Model-Based Meta-Analysis in Newly Diagnosed and Relapsed Refractory Multiple Myeloma to Support Outcome-Efficacy Translations and Extrapolations Across Lines of Therapy
John D. Clements (1), Jon Collins (2), Han Witjes (3), Sandra A. G. Visser (1), Geraldine Ferron-Brady (1)
(1) GSK, USA, (2) ViiV Healthcare, USA, (3) Certara, The Netherlands
Introduction: Drug development for multiple myeloma (MM) is complex for multiple reasons: 1) long durations to demonstrate clinical benefit (overall survival or median progression free survival [mPFS]), 2) differences in treatment benefit in patients with early-stage disease vs patients refractory to treatment, and 3) regional differences in standard of care. In addition, decisions are often made based on early efficacy signals, such as overall response rate (ORR) in patients with more advanced disease, without the benefit of direct comparison against standard of care.
MM has various first line treatment regimens, which may fail and be followed by subsequent lines of therapy (LOT) [1,2]. We describe the development of a model-based meta-analysis (MBMA) framework for MM that provides outcome-efficacy translations and extrapolations across LOT. This supports the establishment of go/no go criteria for decision-making based on early efficacy data, and provides insights into comparative efficacy across LOT. The relationship between ORR and mPFS in relapsed-refractory multiple myeloma (RRMM) was previously explored [3]. Here we expanded the trial inclusion criteria to include newly diagnosed multiple myeloma (NDMM) and develop the relationship between ORR and median prior LOT.
Methods: The analysis used RRMM and NDMM CODEx databases created by Certara from publicly available clinical trial data [4]. The databases contain summary-level trial data with endpoints separated by treatment arm. Single- and parallel-design studies starting in 2010 or later with mPFS and ORR were included in the analysis.
Relationships were modeled for 1) early (ORR) and late (mPFS) efficacy endpoints, and 2) early efficacy endpoint (ORR) and median number of prior LOT. The programming language R [R Statistical Software; R Core Team 2021] was used for MBMA of logit transformed ORR predicting mPFS and subsequently median prior LOT predicting logit transformed ORR. Potential predictors of response included whether patients were newly diagnosed vs relapsed/refractory, presence of a specific drug in the treatment arm (i.e. bortezomib, lenalidomide, pomalidomide, anti-CD38 antibodies, carfilzomib, dexamethasone), median prior LOT, or number of combined therapies within the treatment arm. These relationships were used to predict ORR distributions for first line (1L; as with NDMM) Phase 3 mPFS targets and combinations (step 1) and translated from this 1L patient population to a later line patient population (e.g., 2.5 or 5 median prior treatments) (step 2) to generate thresholds that can be used for decision-making based on emerging data across LOT, in early to late line patients.
Results: The analysis dataset contained 19 studies including 3377 participants and 28 treatment arms in NDMM (excluding transplant eligible), and 83 studies with 8271 participants and 112 treatment arms in RRMM. For the relationship between ORR and mPFS, significant predictors of mPFS included whether patients were newly diagnosed vs relapsed/refractory, number of concurrent assigned treatments, treatment with lenalidomide, and treatment with anti-CD38 antibodies. Similarly, for the relationship between median number of prior LOT and ORR, significant predictors of ORR included number of concurrent assigned treatments and treatment with lenalidomide or anti-CD38 antibodies.
The decision-making framework is illustrated by the following scenario. An efficacy target of 52-months mPFS was set in NDMM for a triplet combination containing lenalidomide, an anti-CD38 antibody, and a novel agent. The MBMA predicted an ORR of 92.2% (95% confidence interval [CI]: 90.0– 94.0%) would need to be observed to reach this mPFS target. In a RRMM patient population with 2.5 median prior treatments, a predicted ORR of 79.3% (95% CI: 74.5–83.6%) would be needed to match this target. If the median number of prior treatments was 5, then the predicted ORR would be 55.5% (95% CI: 48.8– 62.4%), noting that the CIs for 2.5 and 5 median prior treatments are non-overlapping. These ORR target thresholds can be used to assess whether emerging data in later LOT suggest long-term benefit in earlier LOT.
Conclusion: A quantitative MBMA framework for NDMM and RRMM was developed to translate Phase 3 mPFS target to an ORR target in the same setting and in later line settings. These models can be used to inform go/no go thresholds for decision-making on pursuing the initiation of Phase 3 study.
Funding: GSK
References:
[1] Dimopoulos, M.A. Clinical Lymphoma, Myeloma and Leukemia (2022) 22(7). 460-473.
[2] Soekojo, C.Y. and W.J. Chng. Eur J Haematol (2022) 109(5), 425-440.
[3] Teng, Z., et al. CTP:PSP (2018) 11(2), 218-225.
[4] https://codex.certara.com