Modeling hospitalization medical expenditure of the elderly in China

The inpatient medical expenditure data of the elderly typically have semicontinuous characteristics. The main reason is that a fraction of the old people has no hospitalization costs, while the other incurs substantial hospitalization expenses. Therefore, it is difficult to effectively predict the hospitalization expenses of the elderly using the classical linear regression model by ordinary least squares (OLS). To solve the shortcomings of classical linear regression model, this paper attempts to use the logarithmic regression, Tobit, two-part (i.e., binomial-gamma, binomial-inverse Gaussian), and Tweedie models to predict and model the hospitalization medical expenditure of the elderly in China. Based on the Akaike information criterion (AIC), root mean square error (RMSE), and mean absolute percentage error (MAPE) statistical criteria and the Chinese Longitudinal Healthy Longevity Survey (CLHLS) data of 2018, we found that the Tweedie model accurately predicts the rate of no hospitalization expenses. Additionally, we found that age, family income, whether there is insurance, whether there is limited movement, the number of chronic diseases, health status, and residence factors significantly affect the hospitalization expenses of the population of elderly Chinese, which provides a decision-making basis for policy making.