A Risk Score for Predicting the Incidence of Type 2 Diabetes in a Middle-Aged Korean Cohort

Abstract

Methods and Results: Participants without diabetes (6,342 participants, aged 40-69 years) were included and biennial follow ups were conducted. A logistic regression analysis was used to construct the models. The basic model was based on simple information such as age, parental or sibling history of diabetes, smoking status, body mass index, and hypertension, while clinical model 1 was constructed by adding biochemical tests such as fasting plasma glucose, high-density lipoprotein-cholesterol and triglycerides to the basic model; clinical model 2 further added glycated hemoglobin (HbA1c) to clinical model 1. The model accuracy was assessed using area under a receiver operating characteristic (AROC) curve and the Hosmer-Lemeshow statistics. Both net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were calculated to determine the contribution of HbA1c. Two clinical models improved model discrimination (AROC=0.75 and 0.77) when compared with the basic model (AROC=0.65). The addition of HbA1c to clinical model 1 increased AROC by only 0.02 despite its high impact on the prediction of diabetes (odds ratio=2.66). However, the NRI and IDI were significantly improved with the addition of HbA1c. Therefore, a risk score system was developed to estimate the 4-year risk of diabetes based on clinical model 2