Data mining techniques for forecasting the medical resource consumption of patients with diabetic nephropathy

Diabetes has become an important public health issue in the twenty-first century, and dialysis treatment has become a large burden on the National Health Insurance of Taiwan. Diabetic nephropathy(DN) is the leading factor that determines whether patients with diabetes will require dialysis. Statistical data published by the Ministry of Health and Welfare in 2015 indicated that, second only to cancer, chronic kidney failure is the most prevalent disease treated by primary outpatient clinics. In addition, according to the National Health Insurance Administration Ministry of Health and Welfare, 6% of the national health insurance budget was spent to cover the dialysis treatment of ESRD patients. Therefore, in this study, we proposed and developed a forecasting model for the medical resource consumption of DN patients. We used multiple regression, stepwise regression, multivariate adaptive regression splines (MARS), support vector regression, and twostage model (T-SVR). We used a combination of important variables screened out by stepwise regression and MARS to construct the T-SVR model. We screened out the important factors with a significant impact on medical consumption. We then identified the model with the best forecasting performance out of the five data mining techniques. Our results can aid the managers of medical institutions to properly and effectively allocate medical resources and control medical expenses.