Study of cardiovascular disease prediction model based on random forest in eastern China

Abstract

Cardiovascular disease (CVD) is the leading cause of death worldwide and a major public health concern. CVD prediction is one of the most effective measures for CVD control. In this study, 29930 subjects with high-risk of CVD were selected from 101056 people in 2014, regular follow-up was conducted using electronic health record system. Logistic regression analysis showed that nearly 30 indicators were related to CVD, including male, old age, family income, smoking, drinking, obesity, excessive waist circumference, abnormal cholesterol, abnormal low-density lipoprotein, abnormal fasting blood glucose and else. Several methods were used to build prediction model including multivariate regression model, classification and regression tree (CART), Naïve Bayes, Bagged trees, Ada Boost and Random Forest. We used the multivariate regression model as a benchmark for performance evaluation (Area under the curve, AUC = 0.7143). The results showed that the Random Forest was superior to other methods with an AUC of 0.787 and achieved a significant improvement over the benchmark. We provided a CVD prediction model for 3-year risk assessment of CVD. It was based on a large population with high risk of CVD in eastern China using Random Forest algorithm, which would provide reference for the work of CVD prediction and treatment in China.

Figure 1

An illustrative schematic for CVD prediction model.

Figure 2

ROC curves of prediction models for CVD. (a) ROC curve of Multivariate Regression model for CVD. (b) ROC curve of CART model for CVD. (c) ROC curve of Naïve Bayes model for CVD. (d) ROC curve of Bagged Trees model for CVD. (e) ROC curve of Ada Boost model for CVD. (f) ROC curve of Random Forest model for CVD. (g) ROC curve of Framingham Score model for CVD.