Assess the heart disease risk of the Chinese elderly using a predictive model

Abstract

The accelerating aging process worldwide makes chronic diseases the predominant risk for public health, and heart disease is in the top causes of the mortality of the elderly. Studies have verified the interventions can prevent, reduce or delay the onset of chronic diseases. This paper aims to find the domain predictors of heart disease by applying a machine learning technique Extreme Gradient Boosting to 89 predictors extracting from genetic, lifestyle, economic condition, isolation, stressful life events, nutrition and availability of medical service indexes. The individual-level data used is Chinese Longitudinal Healthy Longevity Survey with the time range of 2000 to 2002, and 2011 to 2014. We apply the imputation and oversampling technique to improve the prediction performance and use a step by step parameter tuning process to get the best hyper-parameters needed in the modeling. The fitted predictive model reaches a prediction accuracy of above 90% in the independent test data set. Comparing the first investigated period of 2000 to 2002 with the second period of 2011 to 2014, the predictors associated with economic condition play an important role in the prediction. The nutrition factor, surprisingly, does not contribute significantly to the prediction capability.