Comparative effectiveness of ciltacabtagene autoleucel in CARTITUDE‐1 versus physician's choice of therapy in the Flatiron Health multiple myeloma cohort registry for the treatment of patients with relapsed or refractory multiple myeloma

Abstract Introduction Ciltacabtagene autoleucel (cilta‐cel) is a novel chimeric antigen receptor T‐cell therapy that is being evaluated in the CARTITUDE‐1 trial (NCT03548207) in patients with relapsed or refractory multiple myeloma (RRMM) who received as part of their previous therapy an immunomodulatory drug, proteasome inhibitor, and an anti‐CD38 monoclonal antibody (i.e., triple‐class exposed). Given the absence of a control arm in CARTITUDE‐1, this study assessed the comparative effectiveness of cilta‐cel and physician's choice of treatment (PCT) using an external real‐world control arm from the Flatiron Health multiple myeloma cohort registry. Methods Given the availability of individual patient data for cilta‐cel from CARTITUDE‐1 and PCT in Flatiron, inverse probability of treatment weighting was used to adjust for unbalanced baseline covariates of prognostic significance: refractory status, cytogenetic profile, International Staging System stage, time to progression on last regimen, number of prior lines of therapy, years since diagnosis, and age. Comparative effectiveness was estimated for progression‐free survival (PFS), time to next treatment (TTNT), and overall survival (OS). A range of sensitivity analyses were conducted. Results Baseline characteristics were similar between the two cohorts after propensity score weighting. Patients with cilta‐cel had improved PFS (HR: 0.18 [95% CI: 0.12, 0.27; p < 0.0001]), TTNT (HR: 0.15 [95% CI: 0.09, 0.22; p < 0.0001]), and OS (HR: 0.25 [95% CI: 0.13, 0.46; p < 0.0001]) versus PCT. Cilta‐cel treatment benefit was robust and consistent across all sensitivity analyses. Conclusion Cilta‐cel demonstrated significantly superior effectiveness over PCT for all outcomes, highlighting its potential as an effective therapy in patients with triple‐class exposed RRMM.


INTRODUCTION
Multiple myeloma (MM) is an incurable disease with a high rate of relapse [1]. Treatment often involves sequential lines of therapy (LOTs) with three commonly used classes of agents: immunomodulatory agents (IMiDs), proteasome inhibitors (PIs), and monoclonal antibodies (MoABs) [2,3]. Most patients ultimately become refractory to the major classes of MM therapy, leaving few treatment options and a limited survival. Currently, there is no standard of care for patients with relapsed or refractory multiple myeloma (RRMM) who are tripleclass exposed (to IMiDs, PIs, and anti-CD38 MoABs), with previous evidence showing that this population is treated with at least 336 different regimens that comprise 40 different compounds [4], frequently consisting of continuous triplet therapies [5]. Despite this heavy treatment burden, outcomes for these patients remain poor, with median progression-free survival (PFS) ranging from three to six months and median overall survival (OS) less than 12 months [2,6].
Results indicate substantial activity and duration of response, which appear to be more favorable than currently available treatments [7,8].
However, given the heterogeneity in currently used treatments and the observed poor outcomes in triple class exposed patients, no random- In this study, ITCs were conducted to retrospectively evaluate the comparative effectiveness of cilta-cel from CARTITUDE-1 and physician's choice of treatment from a RW dataset in patients with tripleclass exposed RRMM. A comprehensive overview of the CARTITUDE-1 study has been previously published [7]. The present analysis was based on an updated data cut-off of February 2021, representing a median follow-up of 18

Baseline characteristics for population alignment
Differences between nonrandomized cohorts in baseline characteristics that are prognostic of outcomes may bias comparative effectiveness estimates if left unadjusted [12]. In this study, prognostic factors for adjustment were chosen using a clinician-driven process. First, a list of potential factors was identified a priori by consulting studies from a literature review conducted to identify clinical outcomes in tripleclass exposed RRMM patients. This list was presented to a panel of clinical experts and modified according to their input. The panel was then asked to rank each variable in order of importance for adjustment. To aid in this process, clinicians were provided with univariate regression results showing the prognostic strength of each variable in terms of PFS and OS in CARTITUDE-1. Clinicians were also provided with the standardized mean difference (SMD) for each factor between CARTITUDE-1 and the RW cohort (an SMD ≤ 0.1 was considered a small difference, an SMD > 0.1 and ≤ 0.2 a moderate difference, and an SMD > 0.2 a substantial difference [13]). Clinician rankings were revised iteratively until consensus was achieved. The panel determined that refractory status, cytogenetic profile, ISS stage, time to progression on last regimen, number of prior LOTs, years since MM diagnosis, and age were the minimal set of covariates that should be adjusted for to ensure clinical validity of the analyses. Hence, these variables were adjusted for in the base case analysis. Total plasmacytoma was also among the most important factors but was not available from the FH database and therefore was not included. The remaining identified variables where rank-ordered from most to least important (Table S1).

Outcomes
Outcomes of interest were PFS, time to next treatment (TTNT), and OS.
In CARTITUDE-1, PFS was calculated as the time from the index date to disease progression or death, whichever occurred first. For patients who had not progressed and were alive at data cut-off, data was censored at the last disease evaluation before the start of any subsequent antimyeloma therapy or the retreatment of cilta-cel. Conversely, as progression data may be less strictly monitored and hence more likely to be missing in RW data than in clinical trials, PFS was defined in the RW cohort as the time from the index date to the date of progression, death, or start of next treatment, whichever occurred first, with the date of last follow-up used in censoring. TTNT was defined as the time from the index date to the initiation of the next LOT or death, whichever occurred first. Patients who were still alive and had not initiated a new LOT at the data cut-off were censored at the last date known to be alive. OS was defined as the time from the index date to the date of the patient's death. If the patient was still alive or their vital status was unknown, data were censored at the last date known to be alive (CARTITUDE-1) or the last follow-up date (maximum of last treatment end date or last visit date) (RW cohort). The middle of the month was used as the date of death in the RW cohort because only month and year of death were available in the FH database. Mortality data in the FH database is derived by amalgamating multiple data sources and has been validated against the National Death Index [14].

Statistical methods
Inverse probability of treatment weighting (IPTW) was used to balance baseline characteristics between patient populations [15]. First, propensity scores were calculated using a logistic regression model that predicted assignment in the CARTITUDE-1 cohort as a function of baseline covariates. ATT weighting was applied, with patients in the CARTITUDE-1 cohort kept as observed (i.e., assigned a weight of one) and patients in the RW cohort receiving a weight of p/(1-p), where p is the propensity score predicting inclusion in the CARTITUDE-1 cohort [15]. Patients in the RW cohort with similar characteristics to that of the observed CARTITUDE-1 population received larger weights, thereby balancing the two cohorts. The effective sample size (ESS) was calculated to reflect the impact of weighting on the available information in the individual patient-level data [16].
Estimates of comparative effectiveness were derived for both the unadjusted comparison (i.e., cilta-cel versus physician's choice of treatment prior to IPTW), and the adjusted comparison (i.e., with IPTW). A Cox proportional hazards model (with weights applied for the adjusted comparison) was used to estimate the hazard ratio (HR) and its respective 95% confidence interval (CI) or PFS, TTNT, and OS. The selected covariates were additionally adjusted for in the model for doubly robust results [17]. The cluster-robust sandwich variance-covariance estimator was used to account for within-person clustering of obser-

Adjustment for imbalances between cohorts
The main analysis comprised two patient cohorts: the treated popu-

CARTITUDE-1 COHORT
Patients who received apheresis (N = 113 patients) F I G U R E 1 Flow chart of patient selection [1]. CARTITUDE-1 inclusion criteria required at least three prior LOTs or double refractoriness to an immunomodulatory drug and a proteasome inhibitor; however, all enrolled patients received at least three prior LOTs [2]. CARTITUDE-1 inclusion criterion was creatinine clearance of ≥40 mL/min/1.73 m 2 ; however, all enrolled patients had creatine levels ≤ 2 mg/dL. Abbreviations: ECOG, Eastern Cooperative Oncology Group; LOT, line of therapy; MM, multiple myeloma; N OBS, number of observations; RW, real world cohort registry (i.e., the RW cohort). Patients in the RW cohort contributed a total of 336 observations across all eligible LOTs (Figure 1).
Baseline characteristics before and after adjustment with IPTW for the base case variables are given in Table 1 Figure S1.
Physician's choice of treatment received in all eligible LOTs consisted of 51 different regimens. Treatments received either alone or as part of combination therapies included IMiDs (pomalidomide, lenalidomide, and thalidomide), PIs (carfilzomib, ixazomib, and bortezomib), and MoABs (daratumumab and elotuzumab). See Table 2 for more details on the treatments that comprised physician's choice.

Comparative effectiveness results
Comparative effectiveness estimates for cilta-cel versus physician's choice of treatment in both the unadjusted and adjusted analyses for PFS, TTNT, and OS are shown in Table 3. Prior to adjustment, the HR  (9) 64 (19) a For each treatment, the number and percent represent the patients who received that treatment as a single-agent therapy or in combination with any of the other treatments listed in the subsequent rows. b Received alone or in combination; therefore, the total adds to more than 100% as treatments from the same line of therapy can be counted more than once. c Any one received alone or in combination with either one of the three or other drugs. d "Others" included bendamustine, cisplatin, doxorubicin, etoposide, decitabine, fludarabine, ibrutinib, venetoclax, and clinical study drug.
Abbreviation: RW, real world.  Table 3). The results of the Grambsch-Therneau test [18] for proportional hazards assumption were found to be nonsignificant for each outcome (PFS: p = 0.16; TTNT: p = 0.08; OS: p = 0.27), indicating that the proportional hazards assumption was not violated.

Hazard ratio a (95% CI), p-value for cilta-cel vs. physician's choice of treatment
The exploratory analysis that considered only the first eligible LOT for patients in the RW cohort produced similar results to those from the main analysis (Table 4).

F I G U R E 2
Kaplan-Meier plots for (A) progression-free survival, (B) time to next treatment, and (C) overall survival in CARTITUDE-1 (observed) and the RW cohort (observed and adjusted). Note: Number at risk for the adjusted RW cohort represents the sum of the propensity score weights, not the effective sample size. Adjusted results correspond to the base case analysis which adjusted for refractory status, International Staging System stage, cytogenetic profile, time to progression on last regimen, number of prior lines of therapy, years since multiple myeloma diagnosis, and age. The adjusted curves reflect inverse probability of treatment weighting with average treatment effect in the treated weights (not doubly robust The main analysis included the following specifications: all eligible LOTs from the RW cohort, treated population of CARTITUDE-1, inverse probability of treatment weighting, missing values imputed, and adjustment for base case variables (refractory status, cytogenetic profile, International Staging System stage, time to progression on last regimen, number of prior lines of therapy, years since multiple myeloma diagnosis, and age). For each additional analysis, one of these specifications were modified, as outlined in Table S2

Sensitivity analyses
In the sensitivity analysis that included the enrolled population of

DISCUSSION
Cilta-cel, a novel CAR-T therapy targeted at BCMA, demonstrated early, deep, and durable responses and a manageable safety profile for patients with triple-class exposed RRMM in CARTITUDE-1 [7,19].  after becoming refractory to an index regimen containing an anti-CD38 MoAB [2]. One study of a US-based, RW cohort of triple-class exposed patients reported a median time to discontinuation of 4.2 months (95% CI: 3.1, 5.2) [22].
As in any nonrandomized study, the potential for residual confounding cannot be excluded. However, the availability of individual patient-level data from both cohorts enabled adjustment for imbalances in important prognostic factors. To ensure that the most important clinical factors were balanced between the two populations, an evidence-informed process-which incorporated published literature, clinical opinion, prognostic strength of variables, and baseline differences between the study cohorts-was used to select the covariates for adjustment. In this study, differences in factors included in the base case were minimal after weighting, strengthening the validity of the comparison. Furthermore, compared to other oncology databases, FH provided a comprehensive range of baseline characteristics, optimizing data availability for the present analysis [9]. Of all prognostic factors identified a priori, total plasmacytomas (including extramedullary plasmacytomas) was the only variable that was unavailable for the RW cohort. Moreover, it has been reported that the frequency of extramedullary disease in MM patients increases during the course of the disease [23]. Hence, it is most likely that the patients in the RW cohort had less extramedullary disease given that they were not as heavily pre-treated as the CARTITUDE-1 patients, which is a potential bias against CARTITUDE-1.
The use of RW data presents inherent limitations. For instance, the FH database did not provide data on response outcomes such as overall response rate and complete response or better rate, preventing comparative effectiveness analyses of these outcomes. Information on comorbidities was also unavailable from the FH database. Furthermore, monitoring of patients in RW databases is less rigorous and subject to greater variation than in clinical trials, where patients are regularly and strictly monitored. This is especially relevant for PFS outcomes, as progression data were more likely to be missing for patients in the RW cohort than in CARTITUDE-1. To address this, the earliest of start of a new LOT or disease progression was considered a progression event in the RW cohort, as start of a new LOT may have been more reliably reported than progression. This may have overestimated the time to progression for patients in the RW cohort, as the next treatment is expected to be initiated after progression. Thus, the true benefit conferred by cilta-cel on PFS may be even greater than that observed in the present analyses. Alternatively, the modified PFS definition may have misclassified progression for patients who initiated a new LOT for reasons other than progression; however, at this late stage in a patient's treatment journey, efficacy (i.e., progressive disease) is most often the reason for initiating a new LOT. OS data quality was less of a concern, as methods to assess OS have been validated for the FH MM cohort registry [14].
Future analyses using real-world data can confirm findings from the current analyses. Such studies will also help to better understand the safety of cilta-cel versus physician's choice of treatment, which was outside of the scope of the present study. Furthermore, even though cilta-cel has superior efficacy to real-world conventional treatments in this study period, similar to other CAR-Ts, cilta-cel can only be delivered in certified specialized treatment centers. Thus, real-world evidence can also provide insight into the referral patterns related to CAR-T therapies.

CONCLUSION
The present study assessed the comparative effectiveness of cilta-cel versus treatments received by a similar cohort of patients in RW clinical practice. Cilta-cel demonstrated statistically and clinically superior results for all outcomes studied (PFS, TTNT, and OS), and these were robust across a range of sensitivity analyses. Based on these results, cilta-cel offers substantial clinical benefits for patients with triple-class exposed RRMM compared with physician's choice of conventional treatment.