Matching-adjusted indirect treatment comparison of liso-cel versus axi-cel in relapsed or refractory large B cell lymphoma

Abstract

Background

In the absence of randomized studies directly comparing chimeric antigen receptor T cell therapies, this study used matching-adjusted indirect comparisons (MAIC) to evaluate the comparative efficacy and safety of lisocabtagene maraleucel (liso-cel) versus axicabtagene ciloleucel (axi-cel) in patients with relapsed or refractory large B cell lymphoma (LBCL).

Methods

Primary data sources included individual patient data from the TRANSCEND NHL 001 study (TRANSCEND [NCT02631044]; N = 256 for efficacy set, N = 269 for safety set) for liso-cel and summary-level data from the ZUMA-1 study (NCT02348216; N = 101 for efficacy set, N = 108 for safety set) for axi-cel. Inter-study differences in design, eligibility criteria, baseline characteristics, and outcomes were assessed and aligned to the extent feasible. Clinically relevant prognostic factors were adjusted in a stepwise fashion by ranked order. Since bridging therapy was allowed in TRANSCEND but not ZUMA-1, the initial efficacy and safety analyses included bridging therapy use as a matching factor (TRANSCEND patients who received bridging therapy were removed). Subsequent sensitivity analyses excluded this matching factor.

Results

The initial analysis showed similar MAIC-weighted efficacy outcomes between TRANSCEND and ZUMA-1 for overall and complete response rates (odds ratio [95% confidence interval (CI)], 1.40 [0.56–3.49] and 1.21 [0.56–2.64], respectively) and for overall survival and progression-free survival (hazard ratio [95% CI], 0.81 [0.44–1.49] and 0.95 [0.58–1.57], respectively). MAIC-weighted safety outcomes favored liso-cel, with significantly lower odds of all-grade and grade ≥ 3 cytokine release syndrome (odds ratio [95% CI], 0.03 [0.01–0.07] and 0.08 [0.01–0.67], respectively) and study-specific neurological events (0.16 [0.08–0.33] and 0.05 [0.02–0.15], respectively). Efficacy and safety outcomes remained similar in sensitivity analyses, which did not include use of bridging therapy as a matching factor.

Conclusions

After matching and adjusting for clinically relevant prognostic factors, liso-cel demonstrated comparable efficacy and a more favorable safety profile compared with axi-cel in patients with third- or later-line relapsed or refractory LBCL.

Trial registration: NCT02631044 and NCT02348216

Background

Non-Hodgkin lymphoma (NHL) is a common form of cancer, accounting for 2.8% of new cases and 2.6% of deaths in 2018 from cancer worldwide [1]. Diffuse large B cell lymphoma (DLBCL) is the most common NHL subtype and is included in a group of aggressive lymphomas referred to as large B cell lymphomas (LBCL). DLBCL accounts for approximately 30% of all lymphoid malignancies and 37% of all B cell lymphomas worldwide [2, 3]. Many patients achieve durable remission with chemoimmunotherapy, yet up to 50% eventually have relapsed or refractory (R/R) disease [4]. Treatment options are limited for patients with R/R LBCL and treatment outcomes in the third- or later-line setting are historically poor [5, 6].

Chimeric antigen receptor (CAR) T cell therapies have shown clinical activity in patients with R/R LBCL, with objective response rates (ORR) and complete response rates (CRR) ranging from 52 to 82% and from 40 to 54%, respectively [7,8,9]. The following 3 CAR T cell therapies are available for third- or later-line treatment of LBCL: tisagenlecleucel, axicabtagene ciloleucel (axi-cel), and lisocabtagene maraleucel (liso-cel) [10,11,12,13,14]. All 3 products are CD19-directed CAR T cell therapies; liso-cel and tisagenlecleucel have a 4-1BB costimulatory domain, whereas axi-cel has a CD28 costimulatory domain [7, 8, 15]. Liso-cel is a defined composition CAR T cell product administered at equal target doses of CD8⁺ and CD4⁺ CAR⁺ T cells at the planned dose [7].

As no head-to-head studies comparing CD19-directed CAR T cell therapies have been conducted to inform treatment decisions, methods for indirect treatment comparisons can be used to guide comparison between individual studies. We conducted 2 separate indirect treatment comparisons for liso-cel versus axi-cel and tisagenlecleucel; the comparison with tisagenlecleucel will be reported separately. The analysis reported here used matching-adjusted indirect comparisons (MAIC) to estimate the comparative efficacy and safety of liso-cel from the TRANSCEND NHL 001 study (TRANSCEND) relative to axi-cel using data from the ZUMA-1 study for third- or later-line treatment of R/R LBCL.

Methods

Data sources

The primary data source for liso-cel was individual patient data (IPD), which the authors had access to, from the TRANSCEND study (data cutoff: August 12, 2019). TRANSCEND (NCT02631044) is a phase 1, single-arm, multicenter, seamless design study evaluating the efficacy and safety of liso-cel as third- or later-line treatment for patients with R/R LBCL [7]. Patients with LBCL, which included DLBCL not otherwise specified (de novo or transformed from follicular lymphoma [FL] or other indolent lymphomas), high-grade B cell lymphoma with MYC and BCL2 and/or BCL6 rearrangements with DLBCL histology, primary mediastinal B cell lymphoma, and FL grade 3B, were eligible if they had R/R positron emission tomography (PET)–positive disease after ≥ 2 lines of prior therapy, including treatment with an anti-CD20–targeted agent and anthracycline. Other inclusion criteria included an Eastern Cooperative Oncology Group performance status (ECOG PS) of 0–2, and adequate organ function. Patients with secondary central nervous system (CNS) involvement and prior autologous or allogeneic hematopoietic stem cell transplantation (auto-HSCT; allo-ASCT) were permitted. Patients with primary CNS lymphoma were excluded. Primary endpoints were adverse events (AE) and ORR as assessed by an independent review committee (IRC). Secondary endpoints included CRR, duration of response, progression-free survival (PFS) per IRC assessment, and overall survival (OS).

A literature review identified the ZUMA-1 study as the key source of efficacy and safety data for axi-cel as a treatment option for R/R LBCL. The primary source of data for axi-cel was a peer-reviewed publication reporting summary-level data from the ZUMA-1 study (data cutoff: August 11, 2018) [16]. Secondary data sources for ZUMA-1, which included the YESCARTA European Medicines Agency public assessment report, summary of product characteristics, and Biologics License Application clinical review memorandum [17,18,19], were consulted only when data were unavailable in the primary publication or when additional clarity was required. ZUMA-1 (NCT02348216) was a phase 1/2, single-arm, multicenter, registrational study evaluating the efficacy and safety of axi-cel in patients with R/R LBCL [16]. Patients with histologically confirmed DLBCL, primary mediastinal B cell lymphoma, or DLBCL transformed from FL were eligible if they had R/R disease after 2 systemic lines of therapy, chemotherapy-refractory disease (no response to last line of therapy and/or refractory after auto-HSCT), had previously received an anti-CD20 monoclonal antibody-containing regimen and an anthracycline-containing chemotherapy, and had an ECOG PS of 0–1. Patients were excluded if they had secondary CNS lymphoma or prior allo-HSCT. The primary endpoints were safety for phase 1 and ORR per investigator’s assessment for phase 2. Key secondary endpoints were ORR per IRC, PFS, OS, and duration of response. Among 108 patients who received axi-cel, 101 were evaluable in the phase 2 study.

The TRANSCEND LBCL efficacy set (N = 256), defined as liso-cel-treated patients who had confirmed PET-positive disease before liso-cel administration, was used for liso-cel efficacy outcomes (ORR, CRR, PFS, and OS) [7]. The ZUMA-1 phase 2 modified intention-to-treat set (N = 101), defined as patients enrolled in phase 2 (ie, had PET-positive disease at baseline) who received axi-cel, was used for axi-cel efficacy outcomes [8, 16]. For safety outcomes, the TRANSCEND LBCL-treated set (N = 269), defined as all patients with LBCL who received liso-cel, was used for liso-cel and the ZUMA-1 phase 1/2 safety analysis set, defined as all patients who received axi-cel in both phases 1 and 2, was used for axi-cel (N = 108) [7, 16].

Study comparisons

Comparison of TRANSCEND versus ZUMA-1 study designs, eligibility criteria, and baseline characteristics showed sufficient inter-study similarities to allow for comparison (Table 1). However, imbalances in patient populations enrolled between studies necessitated a MAIC analysis to reduce bias when indirectly comparing liso-cel to axi-cel.

Table 1 Study design characteristics and inclusion criteria for TRANSCEND versus ZUMA-1

Study designs

Both studies were multicenter, single-arm studies, and both used similar lymphodepleting chemotherapy (fludarabine and cyclophosphamide; Table 1), although the doses were higher in the ZUMA-1 study. The studies differed in their allowance of bridging therapy use (ie, not allowed in ZUMA-1) and CAR T cell dosage (defined composition and fixed dose in TRANSCEND, weight-based dose in ZUMA-1). In addition, the liso-cel dose is based on post-thaw measures of viable cell count and CAR⁺ T cell frequency that is used to calculate the corresponding required dose volume to be infused at the administration site, whereas the full weight-based dose of axi-cel is administered, regardless of post-thaw recovery. Eligibility criteria differed between trials for prior allo-HSCT, secondary CNS involvement, degree of impaired renal and cardiac function, and minimum hematologic parameter requirements (Table 1). Notably, the studies also differed in their enrollment processes; ZUMA-1 did not permit enrollment and leukapheresis unless a CAR T cell manufacturing slot was available, whereas TRANSCEND allowed bridging therapy after leukapheresis at the discretion of the treating clinician during the liso-cel manufacturing process [7, 20]. The median (range) time from leukapheresis to product availability was 24 (17–51) days in TRANSCEND and 17 (14–51) days in ZUMA-1 (7, 17).

Patient characteristics

Of 18 baseline patient characteristics reported in both studies, definitions (tumor burden, disease histology, number of lines of prior therapy, and R/R to last therapy), categorization (International Prognostic Index score), and/or minimum/maximum thresholds (creatinine clearance [CrCl] before lymphodepleting chemotherapy, left ventricular ejection fraction [LVEF] at screening, and absolute lymphocyte count before leukapheresis) differed between the studies for 8 patient characteristics (Table 2). For safety outcomes, 11 baseline patient characteristics were considered, including the following 3 unique factors for safety: baseline grade ≥ 3 anemia, neutropenia, and thrombocytopenia. Differences between studies were aligned by reclassifying or recalculating the corresponding variables within the TRANSCEND IPD to match classifications or definitions reported in ZUMA-1. Definitions and/or categorizations for the remaining patient characteristics were similar and, therefore, did not require alignment.

Table 2 Comparison of clinical factors and SMDs before and after MAIC for OS

Definitions of outcome measures

Four efficacy outcomes and 10 selected safety outcomes were assessed in this analysis. The efficacy outcomes evaluated included ORR, CRR, PFS, and OS. For ORR and CRR assessments, although the primary endpoint for ZUMA-1 was investigator-assessed ORR, both studies reported response outcomes by IRC based on PET/computed tomography scans, which were used for comparisons; however, ZUMA-1 used the revised International Working Group criteria [21], whereas TRANSCEND used the more recent Lugano classification [22]. As PET-based assessment was used in both studies and any uncertain responses would be subject to additional testing, inter-study differences in ORR and CRR assessment criteria were anticipated to be minimal. Definitions for PFS and OS were similar. For PFS, both studies captured investigator- and IRC-assessed events and utilized the same censoring rules.

The safety outcomes evaluated CAR T cell AEs of interest, including all-grade and grade ≥ 3 cytokine release syndrome (CRS) by Lee et al. [23] criteria and neurological events (NE), including aphasia and encephalopathy; grade ≥ 3 infections; all-grade hypogammaglobulinemia; and grade ≥ 3 prolonged cytopenia, including anemia, neutropenia, and thrombocytopenia. Both studies had a similar treatment-emergent AE reporting window (90 days after infusion in TRANSCEND and 92 days after infusion in ZUMA-1) and had similar definitions for CRS, infections, and hypogammaglobulinemia, whereas definitions for study-specific NEs, including grouped terms of encephalopathy and aphasia, and prolonged cytopenias varied between studies. Despite potential differences in study-specific NE definitions, NEs reported by each study were investigator-identified neurological AEs considered clinically relevant and related to CAR T cell therapy. In both studies, infections were defined using high-level group terms per the Infections and Infestations system organ class and hypogammaglobulinemia was defined using grouped preferred terms assessed as AEs by investigators. Prolonged cytopenia was defined as grade ≥ 3 anemia, neutropenia, or thrombocytopenia not resolved by Day 29 (TRANSCEND) or by Day 30 (ZUMA-1); however, reporting was based on laboratory values in TRANSCEND and on AEs reported by the investigator in ZUMA-1. Therefore, MAIC was not conducted for prolonged cytopenia and an unadjusted, side-by-side descriptive comparison was done to describe reported data. However, grade ≥ 3 prolonged anemia, neutropenia, and thrombocytopenia were individually reported as AEs by investigators in both studies and were therefore compared via MAIC.

No outcomes were rederived from TRANSCEND for alignment with ZUMA-1.

Statistical analysis

As no common comparator was available, unanchored MAICs were conducted to determine the relative efficacy and safety of liso-cel (TRANSCEND) versus axi-cel (ZUMA-1) [24]. Generalized linear models were used to estimate odds ratios (OR) for binary outcomes (ie, ORR, CRR, and safety outcomes), and Cox proportional hazards models were used to estimate hazard ratios (HR) for time-to-event outcomes (ie, OS and PFS).

Relevant clinical factors for matching and adjusting were identified and ranked via a literature search followed by input from a panel of 5 external clinical experts (Canada, France, Germany, UK, and US) with CAR T cell therapy experience. A final ranked list of clinical factors was derived per outcome by evaluating the strength of association of each clinical factor to each outcome endpoint using statistical random forest models (efficacy endpoints only) and incorporating clinical expert rankings to create an evidence-informed clinical ranking [25,26,27]. A single list of ranked factors was used for all safety outcomes, based on a literature search and consensus among clinical experts.

For each comparison, patients from TRANSCEND were removed from the IPD set if they did not satisfy the eligibility criteria and treatment protocol of ZUMA-1. Specifically, patients with prior allo-HSCT, ECOG PS of 2, secondary CNS involvement, or bridging therapy use were removed from the TRANSCEND IPD set as these patients were ineligible for ZUMA-1. In addition, as FL grade 3B was not expected to differ from the other histologies with respect to safety effects, patients with FL grade 3B were removed from the TRANSCEND IPD set for efficacy outcomes only. As patients in ZUMA-1 were ineligible to receive bridging therapy, initial analyses of liso-cel versus axi-cel involved removing patients from the TRANSCEND IPD set who had received bridging therapy (ie, matching on bridging therapy use). As bridging therapy use could be potentially associated with more aggressive disease or with differences in product manufacturing time between trials, and given the lack of consensus among clinicians regarding bridging therapy use as a matching factor in these analyses, sensitivity analyses were also conducted without matching on bridging therapy use (ie, TRANSCEND patients who received bridging therapy were maintained in the analysis).

After completing the MAIC matching phase, the remaining patients from TRANSCEND were weighted using a method-of-moments propensity score algorithm, which was chosen because only summary-level data were available from ZUMA-1 and this method guarantees an exact balancing of clinical factors between comparison studies of interest. For binary endpoints, estimates of the comparative efficacy of liso-cel versus axi-cel were derived from an intercept-only logistic regression with MAIC adjustment weights. The intercept represents a prediction of the log odds of the outcome of interest if a typical patient from ZUMA-1 had received liso-cel. An estimate of the OR for liso-cel versus axi-cel was derived as the ratio between the estimated odds for liso-cel from weighted IPD and the estimated odds from summary-level data for ZUMA-1. Therefore, the relative probabilities of response or AEs between liso-cel and axi-cel are expressed and presented as ORs. For time-to-event endpoints, comparative efficacy of liso-cel versus axi-cel was estimated as HRs derived from a weighted Cox proportional hazards model with a binary treatment indicator (ie, liso-cel vs axi-cel). To fit this model under usual circumstances, IPD from both trials would be required. In place of IPD for ZUMA-1, pseudo-IPD for PFS and OS were derived by digitizing published Kaplan–Meier survival curves and using the Guyot et al. 2012 [28] approach. In turn, medians and 95% confidence intervals (CI) for ZUMA-1 were estimated from this pseudo-IPD. In the Cox regression model, TRANSCEND IPD values were assigned propensity score weights as defined above, while pseudo-IPD values for ZUMA-1 were left unweighted (weights for pseudo-observations were set to 1).

For a given set of ranked clinical factors, separate MAICs were conducted sequentially, adjusting for 1 additional variable at a time in order of ranked importance. After fitting each model, the performance and suitability of each MAIC model was assessed based on the following criteria: effective sample size (ESS; a proxy for sample size when patients are weighted), distribution of patient weights, summary statistics (assessment of balance between study populations), and assumption of proportional hazards for OS and PFS. Balance was assessed using the absolute value of the standardized mean difference (SMD) for each factor. An SMD ≥ 0.10 was considered indicative of potentially important imbalances between comparisons [29]. For a given factor, a reduction in the SMD after matching and adjusting signified a reduction in imbalance between studies.

Initial analyses of efficacy outcomes were conducted for patients who did not receive bridging therapy (ie, matched for bridging therapy use). Overall, 10 clinical factors were included in this analysis. First, the 5 clinical factors that related to study eligibility and treatment protocol were used as matching criteria (bridging therapy use, disease histology, ECOG PS, secondary CNS involvement, and prior allo-HSCT). An additional 5 clinical factors, which varied across outcomes based on evidence-informed clinical rankings (ORR: tumor burden, R/R status to last therapy, prior auto-HSCT, disease stage, and CrCl; CRR: tumor burden, R/R status to last therapy, bulky disease, prior auto-HSCT, and extranodal disease; PFS: tumor burden, International Prognostic Index [IPI] score, R/R status to last therapy, bulky disease, and CrCl; OS: tumor burden, IPI score, R/R status to last therapy, bulky disease, and age), were then adjusted to reduce residual imbalances between studies among matched patients (Additional file 1: Table S1). Two sensitivity analyses were also performed. Sensitivity analysis 1 (SA1) was the same as the initial analysis except that bridging therapy use was removed as a matching factor (ie, included patients from TRANSCEND who received bridging therapy). This analysis was conducted to help assess the effect of bridging therapy on results, recognizing that bridging therapy use could be related to other factors associated with aggressive disease or the time required to manufacture the cellular product. Sensitivity analysis 2 (SA2) removed bridging therapy use as a matching factor and adjusted for additional factors (number of prior therapies, sex, absolute lymphocyte count, and LVEF). This analysis aimed to assess the effect of balancing additional factors while retaining a reasonable ESS upon excluding the bridging therapy factor.

Initial analyses of safety outcomes were conducted for patients who did not receive bridging therapy. Overall, 9 clinical factors were included: 4 factors related to study eligibility criteria and treatment protocol as matching criteria (secondary CNS involvement, bridging therapy use, ECOG PS, and prior allo-HSCT) and 5 additional factors that were adjusted to minimize differences between patients (baseline grade ≥ 3 anemia, neutropenia, and thrombocytopenia; pre-lymphodepleting chemotherapy tumor burden; and number of prior lines of therapy). A sensitivity analysis was performed, which was the same as the initial analysis except that bridging therapy use was removed from the list of matching factors (ie, included patients from TRANSCEND who received bridging therapy).

All analyses were conducted using R (R Core Team, Vienna, Austria), based on the code outlined in the National Institute for Health and Care Excellence Evidence Synthesis Technical Support Document Series [24].

Results

Clinical factors before and after matching and adjusting

For efficacy and safety outcomes, clinical factor comparison before matching demonstrated that some factors were similar between TRANSCEND and ZUMA-1 studies (Table 2; Additional file 1: Tables S2–S5). For efficacy outcomes, notable differences (SMDs ≥ 0.1) were observed for age, IPI score, ECOG PS, disease stage, tumor burden, secondary CNS lymphoma, extranodal disease, bulky disease, disease histology, number of lines of prior therapy, prior allo-HSCT and auto-HSCT, bridging therapy use, R/R to last therapy, CrCl < 60 mL/min, and LVEF < 50%. In initial and sensitivity analyses, matching and adjusting patients from TRANSCEND to ZUMA-1 produced substantial improvements in the balance of clinical factors between studies, with SA2, which included the greatest number of factors, showing the largest improvements in balance (94% of 18 efficacy factors with SMDs < 0.1). As both sensitivity analyses included patients who received bridging therapy, which was not permitted in ZUMA-1, substantial imbalances in this factor remained after matching and adjusting on other factors. Similarly, for safety outcomes, notable differences were observed for age, ECOG PS, secondary CNS lymphoma, number of lines of prior therapy, prior allo-HSCT and auto-HSCT, bridging therapy use, and baseline grade ≥ 3 anemia, neutropenia, and thrombocytopenia. After conducting the MAIC, 82% of 11 safety factors had SMDs < 0.1, indicating imbalance between studies was minimized and that the majority of factors important to safety outcomes were well balanced after matching and adjusting.

Efficacy outcomes

Response rates

Overall, the MAIC results showed no statistically significant difference in the probability of response (ORR, CRR) between liso-cel and axi-cel (Table 3). Before conducting MAICs, unadjusted ORRs were similar for liso-cel (72.7% [N = 256]) versus axi-cel (74.3% [N = 101]; OR 0.92; 95% confidence interval [CI] 0.54–1.55; P = 0.753). In the initial analysis that matched and adjusted for 10 factors, including bridging therapy use (ie, TRANSCEND patients who received bridging therapy were removed from the data set), the ORR for liso-cel increased to 80.1% (ESS = 42.1) and the probability of overall response between liso-cel and axi-cel remained similar (OR 1.40; 95% CI 0.56–3.49; P = 0.476). In both SA1 (ESS = 150.3), which adjusted for the same factors except bridging therapy use, and SA2 (ESS = 98.9), which adjusted for 17 factors, the ORR for liso-cel was 71.2% and 70.9%, respectively, and the probability of response remained similar between treatments.

Table 3 Summary of ORR and CRR comparisons between the TRANSCEND and ZUMA-1 studies

Likewise, unadjusted CRRs were similar for liso-cel (53.1%) versus axi-cel (54.5%; OR 0.95; 95% CI 0.60–1.50; P = 0.815; Table 3). In the initial analysis, the CRR for liso-cel increased to 59.2% (ESS = 39.6) and the probability of complete response between liso-cel and axi-cel remained similar (OR 1.21; 95% CI 0.56–2.64; P = 0.630). In SA1 and SA2, the CRR for liso-cel was 48.2% and 49.5%, respectively, and probability of complete response remained similar between treatments.

Survival outcomes

The MAIC results showed no statistically significant difference in the rate of disease progression or mortality (PFS) or the rate of mortality (OS) between liso-cel and axi-cel. In unmatched and unadjusted comparisons, median (95% CI) PFS was similar for liso-cel (4.1 [3.0–6.0] months) versus axi-cel (5.8 [3.4–15.0] months; HR 1.20; 95% CI 0.90–1.59; P = 0.219; Table 4). In the initial analysis, the median PFS for liso-cel increased to 6.3 months (ESS = 40.0; 95% CI 3.0–not reached [NR]), and the HR remained similar for liso-cel versus axi-cel (HR 0.95; 95% CI 0.58–1.57; P = 0.847; Fig. 1). In SA1 (ESS = 151.4) and SA2 (ESS = 98.9), the median (95% CI) PFS for liso-cel was 3.5 (3.0–5.9) months and 5.0 (3.0–9.2) months, respectively, and HR (95% CI) remained similar between treatments (SA1, 1.30 [0.96–1.77]; P = 0.095; SA2, 1.16 [0.81–1.66]; P = 0.408).

Table 4 Summary of PFS and OS comparisons between the TRANSCEND and ZUMA-1 studies

Fig. 1

Kaplan–Meier curves for progression-free survival. Kaplan–Meier curves are shown for the initial and sensitivity analysis comparisons with liso-cel versus axi-cel for infused patients. Kaplan–Meier curves for the initial analysis, which matched and adjusted for 10 factors, including bridging therapy use, demonstrated similar cumulative probabilities of progression-free survival for liso-cel and axi-cel. Similar results were observed for sensitivity analysis 1, which was the same as the initial analysis except that bridging therapy use was removed as a matching factor. Axi-cel, axicabtagene ciloleucel; CI, confidence interval; ESS, effective sample size; liso-cel, lisocabtagene maraleucel; N, sample size; NR, not reached

In unmatched and unadjusted comparisons, median (95% CI) OS was 21.1 (13.3–NR) months for liso-cel, with a median follow-up of 17.5 months, and was NR (12.8–NR) for axi-cel and median follow-up was not reported for the OS endpoint, with no significant difference based on HR (HR 1.13; 95% CI 0.81–1.58; P = 0.457; Table 4). In the initial analysis, the median OS for liso-cel was NR (ESS = 38.3; 95% CI 11.6–NR), and the survival rate remained similar for liso-cel and axi-cel (HR 0.81; 95% CI 0.44–1.49; P = 0.506; Fig. 2). In SA1 (ESS = 152.6) and SA2 (ESS = 98.9), the median (95% CI) OS for liso-cel was 19.9 (12.1–NR) months and 21.1 (14.4–NR) months, respectively, and the HR (95% CI) remained similar (SA1, 1.15 [0.80–1.65]; P = 0.454; SA2, 1.04 [0.70–1.56]; P = 0.838).

Fig. 2

Kaplan–Meier curves for overall survival. Kaplan–Meier curves are shown for the initial and sensitivity analysis comparisons with liso-cel versus axi-cel for infused patients. Kaplan–Meier curves for the initial analysis, which matched and adjusted for 10 factors, including bridging therapy use, demonstrated similar cumulative probabilities of OS for liso-cel and axi-cel. Similar results were observed for sensitivity analysis 1, which was the same as the initial analysis except that bridging therapy use was removed as a matching factor. Axi-cel, axicabtagene ciloleucel; CI, confidence interval; ESS, effective sample size; liso-cel, lisocabtagene maraleucel; N, sample size; NR, not reached

Safety outcomes

Safety analyses were conducted using the LBCL-treated set from TRANSCEND and the phase 1/2 safety analysis set from ZUMA-1. In the initial safety analysis that matched and adjusted for 9 factors, including bridging therapy use, liso-cel generally had a lower probability of AEs of interest than axi-cel (Table 5). Specifically, liso-cel was associated with a statistically significantly lower probability of all-grade and grade ≥ 3 CRS, all-grade and grade ≥ 3 NEs per study protocol, all-grade and grade ≥ 3 encephalopathy and all-grade aphasia that were reported as NEs, grade ≥ 3 infections, grade ≥ 3 prolonged anemia, grade ≥ 3 prolonged thrombocytopenia, and all-grade hypogammaglobulinemia. Similar findings were reported in the sensitivity analysis that adjusted for all of the same factors except bridging therapy use. Differences were that the lower probability of grade ≥ 3 prolonged neutropenia with liso-cel was now statistically significant, and the following were no longer statistically significantly lower with liso-cel: grade ≥ 3 prolonged anemia, grade ≥ 3 prolonged thrombocytopenia, and all-grade hypogammaglobulinemia.

Table 5 Summary of MAIC results for safety outcomes of liso-cel versus axi-cel

CRS and NEs

Unmatched and unadjusted rates of CRS were lower for liso-cel versus axi-cel, with a statistically significantly lower probability of all-grade (OR 0.06; 95% CI 0.03–0.12; P < 0.001) and grade ≥ 3 (OR 0.18; 95% CI 0.07–0.50; P = 0.001) CRS with liso-cel (Table 5). After matching and adjusting for 9 clinical factors, the probability of all-grade (OR 0.03; 95% CI 0.01–0.07; P < 0.001) and grade ≥ 3 (OR 0.08; 95% CI 0.01–0.67; P = 0.019) CRS remained statistically significantly lower with liso-cel versus axi-cel. In the sensitivity analysis, the probability of all-grade (OR 0.06; 95% CI 0.03–0.13; P < 0.001) and grade ≥ 3 (OR 0.16; 95% CI 0.06–0.47; P = 0.001) CRS remained significantly lower with liso-cel versus axi-cel.

Unmatched and unadjusted rates of study-specific NEs were lower for liso-cel versus axi-cel, with statistically significantly lower probability of all-grade (OR 0.21; 95% CI 0.13–0.34; P < 0.001) and grade ≥ 3 (OR 0.23; 95% CI 0.13–0.41; P < 0.001) NEs with liso-cel (Table 5). After matching and adjusting, the probability of all-grade (OR 0.16; 95% CI 0.08–0.33; P < 0.001) and grade ≥ 3 (OR 0.05; 95% CI 0.02–0.15; P < 0.001) NEs remained statistically significantly lower with liso-cel versus axi-cel. In the sensitivity analysis, the probability of all-grade (OR 0.21; 95% CI 0.13–0.35; P < 0.001) and grade ≥ 3 (OR 0.20; 95% CI 0.11–0.37; P < 0.001) NEs remained statistically significantly lower with liso-cel versus axi-cel.

Study-specific NEs of encephalopathy and aphasia were analyzed as group terms. The probability of all-grade and grade ≥ 3 encephalopathy (OR 0.42; 95% CI 0.19–0.91; P = 0.028 and OR 0.05; 95% CI 0.01–0.18; P < 0.001, respectively) as well as all-grade aphasia (OR 0.36; 95% CI 0.13–1.00; P = 0.049) was statistically significantly lower with liso-cel versus axi-cel after matching and adjusting for 9 clinical factors (Table 5). In the sensitivity analysis, the probability of all-grade and grade ≥ 3 encephalopathy (OR 0.44; 95% CI 0.26–0.73; P = 0.002 and OR 0.20; 95% CI 0.09–0.41; P < 0.001, respectively) as well as all-grade aphasia (OR 0.51; 95% CI 0.26–1.00; P = 0.049) remained statistically significantly lower with liso-cel versus axi-cel. Unmatched and unadjusted rates of grade ≥ 3 aphasia were lower with liso-cel (1.9%) versus axi-cel (7.4%); however, after matching on bridging therapy use, all patients with grade ≥ 3 aphasia in TRANSCEND were excluded (ie, rate of 0% for liso-cel). Thus, further adjustments based on patient weights were not estimable. In the sensitivity analysis that did not match on bridging therapy use, the probability of grade ≥ 3 aphasia was statistically significantly lower with liso-cel versus axi-cel (OR 0.19; 95% CI 0.06–0.61; P = 0.005).

Use of tocilizumab and corticosteroids to manage CRS and/or study-specific NEs was compared between studies. Notably, for CRS and/or NE management, the overall use of tocilizumab was lower in TRANSCEND than in ZUMA-1 (20% vs 43%, respectively), as was use of corticosteroids (21% vs 27%, respectively).

Additional AEs of interest

The probability of grade ≥ 3 infections was statistically significantly lower with liso-cel versus axi-cel before (OR 0.40; 95% CI 0.23–0.70; P = 0.001) and after (OR 0.19; 95% CI 0.07–0.49; P = 0.001) matching and adjusting (Table 5). Unmatched and unadjusted rates of all-grade hypogammaglobulinemia were similar for liso-cel and axi-cel, with no statistically significant difference in the probability of the event between treatments. After matching and adjusting, the probability of all-grade hypogammaglobulinemia was statistically significantly lower for liso-cel versus axi-cel (OR 0.37; 95% CI 0.14–0.97; P = 0.043). In the sensitivity analysis, the probability of grade ≥ 3 infections remained significantly lower with liso-cel versus axi-cel (OR 0.36; 95% CI 0.20–0.65; P = 0.001), while the probability of all-grade hypogammaglobulinemia was no longer statistically significantly lower for liso-cel (OR 0.71; 95% CI 0.37–1.37; P = 0.311).

An unadjusted comparison between TRANSCEND and ZUMA-1 data was conducted for prolonged cytopenia (grade ≥ 3 anemia, neutropenia, or thrombocytopenia not resolved by Day 29 [TRANSCEND] or by Day 30 [ZUMA-1]). Grade ≥ 3 prolonged cytopenia occurred in 38% of patients in ZUMA-1 (per investigator assessment of AEs) and in 37% of patients in TRANSCEND (per laboratory assessment).

When comparing investigator-reported AEs of grade ≥ 3 prolonged anemia, neutropenia, and thrombocytopenia (at Day 29 for TRANSCEND or Day 30 for ZUMA-1), the probability of an event was lower for liso-cel versus axi-cel. Unmatched and unadjusted probability of AEs of grade ≥ 3 prolonged anemia was not statistically significantly lower for liso-cel versus axi-cel (OR 0.56; 95% CI 0.25–1.24; P = 0.153). After matching and adjusting, the probability of grade ≥ 3 prolonged anemia was statistically significantly lower for liso-cel versus axi-cel in the initial analysis (OR 0.04; 95% CI 0.00–0.32; P = 0.002) but not the sensitivity analysis (OR 0.46; 95% CI 0.20–1.05; P = 0.064).

Unmatched and unadjusted probability of AEs of grade ≥ 3 prolonged neutropenia was statistically significantly lower for liso-cel versus axi-cel (OR 0.47; 95% CI 0.27–0.82; P = 0.007). After matching and adjusting, the probability of grade ≥ 3 prolonged neutropenia was no longer statistically significantly lower for liso-cel versus axi-cel in the initial analysis (OR 0.45; 95% CI 0.19–1.07; P = 0.072) but was in the sensitivity analysis (OR 0.52; 95% CI 0.29–0.91; P = 0.022).

Unmatched and unadjusted probability of AEs of grade ≥ 3 prolonged thrombocytopenia was not statistically significantly different for liso-cel versus axi-cel (OR 0.67; 95% CI 0.39–1.15; P = 0.143). After matching and adjusting, the probability of grade ≥ 3 prolonged thrombocytopenia was statistically significantly lower for liso-cel versus axi-cel in the initial analysis (OR 0.23; 95% CI 0.10–0.57; P = 0.001) but not the sensitivity analysis (OR 0.61; 95% CI 0.35–1.08; P = 0.090).

Discussion

Results from this MAIC of liso-cel versus axi-cel showed that liso-cel had comparable efficacy and a favorable safety profile compared with axi-cel. The feasibility assessment identified 18 clinical factors reported in both TRANSCEND and ZUMA-1 that were available for adjustment and determined that imbalances between studies were mostly small to moderate and suitable for MAIC-based methodology. Therefore, it was determined that MAIC would achieve reductions in between-study imbalances without significant loss in ESS. MAICs were deemed feasible for 4 efficacy outcomes and 9 safety outcomes.

The initial analysis that matched and adjusted for 10 clinical factors (including bridging therapy use) showed that liso-cel and axi-cel have similar efficacy measured as ORR (OR 1.40 [95% CI 0.56–3.49]), CRR (OR 1.21 [95% CI 0.56–2.64]), PFS (HR 0.95 [95% CI 0.58–1.57]), and OS (HR 0.81 [95% CI 0.44–1.49]) as axi-cel. Similar results were observed in sensitivity analyses that excluded bridging therapy use as a matching factor and adjusted for additional potentially prognostic clinical factors.

Adjusted safety outcomes showed statistically significantly lower rates of CRS, study-specific NEs (including those of encephalopathy and aphasia), and grade ≥ 3 infections with liso-cel versus axi-cel. There are several factors that could contribute to the more favorable safety profile observed with liso-cel, including the 4-1BB co-stimulatory domain, which is reportedly associated with a lower incidence of CRS and NEs than CD28-containing construct [30]. In addition, as variability in total and CD8⁺ CAR T cells has been associated with increased toxicity, the administration of liso-cel at a defined composition with consistent total and relative doses of CD8⁺:CD4⁺ CAR⁺ T cells may also contribute to reduced toxicity [31, 32].

Furthermore, as reported per study, tocilizumab and corticosteroid use for management of CRS and/or NEs was also lower in TRANSCEND than in ZUMA-1. Although these interventions could potentially impact the severity of CRS and/or NEs, they would not affect the overall incidence of all-grade CRS or NEs, as they are administered after the onset of the event. As use of tocilizumab and corticosteroids was lower in TRANSCEND, this would not account for the lower incidence of all-grade and lower probability of grade ≥ 3 events associated with liso-cel. Additionally, it is possible that frequency of use of tocilizumab and corticosteroids may have changed over the course of the studies as physicians gained experience treating and managing patients with these AEs. Although we were unable to evaluate this in the analysis, changes in the rates of grade ≥ 3 CRS or NEs over time, as reported per study from early interim analyses to primary data cuts, appear to have been minimal (TRANSCEND interim analysis based on 91 patients versus primary data cut based on 269 patients: grade ≥ 3 CRS was 1% versus 2%, respectively, and grade ≥ 3 NE was 12% versus 10%, respectively; ZUMA-1 interim analysis based on 51 patients versus primary data cut based on 101 patients: grade ≥ 3 CRS was 20% versus 13%, respectively, and grade ≥ 3 NE was 29% versus 28%, respectively) [7, 8, 33, 34].

Our feasibility assessment identified the use of bridging therapy as a notable difference between TRANSCEND and ZUMA-1. ZUMA-1 did not permit bridging therapy, whereas TRANSCEND allowed bridging therapy use at the discretion of the treating clinicians. Use of bridging therapy could potentially be associated with more aggressive disease or with differences between products in the manufacturing time requirement. To explore the potential impact of bridging therapy use, the initial MAIC analyses matched on bridging therapy use and additional sensitivity analyses were conducted without matching on bridging therapy use. No statistically significant differences in efficacy were observed between liso-cel and axi-cel in either analysis.

A similar MAIC was recently conducted to estimate treatment effects with axi-cel versus tisagenlecleucel based on data from ZUMA-1 and JULIET [35]. The study reported that axi-cel had a higher ORR and CRR, longer OS, higher rates of grade 1/2 CRS, and similar rates of grade ≥ 3 CRS and NEs compared with tisagenlecleucel. In the ZUMA-1 and JULIET comparison, matching for bridging therapy use was not possible because 92% of patients in JULIET received bridging therapy and it was not allowed in ZUMA-1 [9, 35]. In contrast, our study was able to match patients for bridging therapy use owing to the large number of patients in TRANSCEND contributing to the IPD (N = 269) and the fact that 110 patients remained after removing the 59% of patients in TRANSCEND who received bridging therapy [7]. Thus, a sufficient number of patients was available to conduct analyses in patients who did not receive bridging therapy (110 from TRANSCEND and 108 from ZUMA-1).

A notable strength of this study was the rigorous, multifaceted process undertaken to identify and rank-order clinically relevant factors. In addition, multiple publications were evaluated to identify the most compatible cohorts from the comparator study to those available in TRANSCEND [8, 16,17,18,19]. Alignments on variable and outcome measure definitions through recalculations or recategorizations with TRANSCEND data were made and documented wherever possible, which further helped to reduce bias in downstream estimates. Finally, sensitivity analyses were conducted to test the robustness of results presented as initial analyses within this study.

Limitations to the current analysis include those inherent to an unanchored MAIC. Despite the methodological rigor, there may be clinical factors unaccounted for in this process. Some clinical factors identified and deemed important through this process were not reported for ZUMA-1 in a format conducive for comparison or alignment with TRANSCEND, such as lactate dehydrogenase, C-reactive protein, and best response to any prior therapy. Thus, some imbalance remained between variables after matching and adjusting that could have influenced point estimates in an unknown direction and magnitude.

As our analysis found comparable efficacy and a more favorable safety profile with liso-cel compared with axi-cel, it could have implications on clinical practice. The lower incidence of CRS and later onset allows outpatient administration at centers with adequate infrastructure; this may decrease hospital utilization [7, 8, 36]. A more favorable safety profile would be advantageous in all patients as well as those with comorbidities. In addition to this indirect treatment comparison, real-world evidence comparing the safety and efficacy of CAR T cell therapies could provide additional insights to help inform clinical decisions. Although analyses of real-world data for liso-cel are not currently available, real-world evidence for axi-cel and tisagenlecleucel has shown that efficacy and safety outcomes in clinical practice are consistent with those reported in clinical studies [37, 38]. In a post-marketing study of axi-cel in 1001 patients with R/R LBCL and ≥ 6 months of follow-up, the best ORR was 70% and CRR was 53%; CRS by Lee 2014 criteria of any grade was reported in 83% of patients (grade ≥ 3: 10%) and NEs (defined as immune effector cell-associated neurotoxicity syndrome) of any grade were reported in 57% (grade ≥ 3: 26%) [38]. These outcomes are consistent with those reported in ZUMA-1 (ORR: 82%; CRR: 54%; any-grade CRS: 93%; grade ≥ 3 CRS: 13%; any-grade NEs: 64%; grade ≥ 3 NEs: 28%) [8].

Conclusions

In summary, we used unanchored MAICs that leveraged IPD from TRANSCEND and summary-level data from ZUMA-1 to derive indirect comparisons while accounting for between-study differences in eligibility criteria and baseline characteristics. Results were similar for initial and sensitivity analyses, supporting the robustness of the findings. Overall, after matching and adjusting for important clinical prognostic factors and treatment-effect modifiers, liso-cel demonstrated similar efficacy and a favorable safety profile compared with axi-cel.