A Review of Semi-Mechanistic HbA1c Models in People with Type 2 Diabetes: Strengths, Limitations, and Applications.

Type 2 diabetes (T2D) is a chronic metabolic disorder associated with high morbidity and mortality, necessitating the urgent need for the research of novel antidiabetic compounds. Population pharmacokinetic/pharmacodynamic (PK/PD) models have been used in T2D drug development to inform dose selection for antidiabetic compounds. However, most population PK/PD models are empirical, and physiological processes (e.g., glucose-mediated hemoglobin glycation) are not routinely incorporated. Conversely, semi-mechanistic models in people with T2D incorporate some relevant physiological processes, enabling prediction of long-term glycated hemoglobin (HbA1c) using only short-term glucose and HbA1c data. Semi-mechanistic T2D models may also have improved ability for extrapolation to multiple antidiabetic drug classes. This narrative review summarizes six published semi-mechanistic models commonly used or discussed for HbA1c prediction in people with T2D: A Dynamic HbA1c EndpOint Prediction Tool (ADOPT), FPG-FSI-HbA1c (FFH), FPG-Hb-HbA1c (FHH), Integrated Glucose-RBC-HbA1c (IGRH), Weight-HbA1c-Insulin-Glucose (WHIG), and Body Weight-Directed Disease Trial (BWDDT) models. An overview of the HbA1c predictive performance is presented for four of these models (ADOPT, FFH, FHH, and IGRH models). Strengths and limitations of the six semi-mechanistic T2D models are evaluated based on model assumptions, HbA1c model predictive performance, predictive accuracy for collected biomarker data, model complexity, and computational considerations. A decision tree framework is provided for model selection. Practical strategies for model development are discussed, including approaches for incorporating different drug effects into the models. Future directions include model validation with early-phase data from various antidiabetic drug classes, and exploration of alternative implementation of disease progression models.