Prediction Models Combining Clinical Measures Identify Participants with Youth-Onset Diabetes Who Maintain Insulin Secretion

With the high prevalence of pediatric obesity, it is challenging to accurately classify diabetes (DM) type, especially in a diverse population of youth. Maintained fasting C-peptide ≥ 0.75ng/ml (250pmol/L) , a level associated with type 2 diabetes and successful insulin withdrawal in adult cohorts, may have implications for treatment and risks of acute complications. We developed clinical prediction models to identify individuals with youth onset DM who will retain endogenous insulin secretion. We studied 2991 youth in the SEARCH study (DM diagnosed ≤ age years) to develop logistic regression based prediction models using clinical measures (race/ethnicity, BMI Z-score, age at diagnosis, triglycerides and HDL-C) with or without islet autoantibodies (Aab) (GADA, IA-2A, ZnT8A) , to identify those with fasting C-peptide ≥ 0.75ng/ml at 3-yr (median 75 months) DM duration. Model covariates (measured median 9 months duration) were chosen based on previous literature and availability. A prediction model using clinical measures was highly accurate in identifying participants with C-peptide ≥75 ng/ml (17% of participants, 3% and 59% of those with and without positive Aab) : area under receiver operator curve (AUC ROC) 0.950 (95%CI 0.938-0.962) , model pseudo-R2 0.69. Adding Aab improved discrimination to AUC ROC 0.976 (0.968,0.983) , pseudo-R2 0.81. In internal validation, optimism was very low, with excellent calibration (slope 0.995-0.999) . Performance remained high in difficult to classify groups: AUC ROC was 0.947 (0.927, 0.966) in those with negative Aab and 0.923 (0.960, 0.984) (model with Aab) in youth with high BMI (Z-score >2) . Discrimination was lower in non-Hispanic white participants (AUC ROC 0.94) compared to participants of other race and ethnic groups (AUC ROC 0.97-0.99) (model with Aab) . In conclusion, prediction models combining routine clinical measures can accurately identify youth with DM who maintain endogenous insulin secretion. Disclosure A.Jones: None. A.H.Williams: None. S.M.Marcovina: None. C.Pihoker: None. J.Divers: None. M.J.Redondo: Advisory Panel; Provention Bio, Inc. B.Shields: None. R.A.Oram: Consultant; Janssen Research & Development, LLC, Research Support; Randox R & D. D.Dabelea: None. W.Hagopian: Research Support; Janssen Research & Development, LLC. E.Lustigova: None. A.S.Shah: None. A.K.Mottl: Advisory Panel; Bayer AG, Board Member; Bayer AG, Research Support; Alexion Pharmaceuticals, Inc., Aurinia, Bayer AG, Boehringer Ingelheim International GmbH, Pfizer Inc. R.Dagostino: Consultant; Aetion, Inc., AstraZeneca, Biogen, Bristol-Myers Squibb Company, Daiichi Sankyo, Merck & Co., Inc. Funding Centers for Disease Control and Prevention (PA numbers 00097, DP-05-069, and DP-10-001) and the National Institute of Diabetes and Digestive and Kidney Diseases. This work is also supported by NIH RO1 awards DK124395 and DK121843-01.