Cardiovascular/stroke risk predictive calculators: a comparison between statistical and machine learning models.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Conventional cardiovascular risk calculators (CVRC) often misestimate cardiovascular disease (CVD) and stroke risk due to limited integration of plaque burden.
• This study addresses the need for improved risk prediction by investigating machine learning (ML)-based CVRC.

Purpose of the Study:

• To compare the predictive performance of ML-based CVRC (CVRC_ML) against statistically derived CVRC (CVRC_Stat).
• To evaluate CVRC_ML and CVRC_Stat using conventional factors and integrated factors including plaque burden.

Main Methods:

• Developed and evaluated 13 types of CVRC_Stat using conventional and integrated risk factors for 10-year CVD/stroke risk stratification.
• Developed CVRC_ML using support vector machine (SVM) with the same risk factors.
• Assessed calculator performance using Area Under the Curve (AUC) from Receiver Operating Characteristic (ROC) analysis.

Main Results:

• CVRC_ML with integrated risk factors achieved an AUC of 0.88 (P<0.001), representing a 42% performance improvement.
• The overall risk-stratification accuracy for CVRC_ML with integrated factors was 92.52%.
• All 13 CVRC_Stat models showed lower AUC values compared to the integrated CVRC_ML.

Conclusions:

• ML-based CVD/stroke risk calculators demonstrate superior predictive ability for 10-year CVD/stroke events.
• The integrated ML model outperformed traditional statistical models, including the AECRS 2.0 integrated model.