Developing an ML-Based Pretest Probability Model of Obstructive CAD in Patients With Stable Chest Pain

Updated pretest probability models (ESC2019, the PTP model supported by the European Society of Cardiology after a pooled analysis; and RF-CL, the risk factor-weighted model) are recommended for initial evaluation of patients with stable chest pain before coronary computed tomography angiography to reduce unnecessary examination by recent guidelines. However, the reliability of those pretest probability models has not been fully investigated, especially in Chinese population.

This study aims to build a machine learning-based pretest probability model in patients with stable chest pain and compare it with ESC2019 and RF-CL model in a Chinese population.

This is an analysis of the Chinese registry in China, with a large scale, foresight, and a multicenter cohort. Obstructive coronary artery disease refers to at least 1 lesion ≥70% diameter stenosis in main branches or ≥50% left main stenosis by coronary computed tomography angiography. A pretest probability model, the C-STRAT (Chinese Registry in Early Detection and Risk Stratification of Coronary Plaques) score, was conducted by an ensemble machine learning algorithm in training data set and compared with other pretest probability models.

In the testing data set, the C-STRAT score gave the best performance in discrimination evaluation (AUC: 0.769; 95% CI: 0.753-0.784). It also performed well in calibration evaluation. The integrated discrimination improvement and net reclassification improvement of the C-STRAT score were positive compared with other pretest probability models.

A high-performance pretest probability model derived from machine learning algorithm was developed based on a multicenter Chinese population and expected to facilitate the decision making for downstream tests. (Chinese Database of National Coronary Plaques Registry; ChiCTR1800015864).