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

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence

Background:

• Postcardiotomy shock (PCS) is a critical complication post-cardiac surgery, requiring timely decisions on temporary mechanical circulatory support (tMCS).
• Traditional risk scores have limitations in managing the complexity of PCS.
• Machine learning (ML) offers a novel approach to integrate extensive perioperative data for improved clinical decision-making.

Purpose of the Study:

• To review the current applications of ML across the entire PCS continuum.
• To summarize essential ML model development aspects, including validation and implementation.
• To highlight the potential and limitations of ML in PCS management.

Main Methods:

• This study is a narrative expert review.
• It evaluates ML applications in risk recognition, phenotyping, tMCS initiation, management, liberation, and prognostication.
• Key aspects of ML model development, such as feature selection and external validation, are summarized.

Main Results:

• ML shows current strength in risk stratification, mortality prediction, and weaning support, especially in venoarterial extracorporeal membrane oxygenation (VA-ECMO) cohorts.
• Most existing ML models are retrospective and lack validation in PCS-specific populations.
• Automated device titration using ML is an emerging but preclinical area.

Conclusions:

• ML can augment, but not replace, clinician judgment in PCS management.
• High-quality, PCS-specific datasets and prospective validation are crucial for future ML advancements.
• Safe, transparent, and equitable implementation requires robust governance frameworks and seamless clinical workflow integration.