Development and testing of new methodical approaches for predicting cardiovascular events in healthy people using machine learning technology based on the «INTEREPID» international research

ABSTRACT

Cardiovascular diseases remained the leading cause of death all over the world in 2023 yr. Prevention is a key issue for the development of current medicine to reduce the burden of this pathology. Up to date, main tools of monitoring are assessment scales of absolute and relative cardiovascular risk. However, more and more researchers consider the use of artificial intelligence in predicting
heart diseases due to development of information technologies.

Objective. To develop and test new methodical approaches for predicting cardiovascular events in healthy people using artificial intelligence.

Material and methods. The study was performed on the basis of «INTEREPID» international research data. The sample consisted of 2,392 thous. participants who have been observed for 4 years, among them 1022 (42.7%) were men, 1369 (57.2%) — women. The analysis included 191 predictors. We used 5 classification algorithms to create prediction models on the Python programming
environment: RandomForestClassifier, GradientBoostingClassifier, ExtraTreesClassifier, XGBClassifier, LGBMClassifier. ROCanalysis was used to evaluate the effectiveness of prediction models.
Results. The most effective algorithm was GradientBoostingClassifier with AUC-0,76. The worst result had ExtraTreesClassifier with AUC-0,68. The most significant risk factors were age, C-reactive protein level in the blood and animal fat consumption.

Conclusion. As a result of the research we were able to obtain a prediction algorithm with relatively good quality of discrimination. Further studies on larger amount of data are necessary to improve this development.