Machine learning prediction in cardiovascular diseases: a meta-analysis.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Machine learning (ML) algorithms are increasingly used for cardiovascular disease (CVD) prediction.
• Assessing the overall predictive performance of various ML algorithms is crucial for clinical application.

Purpose of the Study:

• To systematically evaluate and summarize the predictive capabilities of ML algorithms for major cardiovascular diseases.
• To identify the most effective ML algorithms for conditions including coronary artery disease, heart failure, stroke, and cardiac arrhythmias.

Main Methods:

• A comprehensive literature search was conducted across MEDLINE, Embase, and Scopus databases up to March 15, 2019.
• Included 103 cohorts totaling 3,377,318 individuals to assess ML algorithm predictive ability.
• Meta-analysis was used to pool performance metrics, primarily the area under the curve (AUC).

Main Results:

• For coronary artery disease prediction, boosting algorithms achieved a pooled AUC of 0.88, while custom-built algorithms reached 0.93.
• For stroke prediction, Support Vector Machine (SVM) algorithms showed a pooled AUC of 0.92, followed by boosting (0.91) and Convolutional Neural Network (CNN) (0.90).
• While limited data existed for heart failure and cardiac arrhythmias, SVM algorithms may offer superior performance.

Conclusions:

• ML algorithms, particularly SVM and boosting, exhibit promising predictive performance for cardiovascular diseases.
• Heterogeneity exists among ML algorithms, necessitating careful consideration for specific datasets.
• Findings can guide clinicians in selecting and interpreting ML models for improved cardiovascular risk prediction.