Machine-learning score using CCTA and stress CMR for prediction of cardiovascular events in patients with obstructive CAD

Abstract

Abstract Background In patients with suspected or known coronary artery disease, traditional prognostic risk assessment is based on a limited selection of clinical and imaging findings as coronary computed tomography angiography (CCTA) and stress cardiovascular magnetic resonance (CMR). Machine learning (ML) methods can take into account a greater number and complexity of variables. Purpose To investigate the feasibility and accuracy of ML using stress CMR and CCTA data to predict major adverse cardiovascular events (MACE) in patients with suspected or known coronary artery disease, and compared its performance with existing clinical, CMR or CCTA scores. Methods Between 2008-2020, consecutive symptomatic patients without known CAD referred for CCTA were screened. Patients with obstructive CAD (at least 1 ≥50% stenosis on CCTA) were further referred for stress CMR. Twenty-three clinical, 7 CCTA, and 11 stress CMR parameters were evaluated. ML involved automated feature selection by LASSO, model building with a XGBoost algorithm. The primary composite outcome was MACE defined by cardiovascular death or nonfatal myocardial infarction. The external validation cohort of the ML score was performed in another center (n=274). Results Of 2,210 patients who completed CMR, 2,038 (47% male, mean age 69±12 years) completed follow-up (median 6.8 [IQR 5.9-9.2] years); 281 experienced a MACE (13.8%). The ML score exhibited a higher area under the curve (AUC) compared with ESC risk score, QRISK3 score, Framingham risk score, and CCTA or stress CMR data alone for prediction of MACE (ML score: 0.85 vs ESC score: 0.66, QRISK3 score: 0.64, Framingham score: 0.63, only CCTA score: 0.76, only CMR: 0.82; all p<0.001). Similar results were obtained with AUC of the Precision-Recall (PR) curves (ML score: 0.59 vs. only CCTA score: 0.39, only CMR: 0.45; all p<0.001). The ML score also exhibited a good area under the curve in the external cohort (AUC ROC: 0.85 and AUC PR curve: 0.57). Conclusions The ML score including CCTA and stress CMR data exhibited a higher prognostic value to predict MACE compared with traditional method or traditional scores. In addition, our ML score including CCTA and stress CMR data showed the best performance compared with scores using only CCTA or CMR parameters.ROC et PR curves: ML scorePR curves: ML score