary

This work addresses the limitation of existing reinforcement learning approaches for sepsis treatment, which learn static policies and struggle to adapt dynamically to varying clinical objectives during inference. The study proposes a novel framework that formulates treatment optimization as an inference-time control problem grounded in generative models of electronic health records (EHRs). By constructing patient-specific digital twins, it decouples the learning of state dynamics from policy optimization and integrates model predictive control (MPC) to simulate and plan personalized treatment trajectories within the intervention space. This approach overcomes the adaptability constraints of conventional reinforcement learning and establishes a generalizable framework for dynamic clinical decision-making. Evaluated on a multicenter ICU sepsis cohort spanning eight hospitals, EHR-MPC demonstrates competitive off-policy performance and significantly outperforms existing reinforcement learning baselines in simulated environments.

Abstract

Sepsis is a leading cause of mortality, yet optimal treatment policies remain contested. Existing reinforcement learning (RL) approaches learn fixed strategies for sepsis treatment, limiting adaptability to changing clinical objectives during inference. We propose EHRMPC, a framework that decouples learning patient dynamics from optimizing treatment by training a patient digital twin in the form of a generative electronic health record (EHR) model. The digital twin predicts clinical trajectories under interventions and enables model predictive control (MPC) to optimize treatments via inference-time planning over simulations. We evaluate EHR-MPC on a multicenter ICU sepsis cohort spanning 8 hospitals in the Mass General Brigham health system using both off-policy importance sampling and on-policy simulation-based evaluation. Relative to RL baselines, EHR-MPC achieves comparable off-policy performance and improved simulation performance. Unlike RL, this work frames sepsis treatment optimization as inference-time control over learned patient dynamics, establishing a general framework for decision making with generative clinical models.

Problem

• sepsis treatment
• inference-time control
• treatment adaptability
• clinical decision making
• patient dynamics

Innovation

• generative patient digital twin
• model predictive control
• inference-time planning
• electronic health record (EHR)
• sepsis treatment optimization