LB-035 - TOWARDS MORE ROBUST EVALUATION OF THE PREDICTIVE PERFORMANCE OF PHYSIOLOGICALLY BASED PHARMACOKINETIC MODELS: USING CONFIDENCE INTERVALS TO SUPPORT USE OF MODEL-INFORMED DOSING IN CLINICAL CARE

M. van Borselen¹, L. Sluijterman², R. Greupink³, S. de wildt^3,4; ¹Division of Pharmacology and Toxicology, Department of Pharmacy, Radboud university medical center, Nijmegen, The Netherlands, ²Department of Mathematics, Radboud University, Nijmegen, The Netherlands, ³Division of Pharmacology and Toxicology, Department of Pharmacy, Radboud university medical center -, Nijmegen, The Netherlands, ⁴Department of Neonatal and Pediatric Intensive Care, Division of Pediatric Intensive care, Erasmus MC-Sophia Children's Hospital, Rotterdam, The Netherlands.

Background: With the rise in use of physiologically based pharmacokinetic (PBPK) modeling over the past decade, the use of PBPK modeling to underpin drug dosing for off-label use in clinical care has become an attractive option. In order to use PBPK models for high-impact decisions, thorough qualification and validation of the model is essential to gain enough confidence in model performance. Currently, there is no agreed method for model acceptance while clinicians demand a clear measure of model performance before considering implementing PBPK model-informed dosing. We aim to bridge this gap and propose the use of a confidence interval for the predicted-to-observed geometric mean ratio with predefined boundaries. This approach is similar to currently accepted bioequivalence testing procedures and can aid in improved model credibility and acceptance.

Methods: Two different methods to construct a confidence interval are outlined, depending on whether individual observations or aggregate data are available from the clinical comparator data sets. The 2 testing procedures are demonstrated for an example evaluation of a midazolam PBPK model. In addition, a simulation study is performed to demonstrate the difference between the 2-fold criterion and our proposed method.

Results: Using midazolam adult pharmacokinetic data, we demonstrated that creating a confidence interval yields more robust evaluation of the model than a point estimate, such as the commonly used 2-fold acceptance criterion. In comparison, the 2-fold metric has a false-positive rate of 37% compared to 3.1% for our approach in difficult cases. Additionally, we showed that the use of individual predictions can reduce the number of required test subjects. Furthermore, an easy-to-implement software tool was developed and is provided to make our proposed method more accessible.

Conclusion: With this method, we aim to provide a tool to further increase confidence in PBPK model performance and facilitate its use for directly informing drug dosing in clinical care.