The study aimed to assess the performance of a lifestyle-based prognostic risk model (Diabetes Lifestyle Score) for the prediction of 5-year risk of type 2 diabetes mellitus. The model comprises nine self-reported predictors (sex, age, antihypertensive drugs, body mass index, family history of diabetes, physical activity, fruits, vegetables, and wholemeal/brown bread). We conducted an external validation and update of the model in an Australian cohort including 97,615 residents of New South Wales aged 45 years and older who were free of type 1 and 2 diabetes mellitus at baseline. Of all participants, 4,741 developed type 2 diabetes mellitus over 5 years. We conducted the statistical analyses in RStudio using the programming language R. The area under the receiver operating characteristic curve (AUC) of the original model was 0.726 (95% confidence interval: 0.719, 0.733). After adjusting the calibration intercept and slope, the original model performed reasonably well in the external cohort. The best performance was measured by using the numerical predictors as continuous variables and refitting all coefficients (AUC: 0.741, 95% confidence interval: 0.734, 0.748). The results of the original model after calibration were comparable to those received from the AUSDRISK score which is routinely used in Australian clinical practice. Hence, the lifestyle-based model might be a reasonable alternative for laypersons since the required information is most likely known by these. Further, the risk score may communicate the message about the importance of a healthy diet to reduce the risk of diabetes.
Keywords: Cohort analysis; Diabetes mellitus, type 2; Logistic regression; Risk factor scores; Validation study.
© 2021 The Authors.