Machine-learning-based cardiovascular mortality prediction using a cumulative PM_2.5 exposure metric in high risk...

Area of Science:

• Environmental Health Sciences
• Cardiovascular Medicine
• Biostatistics and Machine Learning

Background:

• Classical risk factors for cardiovascular disease (CVD) are established, but integrating environmental factors like air pollution (AP) into risk prediction models remains limited.
• Machine learning (ML) offers advanced capabilities for risk prediction, yet its application with environmental data for CVD mortality requires further investigation.

Purpose of the Study:

• To quantify the improvement in CVD mortality risk prediction by integrating long-term PM2.5 exposure using the 'ePM-years Index' into ML models.
• To assess the clinical utility and predictive accuracy of ML models incorporating environmental factors compared to traditional risk factors alone.

Main Methods:

• A prospective cohort study (EP-PARTICLES) of 6935 patients, analyzing 21 classical risk factors and CVD mortality.
• Calculation of the 'ePM-years Index' for cumulative long-term PM2.5 exposure and application of ML techniques for risk prediction.
• Feature selection identified key discriminative predictors, and model performance was evaluated using the Matthews Correlation Coefficient (MCC) and Area Under the Curve (AUC).

Main Results:

• The best ML model without the 'ePM-years Index' achieved an MCC of 0.281, correctly classifying 72.59% of patients.
• Incorporating the 'ePM-years Index' significantly improved the ML model's performance to an MCC of 0.657, correctly classifying 92.75% of patients (AUC 0.88-0.91).
• Long-term PM2.5 exposure was linked to increased CVD mortality risk, particularly in high-risk groups.

Conclusions:

• Integrating long-term PM2.5 exposure data via the 'ePM-years Index' substantially enhances CVD mortality risk prediction accuracy using ML.
• ML models incorporating PM2.5 exposure demonstrate superior predictive abilities and clinical utility, improving risk stratification for cardiovascular events.
• Environmental factors should be considered in CVD risk assessment, and ML tools can facilitate the development of more effective clinical prediction models.