External validation and clinical application of the risk prediction model for severe hypoglycemia in type 2 diabetes

Object: We previously developed and internally validated a one-year risk prediction model for severe hypoglycemia (SH) in type 2 diabetes. In this study, we planned an external validation to verify the performance of the previously-developed risk prediction model in a hospital setting. Methods: From December 2017 to December 2019, patients over 30 years old who visited the diabetes center were enrolled. The ROC curve and Harrell C-statistics were compared to identify the discrimination of the developed model. The predicted incidence and actual incidence of SH at 1 year in the development and validation cohort were compared by ranking subjects into deciles of predicted risk. Results: External validation was performed with a total of 2,757 patients. The incidence rate of severe hypoglycemia was 12.6 per 1000 patient-year. The concordance index for SH was 0.854 in the external validation cohort. The sensitivity and specificity of this model prediction in the external validation cohort were 0.714 and 0.853, respectively. However, compared with the model for the development cohort, the incidence rate of SH in each decile group was three times higher than the predicted value. SH risk scores increased in proportion to the age and the duration of diabetes. The risk score for SH was higher in subjects with chronic kidney disease, heart failure, and dementia than those with other major comorbidities. Conclusion: Our SH prediction model showed excellent discrimination when applied to type 2 diabetes subjects in the hospital-based cohort. The results confirm that our risk prediction model for SH can screen the high-risk of SH patients efficiently.