Elranatamab efficacy in MagnetisMM-3 compared with real-world control arms in triple-class refractory multiple myeloma

Multiple myeloma (MM) is a progressive type of malignancy characterized by the rapid accumulation of monoclonal plasma cells mostly in the bone marrow [1]. Patients with MM experience a range of physical symptoms, including pain and fatigue, with many also reporting depression, anxiety and impairment of psychosocial well-being [2,3]. Unfortunately, the long-term prognosis is often poor for these patients, with median survival estimates ranging only from 5 to 7 years [1].

At present, the most common treatment classes used for the management of MM include proteasome inhibitors (PIs), immunomodulatory drugs (IMiDs) and anti-CD38 monoclonal antibodies [4]. Despite these options, many patients with MM will continue to relapse even after exposure to all three treatment classes. For patients with triple-class exposed (TCE; i.e., those patients exposed to ≥1 PI, ≥1 IMiD and ≥1 anti-CD38) and/or triple-class refractory (TCR; i.e., those patients with disease refractory to ≥1 PI, ≥1 IMiD and ≥1 anti-CD38) disease, subsequent management poses a considerable challenge.

Several real-world studies have highlighted the poor outcomes for these patients [5,6]. In the MAMMOTH study, 275 patients from 14 academic institutions in the USA were diagnosed with active MM and had disease refractory to an anti-CD38 (daratumumab or isatuximab) [6]. The objective response rate (ORR) associated with the first regimen after documented refractory disease to an anti-CD38 was 31%, and the median progression-free survival (PFS) and overall survival (OS) in patients were 3.4 and 9.3 months, respectively [6]. Other real-world studies of patients with TCE and/or TCR disease across countries and data sources have reported similar findings. ORRs of approximately 20–30%, median OS of approximately 12 months and median PFS of approximately 3–4 months were typically observed [7–12].

Elranatamab, a humanized bispecific antibody targeting BCMA on myeloma cells and CD3 on T cells, is currently under evaluation for the treatment of patients with relapsed or refractory MM (RRMM). In the single-arm, registrational, phase II MagnetisMM-3 (NCT04649359) trial in patients with TCR MM, elranatamab demonstrated a favorable efficacy profile based on ORR and duration of response [13]. However, given the absence of a control arm, it can be difficult to contextualize the clinical efficacy of elranatamab with what is currently available and used in routine clinical practice. This study sought to address this gap and compare the ORR, PFS and OS of elranatamab from MagnetisMM-3 with real-world external control arms created from electronic health record (EHR) databases.

Methods

Data sources

This retrospective cohort study (NCT05932290) included two cohorts of real-world patients with TCE MM created from USA-based oncology EHR databases: COTA and Flatiron Health (FH). The outcomes of patients from these two cohorts were compared with the outcomes of patients from the MagnetisMM-3 study of elranatamab.

MagnetisMM-3

MagnetisMM-3, an open-label, nonrandomized, phase II trial of elranatamab among patients with TCR MM, is a global study with sites in the USA, Canada, Europe and Asia [13]. Beginning on 2 February 2021, the trial enrolled two independent and parallel cohorts: Cohort A included participants naive to BCMA-directed therapies (N = 123) and Cohort B included participants who have received prior BCMA-directed therapy (N = 64). Per protocol, study participants may be given supportive care for cytokine release syndrome and immune effector cell-associated neurotoxicity syndrome. However, chronic systemic corticosteroids are not permitted for palliative or supportive care. The 14 March 2023 data cut-off of this study, with a median follow-up time of approximately 15 months, was used for the purposes of the analyses. Given the insufficient numbers of patients who have received prior BCMA-directed therapies in real-world data (RWD) sources, the focus of the present analysis was to contextualize only the results of Cohort A.

COTA

The COTA longitudinal database is derived from the EHRs of various healthcare sites, including academic institutions, community centers and hospital systems, representing 500,000 patients from over 200 sites of care in the USA (cotahealthcare.com/data-services/). Data elements are standardized across sources and ontologies to create a single, structured dataset to cover the full longitudinal history of a patient's clinical care. For this study, only patients from the COTA database who had been diagnosed with TCR MM and subsequently initiated a new line of therapy (LOT) between November 2015 and June 2022 were preliminarily identified. We defined refractoriness to a given treatment class as having disease progression (based on International Myeloma Working Group [IMWG] criteria [14] or clinically reported refractory status) while receiving therapy or within 60 days of the last dose in any line, regardless of response.

Flatiron health

The FH database is a demographically and geographically diverse longitudinal database also derived from EHR data. FH includes data from over 280 community cancer centers and academic institutions (~800 sites of care in total) representing more than 2.4 million active US patients with cancer available for analysis ( flatiron.com/real-world-evidence). Only patients from the FH database who had been diagnosed with TCR MM (using the same refractoriness definition as described in the COTA section) and had initiated a subsequent new LOT between November 2015 and August 2021 were preliminarily identified.

Sample

The elranatamab arm in this study (N = 123) included all participants from Cohort A (i.e., BCMA naive) in the MagnetisMM-3 trial. To identify cohorts representing physician's choice of treatment in the real-world setting, we separately imposed the inclusion and exclusion criteria from the MagnetisMM-3 trial on each COTA and FH database (Figure 1). The index date was defined as the date of initiation of elranatamab (in MagnetisMM-3) or physician's choice of treatment regimen (in the RWD sources) after documented TCR MM. Some minor adjustments were made to the inclusion/exclusion criteria, given the differences in data availability/reporting across sources. For example, although the diagnosis of MM in the MagnetisMM-3 trial was made according to IMWG criteria [14], RWD sources relied upon a combination of diagnosis codes, pathology reports, laboratory tests and unstructured data, which may or may not have fit IMWG criteria precisely. Similarly, Eastern Cooperative Oncology Group (ECOG) performance status and laboratory evaluations may have preceded the index date by up to 90 and 30 days, respectively, for RWD sources versus being assessed at the time of index in the MagnetisMM-3 trial.

Outcomes

The outcomes for this study included ORR, PFS and OS. We defined ORR as the proportion of patients with an objective response (partial response or better) based on IMWG criteria [14], which was assessed from the index date until the first documentation of progression, death, or the start of a new anticancer therapy. We defined PFS as the time from the index date until progression was confirmed through IMWG criteria or death due to any cause, whichever occurred first. We defined OS as the time from the index date until the date of death due to any cause.

Confounding variables

Ensuring that baseline factors related to treatment assignment and outcomes are captured is of the utmost importance for the validity of any analysis with external control arms. Therefore, we conducted a systematic literature review to identify variables most strongly and consistently correlated with outcomes in RWD studies conducted among patients with RRMM, which were then validated with clinical input. These variables were prioritized for inclusion in the present study. 57 studies using either univariate (n = 22) or multivariate (n = 35) analyses to predict outcomes were extracted as part of our literature review. Collectively, these studies indicated that the variables with the strongest relationships with future outcomes in an RRMM population were age, sex, cytogenetic risk, number of prior lines of therapy, ECOG performance status, time since initial MM diagnosis, penta-drug refractory status and International Staging System (ISS) disease stage. All these variables were included in the analysis as confounders and were assessed at the index date (i.e., initiation of either elranatamab or physician's choice of treatment) with the exception of ECOG performance status and ISS disease stage in RWD sources. In these cases, the ECOG performance status and ISS disease stage reported closest to the index date were used if they were within 90 days (inclusive) of the index date.

We identified and included additional confounders in the analysis to optimize the balance between participants from MagnetisMM-3 and each RWD source. These confounders were selected based on their clinical importance and relevance to disease prognosis and disease complications: the presence of bone lesions, extramedullary disease (EMD), prior stem cell transplant (SCT), comorbid conditions (assessed using the Charlson Comorbidity Index [CCI]) and levels of serum albumin, calcium, hemoglobin, serum creatinine, bilirubin, aspartate aminotransferase and alanine aminotransferase. In all cases, the value of each confounding variable was measured on or before the index date; if before, the most recent measurement to the index date was used (this was up to 1 year in the case of the CCI calculation). Of note, the presence of EMD was unavailable in the FH database and was therefore excluded from those analyses. All other confounding variables were assessed similarly across sources.

Statistical analyses

To summarize our statistical analysis approach, we first controlled for baseline confounding variables by estimating propensity scores (PSs) using age, sex, race, ISS disease stage, ECOG performance status, time from initial MM diagnosis to index date, bone lesions, EMD (COTA database only), high-risk cytogenetics, CCI score, number of LOTs used prior to index, penta-drug refractory status, SCT and levels of aspartate aminotransferase, alanine aminotransferase, hemoglobin, creatinine clearance, calcium, bilirubin and serum albumin. PSs were then used to calculate inverse probability of treatment (IPT) weights, which balanced the distributions of these confounding variables across treatment groups. Missing data (if ≤30% of data were missing) were imputed using multiple imputation methods to avoid unnecessarily reducing sample size or important variables. Treatment groups were compared on ORR using log-binomial models and on PFS and OS using Cox proportional hazards regression. To test the robustness of the findings, in addition to the IPT weighted analyses, unweighted analysis, doubly robust and propensity score matching results are also presented. Further details on the statistical methods are provided below.

As summarized above, baseline confounding variables were controlled for by using PSs, which represented the probability of initiating elranatamab versus physician's choice of treatment conditional on patient characteristics as described above. The estimated PSs were then used to generate IPT weights, which were subsequently stabilized to avoid outlier values (for elranatamab-treated patients: stabilized IPT weights = P × IPT weights; for physician's choice-treated patients: stabilized IPT weights = [1 – P] × IPT weights). Weights above the 99.5th percentile or below the 0.5th percentile were reset to those values, respectively.

We used standardized mean differences (SMDs) to assess the balance in baseline prognostic characteristics between elranatamab and physician's choice-treated patients before and after applying IPT weights. A characteristic was considered balanced across treatment groups if the SMD was ≤0.20.

Although only patients with complete information on inclusion and exclusion criteria were included in our analyses, data were missing for some confounding laboratory values in the RWD sources. Rather than exclude these patients or these confounding variables, multiple imputations were performed (if ≤30% of data were missing) under the assumption of missingness at random using multiple imputations by chained equations and the fully conditional specification method [15]. The technique is iterative and proceeds via Gibbs sampling if the initial joint distribution defined by the specified conditional distributions exists [16]. Generally, however, rates of missingness were quite low to nonexistent. In the COTA database, the rate of missingness for laboratory values ranged from 5.9 (hemoglobin) to 21.3% (creatinine clearance), with all but two laboratory values (creatine and bilirubin) having <11% missing data. In the FH database, missingness was less common, and no laboratory value had >2% missing data.

We compared the ORRs between treatment groups using unadjusted and IPT weighted (IPTW) log-binomial models. Relative risks (RRs) with 95% CIs were reported, and robust standard errors (SEs) were used for the IPTW analysis. We compared PFS and OS between treatment groups using hazard ratios (HRs) estimated via weighted Cox proportional hazard models. If the proportional hazards assumption for the Cox model was not met, an IPTW restricted mean survival time (RMST) model was applied (as prespecified in the statistical analysis plan) instead of the Cox proportional hazards regression as described by Conner et al. [17].

We conducted a quantification of bias assessment using nullification analysis to assess the potential influence of unmeasured confounding on the observed associations. We computed E-values, which reflect the minimum strength of the association the unmeasured confounder would need to have with the exposure and outcome, conditional on the measured covariates, to fully explain the observed treatment effect [18].

We also performed additional sensitivity analyses using techniques other than IPT weighting. As the first alternative, we employed a doubly robust estimator and fitted a model for treatment (elranatamab vs physician's choice) and a second model for the outcome of interest. If ≥1 of these two models is correctly specified, consistent and unbiased estimates may be obtained [19]. The nonparametric bootstrap was used to estimate p-values and 95% CIs for doubly robust analyses, given 999 bootstrap replicates. In the second alternative, the same PSs generated to calculate the IPT weights were instead used to create PS score-matched treatment groups [20].

Data sharing statement

For original data requests, please contact the corresponding author.

Results

Baseline characteristics

As of the 14 March 2023 data cutoff of MagnetisMM-3, a total of 123 patients with TCR MM were included as part of Cohort A. A total of 239 physician's choice patients with TCR MM initiating a new LOT between November 2015 and June 2022 were identified in COTA, and 152 patients initiating a new LOT between November 2015 and August 2021 were identified in FH. Daratumumab + pomalidomide + dexamethasone (DPd), carfilzomib + pomalidomide + dexamethasone (KPd) and Kd + cyclophosphamide were the most common physician's choice regimens in COTA, while Kd, DPd and elotuzumab + pomalidomide + dexamethasone (EPd) were the most common physician's choice regimens in FH (Supplementary Table 1).

The baseline demographic and clinical characteristics of patients included in the analysis are shown in Tables 1 & 2. The mean age of participants from the elranatamab arm was 67.1 years, and 55.3% were males, which was similar to that of patients from both COTA (mean age, 68.0 years; 54.4% males) and FH (mean age, 69.5 years; 52.6% males). Participants from the elranatamab arm had slightly more treatment experience (mean of 5.2 LOTs prior to index vs 4.9 and 4.0 for COTA and FH, respectively; 70.7% had received a prior SCT vs 57.3 and 36.2%, respectively) and a higher rate of refractoriness (42.3% had penta-drug refractory disease vs 18.8 and 15.1%, respectively). Conversely, participants from the elranatamab arm were less likely to have an ECOG performance status score of 2 than RWD patients (5.7 vs 16.3 and 15.8% for COTA and FH, respectively). No major differences were observed with respect to the proportion of patients with high cytogenetic risk (25.2 vs 20.5 and 25.0% in the elranatamab arm, COTA, and FH, respectively).

Applying multiple imputations by chained equations and IPT weights improved the balance in the distribution of baseline demographic and disease characteristics considerably between participants from MagnetisMM-3 Cohort A and patients from the two RWD sources (Tables 1 & 2). In the COTA database, no variable exhibited an SMD above the threshold of 0.2, and the overall mean SMD was 0.104, suggesting a good overall balance on the specified confounders between the two cohorts. In the FH database, ECOG performance status, time from diagnosis to index date, CCI score, number of LOTs prior to index, penta-drug refractoriness, and prior SCT all remained over the 0.2 threshold. In general, the weighted characteristics favored the FH arm over elranatamab, as these RWD patients had a lower comorbidity burden and less prior treatment experience (conversely, they did have a slightly higher ECOG performance status).

Comparative effectiveness

Participants treated with elranatamab exhibited a higher ORR (61.0%) than patients treated with physician's choice in both COTA (30.5%; RR: 2.00; 95% CI: 1.57–2.53; p < 0.05) and FH (30.3%; RR: 2.01; 95% CI: 1.52–2.67; p < 0.05) prior to adjustment (Figure 2A). Similarly, in the IPTW analysis, participants treated with elranatamab also exhibited a significantly higher ORR than physician's choice in COTA (71.7 vs 31.8%; RR: 2.25; 95% CI: 1.70–2.98; p < 0.05) and FH (57.4 vs 30.5%; RR: 1.88; 95% CI: 1.13–3.12; p < 0.05; Figure 2B).

At the time of 14 March 2023 data cutoff, the median PFS for elranatamab in MagnetisMM-3 had not been reached. Nevertheless, a significantly longer PFS was observed for participants treated with elranatamab than patients treated with physician's choice in the COTA database both prior to weighting (HR: 0.51; 95% CI: 0.37–0.71; p < 0.05) and after IPT weighting (HR: 0.37; 95% CI: 0.22–0.64; p < 0.05; Figure 3). Indeed, the model fit using IPT weights estimated a larger effect of treatment with elranatamab on PFS. Due to violations in the proportional hazards assumption, the unweighted and IPTW comparisons between elranatamab and physician's choice in the FH database were conducted using RMST analyses as prespecified in the statistical analysis plan (Table 3; RMST results for COTA are shown in Supplementary Table 2). At each set time period (9, 12, 15, 18 and 24 months), the average number of months spent in a progression-free state was significantly higher for elranatamab than physician's choice in the FH database (unweighted: 1.52, 2.52, 3.55, 4.55 and 6.54 months more for elranatamab, respectively; all p < 0.05; IPTW: 1.55, 2.48, 3.58, 4.74, and 7.07 months more for elranatamab, respectively; all p < 0.05). The Kaplan–Meier PFS curves for elranatamab compared with physician's choice in the FH database are presented in Supplementary Figure 1.

A significantly longer OS was also observed for patients treated with elranatamab than patients treated with physician's choice in the COTA database in both unweighted (HR: 0.65; 95% CI: 0.47–0.88; p < 0.05) and IPTW analyses (HR: 0.46; 95% CI: 0.27–0.77; p < 0.05; Figure 4). As observed with PFS, the treatment effect was strengthened post IPT weighting. Additionally, the proportional hazards assumption was violated when OS was compared between elranatamab and physician's choice in FH, thus RMST analyses were conducted instead (Table 3). At each time period (9, 12, 15, 18 and 24 months), the average survival for participants treated with elranatamab was significantly longer than that of patients who received treatment with physician's choice in the FH database in the unweighted analyses (0.71, 1.14, 1.70, 2.22 and 3.37 months of additional survival for elranatamab, respectively; all p < 0.05), but only numerically higher in the IPTW analyses (0.02, 0.31, 0.72, 1.14 and 2.34 months of additional survival for elranatamab, respectively; all p = nonsignificant). The average survival was significantly longer for participants treated with elranatamab than for patients treated with physician's choice in the COTA database in both unweighted and IPTW analyses, except for the unweighted results at 9 months (Δ = 0.59 months; p = 0.06; Supplementary Table 2). The Kaplan–Meier OS curves for elranatamab compared with physician's choice in the FH database are presented in Supplementary Figure 2.

Sensitivity analyses

The alternative approach of using doubly robust methods yielded the same conclusions as the main analyses. Elranatamab was associated with significantly longer PFS relative to physician's choice from both COTA (HR: 0.53; 95% CI: 0.32–0.86; p < 0.05) and FH (HR: 0.53; 95% CI: 0.30–0.91; p < 0.05). Similarly, as with the main analysis, elranatamab was associated with significantly longer OS relative to physician's choice from COTA (HR: 0.55; 95% CI: 0.34–0.88; p < 0.05) and numerically longer OS relative to physician's choice from FH (HR: 0.66; 95% CI: 0.37–1.17; p = 0.16). The last alternative approach of using PS-matching methods demonstrated longer PFS and OS for elranatamab across both data sources. Significantly longer PFS and OS were reported for participants treated with elranatamab than physician's choice matched controls from COTA (PFS: HR: 0.57; 95% CI: 0.37–0.87; p < 0.05; OS: HR: 0.53; 95% CI: 0.35–0.81; p < 0.05) and physician's choice matched controls from FH (PFS: HR: 0.41; 95% CI: 0.27–0.62; p < 0.05; OS: HR: 0.60; 95% CI: 0.37–0.97; p < 0.05).

Quantitative bias

Given the lack of published guidance on calculating E-values for RMST analyses, only E-values for the COTA-related analyses are reported. For PFS unweighted and IPTW analyses, E-values were calculated as 2.54 and 3.34, respectively; for OS unweighted and ITPW analyses, E-values were 2.04 and 2.82, respectively. These quantitative bias assessment results suggest that the observed effects could only be explained by an unmeasured confounder if that unmeasured cofounder exhibited an association with both the treatment and PFS (or OS) by a risk ratio of 2.54 or 3.34 (or 2.04 or 2.82), above and beyond the measured confounders.

Discussion

Results from the MagnetisMM-3 trial demonstrated favorable clinical outcomes among participants treated with elranatamab [13]. However, due to the single-arm design, it can be difficult to compare these clinical results with those of other available therapies. Even for studies that do attempt to contextualize clinical results, such as Mol et al. [11], which compared elranatamab with published, aggregated RWD representing physician's choice of therapy among patients with TCE/TCR MM, there are still limitations when baseline factors are not mutually reported between studies. In the case of Mol et al. [11], researchers were unable to adjust for cytogenetic risk, presence of EMD, ECOG performance status and creatinine clearance, as these factors were not reported in the RWD studies. The present study was able to address some of these challenges by leveraging two real-world databases with detailed information on patient characteristics and health history to serve as external controls.

Imposing the same inclusion and exclusion criteria as MagnetisMM-3 onto these RWD sources alone did not sufficiently balance the cohorts as participants from MagnetisMM-3 Cohort A had more treatment exposure/refractoriness (e.g., a higher average number of prior LOTs and a greater proportion of patients having penta-drug refractory disease) and a higher incidence of EMD, among other differences. This reinforces the danger of drawing naive, unadjusted comparisons across studies given the inherent differences in populations. Based on the baseline characteristics of the participants in the MagnetisMM-3 trial, one would expect the outcomes to be poorer than those observed in a typical TCE MM population. Of course, leveraging multiple imputation and IPT weighting methods substantially reduced these differences in our identified confounders, and in the case of COTA, eliminated them entirely.

The main results indicate that BCMA-naive participants treated with elranatamab in MagnetisMM-3 had significantly longer PFS, significantly longer OS, and a significantly higher ORR than patients treated with physician's choice therapies from the COTA database. A similar pattern was observed when comparing BCMA-naive participants treated with elranatamab in MagnetisMM-3 with patients treated with physician's choice therapies from the FH database, although the HRs were higher.

Part of the reason for the smaller differences when comparing elranatamab to physician's choice in the FH database may be due to the inability to statistically balance the cohorts on variables that generally were biased against elranatamab. Even after applying multiple imputation and IPT weighting methods, patients from FH were found to have had received fewer LOTs, had a lower probability of having penta-drug refractory disease, and had a lower comorbidity burden, which would result in better outcomes. The lack of EMD in the FH database may have also resulted in unmeasured confounding. This bias is likely to favor the physician's choice treatments, given the higher EMD prevalence observed in MagnetisMM-3 Cohort A relative to other RWD studies and its importance in predicting clinical outcomes [21,22]. Despite these limitations of the FH database specifically, the totality of the results suggests that patients treated with elranatamab have better outcomes compared with the physician's choice treatments. Considering that the MagnetisMM-3 study is ongoing, the data for PFS and OS will continue to mature.

These results also reaffirm what has been previously published with respect to the outcomes of a TCE MM population. Our observed ORR (~30%) for physician's choice therapies was consistent with that of previously published studies [6–11]. Similarly, the observed median PFS and OS were approximately 4 and 12 months, which is also consistent with what has been reported in prior studies [7–12]. The fact that these outcomes have continued to remain stagnant among patients with TCE MM underscores the desperate need for additional treatment options.

The treatment landscape of MM continues to rapidly evolve beyond IMiDs, PIs and anti-CD38 therapies with the recent approvals of selinexor [23], belantamab [24], idecabtagene vicleucel [25], ciltacabtagene autoleucel [26], teclistamab [27], talquetamab [28] and elranatamab [29], along with additional treatment options under development. Since the clinical trials that have provided the evidence for accelerated approval by the US FDA for many of these therapies in TCE MM populations have relied upon single-arm designs, studies like this one can be informative for contextualizing the clinical efficacy with available treatment options in the real world and aid treatment decision-making.

The comparison of trial data with separate cohorts of real-world patients has several limitations. First, patients without sufficient information in their EHRs may have been excluded from the RWD by the data providers and may never have had the opportunity to be included as part of the external control, which would have introduced a selection bias. Due to the exclusion criteria, it was not possible for an elranatamab patient from MagnetisMM-3 to subsequently appear in either the COTA or FH database as part of the RWD control arm. However, it was technically possible for a patient in USA to appear in the RWD control arm and subsequently enroll in MagnetisMM-3. Due to data privacy and the lack of any personal identifying information, this phenomenon of common patients across arms could not be disallowed but would be extremely unlikely. Further, given that outcomes generally decline over subsequent lines of therapy, the inclusion of any of these patients, if they exist at all, would only bias the results against elranatamab.

Also, applying inclusion and exclusion criteria from a clinical trial to RWD requires some adjustments due to data availability (e.g., EMD in FH) and differences in assessment. Each of these factors may have contributed to reduced comparability between the elranatamab cohort and the physician's choice cohorts. The potential for temporal bias due to differences in time frames between MagnetisMM-3 (February 2021 to March 2023) and the RWD sources (November 2015 to June 2022) exists and could have impacted outcomes. Similarly, supportive care might have varied between a trial environment and real-world clinical practice, although supportive care information were not available in COTA and FH. Unmeasured and residual confounding may remain given the differential availability of information and the imprecision of the data capture of disease characteristics. For example, some myeloma disease markers, such as bone marrow plasma cells, were not complete enough in RWD sources to be included in our study. However, our quantitative bias assessment suggests that any unmeasured confounder would need to have a very strong association (RR >2–3) between the treatment and the outcome to explain the observed effects. The presence of such an unmeasured confounder is extremely unlikely.

Conclusion

Despite these limitations, our study exhibited several strengths. We identified two robust RWD sources that reliably and nearly completely assessed the major confounding variables of interest. The samples from these sources were sufficiently large and sufficiently comparable with the patients from MagnetisMM-3 and were made even more comparable after the imposition of our statistical analysis. This allowed for a robust assessment to contextualize the clinical efficacy of elranatamab in comparison with treatments currently used in real-world clinical practice. Our analytical approach also leveraged multiple distinct approaches and demonstrated that our overall conclusions were insensitive to the choice of any particular statistical technique. In conclusion, these results suggest that, among BCMA-naive patients, those treated with elranatamab in the MagnetisMM-3 trial exhibited significantly and clinically meaningfully better ORR, longer PFS and longer OS compared with patients who have similar characteristics and received their physician's choice of treatment in the real world.