LAUSR

Objective: Develop and assess the performance of clinical/genetic models for quantifying the probability of T1D among diabetes patients with non-classic features. Research Design and Methods: Subjects were defined as having classic features of T1D (ICD codes for T1D not type 2 diabetes [T2D], insulin use, onset ≤20 y; N=380) and T2D (ICD codes for T2D not T1D, not on insulin, onset ≥35 y; N=10,745), or neither, as “non-classic” diabetes (N=11,049) from the UK Biobank. Using simple measures (BMI, HDL, TG, HbA1C) and T1D/T2D polygenic scores, logistic regression models based on clinical (Model-C) and clinical + genetic variables (Model-C+G) were developed in a training set of 80% of the classically-defined T1D and T2D patients, validated in the remaining 20%, and applied to patients with non-classic features. Results: The C-statistic of Model-C and C+G for discriminating T1D and T2D in the training data-set were 0.91 and 0.96, respectively. In the diabetes patients with non-classic features, an increasing trend of insulin use and decreasing trend of oral anti-hyperglycemic drug use were found in patients with increasing probability of T1D by both models, all P-trend Conclusions: We developed clinical/genetic probability models that can assist in determining diabetes type and personalized treatment strategies among diabetes patients with non-classic features. Disclosure L. K. Billings: Advisory Panel; Self; Bayer Inc., Lilly Diabetes, Novo Nordisk, Sanofi. Z. Shi: None. W. Resurreccion: None. A. Qamar: None. C. Wang: None. J. Wei: None. T. I. Pollin: None. M. Udler: None. J. Xu: None.