Machine Learning in Postcardiotomy Shock: Implications for Temporary Mechanical Circulatory Support

Postcardiotomy shock (PCS) is a distinct form of cardiogenic shock that occurs after cardiac surgery and necessitates rapid decisions regarding temporary mechanical circulatory support (tMCS), including escalation, device configuration, and weaning. Machine learning (ML) has emerged as a strategy to integrate high-dimensional perioperative data and augment clinical decision-making beyond traditional risk scores. This narrative expert review evaluates ML applications across the PCS continuum, including early risk recognition and phenotyping, tMCS initiation, management and liberation, and prognostication. This study also summarizes key aspects of ML model development, including feature selection, outcome definition, external validation, and clinical implementation. Current evidence is strongest for risk stratification, mortality prediction, and weaning support, particularly in venoarterial extracorporeal membrane oxygenation cohorts. However, most models are retrospective, derived from heterogeneous shock populations, and lack prospective validation in PCS-specific cohorts. Emerging work on automated device titration is promising but remains preclinical. ML should be considered an adjunct to clinician judgment rather than a substitute for it. Future progress will depend on high-quality PCS datasets, prospective validation, seamless workflow integration, and governance frameworks that ensure safe, transparent, and equitable implementation.