Application of Model-Informed Drug Development for Dose Optimization of Brigimadlin in Patients with Dedifferentiated Liposarcoma
Jurij Aguiar Zdovc1, Kamunkhwala Gausi1, Jurgen Langenhorst1, Anyue Yin1, Chayan Acharya1, Celine Sarr1, Lena E. Friberg1, Mehdi Lahmar2, Rolf Grempler3, Alejandro Pérez-Pitarch3,5, Ulrike Schmid3, Patricia LoRusso4, Reinhard Sailer3, David Busse3
1Pharmetheus AB, 2Boehringer Ingelheim Pharma GmbH & Co.KG, 3Boehringer Ingelheim Pharma, 4Yale University School of Medicine, 5Regeneron
Objectives: The FDA has recently initiated Project Optimus to reform the dose optimization and selection paradigm in oncology drug development, emphasizing the need for greater attention to optimal efficacy and tolerability rather than selecting the maximum tolerated dose. Model-Informed Drug Development (MIDD) can support this endeavor by combining multiple sources of data to predict efficacy and safety outcomes in the overall population and in special patient populations. In this work, pharmacometric modelling was applied to leverage phase I to III data of brigimadlin, a highly potent, oral MDM2-p53 antagonist [1]. The general aim was to support dose selection and further clinical development of brigimadlin in patients with Dedifferentiated Liposarcoma (DDLPS), as well as dose justification for the submission to health authorities. Methods: The analyses used data from three ongoing clinical studies, including a phase I open-label study in advanced solid tumor patients (NCT03449381), a phase II open-label study in predominantly biliary tract cancer patients (NCT05512377) and a sponsor-blinded seamless phase II/III registrational study in DDLPS patients (NCT05218499; the primary endpoint was not met). During early clinical development, PK/PD modeling of longitudinal, continuous dependent variables (brigimadlin plasma concentration, biomarker GDF-15, sum of longest diameters of target lesions [SLD], platelets, neutrophils) was performed based on data from the phase I study to derive categorical outcomes of interest (e.g. tumor response rate of target lesions) supporting the selection of dosing regimens for further clinical development. At this stage, dose reductions and delays were not considered for simulations. During later stage development, pharmacometric modeling was applied to support dose optimization and dose justification for submission to health authorities. Initially the same longitudinal, continuous dependent variables were leveraged as in early clinical development by pooling data from the phase I study and emerging data from the registrational study (data from all solid tumor patients for safety and the biomarker GDF-15 and data from DDLPS patients for SLD). With more mature data from the dose optimization part of the registrational trial and to support the planned submission dossier, these were gradually replaced by categorical exposure-response (ER) analyses of key clinical endpoints. For efficacy, a parametric time-to-event (TTE) model was developed for the primary endpoint progression-free-survival (PFS, pooled across DDLPS patients in all studies). Notably, with long term treatment, dose reductions appeared more frequently and were considered in the modeling strategy via a Markov model. For safety, parametric TTE models were developed for the five most relevant safety endpoints: grade 3/4 thrombocytopenia, grade 3/4 neutropenia, grade 3/4 adverse events (AE) of special interest (AESI), any AE of grade 3/4, and serious adverse events (SAE). NONMEM 7.4 [2] was used for model development and covariate-parameter relationships were evaluated using the stepwise covariate model building procedure (SCM) with adaptive scope reduction (ASR) [3,4]. Simulations of the selected endpoints were performed to compare dose levels. Results: In early clinical development, PK/PD modeling confirmed target engagement and predicted a >25% higher GDF-15 relative change from baseline at 72 h after 45 mg brigimadlin dosing compared to 30 mg. Additionally, a trend was observed for stronger early tumor shrinkage for exposure associated with 45 mg vs 30 mg q3w. These predictions contributed to selecting 30 and 45 mg q3w in the dose optimization part of the registrational study in DDLPS patients. The developed PopPK model revealed that lower body weight and higher total bilirubin were intrinsic factors associated with higher brigimadlin exposure, but these covariates were not deemed clinically relevant in conjunction with the identified ER for both safety and efficacy. This observation supported flat dosing in DDLPS patients. In later stage development, stronger predicted tumor shrinkage after one year of treatment with 45 mg vs 30 mg q3w (-19.5% vs -9.86% relative change from baseline SLD) further supported dose selection of 45 mg brigimadlin q3w in the phase III part of the seamless phase II/III registrational study in DDLPS patients. To support the dose proposed in the label to health authorities, a pooled exposure-PFS analysis was performed, indicating that in DDLPS patients, 45 mg brigimadlin q3w showed longer PFS compared to 30 mg q3w (=2.55 months, after accounting for dose reductions via the developed dose reduction model). The exposure-safety analyses indicated a higher number of events for higher exposure, with two distinct patterns of exposure-safety relationships observed. For AE types related to the class effects of MDM2-p53 antagonists, a steeper exposure-safety relationship was identified (thrombocytopenia, neutropenia, AESI), while a flatter relationship was observed for SAE and any AE. An additional, exposure-independent impact of the covariate “Asian race” (here synonymous with “Asian country”) was identified on the baseline hazard of neutropenia, amongst other safety endpoints. The overall lower body weight (and hence higher exposure) in Asian patients combined with the exposure-independent effect of Asian race resulted in up to 16% higher predicted probability (absolute difference) of experiencing a grade 3/4 neutropenia in Asian patients compared to the overall patient population. Still, the benefit-risk ratio remained positive in all assessed patient subpopulations. Conclusions: This work showcases the application of MIDD approaches to dose optimization during an oncology drug development program, from early to late stage, in the Project Optimus era. The integrated model-based analysis utilizing clinical PK, efficacy, and safety data from several brigimadlin studies supported the selection of 45 mg q3w as the optimal dose in DDLPS patients.
[1] LoRusso P et al. Cancer Discov 2023;13:1802–13 [2] Beal SL, Sheiner LB, Boeckmann AJ, Bauer RJ, editors. NONMEM 7.4 Users Guides. (1989–2019). https://nonmem.iconplc.com/nonmem744. Gaithersburg, MD: ICON plc; 2019. [3] Jonsson EN, Karlsson MO. Automated Covariate Model Building within NONMEM. Pharm Res. 1998 Sep;15(9):1463-8. [4] Jonsson, E. N., Harling, K, PAGE 27 (2018) Abstr 8429 [www.page-meeting.org/?abstract=8429]