Prediction of cardiovascular events using proportional risk models and machine learning models: a systematic review

Relevance. Every year, a large number of people around the world become victims of cardiovascular diseases. To date, the main tools for predicting cardiovascular risk are scales based on proportional
risk models (Cox regression). However, recently many scientists agree that the use of machine learning and artificial intelligence technologies can help to improve the quality of adverse cardiovascular events onset prognosis.

The aim is to conduct a systematic literature review of approaches to the formation of CVD development forecasts based on proportional risk assessment scales and ML methods to identify the most effective methods of data analysis.

Materials and methods: A systematic review of the literature was conducted, which included 58 research papers using methods for assessing cardiovascular risk based on Cox regression and machine
learning technologies.

Results. Predictive capabilities of machine learning are superior to traditional linear methods of data analysis. The average AUC values are 0.82 and 0.75, respectively, p=0.003. It was also possible to
identify the most frequently used and effective prediction algorithms. They turned out to be random forest, gradient boosting and deep learning. However, unlike traditional prediction scales, 80% of the
presented machine learning algorithms did not undergo external validation on independent samples. In addition, the use of machine learning requires a large amount of high-quality digital data.

Discussion. As a result of similar studies analysis conducted by domestic and foreign authors, we were able to confirm that, on average, predictive models built using ML algorithms have advantages
over traditional data analysis methods. Conclusions. Machine learning is a promising method of predicting cardiovascular events, but for its mass use it is necessary to switch to electronic medical records management and aggregation of more qualitative and structured information.