Vincent Hurez 1, Fabian Cardozo 1, Marcal Bravo i Padros 1, Mukti Chowkwale 1, Xiaowen Guan 1, Christine Veyrat-Follet 1, Jason Chittenden 1, Min Zhu 1, Lutz Harnisch 1
1 Regeneron Pharmaceuticals Inc. (Tarrytown, USA)
Introduction:
Excessive cytokine release following administration of T-cell engagers can trigger life-threatening cytokine-release syndrome (CRS) adverse events. Severe CRS symptoms can arise at different cytokine levels in different patients, making them difficult to predict for an individual. A fit-for-purpose mechanistic model of IL-6 cytokine release was previously developed to optimize the step-up regimen of the CD3xCD20 bispecific odronextamab in B-NHL patients and minimize IL-6 release as a surrogate to reduce CRS risk [1]. The model-informed optimized step-up regimen was implemented in subsequent clinical trials, leading to a reduction in the number of severe CRS events [2-4]. This work describes how the model was further improved by creating virtual twins based on data from patients treated with the optimized step-up regimen and establishes a relationship between IL-6 simulations and CRS risk.
Objectives:
• Develop virtual twins matching pharmacokinetics (PK) and IL-6 measurements from patients treated with odronextamab using individual parameter fitting for each virtual twin
• Establish a relationship between simulated IL-6 profiles over the first 4-weeks of individual treatment in virtual twins and CRS events observed in the corresponding real patients
Methods:
The fit-for-purpose IL-6 release model was originally used to simulate various odronextamab step-up dosing regimens in a cohort of virtual twins selected among 1000 plausible virtual patients using a standard Sum of Squared Errors minimization approach [5-6]. After updating the population PK model using new patients’ data [7], a cohort of 233 virtual twins was created matching individual odronextamab concentrations and IL-6 profiles from B-NHL patients treated with the 0.7/4/20 mg odronextamab step-up dosing regimen [8]. Patient’s baseline T-cell, B-cell numbers and tumor burden were used to define virtual twins baseline values. Parameters such as odronextamab binding affinity or number of CD3 receptors per T cell were fixed based on internal data or literature values. Other parameters impacting IL-6 release such as PK parameters, T-cell activation/clearance or IL-6 production and release rates were estimated individually for each virtual twin using Monolix [9] to best fit to individual patient’s data. Odronextamab 0.7/4/20/80 or 160 mg step-up regimens were simulated in the 233 virtual twins to calculate weekly simulated IL-6 peak-concentrations (Cmax) in weeks 1, 2, 3 and 4. A univariate logistic regression analysis was used to investigate the relationship between the weekly simulated IL-6 Cmax and the patients’ weekly observed CRS events.
Results:
Estimating individual parameters for each virtual twin improved the fit to the patient’s PK and IL-6 data thus allowing more accurate predictions for the weekly IL-6 Cmax values. Except for the first week, a significant relationship could not be established between the patients’ weekly IL-6 Cmax values and the occurrence of CRS events, due to the variability in the timing and precision of IL-6 measurements in patients. When using the simulated virtual twin IL-6 Cmax and the weekly CRS events in corresponding patients, the logistic regression analysis indicated that IL-6 Cmax was a significant predictor of CRS risk during weeks 1 to 4 of the step-up. The logistic regression model allows predictions of the incidence rate of grade 1 and grade 2+ CRS in the 233 virtual twins. Model predictions were consistent with the observed grade 2+ CRS incidence reported in B-NHL patients treated with the 0.7/4/20 mg odronextamab step-up:
– Week 1: 6.3% observed vs 6.3% predicted (CI: 3.4-11.0%)
– Week 2: 7.1% observed vs 7.0% predicted (CI: 3.7-12.6%)
– Week 3: 2.7% observed vs 2.9% predicted (CI: 0.9-8.1%)
– Week 4: 1.1% observed vs 1.4% predicted (CI: 0.2-7.4%)
Conclusion:
Using a combination of improved virtual twins and new clinical CRS data, a significant relationship could be established between IL-6 Cmax predictions and CRS risk using the cytokine release mechanistic model. Previous exposure-response models used weekly logistics regression to correlate predicted IL-6:sIL-6R complex with grade 2+ CRS risk in patients treated with another CD3xCD20 TCE [10]. Establishing CRS risk models based on cytokines or other biomarkers that can be simulated in mechanistic models of TCE is critical to select optimal step-up dosing regimens minimizing the risk of CRS. Next steps consist in refining the virtual twin cohort using additional patients’ data and use the CRS risk model to simplify the odronextamab step-up dosing regimen to improve patients’ convenience while maintaining the benefit/risk profile.
References:
[1] Khaksar Toroghi M. et al. Blood (2022) 140(Supplement 1): p. 11930-11931.
[2] Bannerji R. et al. Lancet Haematol. (2022) 9: e327–399.
[3] Topp M.S. et al. Blood (2025) 145(14): p. 1498-1509.
[4] Kim T.M. et al. Ann. Oncol. (2024) 35(11): p. 1039-1047.
[5] Hurez V. et al. ACOP2024 (2024) RC4 session – oral presentation (https://acop2024.eventscribe.net/fsPopup.asp?PresentationID=1464386&mode=presInfo)
[6] Welf E.S. et al. CPT:PSP (2026) manuscript submitted.
[7] Bravo Padros M. et al. CPT Pharmacometrics Syst Pharmacol. (2026) 15(1): p. e70162.
[8] European Medical Agency, Ordspono I Summary of product characteristics. (2025) [https://www.ema.europa.eu/en/documents/product-information/ordspono-epar-product-information_en.pdf]
[9] Monolix 2024R1, Simulations Plus, doi: 10.5281/zenodo.11401936
[10] Bender B. et al. ACOP 2024. Poster T-011. DOI: 10.70534/ZNTH8316
Reference: PAGE 34 (2026) Abstr 12066 [www.page-meeting.org/?abstract=12066]
Poster: Drug/Disease Modelling - Oncology