Cross-sectional and Longitudinal Validation of Risk Prediction Models for Type 2 Diabetes Mellitus

Abstract

Introduction: Early detection of undiagnosed diabetes and prediction of future diabetes are crucial for preventing or delaying the detrimental complications of diabetes. For this purpose, various risk prediction models including a recently published Korean screening score composed of non-laboratory parameters have been developed. We evaluated the validity of the Korean screening score for undiagnosed and incident diabetes in an independent study population. Further, its predictive performance was compared with various other non-laboratory risk prediction models and laboratory parameters. Methods: The data of 26,675 individuals who visited Seoul National University Hospital Healthcare System Gangnam Center for health screening program were reviewed for cross-sectional validation of undiagnosed diabetes and the data of 3,029 individuals with mean 6.2 years of follow-up were reviewed for longitudinal validation of incident diabetes. The predictive performance of the Korean screening score, other 16 previously published risk prediction models and the risk prediction model of laboratory parameters were compared. Results: For the screening of undiagnosed diabetes, the Korean screening score exhibited a sensitivity of 81%, a specificity of 58%, and an area under the curve of receiver operating characteristic curve (AROC) of 0.754. All the other non-laboratory risk prediction models revealed comparable AROC. For the prediction of incident diabetes, the Korean score demonstrated a sensitivity of 74%, a specificity of 54%, and an AROC of 0.696. Relative to the Korean score, the risk prediction by the laboratory parameters - fasting plasma glucose (FPG) and hemoglobin A1c (HbA1c) level - demonstrated a significantly higher AROC (0.838 vs. 0.696, P value <0.001). Combining of FPG, HbA1c and the Korean score data increased AROC by small increment (0.838 vs. 0.849, P value = 0.016) without statistically significant improvement in risk classification (net reclassification index 4.6%, P value = 0.264, integrated discrimination improvement 0.006, P value = 0.006). Conclusions: In conclusion, the Korean screening score is useful for detecting undiagnosed diabetes but inferior to the laboratory parameters for the prediction of incident diabetes.