Background: Developing risk prediction model has been long practiced. Health insurance data records the date and time clinic/hospital visits occur. It is a very good source of tacking patients’ diseases.
Material and Method: This study utilized data from the 1993-1996 Nutrition and Health Survey in Taiwan (NAHSIT 1993-1996) as the baseline and linked it to the health insurance claim data for 10 years’ cardiovascular events. Harrell’s C statistics was used to examine the discrimination of the model, whereas Hosmer-Lemeshow ?2 was used to calibrate the model. In addition, AIC was used in model selection. For cross-validation, the data was divided into 5 parts. Four of them was used for model developing, and the rest was used for testing. The external validation was carried out using another set of data collected in 2002 and repeated in 2007. As the insurance data was for claim, we used the hard events to ensure the accuracy of the events.
Results: Hard events was defined as hospitalized or death due to certain diseases. Coronary heart disease (CHD) and stroke were the outcome. Results showed the models were within reasonable ranges.