Innovative approach for cardiovascular risk predicting by machine learning methods using

The feature of machine learning is the identification in large sets of non-linear relationships data influencing the development, course and outcome of diseases. Objective. To develop an innovative model of the prognostic scale of cardiovascular risk and to determine a range of the model parameters using machine learning methods. Materials and methods. The model was based on a data set of 1217 participants: 354 (29.1% of the sample) had cardiovascular diseases, 863 (70.9%) did not have. Totally, 1915 parameters of different groups were studied: evaluation of the healthcare system, laboratory health indicators, etc. Gradient boosting method was the main approach to develop the machine learning model, SHAP framework was applied. Results. Out of all groups of parameters the most important ones were determined, 9 models were developed, a model consisting of 28 parameters was selected for further study: AUC 0.72, sensitivity 0.71. Elevated internal fat content, waist measurement, waist-to-hip ratio, low forced expiratory volume in 1 second values, total duration of physical activity at work and so on are associated with cardiovascular risk. Conclusion. With the use of machine learning methods, new non-classical indicators affecting the cardiovascular prognosis have been determined: instrumental and laboratory (parameters of spirometry, bioimpedansometry and dynamometry), lifestyle (absence of alcohol consumption, physical activity in different types of activity), healthcare system (receiving recommendations and lifestyle changes) and socio-economic status (absence of work and other things). The prognostic scale of cardiovascular risk has been formed, the range of the model parameters’ values has been defined.