2.1 Overview of Studies in Scope

Pooled efficacy data across four ongoing clinical studies in patients with RRMM, including MagnetisMM-1 (phase 1; NCT03269136), MagnetisMM-2 (phase 1; NCT04798586), MagnetisMM-3 (phase 2; NCT04649359), and MagnetisMM-9 (phase 1/2; NCT05014412) were analyzed. MagnetisMM-1 is a first-in-human phase 1 study where patients with RRMM were administered intravenous (IV) or SC elranatamab [21]. MagnetisMM-2 is a phase 1 study evaluating elranatamab monotherapy in Japanese patients [22]. MagnetisMM-3 is a phase 2 trial of elranatamab monotherapy in patients who are naïve (cohort A) or have previously received (cohort B) prior BCMA-targeted therapy [23]. MagnetisMM-9 is a phase 1/2 open-label trial evaluating an alternative step-up priming regimen, as well as longer dosing intervals with higher doses of elranatamab monotherapy [24].

All the clinical studies reported here are being conducted in accordance with the Declaration of Helsinki and the International Conference on Harmonisation guidelines for Good Clinical Practice. All patients in these studies provided written informed consent. The study protocols and relevant documents were approved by independent institutional review boards or ethics committees at each investigative center. Patient safety was monitored jointly by investigators and safety assessment committees established by the sponsor.

2.2 Study Design and Patients

MagnetisMM-1 was divided into dose escalation phases (part 1 and 1.1) and a dose expansion phase (part 2 A). In part 1, each cycle was 3 weeks in duration and adult patients with advanced RRMM received either IV at doses of 0.1–50 µg/kg or SC at doses of 80–1000 µg/kg QW [21]. In part 1.1 and part 2 A, each cycle was 4 weeks in duration. Patients in part 1.1 received 600 mg/kg SC priming doses of elranatamab in the first cycle followed by 1000 mg/kg SC maintenance doses QW or once every 2 weeks (Q2W). Patients in part 2 A received 44 mg SC priming doses of elranatamab in the first cycle, followed by 76 mg SC maintenance doses QW.

In MagnetisMM-2, elranatamab was administered as a single priming dose of 600 µg/kg SC, followed by 1000 µg/kg QW in Japanese patients with RRMM [22].

MagnetisMM-3 was a two-cohort study. Cohort A enrolled patients with no prior BCMA-targeted treatment, and cohort B enrolled patients with prior BCMA-directed antibody–drug conjugate (ADC) or chimeric antigen receptor T-cell therapy (CAR-T)-cell therapy. It was a phase 2 registration-intent study in patients with IMiD/PI/anti-CD38 antibodies-refractory (also referred to as “triple-class refractory”) myeloma. Patients received SC elranatamab in two step-up priming doses of 12 mg and 32 mg on days 1 and 4 of cycle 1, followed by 76 mg QW in 28-day cycles [23]. If a participant had received QW dosing for at least six cycles and had achieved an IMWG response of a partial response (PR) or better, with responses persisting for at least 2 months, the dose interval was to be changed from QW to Q2W. The first four enrolled patients in MagnetisMM-3 received a single priming dose (44 mg) before implementing the two step-up dose regimen via a protocol amendment.

In MagnetisMM-9, patients with RRMM received SC elranatamab in two step-up priming doses of 4 mg and 20 mg on days 1 and 4 of cycle 1, respectively, followed by 76 mg QW in 28-day cycles (i.e., part 1) [24]. In all studies, elranatamab treatment continued until disease progression, unacceptable toxicity, or withdrawal of consent. Also, in MagnetisMM-1, 2, and 9, patients were allowed to have received prior BCMA-targeted therapy.

2.3 Study Assessments

Blood samples were collected for pharmacokinetic (PK) and sBCMA analyses for MagnetisMM-1 and MagnetisMM-2: predose, during treatment (on days 1, 2, 4, 8, and 15 of cycles 1–2; and day 1 of cycle 3+), and at the end of treatment (EOT); MagnetisMM-3: predose, during treatment (on days 1, 4, 8, 15, and 22 of cycle 1, days 1 and 15 of cycle 2, and thereafter on day 1 of every third cycle), and EOT; MagnetisMM-9: predose, during treatment (on days 1, 4, and 8 of cycle 1, and on day 1 of cycles 2, 4, and 6, and thereafter every third cycle), and EOT. Serum myeloma-protein (M-protein) was assessed on day 1 of each cycle and EOT in MagnetisMM-3.

Disease assessment in all studies occurred on day 1 of every cycle and was determined by an investigator using IMWG response criteria for multiple myeloma [25]. Samples for assessment of M-protein were collected as part of the disease assessment per IMWG response criteria. An objective response (OR) was defined as confirmed stringent complete response (sCR), complete response (CR), very good partial response (VGPR) and PR, from the date of first dose until the first documentation of PD, death, or start of new anticancer therapy, whichever occurred first. The efficacy endpoints (ORR, CRR, PFS, and DOR) assessed by an investigator per IMWG criteria were assessed across all four studies. In MagnetisMM-3, these endpoints, as assessed by blinded independent central review (BICR), defined by IMWG, were also determined.

Longitudinal sBCMA and serum M-protein data were explored via scatter plots under different dosing frequencies (QW, Q2W, and Q4W) in the MagnetisMM-3 study. Pooled efficacy data, as determined by an investigator per IMWG criteria, was analyzed across all four clinical studies to determine the optimal dosing regimen of elranatamab. Only response-evaluable patients were included in this analysis. The following regimens were analyzed:

Dosing regimens with no priming dose: IV: 0.1, 0.3, 1, 3, 10, 30, and 50 µg/kg QW; and SC: 80, 130, 215, 360, 600, and 1000 µg/kg QW.

Dosing regimens with priming dose (SC): (a) one single priming dose of 44 mg on day 1 (followed by 76 mg on day 8) and then 76 mg QW; (b) one single priming dose of 600 µg/kg on day 1 (followed by 1000 µg/kg on day 8) and then 1000 µg/kg QW; (c) one single priming dose of 600 µg/kg on day 1 (followed by 1000 µg/kg on day 8) and then 1000 µg/kg Q2W; (d) step-up priming doses of 12 and 32 mg on days 1 and 4, respectively, followed by 76 mg on day 8 and then 76 mg QW; and (e) step-up priming doses of 4 and 20 mg on days 1 and 4, respectively, followed by 76 mg on day 8 and then 76 mg QW.

The regimen for the MagnetisMM-3 study (i.e., 12/32 two step-up priming dose regimen, followed by 76 mg QW was reduced to 76 mg Q2W starting in cycle 7 in participants who achieved a response, and was further reduced to 76 mg Q4W after at least 24 weeks of Q2W dosing).

2.4 Elranatamab Exposure Metrics

For binary efficacy endpoints (i.e., ORR and CRR) E–R analysis, the considered free (unbounded) PK exposure metric was the average elranatamab concentration on day 28 (Cave,day28) and average elranatamab concentration up to the time of first response or progression/EOT (Cave,event). These free elranatamab summary exposure metrics were obtained using individual post hoc PK parameters from the population elranatamab PK model [26], which adequately described the PK characteristics of elranatamab by a semi-mechanistic target-mediated drug disposition (TMDD) model with first order absorption.

For time-to-event endpoints (i.e., PFS and DOR) E–R analysis, the daily exposure to elranatamab could vary throughout treatment owing to adjustments such as dosing interruptions or dose reductions. To account for these fluctuations in elranatamab exposure, time-varying average elranatamab concentration (Cave,t) was used in the DOR E–R analysis. The Cave,t values up to the times when each efficacy event (i.e., progressive disease) first occurred in a patient was calculated for that patient, as well as for all patients whose first corresponding efficacy event had not happened yet. The longitudinal Cave,t values were calculated using post hoc PK parameters from the population elranatamab PK model [26] and the time up to each event using the mrgsolve simulations in R. All exposure metrics were explored in both the natural and log-transformed scales.

2.5 Modeling Methodologies

The population E–R efficacy analyses were performed using R software version 4.1.3. R software (R Foundation for Statistical Computing, Vienna, Austria) was also used for data manipulation, post-processing, and generation of figures and tables. Summary exposure metrics were merged with the population E–R data files and used to quantify the E–R relationships of the efficacy outcomes.

2.5.1 ORR and CRR Endpoints

For the ORR and CRR E–R analysis, binomial logistic regression using the glm() function in the R programming language was performed. The first step was to build a base model, which defines the regression parameters in the absence of covariate influence except for potential elranatamab exposure metrics:

$$\log it\left( p \right) = \log \left( } \right) = \beta_ + \beta_ \times }$$

(1)

where p is the probability of the event occurring, β0 is the estimated intercept, β1 is the regression coefficient (slope) characterizing the relationship between the log-odds of achieving the efficacy endpoint (ORR or CRR) with the exposure metric, and exposure is the free elranatamab exposure metric being tested (i.e., Cave,day28 in both the natural and log-transformed scales). The standard variance-mean function for binomial logistic regression were parameterized as:

$$V\left( \mu \right) = \mu \left( \right),$$

(2)

where µ is the mean. The most appropriate exposure metric was selected on the basis of multiple factors, including clinical relevance, model fit, statistical significance, and model deviance (D). D was calculated as:

$$D = - 2 \times \ln \left( }}}} }}} \right),$$

(3)

where L0/LF was the ratio of the likelihood of the null model and fitted models. D can be shown to be approximately χ2 distributed with degrees of freedom (df) equal to the difference in the number of estimated parameters between the null and fitted models. An α of 0.01, which equates to a change in D (ΔD) greater than \(_^\) = 6.64, was used to identify the exposure metric.

If more than one exposure metric met the ΔD criteria, the exposure metric with the largest change in D would be selected. This was based on the rationale that smaller values of D indicate a better model fit.

The next step was to develop a full model which described the influence of all potential covariates on the regression parameters. The covariates that were deemed clinically relevant and were considered for E–R model inclusion are shown in Supplementary Table 1. The equation for logistic regression to assess the E–R relationship between elranatamab exposure, potential covariates, and the efficacy endpoints ORR and CRR is shown in Eq. 4:

$$\log it\left( p \right) = \log \left( } \right) = \beta_ + \beta_ \times } + \beta_ \times X_ + \cdots + \beta_ \times X_ ,$$

(4)

where p is the probability of the event occurring, β0 is the estimated intercept, β1 is the regression coefficient (slope) characterizing the relationship between the log-odds of achieving the efficacy endpoint ORR or CRR with the exposure metric, and exposure is the elranatamab exposure metric selected in the previous step, and β2,…βn represents the effect, if any, of each additional covariate on the log-odds of the event occurring.

The final model development started with the full model, containing the parameters from the base model along with every covariate under consideration. The full model, which includes the most appropriate elranatamab exposure metric, would then be subjected to stepwise backward elimination to perform covariate selection. For comparing a larger model to a smaller nested model, the difference in the scaled D (Eq. 3) would be approximately χ2 distributed with df equal to the difference in the number of identifiable parameters. The D values were scaled by the dispersion parameter, which was assumed to be 1 by default.

$$\frac}}} - D_}}} }}}}} - df_}}} }} \sim \chi_}}} - df_}}} }}^ .$$

(5)

This difference would be used to judge whether a covariate should remain in the model during the backward elimination using an α of 0.01. This corresponds to an increase in the deviance greater than \(_^\) = 6.64 with one less degree of freedom. The elimination process would be stopped and the model would be considered final when the removal of any of the remaining covariates results in a ΔD equivalent to p-value > 0.01.

2.5.2 PFS and DOR Endpoints

For the PFS and DOR E–R analysis, firstly, the graphical exploration of the association between free Cave,t and the odds of each efficacy event were conducted. On these selected efficacy event days, the Cave,t distribution in patients without events was represented as a box-and-whiskers plot. Furthermore, the Cave,t values for patients who experienced the efficacy event on that day (i.e., a progression event) were shown as dots on the box plot. To enhance the visualization of the data, a locally weighted scatterplot smoothing line was included for the Cave,t values of patients in both groups (without events and with events).

Secondly, the relationship between different free elranatamab exposure quartiles and PFS or DOR was also graphically explored using Kaplan–Meier (KM) analysis. The model-predicted average free elranatamab (Cave) up to a certain day in the study (i.e., day 28, 42, 56, and 70) were stratified into quartiles and the corresponding PFS or DOR were summarized in a KM plot to assess whether PFS or DOR differed among the exposure metric values.

Lastly, the Cox proportional hazards (PH) model was used to quantify the effects of the elranatamab exposure metric Cave,t and other covariates on efficacy endpoint PFS and DOR. The same set of covariates that were tested for ORR and CRR were also evaluated for PFS and DOR (Supplementary Table 1). The model is described in Eq. 6.

$$h\left( t \right) = h_ \left( } \right) \times \exp \left( \right),$$

(6)

where h(t) is the hazard rate at time t, h0(t) represents the background hazard rate, θ is the coefficient quantifying the effect of covariate on hazard, and Cov is the value of covariate of interest. Initially, a univariate Cox PH model was used to quantify the effects of elranatamab exposure Cave,t and other baseline covariates on safety endpoints (p-value = 0.05). For the elranatamab exposure metric, the exposure metric (either Cave,t or log-transformed Cave,t) resulting in a lower Akaike information criterion (AIC) value in the univariate analysis would be selected and used in the subsequent analyses. The significant covariates identified from the univariable analysis were subsequently included in the multivariable Cox PH analysis. The elranatamab exposure measure Cave,t, no matter if significant in the univariable analysis, was carried forward to the multivariate analyses. Lastly, the final model was developed by Cox PH analysis of significant covariates/exposure metrics identified in the multivariable step (p-value = 0.01). Proportional hazard assumption was tested on the basis of using Schoenfeld residuals using the cox.zph function included in the survival package in R.