Application of Machine Learning Model in Prediction of Adverse Cardiovascular Events

Construction of prognostic models is a promising direction for preventive medicine. The search for new factors affecting cardiovascular health is an important addition to conventional risk scores. The aim of the study was to search for significant cardiovascular risk factors and develop a prognostic model using machine learning in healthy individuals. The analysis was based on an dataset of anamnestic, clinical, paraclinical, socio-economic and other parameters of two stages of the epidemiological study (Research Institute for Complex Issues of Cardiovascular Diseases, Kemerovo), which included 1 217 participants aged 35–70 years. There were 70.9 % (n = 863) and 29.1 % (n = 35) healthy respondents and cardiovascular patients, respectively. A total of 1, 915 features were analyzed using artificial intelligence. We identified 28 significant predictors of the following unfavorable cardiovascular outcomes: angina, myocardial infarction, heart failure, stroke, arrhythmias (atrial fibrillation and/or flutter), etc. Based on these, a prognostic model was developed. It should be noted that the most significant parameters included the forced expiratory volume in one second, internal fat proportion, no alcohol consumption, a change in salt intake after a doctor's recommendation, and no job. The paper determined the significant features that had not previously been recognized as cardiovascular risk factors affecting cardiovascular health. This undoubtedly provides an information gain for conventional prognostic models.