Mathilde Marchand1, Federico Mattiello2, Mark Yan3, Monica Susilo4, Divya Samineni4, Dale Miles4, Chunze Li4, Jin Y. Jin4, Pascal Chanu4
1Certara Drug Development Solutions, 2Biostatistics, Roche, 3Biostatistics, Roche, 4Clinical Pharmacology, Genentech, 5Clinical Pharmacology, Genentech-Roche
Introduction: Model-derived tumor dynamic metrics have shown promise in predicting overall survival (OS) [1,2,3] to support late-stage drug development (DD). However, there is a crucial need to support early decision-making, which generally relies on the objective response rate and progression-free survival. A modeling framework was recently developed to assess the operating characteristics (OCs) of tumor growth inhibition (TGI) metrics to support early decision-making in non-small cell lung cancer (NSCLC) [4]. This work demonstrated that effect size, expressed as geometric mean ratios (GMRs) comparing experimental vs. control arms, can inform early decisions on randomized Phase Ib/II trials [4] to support Phase III un-gating. The objective of this present work was to evaluate the OCs of model-derived M-protein metrics in multiple myeloma (MM) where M-protein is one of the criteria to assess clinical response according to the International Myeloma Working Group Uniform Response. Hence APOLLO, a positive randomized Phase III study comparing daratumumab, pomalidomide and dexamethasone (D-Pd) vs pomalidomide and dexamethasone (Pd) in patients with Relapsed/Refractory Multiple Myeloma (RRMM) was used [5]. Methods: The YODA project [6] allowed access to APOLLO and CASTOR clinical data (NCT02136134, NCT03180736). Using data from the APOLLO trial, we applied the resampling method described in [1] to generate 1000 datasets of 30 patients per arm with a reduced follow-up of 24 weeks to assess the OCs of the M-protein dynamic metrics to quantify the effect size of D-Pd vs Pd. To reproduce a real DD scenario i.e. absence of control in the Phase Ib trial, the experimental group was compared to a large historical control cohort with complete follow-up rather than a small with short follow-up control arm as initially presented [1]. Under the alternative hypothesis, data were bootstrapped from D-Pd and Pd to simulate a positive study. Under the null hypothesis, data were resampled twice from the Pd control arm to mimic a negative study. The model [7] was first fit to the full study dataset using the nlmixr2 package in R [8], then applied to each of the resampled datasets. Patient-level parameters (growth, shrinkage rates) and metrics (M-protein ratio to baseline at 6 to 24 weeks, time-to-growth) were estimated and summarized per dataset and treatment arm. The effect size was expressed as the GMR of experimental (D-Pd)/control (Pd) for all metrics. The probability of exceeding pre-defined (e.g. GMR < 0.8) effect size thresholds were computed for both the alternative and null hypotheses across replicates. These probabilities were visualized using receiver operating characteristic curves, illustrating the correct go rate (D-Pd vs. Pd) against the incorrect go rate (Pd vs. Pd). To evaluate the applicability of this approach in the DD setting, the selected metric and gating criteria were tested on CASTOR [9], comparing daratumumab bortezomib, and dexamethasone (D-Vd) vs bortezomib and dexamethasone (Vd) in patients previously treated MM. Results: Using the full APOLLO dataset, the model parameters were estimated with high accuracy. When applied to the 1000 subsampled datasets, the M-protein ratio to baseline at weeks 6, 8, and 12 demonstrated strong OCs, a GMR < 0.75 gave a correct go rate above 80% and an incorrect go rate < 20%. For the CASTOR trial, the same resampling strategy was applied i.e. 1000 subsamples of small sample size/short follow-up for the experimental arm (D-Vd) vs. full sample size/complete follow-up for the control arm (Vd) to mimic a real early DD decision to potentially un-gate a Phase 3 trial. The model parameters were estimated for both experimental and control arms, and among the 1000 resampled datasets, the proportion meeting the pre-specified condition (GMR < 0.75) was summarized. The GMR for the M-protein ratio to baseline at week 12 was < 0.75 in 88.7% of cases, showing the predictive ability of the selected metric and gating criteria to support a go decision. Conclusion: This analysis suggests that model-derived M-protein dynamic metrics can support early decision-making in MM by assessing drug effects using limited data compared to a large historical control arm. The M-protein ratio to baseline at weeks 6, 8, or 12 emerged as promising metrics for evaluating treatment effects in early DD in MM.
[1] Claret L et al. Model-based prediction of phase III overall survival in colorectal cancer based on phase II tumor dynamics. J Clin Oncol 27:4103-4108 (2009). [2] Claret L et al. A model of overall survival predicts treatment outcomes with atezolizumab versus chemotherapy in non-small cell lung cancer based on early tumor kinetics. Clin Cancer Res (2018) 24, 3292-8. [3] Chanu, P et al. A disease model for Multiple Myeloma developed using Real World Data and validated on Phase 3 clinical trials. PAGE 32 (2024) Abstr 11094 [www.page-meeting.org/?abstract=11094] [4] Bruno, R. et al. Tumor dynamic model-based decision support for Phase Ib/II combination studies: A retrospective assessment based on resampling of the Phase III study IMpower150. Clin Cancer Res 29, 1047-1055 (2022). [5] Dimopoulos et al; Apollo: Phase 3 Randomized Study of Subcutaneous Daratumumab Plus Pomalidomide and Dexamethasone (D-Pd) Versus Pomalidomide and Dexamethasone (Pd) Alone in Patients (Pts) with Relapsed/Refractory Multiple Myeloma (RRMM). Blood 2020; 136 (Supplement 1): 5–6 [6] Krumholz, H. M. et al. A historic moment for open science: the Yale University Open Data Access Project and Medtronic. Ann. Intern. Med. 158, 910–911 (2013). “This study, carried out under YODA Project 2021-4660, used data obtained from the Yale University Open Data Access Project, which has an agreement with JANSSEN RESEARCH & DEVELOPMENT, L.L.C.. The interpretation and reporting of research using this data are solely the responsibility of the authors and does not necessarily represent the official views of the Yale University Open Data Access Project or JANSSEN RESEARCH & DEVELOPMENT, L.L.C..” [7] Stein, W. D. et al. Tumor Regression and Growth Rates Determined in Five Intramural NCI Prostate Cancer Trials: The Growth Rate Constant as an Indicator of Therapeutic Efficacy. Clin Cancer Res 17, 907–917 (2011). [8] Fidler M, et al (2019). Nonlinear Mixed-Effects Model Development and Simulation Using nlmixr and Related R Open-Source Packages. CPT: Pharmacometrics & Systems Pharmacology, 8(9), 621–633. [9] Palumbo A, et al. Daratumumab, Bortezomib, and Dexamethasone for Multiple Myeloma. N Engl J Med. 2016 Aug 25;375(8):754-66.
Reference: PAGE 33 (2025) Abstr 11562 [www.page-meeting.org/?abstract=11562]
Poster: Drug/Disease Modelling - Oncology