Andrew Brandon1, Tadao Tamura1, Arsenii Zats2,3, Shelby Barnett1, Jugal Suthar2,3, Alaric Taylor-Roffey2,3, Gareth Veal1
1Newcastle University Centre for Cancer, Translational & Clinical Research Institute, Newcastle University, 2Vesynta, 3UCL
Introduction: Paediatric oncology patients display rapid and heterogenous organ development and growth rates. This gives rise to significant inter-patient and inter-occasion variability in patient response following administration standardised body surface area or body weight-based dosing. Furthermore, many established chemotherapeutic agents such as vincristine and doxorubicin display clear pharmacological relationships between dose, safety and efficacy. A model-informed precision dosing (MIPD) approach can assist clinicians with dose selection in such situations [1]. A significant challenge lies in the translation of population PK/PD models from popular developmental software (e.g. NONMEM, Monolix, etc) to MIPD software that are intended for use by healthcare professionals in clinical environments. This requires models to be re-written, often in R-based syntax as part of packages such as nlmixr2 [2]. This translation step can introduce performance discrepancies due to, for example, differences in ODE solvers, rounding precision, and human errors during the translation process. It is therefore crucial to check and quantify equivalence to ensure accurate model implementation. Newcastle University Centre for Cancer delivers the UK’s only clinical pharmacology service in paediatric oncology [3]. This research group is looking to trial and compare the performance of an in-house population PK model [4] when implemented on an industry standard development software (NONMEM) versus an under-development commercial MIPD software, designed for bedside use. Objectives: – Develop a verification framework to compare model performance on distinct pharmacology platforms – Validate this framework using simulated and real-world patient datasets – Assess the clinical implications of discrepancies in model outputs across platforms Methods: We compare the performance of the model [4] when deployed across platforms by considering patient covariates. We ran 57 paediatric cancer patients, including 26 neonates and infants patients through the implemented models to generate the critical pharmacokinetic parameters (Cmax and AUC24). This stress test featured both real-world unseen patients from paediatric clinical trials, in addition to simulated patients. To ensure robustness, we also evaluated the impact of varying covariate ranges and extreme patient scenarios. Results: The models tested using real-world patients within both platforms showed high equivalence. This was demonstrated by calculating both RMSE (<0.5% relative difference) and producing and comparing goodness of fit plots. The results from the simulated datasets show that there may be differences at the extremes of the covariate range, even with a “perfectly translated” model due to inherent differences in rounding and ODE solver results in some error between the two. These discrepancies, though minor, highlight the importance of rigorous validation for clinical implementation. Conclusion: A reliable and efficient framework to compare outcomes from two functions was developed. This allowed for easy comparison between model outputs when deployed on different software platforms. This offers a tool to those seeking to validate MIPD software performance before implementing a hosted model in its clinical practice. The findings underscore the need for robust verification processes to ensure patient safety and therapeutic efficacy in precision dosing applications.
[1]: Kaspers, G. J. L. (2023). Model-Informed Precision Dosing to Reduce Vincristine-Induced Peripheral Neuropathy in Pediatric Patients: A Pharmacokinetic and Pharmacodynamic Modeling and Simulation Analysis. Clinical pharmacokinetics. https://doi.org/10.1007/s40262-023-01336-1 [2]: https://nlmixr2.org/ [3]: https://www.ncl.ac.uk/cancer/our-research/late-phase-trials/pharmacology/ [4]: Barnett S et al. Vincristine dosing and therapeutic drug monitoring in neonates and infants. Eur J Cancer. 2022;164:127-136. doi: 10.1016/j.ejca.2021.09.014.
Reference: PAGE 33 (2025) Abstr 11694 [www.page-meeting.org/?abstract=11694]
Poster: Methodology - Model Evaluation