Μπορούν οι σωστές ερωτήσεις να προβλέψουν καρδιακές παθήσεις; Αξιολόγηση προβλεπτικής ικανότητας ερωτηματολογίου και αξιοποίηση στην πρωτοβάθμια φροντίδα.

Abstract

As early as the 1980s the scientific community has identified that personal health behaviors play major role in premature mortality. For this reason, the Centers for Disease Control and Prevention (CDC) which is the national health agency of the United States, has established a behavioral surveillance system, BRFSS. Its purpose, among other things, is monitoring specific chronic diseases and analyzing the prevalence of risk factors in the population using surveys. Using data from the survey conducted in 2015 and machine learning, we aim to create a classification algorithm that is able to accurately predict cases of heart disease and provide a closer look at the importance of the risk factors. Cardiovascular diseases, which include heart attacks, strokes, and heart failure, resulted in 17.9 million deaths (32.1%) in 2015, 80% of whom were considered preventable. This has led many countries to implement educational campaigns in order to increase awareness and take preventive measures. By the end of this thesis, we were able to tune an XGBoost algorithm that can correctly predict 94% of the cases with heart disease and can be used as a screening tool. Furthermore, using the library SHAP not only did we identify risk factors on a population level but also, we take a detailed look at the risk factors of individual persons and the impact of these factors which can be used to create personalized plans indicating which prevention activities each person should take part. This will make the prevention activities much more efficient and of course each patient will be able to get the maximum out of them