Evaluation of Four Semi-Mechanistic Models for Predicting Glycemic Control With a Glucagon Receptor Antagonist in People With Type 2 Diabetes.

Abstract

Glycated hemoglobin (HbA1c) is the gold standard for measuring long-term glycemic efficacy over at least 3 months in Type 2 diabetes (T2D). Being able to predict HbA1c using glucose response from studies of less than 3 months would be useful. Four semi-mechanistic HbA1c models (ADOPT, FFH, FHH, and IGRH) were evaluated for their predictive performance of longer-term HbA1c at 24 weeks of treatment using glucose and HbA1c data up to 4 weeks of treatment. A novel glucagon receptor antagonist (LY2409021) was evaluated in T2D patients for glycemic control. The models were built using LY2409021 pharmacokinetics, glucose, and HbA1c data from a 4-week Phase 1b study. Predictive performance of the models was assessed based on comparing model-estimated and observed HbA1c values from a 24-week Phase 2b study. Metrics for predictive performance included: (a) mean change from baseline HbA1c (ΔHbA1c) at Week 24 between observed and simulated values; (b) mean prediction error (MPE) for bias; and (c) root mean squared error (RMSE) for precision. Overall, the FHH and IGRH models closely predicted the mean ΔHbA1c at Week 24 within 0.1% difference from the observed values in the Phase 2b study. Both models also had reasonable bias (absolute MPE < 0.1%) and precision (RMSE < 0.3%) estimates. Conversely, the ADOPT and FFH models over-predicted the mean reduction in HbA1c by 0.288% and 0.153%, respectively. The FHH and IGRH models featured transit compartments for modeling long delays between glucose and HbA1c. Thus, these models better represented the physiology and provided superior predictive performance.