Sebastian Weber, Juan Gonzalez Maffe, Guillaume Baneyx, Lukas Andreas Widmer, Lada Markovtsova
Novartis Pharma AG
Introduction: Oncology phase I dose escalation (DE) trials enroll cohorts of 3-6 patients to increasing doses of an experimental drug. As patients’ lives are at risk one aims to quickly increase the dose while controlling safety. Dose escalation meetings (DEM) take place once a cohort completes the primary follow-up (e.g., 4 weeks). At a DEM the dose/regimen for the next cohort is selected. At early stages data is sparse and often incomplete. E.g., follow up for the latest cohort is minimal and by operational constrains, only the safety events are on file while other data is not. In this context, advising on the dose/regimen for the next cohort is challenging. Here we introduce a novel integrated population pharmacodynamic (PD) (“integrated popPD”) modeling framework which provides a solution to this challenge and is demonstrated by a re-analysis of a completed trial.
Objectives:
- Inform DEMs in due time by prediction of outcomes for future cohorts at planned doses/regimens
- Avoid overconfident predictions by accounting for uncertainty in key inputs
- Handle diverse set of endpoints (continuous, time to event, binary, count)
- Use of a scientifically plausible longitudinal exposure measure
- Allow missing data, data beyond the primary safety follow-up & literature data
Methods: We use as an example for an integrated popPD modeling framework the DE trial NCT02375958. The trial tested the experimental drug PCA062 administered as infusion to patients with pCAD+ tumors [1]. As the mechanism of action was expected to cause thrombocytopenia, a longitudinal joint PK-PD model for platelet counts was developed at the time and presented at PAGE 2019 [2]. This model was used to predict the proportion of patients with safety events over the duration of multiple cycles.
The key principle for our novel approach is to simplify a full joint PK-PD model. We propose to conceptually start with the full PK-PD model and simplify it as required for a robust population PD model. Here we evaluate to what extent the pharmacometrics (PK) model can be simplified by comparing two models: (1) A PK model focusing on steady state kinetics (1st order absorption & 1-cmt linear elimination) and (2) the respective model restricted to the dosing data only in a K-PD approach. Observed Cmin and Cmax as well as non-compartmental estimates of clearance and volume are used as exposure data in the PK model approach. The simplified PK model is coupled via an effect compartment with a turn-over model (inhibition of drug effect) describing the platelet counts over time. Platelet counts below <50×109/L are classified as grade 3+ events and lead to dose delays/adjustments. Modeling of safety events in a time-to-event framework is presented in [3]. Informative priors are used on most parameters (e.g. disease progression, see [4]), while non-identifiable parameters are fixed to plausible values (e.g. volume for dosing data only model). Model evaluations are performed at every DEM of the trial. Models are fitted in Stan [5] using the brms R package [6].
Results: Both models were informed reliably on each data cut-off and showed reasonable goodness of fit – including for the very first DEM with only 3 patients. The model performance in terms of overall likelihood of the training data preferred the dose only popPD model at earlier stages of the trial while the exposure based popPD model led to a better description of the data starting from the middle of the trial. The respective assessments on the future data are aligned with the relative model performances on the training data.
Conclusions: We have introduced a novel framework addressing the needs of decision making during Oncology phase I DE trials. These trials reassess dose/regimen at each DEM based on sparse and incomplete data. While population PK modeling “popPK” is routinely done, we do suggest to also routinely conduct population PD modeling “popPD”. The key is to use a simplified PK model as longitudinal exposure measure as basis for the popPD model. The simplified PK model must be aligned with known drug pharmacology and sufficiently simple to comply with the trial operationally to be ready in due time for each DEM. The framework avoids overconfidence by accounting for key uncertainties and enables predictions of clinical outcomes under varying dose/regimen to inform decisions at each DEM.
References: [1] Duca M, Lim DW, Subbiah V, Takahashi S, Sarantopoulos J, Varga A, D’Alessio JA, Abrams T, Sheng Q, Tan EY, Rosa MS, Gonzalez-Maffe J, Sand-Dejmek J, Fabre C, Martin M. A First-in-Human, Phase I, Multicenter, Open-Label, Dose-Escalation Study of PCA062: An Antibody-Drug Conjugate Targeting P-Cadherin, in Patients With Solid Tumors. Mol Cancer Ther. 2022 Apr 1;21(4):625-634. doi: 10.1158/1535-7163.MCT-21-0652. PMID: 35131875. [2] J Kim, J Gonzalez Maffe, E Tan, J Sand Dejmek, C Fabre, C Meille. Combining BLRM and safety PKPD models to improve decision making in a phase I dose escalation study: case study of PCA062, an antibody drug conjugate targeting P-Cadherin. PAGE 28 (2019) Abstr 9108 [www.page-meeting.org/?abstract=9108] [3] J Gonzalez Maffe, S Weber, G Baneyx, L Widmer, L Markovtsova. A Bayesian (P)K-TTE model to support early dose escalation decisions in phase 1 Oncology studies. To be submitted to PAGE 32 (2024) (OAK ID: 53364) [4] Bender BC, Quartino A, Li C, Chen S–C, Smitt M, Strasak A, Chernyukhin N, Wang B, Jin J, Girish S, Friberg LE. An integrated pharmacokinetic–pharmacodynamic (PKPD) modeling analysis of T-DM1–induced thrombocytopenia and hepatotoxicity in patients with HER2–positive metastatic breast cancer. PAGE Meeting 2016 [5] Stan Development Team. 2023. Stan Modeling Language Users Guide and Reference Manual, 2.32.2. https://mc-stan.org [6] Bürkner P. C. (2017). brms: An R Package for Bayesian Multilevel Models using Stan. Journal of Statistical Software. 80(1), 1-28. doi.org/10.18637/jss.v080.i01
Reference: PAGE 32 (2024) Abstr 10821 [www.page-meeting.org/?abstract=10821]
Poster: Methodology - New Modelling Approaches