The development of methodological approaches to machine learning use in the patients with miocardial infarction dispensary follow-up

Cardiovascular diseases (CVDs), including myocardial infarction (MI), persist as the primary cause of mortality and a considerable burden on healthcare systems. Although there has been a decrease in mortality rates, the incidence of complications post-treatment remains high. The integration of machine learning (ML) technology offers significant opportunities to enhance risk-based patient follow-up. Objective. To develop and rigorously substantiate methodological and organizational approaches for incorporating machine learning technologies into the follow-up care system for patients who have experienced a myocardial infarction. Additionally, the study seeks to evaluate the potential applications of these technologies for risk-based outpatient medical monitoring. Materials and methods. This study used data obtained from a multicenter observational study of myocardial infarction conducted between 2015 and 2017 across 15 regions of the Russian Federation (n=1096). Within a follow-up period of six months, cardiovascular mortality was recorded in 98 patients, representing 9.7% of observations with a known outcome. To predict risk, gradient boosting was applied to five clinical predictors, using cross-validation and metrics such as AUC and LogLoss. Results. The study developed an ML model for risk-based follow-up care after MI. By adjusting the classification threshold, an AUC of 0.815 was achieved, thereby reducing the probability of overlooking high-risk patients. The model facilitates the early identification of adverse outcomes and promotes the efficient allocation of healthcare resources. Conclusions. A machine learning model for risk-based follow-up care after MI has been developed, enabling the prediction of cardiovascular mortality within six months and enhancing the precision of medical monitoring. Nevertheless, the practical implementation of this model at present requires additional clinical validation, greater interpretability, and a more detailed exploration of organizational and regulatory dimensions.