Development and validation of the elderlies' diabetes risk predictive model using the Chinese data

Abstract

Ageing is closely related to the functional decline and is the predominant causes of the chronic diseases such as cardiovascular disease, stroke and diabetes. Population ageing worldwide accelerates the prevalence of the chronic disease. Ageing China is suffering from the diabetes risk more than other countries according to WHO reports. We adapt a machine learning algorithm Extreme Gradient Boosting to model the incidence rate of diabetes in China using a large amount of individual-level characteristic indexes as predictors. The model performance is guaranteed with a prediction accuracy above 85%, arising from the use of minority class oversampling and a multi-variable grid search technique. We apply the 2000-2002 wave and 2011-2014 wave of the Chinese Longitudinal Healthy Longevity Survey (CLHLS) to investigate how the leading predictors of the diabetes risk change as time pass. The importance of social-economic status, life-style and the access to the medical service rise in the later wave, and the relative importance of isolation and stressful life events which are related to social-psychological health decline in the investigated period, indicating a disparity of the diabetes risk within subgroups of different economic conditions.