Abstract:
Objective To quantitatively evaluate the risk of coronary heart disease based on large scale epidemiological surveillance data.
Methods Epidemiological data, including demographic information and living habits, medical and family history, testing indicators and electrocardiogram indicators, were collected from 19 021 cases with chronic disease by community survey that was conducted by Chinese PLA General Hospital in 2015. Samples with less than 70% data completeness were eliminated. Stepped K-Nearest Neighbor method was used to fill the missing value, Adaboost algorithm was used to assess the risk of coronary heart disease, and 10-fold crossover method was applied for model validation.
Results Age, duration of hypertension, dyslipidemia, presence of other comorbidities, duration of diabetes and low-density lipoprotein cholesterol were important indicators for evaluating the incidence of coronary heart disease. The recall rate, accuracy, AUC and F1 values of the model for evaluating the risk of coronary heart disease were 0.727, 0.741, 0.796 and 0.796, respectively.
Conclusion Our model can provide personalized prediction of the risk of coronary heart disease.
下载: