Clinical Applications of Data Science and Machine Learning in the Pediatric Cardiac Intensive Care Unit

Area of Science:

• Pediatric Cardiac Intensive Care
• Machine Learning Applications
• Clinical Decision Support

Background:

• Machine learning (ML) is increasingly applied in pediatric cardiac intensive care units (CICUs).
• Studies focus on predicting critical outcomes like mortality, cardiac arrest, and low cardiac output syndrome (LCOS).
• Evaluating algorithm performance, validation rigor, and readiness for clinical integration is crucial.

Purpose of the Study:

• To synthesize and critically appraise ML applications in pediatric CICU.
• To assess algorithm performance, validation, and clinical decision-support readiness.
• To identify gaps and future directions for ML in pediatric cardiac care.

Main Methods:

• Scoping review of studies from 2015-2025 in PubMed and PubMed Central.
• Included studies on congenital heart disease (CHD) or CICU populations with >90,000 pediatric encounters.
• Analyzed endpoints including mortality, cardiac arrest, LCOS, acute kidney injury (AKI), and postoperative complications.

Main Results:

• Tree-based ensembles and gradient boosting achieved AUROC of 0.83-0.97, outperforming traditional scores.
• Deep learning models showed similar accuracy using EHR or physiologic data.
• Limited calibration ( <1/3 studies) and external validation (4 studies) were noted; explainability tools improved interpretability.

Conclusions:

• Pediatric CICU ML models demonstrate high predictive power but lack robust calibration and validation evidence.
• Further development requires multicenter data, standardized reporting, interpretable models, and pragmatic trials.
• Translation to bedside use necessitates demonstrating clear clinical benefit and safety.