Imitation learning for clinical decision support in pediatric ECMO

2026-05-15

AI Summary

This study addresses the challenges of modeling clinical decision-making in pediatric extracorporeal membrane oxygenation (ECMO) therapy, where high complexity and scarce data hinder conventional approaches. The problem is formalized as an imitation learning task without explicit action labels, aiming to infer implicit clinical interventions from observed patient trajectories. For the first time, TabPFN—a Transformer-based model designed for tabular data—is introduced to this domain and evaluated against established baselines such as XGBoost and multilayer perceptrons (MLPs) for action prediction. Experimental results on real-world pediatric ECMO data demonstrate that TabPFN significantly outperforms traditional methods, highlighting its potential as a robust benchmark for clinical decision support. This work establishes a novel paradigm for modeling medical decisions under conditions of limited labeled data.

Abstract

Pediatric critical care is a dynamic, high-stakes process involving constant monitoring and adjustments in life-saving treatments. Modeling these interventions is crucial for effective decision support. To address the challenges of high complexity and data scarcity in pediatric Extracorporeal Membrane Oxygenation (ECMO), we frame clinical decision-making as learning to act from trajectories, i.e., imitation learning that learns action models from observational data, with a key feature that actions are not directly observed. We consider TabPFN, a recent transformer-based approach for tabular data, and traditional baselines including XGBoost and Multi-Layer Perceptrons(MLPs) on real-world pediatric ECMO data to learn the action models. We find that the TabPFN-based approach consistently outperforms these classical baselines, supporting its use as a strong clinician-behavior baseline for pediatric ECMO decision support.

Problem

• imitation learning
• pediatric ECMO
• clinical decision support
• data scarcity
• action modeling

Innovation

• imitation learning
• TabPFN
• pediatric ECMO
• clinical decision support
• action modeling