Simple Models to Assess the Risk of Developing Diabetes

• Activity Number: 17
• Type: Topic Contributed
• Date/Time: Sunday, August 8, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ken Williams*+ and Steven E. Stern and Michael P. Stern
• Companies: University of Texas, San Antonio and Australian National University and University of Texas, San Antonio
• Keywords: diabetes ; incidence ; prediction ; logistic regression ; recursive partitioning ; score

Abstract

Available models which can identify individuals likely to benefit from diabetes preventive treatment are not commonly applied because they require calculations or a two-hour oral glucose tolerance test. We therefore sought ways to easily identify in a clinic, individuals likely to develop diabetes. Of 3,228 subjects examined in the San Antonio Heart Study between 1979 and 1988 and again after seven to eight years, 295 developed diabetes. A logistic regression model from this study estimated the probability of developing diabetes as a function of age, gender, ethnicity, fasting glucose, blood pressure, HDL cholesterol, body mass index, and family history. We dichotomized each continuous predictor and rescaled the coefficients so they totaled from 0 to 100 points to provide a simple risk assessment score sheet. The area under the receiver operating characteristics curve of this score was 0.831 vs. 0.836 (p = 0.51) for the continuous model. We also used recursive partitioning to develop a decision tree model which provided equivalent sensitivity at high specificity from simple decision rules, e.g., treat if fasting glucose > 100 and HDL < 50 mg/dl (sensitivity = 30%, specificity = 97%).