Sex-specific cardiovascular disease risk prediction using statistical learning and explainable artificial...

Area of Science:

• Cardiology
• Biostatistics
• Public Health

Background:

• Current cardiovascular disease (CVD) risk prediction models, such as NORRISK 2, explain only a modest proportion of CVD incidence.
• Existing models may not adequately capture sex-specific risk factors, necessitating improved prediction tools.

Purpose of the Study:

• To develop improved sex-specific models for predicting 10-year CVD risk.
• To establish sex- and age-specific thresholds for CVD intervention.

Main Methods:

• Analysis of data from 31,946 participants (40-79 years) without prior CVD.
• Development of prediction models using extreme gradient boosting (XGBoost) and logistic regression, with separate models for each sex.
• Evaluation of models using receiver-operating characteristic (ROC) and precision recall (PR) curves.

Main Results:

• XGBoost models significantly improved CVD risk prediction for males compared to NORRISK 2 and SCORE2, indicated by increased ROC-AUC and PR-AUC.
• For females, neither XGBoost nor logistic regression models showed significant improvement over existing models.
• Exploration of age- and sex-specific thresholds demonstrated improved sensitivity compared to NORRISK 2 thresholds.

Conclusions:

• Statistical learning and sex-specific risk factors yield improved CVD risk prediction models for males.
• Implementing sex-specific intervention thresholds may enhance CVD prevention efforts for both sexes.