PI-098 - PBPK MODEL REPRODUCIBILITY: LEARNINGS FROM THE DEVELOPMENT OF A PBPK MODEL REPOSITORY FOR GLOBAL HEALTH

M. Bergagnini-Kolev¹, j. Ning², H. Jones³, L. Almond⁴; ¹Simcyp/Certara, WA, USA, ²Simcyp/Certara, N/A, China, ³Certara, MA, USA, ⁴Certara UK, N/A, UK.

Background: A repository of freely available PBPK models was developed to aid optimization of dosing regimens and decision-making across drug development in global health. The aim of this work was to use published PBPK models in the HIV therapeutic area to assess reproducibility and common barriers that limit it.

Methods: The availability of published PBPK models for 19 prioritised HIV drugs was assessed. Models were recreated in Simcyp and checked for reproducibility (within 1.5-fold of published simulated results). In cases of discrepancies, the software platform, version and omissions in published information were assessed for common themes. Opportunities for model updates and extension were also investigated.

Results: Of the 19 HIV drugs, 14 had at least 1 published model and there was a total of 31 PBPK models identified. Of those models, 20 used commercial platforms (Simcyp, PK-Sim) and 11 were bespoke PBPK models coded in other software (e.g. Matlab, RStudio). None of the bespoke models could be recreated directly in Simcyp, likely due to significant differences in model structure. Of the models developed in commercial platforms, 16 could be reproduced in Simcyp V18 within 1.5-fold of the published model results. Of the reproducible models, 3 were refined to improve recovery of clinical observations and 2 to extend the scope of model application. Models were developed for drugs that had no reproducible model. In cases where multiple models were published for the same drug, verification datasets were rarely consistent. Barriers to reproducibility included different model platforms and insufficient description of input or verification data.

Conclusion: Although it was possible to reproduce many of the models using tables of input parameters, it was still challenging and time consuming to find the source of discrepancies and to make sure the model was effectively reproduced. Following our analysis we would recommend that published model files are shared, either in a repository or as part of the publication. This would increase reproducibility, efficiency and modeler confidence. Clear description of model performance, purpose and assumptions are also required to facilitate the effective use of ‘off-the-shelf’ models to reduce duplication of effort and support wider application of PBPK to impact decision making in drug development and patient care.