Cardiometabolic risk factors in predicting obstructive coronary artery disease in patients with non-ST-segment elevation acute coronary syndrome

Aim. To develop predictive models of obstructive coronary artery disease (OPCA) in patients with non-ST-segment elevation acute coronary syndrome (NSTE-ACS) based on the predictive potential of cardiometabolic risk (CMR) factors. Material and methods. This prospective observational cohort study included 495 patients with NSTE-ACS (median age, 62 years; 95% confidence interval [60; 64]), who underwent invasive coronary angiography (CAG). Two groups of persons were identified, the first of which consisted of 345 (69,6%) patients with OPCA (stenosis ≥50%), and the second — 150 (30,4%) without OPCA (<50%). The clinical and functional status of patients before CAG was assessed including 29 parameters. For data processing and analysis, the Mann-Whitney, Fisher, chi-squared tests and univariate logistic regression (LR) were used. In addition, for the development of predictive models, we used multivariate LR (MLR), support vector machine (SVM) and random forest (RF). The models was assessed using 4 metrics: area under the ROC-curve (AUC), sensitivity, specificity, and accuracy. Results. A comprehensive analysis of functional and metabolic status of patients made it possible to identify the CMR factors that have linear and nonlinear association with OPCA. Their weighting coefficients and threshold values with the highest predictive potential were determined using univariate LR. The quality metrics of the best predictive algorithm based on an ensemble of 10 MLR models were as follows: AUC — 0,82, specificity and accuracy — 0,73, sensitivity — 0,75. The predictors of this model were 7 categorical (total cholesterol (CS) ≥5,9 mmol/L, low-density lipoprotein cholesterol >3,5 mmol/L, waist-to-hip ratio ≥0,9, waist-to-height ratio ≥0,69, atherogenic index ≥3,4, lipid accumulation product index ≥38,5 cm*mmol/L, uric acid ≥356 pmol/L) and 2 continuous (high density lipoprotein cholesterol and insulin resistance index) variables. Conclusion. The developed algorithm for selecting predictors made it possible to determine their significant predictive threshold values and weighting coefficients characterizing the degree of influence on endpoints. The ensemble of MLR models demonstrated the highest accuracy of OPCA prediction before the CAG. The predictive accuracy of the SVM and RF models was significantly lower.