The gradient boosting model predicted adverse outcomes with an AUROC of 0.822 in a cohort of 1439 CAD patients over 45 months, identifying critical prognostic features.
Does machine learning analysis of echocardiographic data accurately predict adverse outcomes in patients with obstructive and non-obstructive coronary artery disease?
Machine learning models utilizing echocardiographic features like global longitudinal strain and myocardial work can accurately predict adverse clinical outcomes in patients with both obstructive and non-obstructive coronary artery disease.
Absolute Event Rate: 0% vs 0%
BACKGROUND: Research on echocardiographic prognostication in non-obstructive coronary artery disease (CAD) is limited, mainly accompanied by different outcomes from obstructive CAD. This study developed machine learning (ML) models to predict clinical outcomes for both CAD types, highlighting crucial prognostic indicators. METHODS: A prospective analysis was conducted using echocardiographic data from patients undergoing coronary angiography. ML algorithms extracted key features, and six classifiers assessed prognostic accuracy using the area under the receiver operating characteristic curve (AUROC). The SHapley Additive exPlanations (SHAP) method identified the most influential prognostic features. External testing was performed using an independent dataset. RESULTS: The cohort included 1439 patients, with 487 (33.8 %) classified as having non-obstructive CAD. Over a median follow-up period of 45 months, 319 (22.2 %) adverse events were recorded. The gradient boosting model outperformed others, achieving an AUROC of 0.822 (95 % confidence interval (CI): 0.816-0.827). In the obstructive CAD group, quadratic discriminant analysis demonstrated the highest performance (AUROC: 0.794; 95 % CI: 0.775-0.812), while ridge regression performed best for the non-obstructive group (AUROC: 0.856; 95 % CI: 0.829-0.883). Key predictive variables included global longitudinal strain (GLS) and global systolic constructive work, highlighting the importance of plasma brain natriuretic peptide alongside various GLS and myocardial work sub-parameters for distinguishing CAD types. The best-performing ML models also exhibited good performance in the external validation datasets (AUROC: 0.802, 0.819 and 0.869). CONCLUSIONS: The machine learning approach accurately predicted adverse outcomes in CAD patients and identified critical echocardiographic variables specific to obstructive and non-obstructive forms, promoting CAD prognosis. TRIAL REGISTRATION: ClinicalTrials.govNCT03905200 (Registered 5 April 2019).
Guo et al. (Thu,) reported a other. The gradient boosting model predicted adverse outcomes with an AUROC of 0.822 in a cohort of 1439 CAD patients over 45 months, identifying critical prognostic features.