Development and internal validation of a probability-based coronary artery calcium prediction model to refine cardiovascular risk
B S Yeoh, A Loch, M I Zakaria, M N Md Shah, R R A Raja AmanAbstract
Background
Cardiovascular risk stratification in Southeast Asia (SEA) faces two challenges: Western-derived risk scores frequently misclassify individuals, and coronary artery calcium (CAC) scoring—a validated risk modifier—is not widely available. Consequently, asymptomatic adults most likely to benefit from definitive assessment are difficult to identify, despite substantial regional atherosclerotic cardiovascular disease (ASCVD) burden.
Purpose
To develop and internally validate a probability-based CAC prediction model to refine cardiovascular risk classification and optimise targeted testing among asymptomatic adults in SEA.
Methods
This cross-sectional retrospective study included 531 asymptomatic adults without known ASCVD who underwent health screening at two hospitals in a major SEA city between 2018 and 2023. Ten-year cardiovascular risk was estimated with the Framingham Risk Score (FRS) and SCORE2 Asia-Pacific. Presence of CAC was defined as an Agatston score >0. Pre-specified predictors included age, sex, diabetes mellitus, hypertension, statin use, ever-smoking, and the triglyceride-to-high-density lipoprotein cholesterol ratio. A multivariable logistic regression model with least absolute shrinkage and selection operator (LASSO) penalisation was developed. Discrimination was assessed using the area under the receiver-operating characteristic curve (AUC) with internal validation via 1,000 bootstrap resamples; calibration was evaluated using the Brier score and calibration plots. Sensitivity, specificity, and number needed to screen (NNS) were calculated for detecting CAC >0: (1) applying the CAC model to all participants, (2) conventional intermediate-risk testing, and (3) extended screening of low-risk adults with high CAC probability.
Results
Among 531 participants (mean age 52.7 years [SD 11.3]; 351 [66.1%] male), 276 (52.0%) had CAC >0. The LASSO-selected model retained all pre-specified predictors and demonstrated moderate-to-good discrimination (apparent AUC 0.804, 95% CI: 0.767–0.841; optimism-corrected AUC 0.799). Calibration was acceptable (Brier score 0.18), with plots showing close agreement between predicted and observed probabilities. At the threshold maximising the Youden index (0.44), the model achieved sensitivity of 79.3%, specificity of 67.8%, and NNS of 1.4. FRS 10–20% identified 79 CAC-positive individuals (sensitivity 52.3%; specificity 71.1%; NNS 1.9), and SCORE2 Asia-Pacific 5–10% identified 59 (sensitivity 83.1%; specificity 38.5%; NNS 2.1). Extended screening with the CAC model improved sensitivity (FRS 69.5%; SCORE2 Asia-Pacific 90.1%) without compromising specificity or NNS, detecting additional CAC cases missed by established risk scores.
Conclusion
This internally validated CAC model robustly stratifies subclinical ASCVD burden, guiding targeted testing in SEA adults. External validation and long-term outcome studies would support integration into preventive cardiology frameworks.Figure1_ROC_CAC_DiscriminationTable1_CAC_Testing_Strategy_Performance