Comparative Evaluation of Pre-Test Probability Models for Coronary Artery Disease with Assessment of a New Machine Learning-Based Model.

Objective: This study aimed to validate pivotal pre-test probability (PTP)-coronary artery disease (CAD) models (CAD consortium model and IJC-CAD model).

Methods: Traditional PTP models-CAD consortium models: two traditional PTP models were used under the CAD consortium framework, namely CAD1 and CAD2. Machine learning (ML)-based PTP models: two ML-based PTP models were derived from CAD1 and CAD2, and used to enhance predictive capabilities [ML-CAD2 and ML-IJC (IJC-CAD)]. The primary endpoint was obstructive CAD. The performance evaluation of these PTP models was conducted using receiver-operating characteristic analysis.

Results: The study included 238 participants, among whom 157 individuals (65.9% of the total sample) had CAD. The IJC-CAD model demonstrated the highest performance with an area under the curve (AUC) of 0.860 [95% confidence interval (CI): 0.812-0.909]. Following this, the ML-CAD2 model exhibited an AUC of 0.814 (95% CI: 0.758-0.870), CAD1 showed an AUC of 0.767 (95% CI: 0.705-0.830), and CAD2 had an AUC of 0.785 (95% CI: 0.726-0.845). Each of the PTP models was adjusted to have a CAD score cutoff that classified cases with a sensitivity of over 95%. The respective cutoff values were as follows: CAD1 and CAD2 >12, ML-CAD2 >0.380, and IJC-CAD >0.367. All PTP models achieved a CAD sensitivity of over 95%. Similar to the AUC performance, the accuracy of the PTP models was highest for IJC-CAD, reaching 80.3%. The accuracy of ML-CAD2 was 77.7%, while that for CAD1 and CAD2 was 74.8% and 75.2%, respectively.

Conclusions: ML-CAD2 and IJC-CAD showed superior performance compared to traditional existing models (CAD1 and CAD2).