Machine Learning-Based Prediction Models for Healthcare Outcomes in Patients Participating in Cardiac Rehabilitation:...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Cardiac rehabilitation (CR) significantly reduces mortality and morbidity in cardiovascular disease patients.
• Machine learning (ML) is increasingly applied to predict healthcare outcomes, including within CR.

Purpose of the Study:

• To critically appraise existing ML-based prognosis predictive models in CR.
• To identify research gaps in ML applications for CR outcome prediction.

Main Methods:

• Systematic literature search across major databases (Scopus, PubMed, Web of Science, Google Scholar) up to January 2024.
• Extracted data on clinical features, outcomes, model development, validation, and performance metrics.
• Assessed study quality using IJMEDI and Prediction Model Risk of Bias Assessment Tool.

Main Results:

• 22 ML models from 7 studies were analyzed, mostly with small patient cohorts (41-227).
• Models predicted various CR progression stages with good performance (AUC 0.82-0.91, sensitivity 0.77-0.95).
• No models underwent calibration or external validation; most had bias concerns.

Conclusions:

• Current ML models for CR show potential but are not ready for clinical implementation due to lack of validation and bias.
• Further research requires external validation of existing models and development of new models using robust methods on larger datasets.
• Future models should target diverse clinical outcomes within CR to enhance predictive capabilities.