Predicting the Risk of Coronary Heart Disease and Diabetes in Taiwan

An accurate risk prediction model can be used to prevent diseases and promote health. Recently, many methods in selecting models have been developed. Discrimination and calibration are two major elements for evaluating the performance of prediction models. Measures of the discrimination performance of models are derived from Harrell’s C statistic, and the calibration abilities are derived from Hosmer-Lemeshow test. We used the Nutrition and Health Survey in Taiwan (NAHSIT, 1993-1996) linked to National Health Insurance database and mortality records (1996-2006) to construct the disease risk prediction models for coronary heart disease, diabetes, and chronic kidney disease. Results show that age, systolic blood pressure, triglyceride, uric acid, and BMI are major risk factor for CHD, and age, fasting glucose, triglyceride, and the waist circumferences are major risk factors for diabetes. Our models can predict the probability of having CHD or diabetes in 10 years given risk factors. The values of C statistic and H-L test for CHD risk prediction model are 0.67 and 13.68 for men, and 0.68 and 10.21 for women. The values of C statistic and H-L test for diabetes risk prediction model are 0.7 and 10.65 for men, and 0.73 and 22.15 for women. That indicates the models fit the data well. We will use another data set to validate the results.