Comparison of outcomes with elranatamab and real world treatments in the UK for triple class exposed relapsed and refractory multiple myeloma

Abstract

Background

Patients with triple class exposed (TCE), relapsed and refractory (RR) multiple myeloma (MM) have limited treatment options and poor prognosis. Elranatamab, a bispecific BCMA-targeted antibody, is an investigational treatment for RRMM with demonstrated efficacy and safety in MagnetisMM-3, a single-arm, multi-centre, phase-2 study. This study aimed to characterise outcomes for real world TCE RRMM patients and to estimate the treatment effect of elranatamab compared to treatments available in routine clinical care for TCE RRMM in the NHS.

Methods

A retrospective, observational, external control arm (ECA) study combining participants from a single arm, multi-centre phase 2 study, MagnetisMM-3, receiving elranatamab to compare patient characteristics and median survival using a comparator cohort of TCE RRMM patients treated with real world regimens in five UK centres between 2015 and 2023. Both naive and adjusted treatment effect estimates for progression free survival (PFS) and overall survival (OS) were obtained using inverse probability of treatment weighted (IPTW) Cox proportional hazards models and differences in restricted mean survival time (dRMST). Quantitative bias analysis was used to assess the robustness of effect estimates to unmeasured confounding.

Results

From a total of 5,535 patients identified with a diagnosis of MM, 81 were identified as eligible for inclusion in the ECA. A total of 13 different regimens were recorded as being initiated from the real world RRMM at index date, the most common regimen was pomalidomide + dexamethasone (48.15%). Clinical outcomes in the ECA were poor (median PFS 3.71 months [95% confidence interval (CI) 2.73–4.73], median OS 11.00 months [8.02–18.10]). In unadjusted analyses the elranatamab cohort had significant improvements in PFS (dRMST 6.95 months [4.08–9.61]) and OS (Hazard Ratio (HR) 0.66 [0.45–0.96]). Adjusted analyses showed similar effects for PFS (dRMST 6.45 [3.05–9.45]) but were equivocal for OS (HR 0.75 [0.46–1.26]).

Conclusion

This study provides recent real world evidence of poor outcomes in TCE RRMM in the UK. PFS was longer among patients who received elranatamab compared with treatments for TCE RRMM in routine UK clinical practice.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-025-14624-9.

Background

Multiple myeloma (MM) is an incurable plasma cell malignancy where treatment aims to extend survival and manage disease symptoms [1]. Response to treatment typically decreases with each relapse, as the disease becomes increasingly refractory to available classes of therapy [2, 3]. Over the course of their treatment, patients with MM typically receive regimens consisting of one or more drugs from the proteasome inhibitor (PI), immunomodulatory agent (IMiD), and anti-CD38 monoclonal antibody (mAb) classes. Heavily pre-treated patients with RRMM become triple class exposed (TCE) after receiving one or more therapies from each class. Patients with TCE MM have particularly poor outcomes, with median progression-free survival (PFS) in real world cohorts observed to be 4.6 to 6.2 months and median overall survival (OS) to be between 12.4 and 14.2 months [4, 5].

Several novel bispecific therapies [6–8] have demonstrated promising efficacy in treating RRMM in clinical trials. Elranatamab is a humanised bispecific antibody targeting B-cell maturation antigen (BCMA) on myeloma cells and CD3 on T cells [9]. This treatment is indicated as a monotherapy for adult patients with RRMM who have received at least three prior therapies, including a PI, an IMiD and an anti-CD38 mAb, and have demonstrated disease progression on the last therapy [10] and has recently been recommended by the National Institute for Health and Care Excellence for use in England for this patient group (October 2024) [11]. The efficacy and safety of elranatamab were demonstrated in MagnetisMM-3 (NCT04649359), a single-arm, multi-centre, phase-2 study in RRMM [9, 12]. As MagnetisMM-3 does not have a comparator arm, external data sources are required to contextualise the treatment effects of elranatamab. Such analyses have previously been performed to estimate the effectiveness of elranatamab relative to therapies used to treat triple-class–refractory (TCR) MM in the United States [13].

Given variation in treatment pathways internationally, an external control arm (ECA) study was constructed using national health service (NHS) electronic health records (EHRs) to generate UK-relevant evidence of the relative effectiveness of elranatamab against treatments available in real world UK clinical practice for RRMM. The study tested the null hypotheses of no difference in PFS or OS between patients who received elranatamab in MagnetisMM-3 to patients who received real world treatment in the UK.

Methods

Data sources

MagnetisMM-3 study

This study used individual patient data for elranatamab treatment from the 15-month data cut of the MagnetisMM-3 study (data cut-off date of 14th April 2023), which started enrolment in February 2021 [9]. MagnetisMM-3 included two cohorts, with one cohort consisting of patients who were naive to BCMA targeted therapy (Cohort A), and the other consisting of patients who had prior exposure to BCMA targeted therapy (Cohort B). Only patients in Cohort A of MagnetisMM-3 (n = 123), were included in analyses, as no BCMA-targeted therapies were routinely available in the UK at the time of this study.

Real world data

To construct the real world ECA RRMM cohort, de-identified, individual patient level EHR data were used. Arcturis accesses EHR data generated as part of routine clinical care through strategic research agreements with a number of NHS centres. Data in this study were accessed from the research database held by Arcturis, from four English NHS centres (Chelsea and Westminster NHS Foundation Trust, Oxford University Hospitals NHS Foundation Trust, Hampshire Hospitals NHS Foundation Trust and Milton Keynes University Hospital NHS Foundation Trust) and one Scottish NHS centre through the provision of a pseudonymised dataset by the West of Scotland Safe Haven at NHS Greater Glasgow and Clyde (REC approval 22/WS/0163). EHR data from January 2000 to August 2023 across the five NHS centres were curated for the study. Based on national statistics (NHS Digital, 2021), the real world study dataset captured 9.7% of incident MM cases in England (9.8% male, 9.5% female) [14]. Data from each NHS centre were cleaned and harmonised by mapping to common concepts, including laboratory tests names and medications, to create a single analysis dataset.

Study eligibility

Real world patients were included in analyses if they met all of the following criteria for RRMM: had one or more records of the International Classification of Disease-10th revision (ICD-10) diagnosis code for MM (C90.0) at any point in their EHR; received at least one PI, at least one IMiD, and at least one anti-CD38 mAb; had documented disease progression (according to the International Myeloma Working Group [IMWG] definition) within 60 days of the last dose, or during treatment with that drug class; and had at least one subsequent line of therapy (LOT) after becoming TCE. Additionally, patients had to meet the eligibility criteria for MagnetisMM-3 [9], as applied in that study or based on a real world approximation where necessary (Table S1). Patients who received a BCMA-targeted therapy after index date were censored at initiation of this therapy. Patients with RRMM were identified between January 2015 to August 2023, reflecting the period in which anti-CD38 mAb therapies have been available in the UK. The index date was defined as the start of elranatamab initiation for patients in the MagnetisMM-3 study and initiation of the subsequent LOT following TCE eligibility in the real world cohort.

Outcomes

In MagnetisMM-3, disease progression was assessed using the IMWG definition of progression [15]. Whilst no data field specifically reported disease progression status in the EHRs, many of the tests required to identify IMWG defined disease progression (e.g., serum protein electrophoresis [SPEP], Kappa/Lambda Free-Light Chain [FLC], and immunoglobulin-A [IgA]) were available for analysis. Therefore, a real world proxy of IMWG-defined PFS was applied (real world PFS) in the real world cohort, where progression in any biomarker used in the IMWG definition of progression was taken as indicative of progressive disease [15] (see Additional file 1). Assessment of real world PFS did not include information for urine protein electrophoresis (UPEP), plasma cell percentage, hypercalcaemia solely attributed to the plasma cell proliferative disorder, or development of new bone lesions or soft tissue plasmacytomas, as these measures are not typically recorded routinely in clinical care. OS was defined as the time from index date up to the date of death, with administrative censoring at the last recorded death date for the corresponding NHS centre where the patient was treated.

Statistical analysis

Baseline demographic and clinical characteristics of the MagnetisMM-3 and real world cohorts were descriptively summarised. To mitigate confounding when comparing outcomes in the MagnetisMM-3 and real world cohorts, inverse probability treatment weights (IPTW) were used to balance measured confounding covariates between the study cohorts. The treatment weight for each patient was derived by estimating a propensity score (PS) and then creating stabilised treatment weights using the PS for each patient. The variables included in the PS model were: sex; age; international staging system (ISS) stage [16]; cytogenetic risk; Eastern Cooperative Oncology Group (ECOG) performance status; MM duration, defined as the time between index date and the earliest date of first MM therapy or first ICD-10 MM diagnosis code; and number of previous LOTs. These covariates were identified from a prior ECA analysis using data from MagnetisMM-3 in which a systematic literature review and expert clinical guidance were used to identify confounding covariates [13]. Extreme weights were trimmed at the 1st and 99th percentiles.

Weighted comparisons of PFS and OS between the MagnetisMM-3 and real world cohorts were undertaken using Cox proportional-hazards (Cox-PH) models to obtain a hazard ratio (HR) for elranatamab treatment. Additionally, the weighted difference in restricted mean survival time (dRMST) [17] was calculated to enable comparative effectiveness analysis where non-proportional hazards were detected using the Schoenfeld residual test. Unweighted analyses were also performed by excluding the weights from Cox-PH estimation and dRMST calculations.

Missing data was handled with multiple imputation by chained equations (MICE) and bootstrapping was used to estimate the 95% confidence interval for the estimated HR and dRMST. Covariate balance after weighting was assessed using standardised mean differences (SMDs) and covariate balance plots.

As data arising from routine clinical care was used to generate the ECA cohort, an a priori sample size calculation was not possible, since the number of patients that would meet the eligibility criteria of MagnitisMM-3 could not be determined ahead of any formal analyses. Additionally, because the distribution of propensity scores in the ECA and MagnetisMM-3 populations was unknown prior to analysis, the eventual effect of inverse probability of treatment weighting (IPTW) on the effective sample size also could not be predicted. Consequently, there was insufficient information to conduct a pre-specified sample size or power calculation.

Sensitivity analyses were undertaken to quantify the impact of unmeasured confounding on the estimated treatment effect for elranatamab, using quantitative bias analysis (QBA). Given the lack of a published consensus on how to perform QBA for dRMST analyses, the E-Value [18] QBA method was applied to the estimated HR only [19]. Further sensitivity analyses using an expanded PS model consisting of a larger selection of covariates were performed to assess the robustness of results to the choice of PS covariates. See Additional file 1 for further information on the statistical methods used.

All analyses were performed in R [20] version 4.0.2, using WeightIt [21] for IPTW, cobalt [22] for covariate balance assessment, MICE [23] for imputation, EValue [18] for QBA, and the survival [24] package for Cox-PH modelling. This study is registered on the Open Science Framework (OSF) Real World Evidence Registry (10.17605/OSF.IO/2J7QE).

Results

Patient characteristics

A total of 5,535 patients with a diagnosis code for MM were identified in the real world dataset. Of these, 475 patients were exposed to all 3 drug classes, 201 also received a subsequent LOT after becoming TCE and among these, 115 had disease progression (real world proxy) on their last therapy prior to their index date. After applying the MagnetisMM-3 eligibility criteria, 81 patients in the real world cohort were retained for analyses (Fig. 1) alongside 123 patients in MagnetisMM-3 Cohort A (Table 1).

Fig. 1.

Patient flow for ECA cohort construction using magnetisMM-3 eligibility criteria [12]

Table 1.

Cohort characteristics for the ECA (real world) and MagnetisMM-3 cohorts

Variable	ECA, real world cohort (N = 81)	MagnetisMM-3 study Cohort A (N = 123)	SMD (Weighted SMD) [Sensitivity Analysis Weighted SMD]
Age (Years), Mean (SD)	67.56 (10.61)	67.07 (9.45)	−0.05 (−0.01) [0.03]
Sex, n (%)
Male	46 (56.79)	68 (55.28)	−0.02 (0.03) [−0.01]
Ethnicity, n (%)
White	59 (72.84)	72 (58.54)	−0.14 [−0.10]
Asian	≤ 5	16 (13.01)	0.11 [0.06]
Black	≤ 5	9 (7.32)	0.06 [0.04]
Unknown/Other	19 (23.46)	26 (21.14)	−0.02 [0.00]
BMI (kg/m2), Mean (SD)	27.02 (5.63)	26.62 (5.43)	−0.072 [−0.01]
BMI Missing, n (%)	31 (38.27)	0 (0)	-
Haemoglobin (g/dL), Mean (SD)	10.97 (1.92)	10.37 (1.74)	−0.33 [−0.09]
Haemoglobin Missing, n (%)	11 (13.58)	0 (0)	-
Platelets (109/L), Mean (SD)	184.06 (91.59)	158.99 (83.29)	−0.29 [−0.16]
Platelets Missing, n (%)	11 (13.58)	0 (0)	-
ANC (109/L), Mean (SD)	3.63 (2.38)	2.87 (1.50)	−0.38 [−0.20]
ANC Missing, n (%)	11 (13.58)	0 (0)	-
ALT (u/L), Mean (SD)	20.50 (15.69)	20.09 (16.62)	−0.03 [−0.12]
ALT Missing, n (%)	11 (13.58)	0 (0)	-
AST (u/L), Mean (SD)	19.17 (9.71)	24.62 (12.94)	0.48 [0.03]
AST Missing, n (%)	51 (62.96)	≤ 5	-
eGFR < 40 ml/min/1.73m2, n (%)	≤ 5	11 (8.94)	0.08 [0.03]
eGFR Missing, n (%)	11 (13.58)	0(0)	-
Charlson Comorbidity Index
Index, Median (Q1, Q3)	1 (0, 2)	0 (0, 1)	-
Index = 0, n (%)	30 (37.04)	63 (51.22)	0.14 [0.06]
Index = 1, n (%)	27 (33.33)	50 (40.65)	0.07 [0.01]
Index = 2, n (%)	15 (18.52)	8 (6.50)	−0.12 [−0.04]
Index = 3+, n (%)	9 (11.11)	≤ 5	−0.09 [−0.03]
Stem cell transplant prior to index date, n (%)	30 (37.04)	87 (70.73)	0.34 [0.13]
ECOG performance status, n (%)
0	20–25*	45 (36.59)	−0.11 (−0.04) [−0.04]
1	20–25*	71 (57.72)	0.16 (0.02) [0.03]
2	≤ 5	7 (5.69)	−0.04 (0.04) [0.02]
Missing	31 (38.27)	0 (0)	-
ISS, n (%)
Stage 1	≤ 5	35 (28.46)	0.27 (0.22) [0.15]
Stage 2	5–10*	47 (38.21)	0.01 (−0.20) [−0.08]
Stage 3	5–10*	24 (19.51)	−0.27 (−0.03) [−0.07]
Missing	65 (80.25)	17 (13.82)	-
Cytogenetic Risk, n (%)
Standard	NA	83 (67.48)	-
High	NA	31 (25.20)	-
Missing	81 (100)	9 (7.32)	-
MM duration (Months), Mean (SD)	57.17 (35.53)	78.71 (45.87)	0.3 (0.18) [0.22]
Year of MM diagnosis, Median (Q1, Q3)	2016 (2013, 2018)	2015 (2012, 2018)	-
LoT immediately prior to index date, n (%)
Includes PI	29 (35.80)	54 (43.90)	-
Includes IMiD	11 (13.58)	38 (30.89)	-
Includes anti-CD38 mAb	77 (95.06)	47 (38.21)	-
Other	0 (0)	22 (17.89)	-
Duration of prior therapy (Years), Median (Q1, Q3)	7.13 (4.37, 12.78)	5.49 (2.48, 10.35)	0.01 (−0.02) [−0.02]
Prior LoT Count (n), Median (Q1, Q3)	4 (3, 4)	5 (4, 6)	0.86 (0.31) [0.40]

Standardised mean differences (SMD) before and after the application of IPTW are presented alongside summary statistics for each characteristic employed in the propensity score model. SMD is not presented for covariates not included in the main propensity score model or extended propensity score model. SMD obtained using the extended propensity score model are only presented for covariates included in the sensitivity model, SMD obtained using the main propensity score model are only presented for covariates included in the main propensity score model. SMD is calculated using only observed data points (i.e., without consideration of imputed datapoints applied in comparative effectiveness analyses). Sensitivity Weighted SMD is obtained using an extended propensity score model, any data from any bootstrap/imputation replicates that resulted in extreme weights (i.e., > 100) were excluded from the calculation of sensitivity weighted SMD. *Small values less than or equal to 5 are suppressed for both the real world cohort and MagnetisMM-3, with categorical variables in the real world cohort masked to prevent back-calculation of small values.

ALT Alanine aminotransferase, ANC Average Neutrophil Count, AST Aspartate Aminotransferase, BMI Body Mass Index, ECOG Eastern Cooperative Oncology Group, eGFR Estimated Glomerular Filtration Rate, ISS International Staging Score, IMiD Immunomodulatory drug, LoT Line of Therapy, mAb Monoclonal antibodies, MM Multiple myeloma, PI Proteasome inhibitor, Q Quartile, SD Standard Deviation

The two cohorts had similar mean age (67.56, SD 10.61 in the real world cohort compared with 67.07, SD 9.45 in MagnetisMM-3), percentage of male patients (56.79% compared with 55.28%) and similar median comorbidity status measured by the Charlson Index (1, Interquartile range [IQR] 0–2 compared with 0, IQR 0–1) (Table 1). Duration of MM was shorter in the real world cohort than in MagnetisMM-3 (mean 57.17-months, SD 35.53 compared with 78.71-months, SD 45.87) but median year of MM diagnosis was similar in both cohorts (2016, IQR 2013–2018 compared with 2015, IQR 2012–2018). The number of prior therapy lines at index date was 4 (IQR 3–4) in the real world cohort and 5 (IQR 4–6) in MagnetisMM-3. Almost all real world patients received an anti-CD38 mAb in the LOT prior to index date (95.06%) compared to a smaller proportion in MagnetisMM-3 (38.21%). All real world patients received either a PI, IMiD, or anti-CD38 mAb in the LOT prior to index, whereas 17.89% of MagnetisMM-3 participants did not receive a therapy belonging to any of these classes in their prior LOT. Relatively high levels of missing data were present in the real world cohort for ECOG status (38.27%) and ISS (80.25%). Cytogenetic risk score was not routinely recorded in real world EHRs and therefore could not be included in analyses. Maximum follow up was 25.1 months in both cohorts, in line with the maximum follow-up available from the MagnetisMM-3 Study.

Treatment received at index date

Amongst the 81 patients in the real world cohort, 31 (48.15%) received pomalidomide and dexamethasone at index date, 12 (14.81%) received ixazomib, lenalidomide, and dexamethasone, 10 (12.35%) received carfilzomib and dexamethasone, and 20 (24.69%) received different regimens for which there were n ≤ 5 recipients per regimen. A total of 13 different regimens were initiated in the real world ECA cohort at index date. All patients in the MagnetisMM-3 cohort received elranatamab as per the trial protocol [9, 12].

Comparative progression-free survival and overall survival

Median PFS in the real world cohort was 3.71 months (95% confidence interval [CI] 2.73–4.73) and median OS was 11.00 (8.02–18.10) (Fig. 2). In MagnetisMM-3, median PFS was not reached (NR) (95% CI 10.11-NR) and median OS was also NR (95% CI 13.85-NR) at the data cut-off date of 14th April 2023 (Fig. 3). As shown in Fig. 4, significant improvements in PFS were observed in unweighted analyses for elranatamab compared with real world treatments (HR 0.48 [0.33–0.70]; dRMST 6.95 months [4.08–9.61]). However, the HR results should be treated with caution as the null-hypothesis of proportional hazards was rejected (p = 0.002). Significant improvements in OS were also observed for elranatamab (HR 0.66 [0.45–0.96]; dRMST 3.01 months [0.32–5.70]), with the proportional hazards assumption upheld (p = 0.32). Weighted analyses showed significant improvements in PFS associated with elranatamab treatment (HR 0.52 [95% CI 0.35–0.80] and dRMST 6.45 months [3.05–9.45]. Weighted results for OS were not statistically significant (HR 0.75 [0.46–1.26] and dRMST 2.02 months [−1.43-5.27]). As with the unadjusted analyses, proportional hazards were violated for PFS (p = 0.003), but not OS (p = 0.34).

Fig. 2.

Kaplan-Meier OS Curves: Unweighted (a) vs. IPTW Weighted (b). Vertical bars denote the date of censoring for censored observations. Shaded regions around each curve represent the 95 % confidence interval for the estimated survival probability. The vertical dashed line indicates the median survival time, if this was reached during the study period. Separate survival distributions are presented for the real-world and MagnetisMM-03 study cohorts. The table below presents the number at risk in each cohort through time

Fig. 3.

Kaplan-Meier PFS Curves: Unweighted (a) vs. IPTW Weighted (b). Vertical bars denote the date of censoring for censored observations. Shaded regions around each curve represent the 95% confidence interval for the estimated survival probability. The vertical dashed line indicates the median survival time, if this was reached during the study period. Separate survival distributions are presented for the real-world and MagnetisMM-03 study cohorts. The table below presents the number at risk in each cohort through time

Fig. 4.

Comparative Treatment Effects for PFS and OS: Unweighted, IPTW Weighted, and Sensitivity Analysis. Points represent the point-estimate of the hazard ratio (HR) on the left panel, and difference in restricted mean survival time (dRMST) on the right panel. Horizontal lines represent the 95% confidence interval (CI). The dotted lines represent the null hypothesis of no reduction in hazard (left panel), and no improvement in RMST (right panel)

Calculated SMDs (Table 1) and covariate balance plots (Additional file 1 - Supplementary Figure S1) show that IPTW improved balance across all covariates except for MM duration and number of prior LOTs for both PFS and OS.

Sensitivity analyses for progression-free survival and overall survival

The E-values calculated for the IPTW adjusted HR for PFS and OS at the point estimate were 2.50 and 1.74, respectively (Figure S2). For PFS, the E-value calculated at the 95% CI limit closest to the null was 1.61. The corresponding calculation for OS was redundant as the CI encompassed the null.

Sensitivity analyses in which an expanded PS model was applied were congruent with the main weighted analyses. For the expanded PS model, a significant improvement in PFS was observed for elranatamab compared to real world treatments (HR 0.50 [0.31–0.82]; dRMST 6.75 months [2.95–10.20]), but results for OS were equivocal (HR 0.83 [0.47–1.42]; dRMST 1.28 months [−2.40-5.12]). As with the main analyses, IPTW improved covariate balance across all outcomes, but residual imbalance remained in MM duration, number of prior LOTs, neutrophils, and platelets, with smaller imbalances present in alanine aminotransferase, prior stem cell transplant, and ISS stage (Table 1). The variation of SMD observed across all imputations and bootstrap resamples is provided in Figure S3.

Discussion

Here, we constructed a real world cohort of RRMM patients in the UK based on the MagnetisMM-3 eligibility criteria. These patients received treatments used in routine clinical practice and compared their outcomes to patients in the MagnetisMM-3 study who received elranatamab, a novel BCMA-directed antibody. We found that PFS was consistently and substantially longer in the elranatamab cohort than the real world cohort of UK patients (HR 0.52 [0.35–0.80]; dRMST 6.45 months [3.05–9.45]). Weighted results for OS were not statistically significant (HR 0.75 [0.46–1.26]; dRMST 2.02 months, [−1.43-5.27]) but analyses may have been affected by the immaturity of OS data from the 15-month MagnetisMM-3 data cut. Re-analysis of OS using a later data cut from MagnetisMM-3, which would enable a later dRMST restriction time point, may improve the precision of the estimated effect of elranatamab on survival. The comparative PFS results in this study are congruent with a study comparing outcomes in MagnetisMM-3 with two US real world cohorts of patients treated with physician’s therapy choice (PTC); COTA (HR 0.37 [0.22–0.64]) and Flatiron Health (dRMST at 24 months 7.07 [3.38–10.76]) [13]. When assessing OS, Costa et al. (2024) found that the elranatamab cohort experienced significantly longer OS than PTC in the COTA real world cohort (HR 0.46 [0.27–0.77]) but similar to our study, OS was numerically but not statistically greater in the Flatiron Health cohort (dRMST at 24 months 2.34 [−1.15 to 5.84]). These comparable results were found despite heterogeneity in treatment options for RRMM in the US and the UK, providing evidence on the generalisability of elranatamab efficacy results across different healthcare settings and countries.

Poor survival outcomes were observed in the real world UK RRMM cohort of our study, with median PFS of 3.71 months (95% CI 2.73–4.73) and median OS of 11.00 months (95% CI 8.02–18.10). These results are comparable to other studies of real world RRMM. For example, the KarMMa-RW study [25] compared idecaptagene vicleucel (ide-cel) to real world treatments in Europe and the US and reported median PFS and OS of 3.5 months and 14.7 months, respectively. In the international LocoMMotion study of real world treatments [4], median PFS was 4.6 months (95% CI 3.9–5.6) and median OS was 12.4 (10.3–NR). Among the Canadian cohort in Visram et al. who received real world treatments [3], the median PFS and OS were 4.4 months (95% CI 3.6–5.3) and 10.5 months (95% CI 8.5–13.8), respectively. The consistency in, and persistence of, poor outcomes among heavily pre-treated patients, both within the UK and globally, demonstrates ongoing unmet need in this patient group and the importance of effective novel therapies to meet this need.

This study benefited from access to up to date, de-identified, patient-level EHRs collected as part of routine clinical practice in the UK. With granular clinical data from five NHS centres, it was possible to apply the eligibility criteria from the MagnetisMM-3 study to identify real world patients with characteristics comparable to the MagnetisMM-3 cohort. The application of IPTW mitigated the influence of confounding and covariate imbalance to provide robust estimates of the effect of elranatamab on OS and PFS, relative to treatments received by the UK RRMM population. This study also benefitted from access to longitudinal biochemistry test results to identify progression events. Through application of a real world proxy of IMWG defined PFS [15], a direct comparison of endpoints based on biochemical progression was possible (see Additional file 1 - Supporting Methods for more detail). Alternative, therapy based endpoints, such as time to next treatment (TTNT) or time to treatment discontinuation (TTD), have been used as PFS proxy endpoints in other studies integrating real world data, including ECAs [26–28], but delays in cessation of treatment or initiation of subsequent treatment mean that a direct assessment of real world biochemistry more closely aligns with the PFS endpoint in the MagnetisMM-3 study. Previous descriptive studies of outcomes for TCR patients in England lacked access to granular individual patient level data, restricting these analyses to the assessment of OS and TTNT as a proxy for PFS [26]. Additionally, as the index date was defined as the start of elranatamab initiation for patients in the MagnetisMM-3 study and initiation of the subsequent LOT following TCE eligibility in the real world cohort, the potential impact of immortal time bias was minimised.

However, there were potential limitations associated with using real world health data and the selected statistical approaches in this study. Ethnicity status was unknown for approximately one-quarter to one- fifth of the real world cohort, although this level of unknown ethnicity status was comparable with MagnetisMM-3. Despite a lack of national data summarising the ethnicity, age, or sex of TCE patients in the UK, it is possible to compare the real world cohort in this study to other studies of MM in England to assess demographically representativeness. A previously published study of TCE MM patients in England using national cancer registry data (Elsada et al., 2021), reported a similar cohort profile to the real world cohort analysed here in terms of age at index date (70.5 [SD 9.3], compared with 67.56 [10.61] in this study) and sex (58.7% male, compared with 56.79% male in this study). This suggests that the cohort analysed here may be representative of the national TCE MM population. However, the proportion of patients with unknown ethnicity reported by Elsada et al. (2021) was considerably smaller than observed in this study (23.5%, compared with 1.6% in this study), limiting the ability to assess the representativeness of the composition of ethnicity in the real world cohort.

Our real world definition of PFS lacked some biochemistry data items from the IMWG definition (but were present in the MagnetisMM-3 dataset) and therefore misclassification bias cannot be excluded. Related to this, some MagnetisMM-3 eligibility criteria could not be assessed in the real world cohort, or were assessed using proxies, due to the lack of routine measurement and recording in the NHS. Whilst this could explain some of the residual imbalances between the cohorts, relatively few criteria were not fully assessed and therefore unlikely to be a major contributing factor the residual imbalances observed. There were structural differences between the two cohorts as MagnetisMM-3 recruited participants from multiple countries with different treatment pathways that may not reflect the care received by the real world UK cohort. For example, this was reflected in the median number of prior LOTs, which was higher in the MagnetisMM-3 cohort (5 LOTs, IQR 4–6) than in the real world cohort [4 (3, 4)]. Median year of first MM diagnosis was comparable between the two cohorts (2016 [IQR 2013–2018] in the real world cohort compared with 2015 [2012, 2018] in MagnetisMM-3), but there was residual imbalance in MM duration (median 57.15 months [SD 35.53] compared with median 78.71 months [SD 45.87]). This, and the difference in the proportion of patients receiving either an anti-CD38 mAb in the line immediately prior to index date (95.06% in the real world cohort compared to 38.21% in MagnetisMM-3) or a therapy outside of the three classes that define TCE (0% in the real world cohort and 17.89% in MagnetisMM-3), may be indicative of differences in disease aggressiveness between populations, but they most likely reflect global differences in the timing of new treatment approvals and consequently, where in the treatment pathway patients become TCE. These residual imbalances may impact causal interpretation of estimated treatment effects; however, it is likely that any bias introduced by these imbalances was toward the null. It was also the case that MagentisMM-3 patients were explicitly TCR, whereas entry into the real world cohort required patients to attain TCE status and to be refractory to at least their most recent therapy line prior to index. Assessment of refractoriness to prior therapy lines in the real world cohort was impacted by missingness and lower data granularity further back in EHR, which necessitated the application of less stringent eligibility criteria. Due to reimbursement eligibility requirements and therapy sequencing in the UK, it is likely that patients in the real world cohort will also be TCR.

In Costa et al. similar residual imbalances were proposed as a potential cause of a numerically greater but statistically non-significant benefit in OS among the MagnetisMM-3 cohort who received elranatamab over patients in the real world cohort who were treated with physicians choice of therapy [13]. Cytogenetic risk is a key potential confounder that could not be included in these analyses due to limited testing in the NHS, however QBA analysis indicated that an outcome or exposure effect of a magnitude of 1.61 would be required to nullify the PFS results presented here. These QBA results suggest that unmeasured confounding was unlikely to affect the estimated treatment effects of elranatamab, but the observational nature of this study means that other sources of unmeasured confounding may remain. Multiple imputation depends on missing data being distributed at random otherwise bias may be introduced into the analyses, although it is not possible to determine how the presence of non-randomly distributed missing data may affect comparative effectiveness estimates from the data alone [29].

Conclusions

We found that among TCE RRMM patients in the UK, PFS and OS were comparable with results from multiple international real world studies, despite known differences between countries in available treatment options and consequently time to reach TCE after diagnosis. By using EHRs from multiple NHS centres that reflect recent clinical practice and overlap with the MagnetisMM-3 study period, we demonstrated that elranatamab has a strong PFS benefit over existing treatments for RRMM in the UK. The integration of new therapeutic options, alongside holistic clinical management incorporating clinical, nutritional and mental health care, are critical in the management of triple-class refractory MM. Given the immaturity of OS data from MagnetisMM-3, the precision of comparative OS effect estimates may be improved by re-analysis using full follow up data when available.

Abbreviations

ALT

Alanine Aminotransferase

AST

Aspartate Aminotransferase

BCMA

B-cell maturation antigen

BICR

Blinded Independent Central Review

BI

Bootstrap-then-Impute

BMI

Body Mass Index

CCI

Charlson Comorbidity Index

CI

Confidence Interval

Cox-PH

Cox proportional-hazards

dRMST

differences in Restricted Mean Survival Time

ECA

External Control Arm

ECOG

Eastern Cooperative Oncology Group

eGFR

Estimated Glomerular Filtration Rate

EHR

Electronic healthcare Records

FLC

Free-Light Chain

HR

Hazard Ratio

ICD-10

The International Classification of Diseases – 10th Revision

IgA

Immunoglobulin-A

IMiD

Immunomodulator

IMWG

International Myeloma Working Group

IPTW

Inverse Probability of Treatment Weight

IQR

Interquartile Range

ISS

International Staging System

LOT

Line Of Therapy

mAB

Monoclonal Antibody

MICE

Multiple Imputation by chained equations

MM

Multiple myeloma

NHS

National Health Service

NR

Not reached

OS

Overall survival

OSF

Open Science Framework

PFS

Progression Free Survival

PS

Propensity Score

PTC

Physician’s therapy choice

QBA

Quantitative bias analysis

RMST

Restricted Mean Survival Time

RR

Relapsed and refractory

RRMM

Relapsed and Refractory Multiple Myeloma

RWD

Real World Data

SCT

Stem Cell Transplant

SD

Standard Deviation

SMD

Standardised Mean Difference

SPEP

Serum protein electrophoresis

TTNT

Time to Next Treatment

TTD

Time to Treatment Discontinuation

TCE

Triple Class Exposed

TCR

Triple Class Refractory

UK

United Kingdom

UPEP

Urine protein electrophoresis

Authors’ contributions

CT, JOR, LC, CD, FT, JW, AP contributed to study design. CT, JOR, LC, FT, JW and AP were responsible for data acquisition. JOR, LC, FT, JW and AP contributed to data analysis. CT, JOR, LC, CD, FT, JW, AP, and TP contributed to interpretation of the results. JOR wrote the original draft of the manuscript. CT, JOR, LC, CD, FT, JW, AP, SW, and TP reviewed manuscript drafts and approved the final version of the manuscript.

Funding

This study was sponsored by Pfizer.

Data availability

The datasets generated and/or analysed during the current study are not publicly available due to the data sharing agreements with data providers not permitting the re-use of data by the wider research community but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study uses anonymised data collected by NHS centres as part of routine care. The contributing centres to this study were Oxford University Hospital NHS Foundation Trust, Chelsea and Westminster Hospital NHS Foundation Trust, Hampshire Hospitals NHS Foundation Trust and NHS Greater Glasgow and Clyde.

The study was submitted to and approved by the Local Privacy Advisory Committee of the West of Scotland Safe Haven research database at NHS Greater Glasgow and Clyde under NHS Research Ethics Committee (REC) approval 22/WS/0163, and adheres to the Declaration of Helsinki. Informed consent was not collected for the study as UK law (UK GDPR and the Data Protection Act 2018) does not require consent for the collection and processing of anonymised data.

Consent for publication

Not applicable.

Competing interests

CT, TP and SW are employees of Pfizer and may hold stock or stock options. CD was an employee of Pfizer at the time of the study and may hold stock or stock options. JOR, LC, FT, JW, AP are employees of Arcturis Data Limited which received funds from Pfizer to conduct the study and develop the manuscript. LC has also received personal consulting fees from Pfizer and is a member for a safety monitoring board for the George Institute for Global Health. KR has received research support from Celgene (BMS), Takeda, Janssen, Amgen and GSK; speaker fees from Celgene, Takeda, Sanofi, EUSA Pharma/Recordati Rare Diseases, Menarini Stemline, Janssen, Pfizer and GSK; advisory board support from Celgene, Takeda, Janssen, Amgen, AbbVie, Sanofi, Oncopeptides, Karyopharm, GSK, EUSA Pharma/Recordati Rare Diseases and Pfizer; and travel fees from Takeda, Amgen and Menarini Stemline.