Pediatric ECMO care improved with safer offline learning method
Support-Constrained Offline Policy Improvement for Pediatric ECMO from Limited Clinical Data
Machine Learning
Summary
Treating children on ECMO is tricky because doctors need to keep kids stable but also avoid long-term brain problems. The authors created a new computer method that carefully learns from past doctor decisions without trying risky new actions. This method improves treatment plans by sticking to what is known and safe, while still making better decisions than just copying doctors. Tests on real ECMO data showed this method works better and agrees more with clinician choices than other approaches.
What this means in practice
- •For hospital clinical teams: Improve ECMO treatment decisions for pediatric patients by using offline models that enhance clinician policies safely from existing data.
- •For medical device software developers: Develop decision support tools for ECMO management that avoid risky recommendations by constraining learning to well-supported actions.
Authors
Fateme Golivand Darvishvand, Saurabh Mathur, Nikhilesh Prabhakar, Michael A. Skinner, Ameet Soni, Neel Shah, Ethan L Sanford, Lakshmi Raman, Sriraam Natarajan
Abstract
Managing children on ECMO (Extracorporeal Membrane Oxygenation) requires balancing short-term physiological stability with the long-term goal of preventing neurological injury. Standard reinforcement learning can propose risky, out-of-distribution actions, while imitation learning is limited to reproducing clinician behavior. We introduce Support-Constrained Fitted Q-Iteration (FQI-SC), an offline reinforcement learning framework that cautiously improves upon expert behavior by restricting optimization to empirically supported state-action pairs and defaulting to the clinician baseline in low-data regions. On 78 pediatric trajectories totaling over 6,000 hours of clinical data, we evaluate our framework using Fitted Q-Evaluation (FQE) and a behavioral disagreement. Additionally, we formalize the performance-safety trade-off by introducing a unified Balance Score ($S(π)$). Our empirical evaluation shows that support-constrained policy improves the Balance Score by 38\% relative to FQI, improves estimated value by 1.28 FQE units over imitation learning, and reduces disagreement with clinician actions by 71\% relative to FQI.