Mayu Osawa, PhD: No financial relationships to disclose
Background/Objectives: Overall survival (OS) is the gold standard for assessing the benefit of new drugs and treatments in interventional cancer clinical trials. However, the maturation of OS data often requires a protracted timeline. Tumor growth dynamic modeling has been widely used to predict late-stage outcomes and support decision-making in solid tumors [1]. In hematology, recent research suggests that M-protein may serve as a surrogate biomarker linked to progression-free survival (PFS) in patients with relapsed and refractory multiple myeloma (RRMM) [2]. We developed a model platform that uses longitudinal M-protein data to predict OS in RRMM across various drug modalities and treatment regimens, aiming to support decision-making.
Methods: Given the reported positive correlation between PFS and OS in clinical trials of RRMM [3], we leveraged the previously established model that links longitudinal M-protein data to PFS [4], using it as a starting point for the OS model. The covariate effects were reassessed by re-estimating the parameter estimates. A total of 2449 RRMM subjects from 6 Phase II/III studies, involving 12 treatment combinations, were included in model development. Model performance was assessed using visual predictive checks with 500 simulated datasets. The developed model was then externally validated using data from a Phase 3 study (n=682) not included in the model development. As part of the external validation, we assessed model performance with both full and truncated M-protein data (at 1, 3, and 6 months) to evaluate the predictive power of early M-protein data. In addition, hazard ratios (HR) were simulated based on 500 replicates for each study and compared with the observed HR.
Results: The covariate effects, except for treatment group, were consistent with those from the previously established model. Due to the lack of significance, the treatment effect was removed from the final OS model. The final OS model showed good predictive performance across all drug modalities and treatment regimens. The predicted OS in the external validation study was in agreement with the observed OS, even when using early M-protein data (at least 3 month). Additionally, the observed HRs fell within the 95% prediction intervals for each study, further supporting the robustness of the model.
Conclusions: The model platform described in this study demonstrates strong predictive performance across different drug modalities and treatment regimens. This platform can be used to support decision-making and provide data augmentation for future clinical trials, as well as the development of new drugs and dosing regimens in RRMM, by simulating OS outcome.
Citations: [1] Bruno R et al. Clin Cancer Res. 2020;26(8);1787-1795 [2] Cheng Y et al. eJHaem. 2022; 3(3);815-827 [3] Dimopoulos et al. BMC Cancer. 2024, 24(1):541 [4] Osawa et al. ACoP 2024.10.70534/NBPJ3182
