Jen-Hao Wu1,2, Kim P J Schellekens1,2, Charlotte Rigaud3, Véronique Minard-Colin3, Reineke A Schoot2, Gertruud Bakker2, Maaike Boonstra-Schelfhorst2, Erica Brivio2, Francisco Bautista2, Michael J. Hanley4, Samantha Lampron4, Naoko Murakami4, Florin Vranceanu4, Karey Kowalski5, Brian Jermain5, Aurelia H M de Vries Schultink6, Hinke Huisman Siebinga7, Michel Zwaan1,2, Alwin D.R. Huitema2,6,7
1Department of Pediatric Oncology, Erasmus MC-Sophia Children’s Hospital, 2Princess Máxima Center for Pediatric Oncology, 3Department of Children and Adolescent Oncology Gustave Roussy Cancer Campus, 4Takeda Development Center Americas, 5Pfizer Inc, 6Department of Clinical Pharmacy, University Medical Center Utrecht, 7Department of Pharmacy & Pharmacology, Netherlands Cancer Institute
Introduction Phase I trials aim to establish the recommended phase II dose (RP2D) by assessing drug safety [1]. In pediatric oncology, defining the RP2D often involves alignment with adult pharmacokinetic (PK) targets [2,3]. While non-compartmental analysis (NCA) [4,5] is widely used in these phase I trials for its simplicity, it falls short in describing PK with sparse samples or limited patient numbers [4–6]. Its reliance on intensive sampling for accuracy and its limited extrapolation capability to other dosing regimens constrain the utilities of NCA in vulnerable populations [2,7]. Developing a standalone pediatric population PK (popPK) model is often infeasible with sparse phase I data. Yet, incorporating prior adult PK knowledge in pediatric popPK model development via the frequentist prior framework (NONMEM $PRIOR subroutine) [8,9] offers a potential alternative to characterize pediatric PK at an early trial stage. This approach exploits prior information and phase I trial data without requiring granular adult data or imposing additional burdens on trial participants. We implemented the frequentist prior population modeling approach in two completed pediatric dose-finding component of the trials that employed NCA to estimate drug exposure for RP2D selection. Given the limitations of NCA, we examined the feasibility of the modeling approach in pediatric populations to generate comprehensive PK insights and compared its performance with NCA in estimating exposure. Methods PK data were obtained from two pediatric dose-finding trials: ITCC-098 (brigatinib, tablets) in ALK+ malignancies and ITCC-054 (bosutinib, tablets/capsules) in Philadelphia chromosome+ chronic myeloid leukemia. In the brigatinib trial, two weight-bin-based dose levels (DLs) were tested. Patients received a lower daily dose for a 7-day lead-in phase then the full dose for the following 28-day cycles. Three steady-state samples were collected on day 22 for patients <18 kg (pre-dose, 2, 4h post-dose) and six for patients =18 kg (pre-dose, 0.5, 1, 2, 4, 6h post-dose). In the bosutinib trial, three DLs were tested in 28-day cycles. Steady-state samples on day 14 (pre-dose, 1, 3, 6, 8, 24h post-dose) were collected. The RP2D was determined based on safety assessments and NCA-estimated exposure, to identify the DL that achieved an exposure level (AUCt,ss; ±20%) comparable to that observed in adults receiving the approved dosing regimen. Pediatric popPK model development followed a structured workflow. First, fixed allometric scaling was implemented in an adult reference model. The NONMEM $PRIOR NWPRI subroutine was then used to specify hyperparameters resembling prior PK knowledge. Lastly, additional variability was tested. Uninformative priors on fixed effects (THETAs, with large variance, e.g., 107) and between-subject variability (OMEGAs, with degree of freedom (DF) = OMEGA-block dimensions + 1) were first applied to ensure estimation was driven by pediatric data. Afterward, on parameters with estimation difficulties, informative priors were applied (THETAs prior with variance/covariance matrix from the reference model and OMEGAs prior with calculated DF [9]), with no priors applied to residual unexplained variability. Estimation difficulties were defined as parameters displaying implausible values (=50-fold different from the reference value), unstable estimates (RSE =50%), shrinkage =40%, zero gradient or rounding errors. To compare exposure estimation performance, PK profiles (6000 patients per DL) were simulated using the two respective pediatric popPK models to obtain the true AUCt,ss. Following the sampling schedule of each protocol, 1000 simulated trials (6 patients per DL) were generated. Within each trial, AUCt,ss was estimated using both NCA and a re-developed popPK model with the same $PRIOR subroutine setting. To examine the capacity of both methods to generate evaluable exposure estimates to avoid failed PK analysis which leads to additional enrollment demand and potential trial delay, the following criteria were applied: For NCA, an evaluable exposure estimate required a patient with a trough sample and =1 sample in both absorption and elimination phases; the modeling approach required a successful model re-development. To examine the estimation accuracy, the power of each method was defined as the percentage of evaluable AUCt,ss estimates falling within ±20% of the true value. A sensitivity analysis was conducted to assess the performance of both approaches under real-world scenarios, where missingness during sample collection was introduced (started at 0%, increased by 10% per hour, and capped at 60%). PopPK analyses and simulations were performed in NONMEM 7.5. Results PK data from the brigatinib trial (10 patients, 95 samples) and the bosutinib trial (26 patients, 235 samples) were analyzed. Prior PK information was sourced from their respective adult popPK models [10,11]. In both pediatric models, THETAs and OMEGAs of central PK parameters (clearance (CL) and central volume of distribution (V1)) and parameters without estimation difficulties were estimated with uninformative priors. Informative priors were applied to the rest. The brigatinib pediatric PK model was a three-compartment model with a two-transit-compartments input and fixed allometric scaling. Between-occasion variability (BOV) was included on V1/F. The pediatric apparent CL was 14.0 L/h (RSE: 14.6%) for a 70-kg patient, 32.1% higher than the adult value of 10.6 L/h. The bosutinib pediatric PK model was a two-compartment model with first-order absorption, absorption lag time and fixed allometric scaling. BOV was included on CL. The pediatric CL was 75.6 L/h (RSE: 7.50%) for a 70-kg patient, 34.4% higher than the adult CL of 56.3 L/h. In simulated trials, the modeling approach generated 100% evaluable estimates for brigatinib (n=12,000) and bosutinib (n=18,000), whereas NCA yielded evaluable estimates in only 81.6% (9,788) and 76.5% (13,765) of cases, respectively. The power among evaluable estimates was higher with the modeling approach compared to NCA: 84.0% vs. 66.5% for brigatinib and 84.4% vs. 80.2% for bosutinib. Under missing data scenarios, the modeling approach still generated 100% evaluable estimates, while NCA performance deteriorated, yielding evaluable estimates in only 56.8% (6,811) for brigatinib and 40.9% (7,357) for bosutinib. And the modeling approach maintained superior power compared to NCA: 80.4% vs. 60.2% for brigatinib and 73.8% vs. 71.1% for bosutinib. Conclusions This study highlights the feasibility and value of the frequentist prior modeling approach as an alternative to NCA to characterize pediatric PK at an early trial stage by leveraging existing knowledge and limited dose-finding trial data in pediatric populations. This approach can also be applied to other vulnerable populations with restricted sampling opportunities. The higher pediatric CL relative to adults supports findings in both trials that a higher dose was required to achieve adult exposure [12,13]. The modeling approach outperformed NCA in exposure estimation, especially in realistic scenarios with sparse and missing data. The smaller reduction in the numbers of evaluable estimates and the higher power with missing data suggest this method could mitigate the risk of trial delay due to failed PK analysis and enable reduced sampling per patient in future trials. Moreover, its utilities extend to refining dosing regimens, for example our model-informed dose proposal for brigatinib age-appropriate liquid formulation, which has been amended in the pediatric trial protocol.
[1] Le Tourneau C, Lee JJ, Siu LL. Dose Escalation Methods in Phase I Cancer Clinical Trials. JNCI: Journal of the National Cancer Institute 2009;101:708–20. https://doi.org/10.1093/jnci/djp079. [2] European Medicines Agency. Guideline on the role of pharmacokinetics in the development of medicinal products in the paediatric population n.d. [3] Paoletti X, Geoerger B, Doz F, Baruchel A, Lokiec F, Le Tourneau C. A comparative analysis of paediatric dose-finding trials of molecularly targeted agent with adults’ trials. Eur J Cancer 2013;49:2392–402. https://doi.org/10.1016/j.ejca.2013.02.028. [4] Gabrielsson J, Weiner D. Non-compartmental Analysis, 2012, p. 377–89. https://doi.org/10.1007/978-1-62703-050-2_16. [5] Patel K, M.J. Kirkpatrick C. Pharmacokinetic Concepts Revisited – Basic and Applied. Curr Pharm Biotechnol 2011;12:1983–90. https://doi.org/10.2174/138920111798808400. [6] Straughn AB. Limitations of Noncompartmental Pharmacokinetic Analysis of Biotech Drugs. Pharmacokinetics and Pharmacodynamics of Biotech Drugs, Wiley; 2006, p. 181–8. https://doi.org/10.1002/9783527609628.ch7. [7] Batchelor HK, Marriott JF. Paediatric pharmacokinetics: key considerations. Br J Clin Pharmacol 2015;79:395–404. https://doi.org/10.1111/bcp.12267. [8] Gisleskog PO, Karlsson MO, Beal SL. Use of prior information to stabilize a population data analysis. J Pharmacokinet Pharmacodyn 2002;29:473–505. https://doi.org/10.1023/A:1022972420004. [9] Chan Kwong AH-XP, Calvier EAM, Fabre D, Gattacceca F, Khier S. Prior information for population pharmacokinetic and pharmacokinetic/pharmacodynamic analysis: overview and guidance with a focus on the NONMEM PRIOR subroutine. J Pharmacokinet Pharmacodyn 2020;47:431–46. https://doi.org/10.1007/s10928-020-09695-z. [10] Gupta N, Wang X, Offman E, Prohn M, Narasimhan N, Kerstein D, et al. Population Pharmacokinetics of Brigatinib in Healthy Volunteers and Patients With Cancer. Clin Pharmacokinet 2021;60:235–47. https://doi.org/10.1007/s40262-020-00929-4. [11] Garrett M, Knight B, Cortes JE, Deininger MW. Population modeling of bosutinib <scp>exposure-response</scp> in patients with newly diagnosed chronic phase chronic myeloid leukemia. Cancer Med 2023;12:17981–92. https://doi.org/10.1002/cam4.6439. [12] Brivio E, Pennesi E, Willemse ME, Huitema ADR, Jiang Y, van Tinteren HDR, et al. Bosutinib in Resistant and Intolerant Pediatric Patients With Chronic Phase Chronic Myeloid Leukemia: Results From the Phase I Part of Study ITCC054/COG AAML1921. Journal of Clinical Oncology 2024;42:821–31. https://doi.org/10.1200/JCO.23.00897. [13] Rigaud C, Schellekens K, Damm-Welk C, Minard-Colin V, Bakker G, Boonstra-Schelfhorst M, et al. Brigatinib monotherapy in children with R/R ALK + ALCL, IMT, or other solid tumors: Results from the BrigaPED (ITCC-098) phase 1 study. . Journal of Clinical Oncology 2024;42:10017–10017. https://doi.org/10.1200/jco.2024.42.16_suppl.10017.
Reference: PAGE 33 (2025) Abstr 11753 [www.page-meeting.org/?abstract=11753]
Poster: Drug/Disease Modelling - Oncology