We develop, parameterize, and validate a model for tumor growth inhibition using in vivo mouse data and study the effects of modeling uncertainty and inter-individual variability on drug candidate efficacy predictions.
Modeling population pharmacokinetics/pharmacodynamics (PK/PD) is an essential step in the design of any therapeutic. However, selecting and parameterizing a model, for the purposes of understanding the fundamental biology and guiding clinical studies, given experimental results from a small sample of individuals, remains challenging. For certain combinations of drug, dosage, tumor model, and animal, the model is not fully identifiable, with widely varying parameter values consistent with all the data. Further, given a single drug, different individuals will exhibit variability in distribution, clearance, efficacy, and side effects. We refine and parameterize a PK/PD model from in vivo mouse data and a panel of proposed therapeutics and examine the distributions of parameters consistent with the data and prior knowledge of the system. We explore optimal experimental design strategies for selecting the appropriate model structure and maximizing model identifiability with respect to clinically relevant metrics of interest. A greedy algorithm consisting of sequentially incorporating existing knowledge, predicting experiment outcomes, and performing and evaluating the designed experiments is developed and analyzed. In collaboration with the Pharmacokinetics, Dynamics & Metabolism group, Pfizer.