ALGORITHM TO STRATIFY THE RISK OF MYOCARDIAL INFARCTION IN PATIENTS WITH ACUTE CORONARY SYNDROME AT PRIMARY EXAMINATION

Abstract

The episode of acute coronary syndrome is most often preceded by the development of systemic and local inflammation, which plays a significant role in the pathogenesis of the disease. General clinical blood analysis, directly or indirectly reflecting systemic pathological processes in the patient’s body based on quantitative and morphological assessment of blood composition, is one of the most affordable methods of laboratory diagnostics in modern public health. Taking into account the growing number of digital data obtained by diagnosticians from analytical systems, there is a growing potential for the use of machine learning methods to increase the effectiveness of provided diagnostic information in the interests of the patient. The aim of this study was to create an algorithm for stratifying the risk of myocardial infarction based on the methods of machine learning in patients with acute coronary syndrome at primary examination. A prospective pilot study was conducted. In total 307 patients with acute coronary syndrome (169 men and 138 women) were examined. The average age of patients was 68.6 ± 12.5 years. Retrospectively, the patients were divided into two groups: the main group - patients with the final diagnosis “Myocardial infarction” and the control group with the diagnosis “Unstable angina pectoris”. All patients at hospitalization at the primary laboratory examination along with the study of the concentration of cardiac troponin I by a highly sensitive method were examined by a general clinical blood analysis on an automatic hematological 5-diff analyzer. As a result of the application of the ensemble method as a method of machine learning and artificial neural networks as 6 independent models of the ensemble it was possible to achieve the area under the ROC curve = 0.77 on the test set when assessing the quality of patient stratification. Taking into account the volume of the training sample in 214 patients and the results of similar studies, the achieved stratification quality can be considered acceptable and promising for further accumulation of the database with the purpose of additional training of the developed algorithm and improvement of the disease prognosis accuracy characteristics.