Alieke Bos 1, Neeltje Steeghs 2,3, Alwin Huitema 1,4,5, Hinke Huisman-Siebinga 1
1 Department of Pharmacy & Pharmacology, Netherlands Cancer Institute-Antoni van Leeuwenhoek Hospital (Amsterdam, The Netherlands), 2 Department of Medical Oncology, Netherlands Cancer Institute-Antoni van Leeuwenhoek Hospital (Amsterdam, The Netherlands), 3 Department of Medical Oncology, University Medical Center Utrecht, Utrecht University (Utrecht, The Netherlands), 4 Department of Clinical Pharmacy, University Medical Center Utrecht, Utrecht University (Utrecht, The Netherlands), 5 Princess Máxima Center for Pediatric Oncology (Utrecht, The Netherlands)
Objectives: The multikinase inhibitor lenvatinib is poorly tolerated by differentiated thyroid cancer (DTC) and metastatic renal cell carcinoma (mRCC) patients in clinical practice, with 68.8% experiencing dose-limiting toxicities (DLTs) [1]. An initial exposure-toxicity relationship was explored in this population. However, a comprehensive analysis is desirable, since Ctrough levels were log-linearly extrapolated and time-to-DLT was disregarded. Therefore, we aimed to quantify the risk of any grade DLT in the first year of treatment as a function of time-varying lenvatinib exposure in real-world DTC and mRCC patients using a repeated time-to-event (RTTE) model.
Methods: Data from 35 DTC and 29 mRCC patients with 358 lenvatinib plasma concentrations were included [1]. First, a population pharmacokinetic (popPK) model was developed. Given the sparse blood sample data, a frequentist prior approach (NONMEM PRIOR subroutine) was employed to support parameter estimation. The reference popPK model consisted of three compartments and included tumor type and body weight as covariates [2,3]. Model evaluation was based on assessment of parameter precision, goodness-of-fit diagnostic plots and a visual predictive check (VPC). Individual daily Ctrough and AUC0-24h were derived from the final model based on Maximum a Posteriori Bayesian estimation. Second, a RTTE model was developed to relate DLTs to time-varying lenvatinib exposure. Different baseline hazard parametrizations (exponential, Weibull, Gompertz, log-normal and log-logistic) were explored. Additionally, tumor type was examined as covariate on the hazard function. The effect of lenvatinib Ctrough and AUC0-24h on the hazard was tested with linear, exponential and (sigmoid) Emax relationships. The final RTTE model was selected based on scientific plausibility, objective function value (OFV) for nested models, estimated parameter precision, kernel-based visual hazard comparisons (kbVHC) and Kaplan-Meier VPCs (KM-VPC) and also evaluated with a posterior predictive check (PPC) of annualized DLT rates. Finally, simulations were performed to evaluate the effect of different dose levels on the 4-week DLT rate, where >20% of patients experiencing ≥1 DLT was considered clinically relevant. Evaluated doses (once daily) were 8, 10, 14 and 18 mg for mRCC patients and 10, 14, 20 and 24 mg for DTC patients. Steady state (ss) exposure was simulated for 8,000 typical patients using the final popPK model. The 80% prediction intervals were used to simulate repeated DLTs. The 4-week DLT rate was separately simulated for 1,000 patients with Ctrough,ss of 88 ng/mL to assess the current toxicity threshold [4]. Analyses were performed in NONMEM (v7.5) using the FOCE-I and the Laplace estimation method for popPK and RTTE model development, respectively.
Results: The parameters of the final popPK model were majorly informed by the prior model, except for clearance that was estimated at 6.35 L/h (RSE 3.88%). The observed PK data was adequately captured by the final popPK model. In total, 36 patients (21 DTC and 15 mRCC) experienced 60 DLTs and 22 patients (7 DTC and 15 mRCC) discontinued treatment prematurely due to disease progression or other reasons. The Weibull function described the baseline hazard best and was included in the final RTTE model. The scale (λ) and shape (γ) parameter were estimated at 0.146/year (RSE 52.1%) and 1.01 (RSE 13.0%), respectively, with 104% (CV% with RSE 57.2%) interindividual variability for λ. Including tumor type as covariate did not significantly improve the model fit (dOFV -1.05). The effect of lenvatinib exposure on the hazard was best described by Ctrough using an exponential relationship. The exposure coefficient was estimated at 0.0482/(ng/mL) (RSE 22.9%), corresponding to a 1.62-fold higher hazard for a 10 ng/mL increase in Ctrough. Model evaluations showed that the final RTTE model could adequately capture DLT probability based on lenvatinib Ctrough, particularly for DTC patients and the first two events. Simulations showed that 21.8% of mRCC patients and 25.3% of DTC patients experienced ≥1 DLT within 4 weeks of treatment start with registered doses. The current toxicity threshold Ctrough,ss of 88 ng/mL resulted in 55.5% of simulated patients experiencing ≥1 DLT, with a median 4-week DLT rate of 2.
Conclusions: The final RTTE model adequately described the observed DLT data as a function of lenvatinib exposure. The model shows potential to be used in clinical practice to support lenvatinib dose decision making, showing an increased risk for DLTs at registered doses. The model also supports the current toxicity threshold.
References:
[1] Meertens M et al. Cancer Chemother Pharmacol. 2025 Jan 17;95(1):25.
[2] Gupta A et al. Br J Clin Pharmacol. 2016 Jun;81(6):1124-33.
[3] Majid O et al. CPT Pharmacometrics Syst Pharmacol. 2024 Jun;13(6):954-969.
[4] Nagahama M et al. Med Oncol. 2019 Mar;36(5):39.
Reference: PAGE 34 (2026) Abstr 12160 [www.page-meeting.org/?abstract=12160]
Poster: Drug/Disease Modelling - Oncology