Machine learning methodologies versus cardiovascular risk scores, in predicting disease risk

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Cardiovascular Disease (CVD) risk estimation scores are crucial for primary prevention.
• The performance of existing CVD risk scores remains a concern.
• This study investigates machine learning (ML) for CVD prediction.

Purpose of the Study:

• To explore the potential of ML methodologies for CVD prediction.
• To compare the performance of ML models against the established HellenicSCORE tool.
• To identify optimal ML approaches for CVD risk assessment.

Main Methods:

• Utilized data from the ATTICA prospective study (n=2020 adults).
• Trained and evaluated three ML classifiers (k-NN, random forest, decision tree) against 10-year CVD incidence.
• Compared ML performance with the HellenicSCORE tool using varying training datasets and bootstrapping techniques.

Main Results:

• ML methodologies achieved comparable outcomes to HellenicSCORE.
• HellenicSCORE: Accuracy 85%, Specificity 20%, Sensitivity 97%, PPV 87%, NPV 58%.
• ML (Random Forest best): Accuracy 65-84%, Specificity 46-56%, Sensitivity 67-89%, PPV 89-91%, NPV 24-45%.

Conclusions:

• Machine learning classification offers a viable alternative for CVD prediction.
• ML methodologies provide comparable results to traditional risk prediction scores.
• ML approaches present potential advantages for CVD risk assessment in primary prevention.