Electrocardiographic, echocardiographic and lipid parameters in predicting obstructive coronary artery disease in patients with non-ST elevation acute coronary syndrome

Aim. To assess the predictive potential of electrocardiographic (ECG), echocardiographic, and lipid parameters for predicting obstructive coronary artery disease (oCAD) in patients with non-ST-elevation acute coronary syndrome (NSTE-ACS) prior to invasive coronary angiography (CA). Material and methods. This prospective observational cohort study included 525 patients with NSTE-ACS with a median age of 62 years who underwent invasive coronary angiography. Two groups were distinguished, the first of which consisted of 351 (67%) patients with oCAD (stenosis 50%), and the second — 174 (33%) without oCAD (<50%). Clinical and functional status of patients before CAG was assessed by 40 indicators. Mann-Whitney, Fisher, chi-squared, univariate logistic regression (LR) methods were used for data processing and analysis, while miltivariate LR (MLR), gradient boosting (XGBoost) and artificial neural networks (ANN) were used to develop predictive models. The quality of the models was assessed using 4 following metrics: area under the ROC curve (AUC), sensitivity (Se), specificity (Sp), and accuracy (Ac). Results. A comprehensive analysis of ECG, echocardiography and lipid profile parameters made it possible to identify factors that had linear and non-linear association with oCAD. LR were used to determine their weight coefficients and threshold values with the highest predictive potential. The quality metrics of the best predictive algorithm based on MLR were 0,81 for AUC, 0,74 for Sp and Ac, and 0,75 for Se. The predictors of this model were 4 categorical parameters (left ventricular (LV) ejection fraction of 42-60%, global LV longitudinal systolic strain <19%, low-density lipoprotein cholesterol >3,5 mmol/l, age >55 years in men and >65 years for women). Conclusion. The prognostic model developed on the basis of MLR made it possible to verify oCAD with high accuracy in patients with NSTE-ACS before invasive CA. Models based on XGBoost and ANN had less predictive value.