Identification of treatment effect modifiers and prognostic factors in newly diagnosed and relapsed or refractory multiple myeloma

Complex treatment landscapes in MM pose practical challenges in directly comparing the effectiveness of novel regimens versus all relevant standard of care regimens using head-to-head clinical trials. In the absence of head-to-head data, ITC is a way to compare the relative effectiveness of alternative treatments that have been studied in separate clinical trials. Prior to conducting ITCs, patient characteristics that may affect outcomes or influence the magnitude of the treatment effect should be identified and adjusted for via statistical methods to minimize bias. This work aimed to identify potential TEMs on a statistical basis using summary-level data from published RCTs in MM. A literature review of previously published PAICs in MM was also conducted to validate TEM findings and explore candidate PFs.

Age, sex, ISS/r-ISS stage and cytogenetic risk were consistently identified via RCTs as likely TEMs for PFS and were validated by the augmented review of PAICs in MM across all MM populations evaluated herein. Cytogenetic risk was also identified as a likely TEM for OS via RCTs in NDMM TIE and RRMM 1PL. Additional covariates of age, ISS/r-ISS stage and geographical region were identified for RRMM 1PL. Age, ISS/r-ISS stage and cytogenetic risk were validated as TEMs of OS by the augmented review of PAICs.

TEMs may vary depending on treatments investigated, MM population or outcome for assessment [6,7]. Patients with RRMM are known to have worse prognosis compared with those with NDMM due to disease progression, and similarly between patients with NDMM TIE versus NDMM TE due to comorbidities and other risk factors [6,7]. Hence, the current approach delineated each MM population as per NCCN and ESMO definition and then assessed TEMs within each subpopulation independently to minimize bias. Additionally, current therapies for MM consist of regimens composed of drugs with different mechanisms of action [1,5–7]. Given the treatment-specific nature of effect modification, certain covariates may only be consistently identified as a TEM within trials investigating a specific drug. For instance, age, sex and ISS/r-ISS stage were consistently identified as likely TEMs of PFS across RCTs in NDMM TE patients investigating the addition of an PI with a corticosteroid or chemotherapy to the control regimen [30,31], whereas results of the TEM assessment were mixed for the covariates of age and ISS/r-ISS stage across RCTs evaluating the addition of an anti-CD38 mAb to the control regimen [32–35]. Additionally, effect modification could be of varying magnitudes within a specific class of drugs. Among RCTs investigating regimens consisting of the same treatment classes but different drugs across the control and treatment arms, effect modification was not consistently identified for the same covariate across all RCTs in NDMM TIE [36,37]. Potential interaction effects may also explain this discrepancy, considering patients with NDMM TIE have comorbidities that render them ineligible for SCT [1] and no subgroup analyses by comorbidities were available for further evaluation. Moreover, covariates could have varying additive or multiplicative effect modification [38]. For continuous variables such as laboratory values, effect modification may have only been detected from subgroup analyses with specific numerical cut-offs.

In general, the literature review of PAICs validated the TEMs identified from the literature review of RCTs. The majority of PAICs across all MM populations adjusted for covariates that were identified as likely TEMs of PFS or OS from RCTs. Few discrepancies between covariates identified in PAICs and those identified as potential TEMs from subgroup analyses in RCTs were noted, apart from covariates related to baseline laboratory values (e.g., hepatic function and LDH). Overall, published PAICs identified in this review adjusted for a broader pool of covariates than those identified from RCTs. This could be due to the incorporation of additional covariates specific to the treatment under investigation and deemed relevant to clinical practice considering covariate selection for PAICs include statistical ranking to identify PFs based on unpublished IPD and consultations with medical experts, respectively. Nevertheless, there was a significant overlap between the covariates identified in the PAICs list and the RCTs lists, primarily with covariates related to patient and disease characteristics. Additionally, certain covariates adjusted for in published PAICs may potentially be PFs, which were largely indistinguishable from TEMs in our assessments due to either the lack of differentiation in reporting, or the lack of data to assess whether a covariate was a PF or TEM from the PAICs. Exceptions to this include the covariates of age, sex, ISS/r-ISS, ECOG PS, cytogenetic risk, and creatinine clearance, which could potentially be considered PFs since they were adjusted for in both anchored and unanchored analyses for NDMM TIE. Even though only two anchored analyses were identified, evidence supporting these covariates as PFs are established in literature [39–45]. Despite some differences between covariates adjusted for in PAICs and those identified from subgroup analyses in RCTs, the considerable overlap across survival outcomes and different MM populations provide evidence that the covariates adjusted for in published PAICs identified herein could serve as an initial set of potential covariates for adjustment or subgroup analyses in future PAICS or RCTs, respectively.

While there was overlap in covariates identified for OS versus those identified for PFS, there were also a few discrepancies. The inconsistency across outcomes may primarily be driven by the higher frequency of subgroup analyses reported in RCTs for PFS than for OS. This aligns with findings from another study investigating effect modification, but solely for the covariates of refractory status and number of LOTs in RRMM [46]. A larger evidence base for PFS is also expected as PFS is an accepted surrogate end point of OS for accelerated regulatory approval of certain targeted therapies and combination regimens that demonstrate improved survival outcomes [47]. Although fewer TEMs were identified for OS, it does not preclude other variables as potential candidates for TEM.

Overall, this study had several strengths. Various data sources were leveraged to inform the evidence base, including both RCT and PAIC study designs which can more reliably capture TEMs and PFs. Searches of published literature included historical and contemporary data (i.e., since the years 1996 for RCTs and 2013 for PAICs) to ensure the identification of potential TEMs that were more recently identified as an important risk factor (i.e., EMD/plasmacytomas) in addition to those commonly reported in RCTs evaluating established therapies. Analysis approaches also consider potential heterogeneity of TEMs within MM. Considering TEMs can vary by MM population, including a mixed population for TEM assessment may introduce bias. By delineating each MM population per NCCN and ESMO treatment guideline definitions, and assessing TEMs within each subpopulation independently, the current approach minimizes uncertainty, allowing the results to be valuable for diverse stakeholders involved in healthcare decision-making. Assessments of TEM were conducted with established statistical methods [26,27], and validated iteratively by a panel of statistical experts. Additionally, clinical expert opinion reported in published PAICs and HTAs were included to inform which covariates might be of greater clinical relevance.

To our knowledge, there is no other study that reviews both TEMs and PFs across four MM populations and determine TEMs through a statistical approach. However, three publications in RRMM populations have reviewed potential TEMs or PFs [20,21], or evaluated effect modification in selected covariates [46] at the time when this manuscript was written. Towle et al. was a conference poster focused on identifying both potential TEMs or PFs in unanchored ITCs of TCE RRMM [21]. Kumar et al. aimed to review potential PFs broadly for both response and survival outcomes reported in either observational studies or clinical trials of RRMM [20]. Although methods vary between our study and the two aforementioned studies, the covariates identified were largely aligned. Compared with Kumar et al. [20], our study also captured baseline covariates related to demographics (age and sex), disease characteristics (cytogenetic risk, ECOG PS, ISS/r-ISS stage, EMD/plasmacytomas, time from diagnosis), prior treatment history (prior therapy with IMiD, prior therapy with SCT and prior LOTs) and biochemical laboratory values (albumin, LDH and β2-microglobulin) as potential PFs of OS and PFS in RRMM 1PL. Covariates captured in Towle et al. were consistent with Kumar et al. and additionally included refractory status, time to progression on last LOT, hemoglobin, creatinine clearance, average duration of prior therapies and race [21]. Our study identified all covariates noted in Towle et al. [21]. Rose et al. evaluated effect modification of refractory status and prior LOTs in RRMM using a random-effects meta-analysis approach [46]. Effect modification by these two covariates were considered weak and thus unlikely to bias NMA estimates [46]. This contradicts with our findings, in which number of prior LOTs and refractory status to IMiDs were considered likely TEMs. However, conclusions from Rose et al. were based on broad covariate definitions, which does not consider the treatment-specific nature of TEMs. In our analysis, refractory status was only considered a likely TEM in the context of IMiDs. Additionally, Rose et al. did not acknowledge how varying magnitudes of effect modification on the additive or multiplicative scale could influence detection of TEMs.

Despite these strengths, this study also had limitations. TEM assessments were limited to known covariates that have been considered in published clinical trials and thus may not have necessarily identified unknown covariates that are potential TEMs. Published data are also strictly summary-level, which prevented further evaluation of the magnitude of effect modification on either the additive or multiplicative scale. Additionally, although covariates selected for adjustment were reported, whether these covariates were considered TEMs or PFs was not disclosed in included PAICs of MM populations except one [29]. Concluding whether covariates were TEMs or PFs based on methodology from NICE TSDs [11,14,28] were not feasible considering anchored PAICs are rarely conducted in MM due to lack of a common comparator regimen across RCTs. Furthermore, covariates identified as TEMs based on historical data may not necessarily be considered TEMs in future PAICs given the treatment-specific nature of TEMs. Thus, while this analysis could potentially inform preliminary covariate selection prior to the conduct of future PAICs, de novo analysis of newly collected data should be performed alongside consultations with subject-matter experts to ensure all relevant covariates are adjusted.

Several opportunities for future research are possible. Availability of IPD or real-world evidence can permit investigations into the prognostic value of candidate PFs and each TEM’s magnitude of effect modification, either on additive scale or on a multiplicative scale. Additionally, emerging understanding of covariates associated with poorer prognosis, such as EMD/plasmacytomas [48], and the incorporation of additional factors to assign risk to MM patients [49,50], may also lead to increased reporting of subgroup analyses in pivotal publications for covariates that were either sparsely identified or lacking entirely in the current study. As the treatment landscape in MM is constantly evolving, increasing evidence generation from both RCTs and PAICs are expected, providing available data and thus opportunities to investigate TEM's treatment-specific nature and outcome associations.

Conclusions from this work can be leveraged by various stakeholders who rely on accurate estimates of comparative effectiveness to provide guidance on MM treatment guideline updates. In line with best practices published by NICE and IQWiG, a priori identification and adjustment of TEMs and PFs strengthen the validity of results from comparative estimates in PAICs and subgroup or stratification analyses in RCTs [51,52]. The current work defines an initial set of potential TEMs and PFs that can be used to establish a standard prespecified list, which can then be shared with a panel of clinical experts in MM to validate, supplement or refine prior to a formal statistical analysis. Following these steps, data analyses using IPD can further validate potential or identify new TEMs and PFs, assess the degree of collinearity (or redundancy) among TEMs and PFs and assess population differences in TEMs and PFs across the evidence. Findings from this research can contribute to increased transparency and standardization in the selection process to support comparability of different PAICs and future clinical trial design in terms of preplanned subgroups analyses as well as stratification analyses.

The use of a historical data-informed approach allowed the identification of a list of potential TEMs and PFs to use in conjunction with data analysis and clinical validation during the development of future studies in MM populations of interest. Given the emergence of new diagnostic tools and treatment classes for MM, future efforts should focus on ensuring selected covariates remain accurate with updated clinical opinion prior to conducting analyses. This research can help increase transparency in the identification process of TEMs and PFs in designs of preplanned subgroup and stratification analyses in future RCTs and improve comparability of results from different PAICs.

All authors participated in the conception and design of the study. K Chiang and IA Samjoo contributed to the literature review, data collection and data analysis. J He and X Lin contributed to data analysis. J Maringwa contributed to data collection. All authors contributed to the interpretation of the data and critically reviewed for importance of intellectual content for the work. All authors were responsible for drafting or reviewing the manuscript and for providing final approval. All named authors meet the International Committee of Medical Journal Editors (ICMJE) criteria for authorship for this article, take responsibility for the integrity of the work and have given their approval for this version to be published.

The authors acknowledge Joanna Bielecki who developed, conducted and documented the database searches, Muhaimen Siddiqui for providing statistical expertise and Maxwell Bezaire and Alexandra Hall for assisting with the literature review and data collection. J Bielecki and M Siddiqui are employed by EVERSANA™, Canada. M Bezaire and A Hall are former employees of EVERSANA™, Canada. In addition, the authors acknowledge Annette Lam, Eric Ammann, Joris Diels and Joao Mendes from Johnson and Johnson who provided valuable comments for this work.

This work was sponsored by Johnson and Johnson.

J He, X Lin, J Kwong, J Maringwa, L Yang, S Nair, M Hashim, MJ Keng are employees of Johnson and Johnson. J He, X Lin, J Kwong, L Yang, M Hashim, S Nair also own Johnson and Johnson stock. K Chiang and IA Samjoo are employees of EVERSANA™, Canada, which was a paid consultant to Johnson and Johnson in connection with the development of this manuscript. The authors have no other competing interests or relevant affiliations with any organization or entity with the subject matter or materials discussed in the manuscript apart from those disclosed.

Medical writing support was provided by K Chiang and IA Samjoo from EVERSANA^™, Canada and funded by Johnson and Johnson.

This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.

This work is licensed under the Attribution-NonCommercial-NoDerivatives 4.0 Unported License. To view a copy of this license, visit https://creativecommons.org/licenses/by-nc-nd/4.0/