IMI - Oral biopharmaceutics tools project - Evaluation of bottom-up PBPK prediction success part 2: An introduction to the simulation exercise and overview of results.

Orally administered drugs are subject to a number of barriers impacting bioavailability (F), causing challenges during drug and formulation development. Physiologically-based pharmacokinetic (PBPK) modelling can help during drug and formulation development by providing quantitative predictions through a systems approach. The performance of three available PBPK software packages (GI-Sim, Simcyp®, and GastroPlus™) were evaluated by comparing simulated and observed pharmacokinetic (PK) parameters. Since the availability of input parameters was heterogeneous and highly variable, caution is required when interpreting the results of this exercise. Additionally, this prospective simulation exercise may not be representative of prospective modelling in industry, as API information was limited to sparse details. 43 active pharmaceutical ingredients (APIs) from the OrBiTo database were selected for the exercise. Over 4000 simulation output files were generated, representing over 2550 study arm-institution-software combinations and approximately 600 human clinical study arms simulated with overlap. 84% of the simulated study arms represented administration of immediate release formulations, 11% prolonged or delayed release, and 5% intravenous (i.v.). Higher percentages of i.v. predicted area under the curve (AUC) were within two-fold of observed (52.9%) compared to per oral (p.o.) (37.2%), however, F and relative AUC (F) between p.o. formulations and solutions were generally well predicted (64.7% and 75.0%). Predictive performance declined progressing from i.v. to solution and immediate release tablet, indicating the compounding error with each layer of complexity. Overall performance was comparable to previous large-scale evaluations. A general overprediction of AUC was observed with average fold error (AFE) of 1.56 over all simulations. AFE ranged from 0.0361 to 64.0 across the 43 APIs, with 25 showing overpredictions. Discrepancies between software packages were observed for a few APIs, the largest being 606, 171, and 81.7-fold differences in AFE between SimCYP and GI-Sim, however average performance was relatively consistent across the three software platforms.