Predictive performance of machine learning compared to statistical methods in time-to-event analysis of...

Area of Science:

• Medical Informatics
• Biostatistics
• Cardiology

Background:

• Cardiovascular disease (CVD) is the leading global cause of mortality.
• Accurate CVD risk prediction is crucial for effective prevention strategies.
• Machine learning (ML) models show promise in various fields, including healthcare.

Purpose of the Study:

• To systematically review and compare the predictive performance of ML models versus statistical models for CVD time-to-event outcomes.
• To assess the discrimination and calibration of both ML and statistical models in CVD risk prediction.
• To identify which modeling approach offers superior accuracy for time-to-event data with censoring.

Main Methods:

• Systematic review of original research articles on prognostic prediction studies for CVD.
• Inclusion of studies developing or validating prognostic models with at least a 12-month follow-up.
• Adherence to the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CRD42023484178) checklist.

Main Results:

• This section will present the comparative performance metrics (discrimination and calibration) of ML and statistical models based on the reviewed studies.
• Key findings on the relative strengths and weaknesses of each model type for CVD risk prediction will be detailed.

Conclusions:

• The review will conclude on the superiority of ML or statistical models in predicting CVD time-to-event outcomes.
• Recommendations for clinical practice and future research in CVD risk prediction will be provided based on the findings.