New Model for Predicting the Presence of Coronary Artery Calcification

Area of Science:

• Cardiology
• Medical Imaging
• Predictive Analytics

Background:

• Coronary artery calcification (CAC) is a marker of coronary atherosclerosis and a significant risk factor for cardiovascular disease (CVD).
• Measuring CAC score aids in cardiovascular risk stratification, especially when risk assessment is uncertain.
• The absence of CAC is linked to a lower incidence of CVD.

Purpose of the Study:

• To develop and evaluate predictive models for the presence of coronary artery calcification (CAC).
• To assess the discrimination and calibration power of different machine learning models in predicting CAC.
• To identify a practical model for real-world clinical implementation.

Main Methods:

• Retrospective study utilizing data from 3,302 patients aged 40-75 from two cohorts.
• Data split into 80% training and 20% testing sets, with ten-fold cross-validation.
• Prediction models built using logistic regression (LRM), classification and regression tree (CART), conditional inference tree (CIT), and random forest (RF).

Main Results:

• All models demonstrated acceptable predictive accuracy: LRM (70.71%), CART (71.32%), CIT (71.32%), and RF (71.02%).
• Decision tree models (CART and CIT) achieved reasonable accuracy without excessive complexity.
• The models effectively predicted the presence or absence of CAC (CAC score = 0 or not).

Conclusions:

• Decision tree models, specifically CART and CIT, provide a practical and reasonably accurate method for predicting coronary artery calcification.
• These models can be implemented in clinical practice for improved cardiovascular risk stratification.
• Further validation may enhance the utility of these CAC prediction models.