Network meta‐analysis of randomized trials in multiple myeloma: Efficacy and safety in frontline therapy for patients not eligible for transplant

Abstract The treatment scenario for newly‐diagnosed transplant‐ineligible multiple myeloma patients (NEMM) is quickly evolving. Currently, combinations of proteasome inhibitors and/or immunomodulatory drugs +/− the monoclonal antibody Daratumumab are used for first‐line treatment, even if head‐to‐head comparisons are lacking. To compare efficacy and safety of these regimens, we performed a network meta‐analysis of 27 phase 2/3 randomized trials including a total of 12,935 patients and 23 different schedules. Four efficacy/outcome and one safety indicators were extracted and integrated to obtain (for each treatment) the surface under the cumulative ranking‐curve (SUCRA), a metric used to build a ranking chart. With a mean SUCRA of 83.8 and 80.08 respectively, VMP + Daratumumab (DrVMP) and Rd + Daratumumab (DrRd) reached the top of the chart. However, SUCRA is designed to work for single outcomes. To overcome this limitation, we undertook a dimensionality reduction approach through a principal component analysis, that unbiasedly grouped the 23 regimens into three different subgroups. On the bases of our results, we demonstrated that first line treatment for NEMM should be based on DrRd (most active, but continuous treatment), DrVMP (quite “fixed‐time” treatment), or, alternatively, VRD and that, surprisingly, melphalan as well as Rd doublets still deserve a role in this setting.

In this scenario, current first line treatments for NEMM include the combination of daratumumab + bortezomib, melphalan and prednisone (DrVMP) or lenalidomide and dexamethasone (DrRd) in Europe, while melphalan-free regimens such as Rd + bortezomib (VRD) or DrRd are the preferred regimens in the USA. 2 However, the lack of direct head-to-head comparisons between approved regimens and the recent introduction of monoclonal antibodies, further complicated the decision-making regarding frontline strategy for NEMM. To overcome these limitations, we adopted an approach based on network meta-analysis (NMA) (a recently introduced Bayesian statistical methodology that allows combining direct and indirect evidence to rank the different treatments according to their efficacy and safety 1,5 ), to identify regimens with the highest probability of being the most efficacious and safest in this setting.

| Search strategy
Relevant publications have been identified through an electronic search of the main relevant databases including PubMed, Embase, Ovid, Cochrane, and proceedings from the major international meetings in hematology and oncology. The following search terms were used: "multiple myeloma", "Clinical Trials", "Phase III", "Phase II", "Randomized Controlled Trials", "untreated", "transplant ineligible".
All titles were screened and selected abstracts were reviewed. The related-articles function, article references, and Google Scholar were also screened for other applicable publications and were used for searching related studies, abstracts, and citations. Published articles were considered for the analysis if written in English only. The last date of the search was 25 November 2021. A systematic review was performed according to the guidelines and recommendations from the preferred reporting items for systematic reviews and network metaanalyses (PRISMA) checklist. 7

| Inclusion criteria
Retrieved studies were included into the final analysis if the following criteria were met: (1) they had to involve NEMM (transplant notplanned); (2) they should be randomized controlled trials, with or without blinding; (3) they could be abstracts, only if they sufficient information on study design, characteristics of participants, interventions, and outcomes were available; (4) they should include patients who received an unconventional or new regimen in the experimental arm, and a standard regimen in the control arm; (5) all trials should have been performed starting from the introduction of the so called "novel agents": IMiDs and PI.

| Exclusion criteria
Studies were excluded from the analysis if they were not comparative, if outcomes of interest were not reported, if the methodology was not clearly reported, if included patients eligible for autologous stem cell transplant (without non-ASCT subgroup analyses) or relapsed after a frontline therapy.

| Data extraction and quality assessment
Three reviewers (C.B., R.A. and E.G.) independently reviewed published literature according to the above predefined strategy and criteria. Each reviewer extracted from each selected study the following data: title and reference details (first author, year), study population characteristics (number of patients in study, number of 988 -BOTTA ET AL. patients in each treatment), type of interventions, and outcome data.
For each trial, we evaluated hazard ratios (HRs) of progression-free survival (PFS); overall survival (OS); odds ratio (OR) of overall response rate (ORR), complete response (CR); and risk ratio (RR) for safety (evaluation of the most common grade 3-4 toxicity). If the HR of survival curves was not reported, it was derived from the graph by using the method of Tierney et al. 8 All data were recorded independently in separate databases by all 3 reviewers and were compared just before the final analysis to limit selection bias. The final database was also reviewed an additional investigator (M.S.).
Duplicates were removed and any disparity clarified.
All the selected studies were assessed for quality according to the Cochrane Handbook for Systematic Reviews of Interventions, as described elsewhere 1,9 by computing a score based on the following items (1 point for each of them): method of randomization, allocation concealment, blindness, withdrawal or dropout, and adequacy of follow-up. Visual inspection of funnel plots were used to assess the presence of publication bias.

| Network meta-analysis
We performed a NMA by using a Bayesian approach to compare the different therapeutic regimens simultaneously. The analysis was performed in STATA software by using the mvmeta package.  12,13 This allowed us to identify clusters of regimens with similar profiles of efficacy/safety rather than the "best" treatment.

| Study selection and quality assessment
As shown in the PRISMA flow chart in Figure 1, with our search strategy we retrieved a total of 2579 studies. Of them, 27 studies, including a total of 12,935 patients were included in the final analysis (Table 1).  Almost all the trials included all the variables necessary to perform the whole analysis, and all the missing information where retrieved from other meta-analysis, calculated from reported data, or obtained from updated analyses (e.g., OS data were often presented when a longer follow-up was available). 46 No significant inconsistency or loop-specific heterogeneity were found in our NMA (data not shown). Clarithromycin + RD (ClRD), ixazomib + RD (IRD)) to be compared (as reported in Table 1), linked by nine triangular loops.

| Quadruplet and mAbs containing-regimens consistently improve patients' outcome
Each group was subsequently compared against all other groups through a Bayesian NMA, and efficacy results for PFS and safety, using the MP regimen as comparator, are shown in Figure 1B  Regarding safety, regimens combining melphalan and lenalidomide delivered the highest toxicity to patients, while other regimens failed to demonstrate important differences.

| DrRD and DrVMP could guarantee the best outcome for NEMM
Network meta-analysis has the possibility to calculate the probability of each regimen evaluated of being the best or the worst as well as the probable "position" within a ranking of all regimens. In Figure 3A the probability distribution of being the regimen placed at the "x" position in the PFS rank is showed. DrRd has a 58.6% probability of being the best regimen according to this outcome, immediately followed by DrVMP (25.3%) and VRD (9.7%). Figure 3B, which reports the cumulative probabilities, confirmed these results: indeed, in the

| PCA analysis identified the best regimens according to needed outcomes
To overcome the limit of using a simple and not weighted "average" of the SUCRA score, we applied a dimensionality reduction approach known as "principal component analysis," PCA, to distribute in a plane all the 23 evaluated regimens. The distance between each point depends upon the difference in the "profile" of SUCRA scores. By using this approach we were able to unbiasedly cluster all the

| MRD assessment further support NMA results
Currently, the absence of detectable minimal residual disease (MRD), especially if sustained, is considered the best surrogate marker of OS. 50 Along this line we retrieved the rates of MRD negativity in each study that investigated/disclosed this endpoint. Unfortunately, 4 studies only reported these results (Table 2). Interestingly, both DaraRD and DaraVMP reported similar MRD negativity rates, a result that further supports the conclusion of our NMA. No data regarding the SWOG5077, and specifically, the VRD regimen, were reported in any other study on NEMM patients.

| DISCUSSION
The landscape of first line treatment for NEMM has dramatically -993 according to one specific end-point only, and an "average" score, by mixing results obtained in different aspects, could not be able to capture the overall efficacy/safety profile of a regimen. 5 On these bases, we used a dimensional reduction approach (principal component analysis) and the k-means derived algorithm partitioning around medoids to group the different treatments according to their efficacy and safety profiles. 13 Therefore, we obtained three groups: one efficacy-driven group, a second "alternative" group and a third "bad"  Our work presents some limitations that should be carefully taken into account: first, all data were retrieved or calculated from published studies rather than from individual patients'; second, potential biases can be produced by the heterogeneity of the agents, patient populations as well as the long timeframe included in the analysis: to reduce this factor, we tried to limit the timeframe to the latest 20 years, that is, from the introduction of modern drugs (IMiDs and PIs). Finally, this work should be considered a snapshot of current evidence that could quickly evolves with the introduction of new drugs in the frontline setting.

| CONCLUSION
Overall, this is, to our knowledge, the first NMA which use a dimensionality reduction approach to group treatments according to