(M-019) Predictors Of Insulin Dose In Individuals With New-Onset Type 1 Diabetes

Alejandro Lozano, B.S. – Data Manager II, Critical Path Institute; Frank Walker, B.S. – Data Manager III, Critical Path Institute; Erin Beach, B.S. – Senior Project Coordinator, Critical Path Institute; Brianna Greeno, B.S. – Senior Project Manager, Critical Path Institute; Paul Belmonte, PhD – Scientific Director, Type 1 Diabetes Consortium, Critical Path Institute; Joseph Hedrick, PhD – Executive Director, Type 1 Diabetes Consortium, Critical Path Institute; Yi Zhang, PhD – Director of Pharmacometrics, Critical Path Institute; Shu Chin Ma, PhD – VP of Model-Informed Drug Development & Quantitative Medicine, Critical Path Institute

Objectives: Type 1 diabetes is an autoimmune disease characterized by destruction of pancreatic beta cells, requiring these individuals to be completely dependent on exogenous insulin. Intensive insulin therapy is highly burdensome for patients and caregivers and is associated with the risk of fatal hypoglycemia. Residual endogenous insulin secretion is associated with reduced exogenous insulin needs and improved outcomes [1]. Our objective was to understand the predictors of exogenous insulin use in new-onset T1D patients, defined as less than 100 days from diagnosis. Although reduction in insulin dose is not a validated endpoint, these predictors can provide meaningful information in clinical trial development.

Methods: Subject-level data from 2048 individuals with new-onset T1D were pooled from 17 randomized clinical trials. Data were supplied from the NIDDK Central Repository (TN05, TN08, TN09, TN14, TN19), GlaxoSmithKline (DEFEND1, DEFEND2, GSKALB), ExTOD, NIAID (EXTEND), Diamyd, MacroGenics (Protégé), Immune Tolerance Network (RETAIN, T1DAL), Janssen Research and Development, LLC (T1GER), Yale University (AbATE) and UCSF. Weight-adjusted total daily insulin values were modeled as averages (3-7 days) and square root-transformed. The model was trained and validated using an 80/20 data split. Stepwise covariate model building was used to test the following covariates: baseline (insulin, age, BMI, disease duration), sex, race, ethnicity, and HLA genotypes. In addition, a drug effect model was developed by grouping active treatment arms by mechanisms of action. Model performance was guided by standard goodness-of-fit measures and visual predictive checks.

Results: Insulin use was best described by a quadratic function with additive inter-individual and residual unexplained variability. After the development of a symptomatic drug effect model to account for changes in exogenous insulin use in the presence of additional therapies (deltaAIC: -63.1), significant predictors were as follows: A0 (baseline age, baseline disease duration, race, and sex), A1 (baseline disease duration and race). The change in AIC after the addition of covariates was -244.4. Visual predictive checks on the training and validation dataset showed good performance.

Conclusions: Baseline age, disease duration, race and sex were predictors on exogenous insulin use after accounting for changes based on additional therapies. Understanding these predictors of exogenous insulin use can help inform the design of future clinical trials. Our next steps are to utilize this insulin model in combination with HbA1c and c-peptide as a multivariate model to characterize the interactions between each endpoint. These models will serve as the basis for an interactive clinical trial simulation tool that allows sponsors to specify ranges of predictive covariates, sample size and anticipated drug effect to determine clinical trial power.

Citations: [1] Mireia Fonolleda, Marta Murillo, Federico Vázquez, Joan Bel, Marta Vives-Pi; Remission Phase in Paediatric Type 1 Diabetes: New Understanding and Emerging Biomarkers. Horm Res Paediatr 21 November 2017; 88 (5): 307–315. https://doi.org/10.1159/000479030.