Artificial intelligence guided tools in pediatric extracorporeal membrane oxygenation: Implications for clinical practice

Abstract

Introduction

Pediatric extracorporeal membrane oxygenation (ECMO) is a highly complex therapy characterised by rapidly evolving physiology, high complication rates, and substantial cognitive demands on multidisciplinary teams. Management requires continuous integration of physiologic, laboratory, ventilatory, and circuit-derived data to support time-critical decisions related to gas exchange, haemodynamics, anticoagulation, transfusion, circuit surveillance, and weaning. Despite technological advances and growing international standardisation, morbidity and mortality continue to be driven by neurologic injury, bleeding and thrombosis, and clinical factors influencing successful liberation from ECMO support.

Aim

To review contemporary artificial intelligence (AI) and machine learning (ML) applications relevant to pediatric ECMO and evaluate their potential role as clinically aligned decision-support tools for clinicians and ECMO practice.

Methods

A narrative review of English-language literature (2020–2026) was conducted using PubMed, MEDLINE, and Scopus to identify AI/ML studies, systematic reviews, and guideline documents addressing pediatric ECMO outcomes, complication prediction, circuit surveillance, and weaning or decannulation support, as well as methodological and governance frameworks relevant to clinical implementation.

Results

Most ECMO-AI applications employ supervised learning or deep learning approaches integrating demographic, clinical, laboratory, and ECMO device time-series data. Recent studies demonstrate emerging promise for neurologic risk prediction, bleeding and transfusion risk forecasting, circuit anomaly detection, and dynamic assessment of weaning readiness. Progress has included registry-based analyses, externally validated pediatric models, and time-series frameworks aligned with clinical workflows. However, heterogeneity in data quality, outcome definitions, and validation strategies remains substantial.

Conclusion

AI-guided tools in pediatric ECMO are transitioning from early feasibility studies toward more mature, clinically aligned decision-support applications. With rigorous validation, transparent reporting, and strong governance, AI has the potential to enhance situational awareness and consistency of care while preserving the central role of expert multidisciplinary judgement.