Machine Learning Efficiency in Predicting Obstructive Coronary Artery Disease in Patients with Non-ST Elevation Acute Coronary Syndrome in the First Hours of Admission

TL;DR

This study shows that machine learning models, especially stochastic gradient boosting, can accurately predict obstructive coronary artery disease in patients with non-ST elevation acute coronary syndrome within the first few hours of hospital admission.

Contribution

The study introduces high-accuracy machine learning models for early prediction of obstructive coronary artery disease in NSTE-ACS patients using clinical data from the first hours of admission.

Findings

Stochastic gradient boosting models achieved the highest accuracy (AUC 0.949) in the third scenario with data collected 3 hours after admission.

Anthropometric indicators and echocardiographic parameters were key predictors of obstructive coronary artery disease.

Risk stratification using these models can help guide optimal myocardial revascularization strategies.

Abstract

The aim of the study was to assess the accuracy of prognostic models for obstructive coronary artery disease (OCAD) in the first hours of admission in patients with non-ST segment elevation acute coronary syndrome (NSTE-ACS).

The study involved 610 patients with low- and intermediate-risk NSTE-ACS (Me — 62 years). Based on invasive coronary angiography findings the patients were divided into 2 groups: the first — 363 (59.5%) patients with OCAD (coronary artery luminal occlusion ≥50%), the second — 247 (40.5%) patients without coronary obstruction (<50%). Clinical and functional status was assessed using 62 parameters available at the early hospitalization including: clinical and demographic, anthropometric, laboratory, electrocardiographic and echocardiographic data.

OCAD predictive models were developed using machine learning methods: multifactorial logistic regression, random forest, and stochastic gradient boosting (SGB). The models contained the sets of predictors identified during the initial medical examination in the hospital (the first scenario), after 1-hour observation (the second scenario), and 3 h later (the third scenario). The quality of the models was assessed using six metrics. The impact degree of individual predictors on the study endpoint was determined by the Shapley method of additive explanation (SHAP). OCAD probability stratification was performed by distinguishing the categories of low, medium, high and very high risk.

Based on machine learning methods, OCAD predictive models were developed, among which the best quality metrics were demonstrated by SGB models with the sets of predictors corresponding to three prognostic scenarios (the area under ROC curve: 0.846, 0.887, and 0.949, respectively). Using the SHAP method, we identified the factors with a dominant impact on OCAD, which included the anthropometric indicators (waist circumference, hip circumference, and their ratio) — in the first and second prognostic scenarios; and global longitudinal systolic strain of the left ventricle — in the third scenario. Based on SGB model data there were distinguished the categories of low, medium, high and very high risk of OCAD, their digital ranges depended on the prognostic scenarios.

The prognostic OCAD models developed based on SGB enable to highly accurately assess the degree of coronary damage in NSTE-ACS patients in the first hours of hospitalization. The highest accuracy of OCAD prediction was demonstrated by the models of the third scenario, the structure of which, in addition to anamnestic, anthropometric and ECG data, included clinical and biochemical blood parameters and echocardiographic indicators. Thus, OCAD risk stratification using the mentioned models can be a useful tool in selecting the optimal myocardial revascularization strategy.

Figures