Comparative efficacy of tisagenlecleucel and lisocabtagene maraleucel among adults with relapsed/refractory large B-cell lymphomas: an indirect treatment comparison

Abstract

This study compared overall survival (OS), progression-free survival (PFS), complete response rate (CRR), and overall response rate (ORR) of tisagenlecleucel (tisa-cel) and lisocabtagene maraleucel (liso-cel) in relapsed or refractory large B-cell lymphomas (r/r LBCL). Using matching-adjusted indirect comparison (MAIC), individual patient-level data from JULIET (tisa-cel) were weighted to match the patient population in TRANSCEND (liso-cel). The main analysis compared infused JULIET patients (N = 106) with the TRANSCEND efficacy-evaluable set (EES) (N = 256 [infused]). After adjustment, OS, PFS, and the CRR were comparable between tisa-cel and liso-cel EES patients. The estimated adjusted 2-year OS, 2-year PFS, ORR, and CRR were 45.6, 38.2, 62.9, and 47.7%, respectively, for tisa-cel vs. 43.8, 42.1, 72.7, and 53.1% for liso-cel. A scenario analysis compared JULIET patients to the TRANSCEND primary analysis set (PAS) (N = 133). ORR was significantly higher in the TRANSCEND PAS compared with matched JULIET patients, but no significant differences in CRR were observed.

Keywords:

• Lisocabtagene maraleucel
• matching-adjusted indirect comparison
• relapsed or refractory large B-cell lymphoma
• tisagenlecleucel

Introduction

Large B-cell lymphomas (LBCL) include diffuse LBCL (DLBCL), transformed follicular lymphoma (tFL), and high-grade B-cell lymphoma (HGBCL) [1]. DLBCL is the most common type of non-Hodgkin lymphoma globally, comprising ∼40% of cases [2]. In the United States (US), the annual incidence rate of new DLBCL cases is ∼5.6 per 100,000 [3]. Adults newly diagnosed with DLBCL are typically treated with combination chemotherapy and the monoclonal antibody rituximab, but ∼30–40% develop relapsing or refractory (r/r) disease after first-line treatment [4,5]. Approximately 50% of patients with r/r DLBCL respond to second-line chemotherapy, and of these, about half proceed to autologous hematopoietic stem cell transplantation (ASCT) [6–10], recognizing that not all patients are candidates for ASCT.

Before the introduction of chimeric antigen receptor (CAR) T-cell therapies, few treatment options existed for patients with r/r DLBCL who had failed two lines of therapy and were ineligible or unwilling to undergo ASCT, or who relapsed early after ASCT. For these patients, therapeutic options were primarily limited to salvage chemotherapy (SC), palliative therapy, and/or best supportive care [5,10–13]. SCHOLAR-1, a comprehensive analysis of pooled outcomes from several large studies of SC in r/r DLBCL, reported that the overall response rate (ORR) was 26% and only 7% of patients achieved CR [5].

CAR T-cell therapies for r/r LBCL use engineered T-cells prepared from patients’ T-cells to target the CD19 antigen expressed by both normal and malignant B-cells, and have transformed the treatment landscape [7,14–16]. The three CAR T-cell therapies currently approved by the US Food and Drug Administration (FDA) for r/r DLBCL include axicabtagene ciloleucel (axi-cel; which uses the CD28 costimulatory domain, approved in October 2017) [17], tisagenlecleucel (tisa-cel; which uses the 4-1BB costimulatory domain, approved in May 2018) [18], and lisocabtagene maraleucel (liso-cel; which uses the 4-1BB costimulatory domain with 1:1 co-administration of CD4:CD8 CAR T-cells, approved in February 2021) [19]. All three CAR T-cell therapies have demonstrated clinical efficacy, separately assessed in large single-arm clinical trials: the ZUMA-1 trial of axi-cel (CR rate [CRR]: 54.5%, 24-month progression-free survival [PFS]: 35.6%), the JULIET trial of tisa-cel (CRR: 41.4%, 24-month PFS: 32.7%), and the TRANSCEND trial of liso-cel (CRR: 53.1%, 24-month PFS: 42.1%) per independent review committee (IRC) assessment in all three trials [7,14,15,20].

Given the recent FDA approval of liso-cel, evidence regarding the comparative efficacy and safety of the two 4-1BB CAR-T products, tisa-cel, and liso-cel, could be valuable for treatment selection in r/r LBCL. In the absence of head-to-head randomized controlled trials, indirect treatment comparisons can be conducted. However, differences in the patient populations and study designs exist across ZUMA-1, JULIET, and TRANSCEND, which are essential to assess and adjust for when indirectly comparing effectiveness. Previously, Oluwole et al. conducted an indirect comparison of efficacy between axi-cel and tisa-cel using data from the JULIET and ZUMA-1 trials [21], but Zhang et al. countered that it was difficult to draw conclusions given the major differences between the trials that cannot be accounted for in any indirect treatment comparison [22,23]. The current study evaluated the feasibility of indirectly comparing tisa-cel and liso-cel using JULIET and TRANSCEND data, and compared efficacy outcomes among patients with r/r DLBCL using matching-adjusted indirect comparison (MAIC).

Methods

Study population and data source

Individual patient-level data from the JULIET trial of tisa-cel (ClinicalTrials.gov identifier: NCT02445248; February 2020 data cut) [7,20] and published aggregate data from the TRANSCEND NHL 001 trial of liso-cel (NCT02631044) [14] were used. Both trials assessed efficacy and safety outcomes of CAR-T therapies in patients with r/r DLBCL (not otherwise specified), tFL, and HGBCL. TRANSCEND also included patients with primary mediastinal large B-cell lymphoma (PMBCL), DLBCL transformed from indolent lymphomas other than FL, and FL grade 3B. Both trials allowed the use of bridging treatment during CAR T-cell manufacturing. While all TRANSCEND patients received fludarabine-based lymphodepletion chemotherapy (LDC), JULIET permitted the use of bendamustine as LDC in patients who could not tolerate fludarabine and/or cyclophosphamide (Flu/Cy) and the omission of LDC if the patient met certain criteria (i.e. a white blood cell count ≤1000 cells/uL within one week before infusion).

The feasibility assessments identified some differences in patient population and study design between JULIET and TRANSCEND (Supplemental Table 1). Specifically, TRANSCEND included a broader patient population compared to JULIET. In addition, retreatment with CAR-T was allowed in TRANSCEND (16 of 269 treated patients were re-treated) but not in JULIET, and the grading and management of cytokine release syndrome (CRS) differed between the trials. Despite these differences, an indirect treatment comparison of efficacy outcomes between tisa-cel and liso-cel, bearing limitations, was deemed feasible.

Of the 115 patients infused with tisa-cel in JULIET, 106 were included in the current study after excluding eight patients who did not receive LDC and one patient with large cell neuroendocrine carcinoma who was misclassified as having LBCL. In the primary analysis, JULIET patients not receiving LDC were excluded, since all TRANSCEND patients received LDC. Among the 269 patients who received liso-cel in TRANSCEND, 256 were included in the efficacy-evaluable set (EES) with efficacy results reported for comparison, as documented in Abramson et al. [14].

Outcome measures

Efficacy outcomes including ORR, CRR, PFS, and overall survival (OS) were compared between tisa-cel and liso-cel. ORR was defined as the proportion of patients who achieved CR or partial response per the Lugano Criteria (assessed by an IRC) for both JULIET and TRANSCEND [24,25]. PFS and OS were assessed from the time of infusion in both trials. Kaplan-Meier (KM) curves of OS and PFS for TRANSCEND EES patients were informed from Abramson et al. (Figure 3 in that publication) [14], and reconstructed using Engauge Digitizer software (version 4.1) [26] and the approach used in Guyot et al. [27]. PFS and OS KM curves provide a continuous measurement of treatment efficacy over time, in addition to the initial response to treatment.

Baseline factors adjusted for in the main MAICs

Potential confounders identified and confirmed by input from clinical experts were adjusted for in the MAIC. In addition, factors considered important and assessed as subgroups in Abramson et al. (e.g. use of bridging treatment, the sum of product diameter [SPD], age <65 years, and refractoriness to chemotherapy) were included [14].

Ultimately, the following factors available in both JULIET and TRANSCEND were adjusted for in the MAIC: demographics (age and sex); disease characteristics (histology, Eastern Cooperative Oncology Group Performance Status [ECOG PS], left-ventricular ejection fraction [LVEF], SPD, lactate dehydrogenase [LDH]); prior treatments (prior stem cell transplant [SCT], number of prior lines of therapy, received bridging chemotherapy); and refractory status to prior therapies (patients refractory to the last treatment or who relapsed <12 months after ASCT [labeled as chemotherapy-refractory in TRANSCEND] and never achieved CR with prior therapy).

Statistical methods (MAIC)

To balance the average baseline characteristics between JULIET and TRANSCEND patients using the MAIC approach, a propensity score model was used to assign statistical weights for JULIET patients based on the baseline factors adjusted in the MAIC [28–30]. After adjusting for differences in baseline characteristics, efficacy outcomes were compared using statistical tests that incorporated the propensity score weights. This allowed a comparison of tisa-cel and liso-cel in a balanced patient population. For ORR and CRR, Wald tests were used to compare the differences in response rates and relative risk of response. For PFS and OS, the log-rank test was used to compare the weighted KM curves of JULIET with observed KM curves of TRANSCEND; the weighted Cox proportional hazards model was used for hazard ratio (HR) estimation. Effective sample size after adjustment in the MAIC was reported for the tisa-cel-infused patients, which could indicate the extent of overlap between JULIET and TRANSCEND patients. Unadjusted comparisons without weighting were also reported.

All analyses were conducted using SAS 9.4 software (SAS Institute Inc., Cary, NC, USA) and R software version 3.4.2 (R Foundation for Statistical Computing). Statistical significance was considered at a level of 0.05.

Scenario and sensitivity analyses

In the scenario analysis, JULIET patients were compared to the TRANSCEND primary analysis set (PAS) (N = 133), a subset of the TRANSCEND EES. The TRANSCEND PAS included patients from the EES who were treated at dose level 2 (100*10⁶ CAR T-cells) and excluded patients with ECOG PS of 2, prior allogeneic SCT (allo-SCT), PMBCL, FL grade 3B, or transformation from indolent lymphoma other than FL. Because baseline data were not reported for the TRANSCEND PAS in Abramson et al., baseline summary data were assumed to be the same as those for the TRANSCEND dose level 2 population (N = 177). ORR and CRR were statistically compared, but such comparisons were not conducted for OS or PFS because KM curves were not available for the TRANSCEND PAS. The list of factors adjusted for in the scenario analysis remained the same as those used for the main analysis.

Two sensitivity analyses were performed in addition to the main results. The first sensitivity analysis matched the LDC regimen between the two trials. Specifically, only the subset of JULIET patients who received fludarabine-based LDC was included in the sensitivity analyses, as all TRANSCEND patients received fludarabine-based LDC. The factors adjusted for in the first sensitivity analysis remained the same as those in the main analysis. The second sensitivity analysis limited the adjusted prognostic factors to those that were significantly different between JULIET patients and the TRANSCEND EES at baseline, which included age, histology, ECOG PS, prior SCT, and use of bridging chemotherapy.

Results

Study population and baseline characteristics

A total of 106 tisa-cel-infused patients in JULIET and 256 patients in the TRANSCEND EES were included in the analysis. Before matching, JULIET had significantly higher proportions of patients younger than 65 years (76.4 vs. 57.8%) and with DLBCL histology (66.0 vs. 51.2%), ECOG PS of 0 (57.5 vs. 40.9%), prior SCT (47.2 vs. 34.0%), and bridging chemotherapy (89.6 vs. 58.6%; all p < 0.05) compared to the TRANSCEND EES (Table 1). After matching, baseline characteristics were balanced. The effective sample size for JULIET after weighting was N = 29, which was a 73% reduction from the original sample size.

Table 1. Baseline characteristics of JULIET patients and the TRANSCEND EES, before and after matching.

	Before matching			After matching
	JULIET (N = 106)	TRANSCEND EES (N = 256)	p	JULIET (N = 106; effective sample size = 29)	TRANSCEND EES (N = 256)	p
Age < 65	76.4%	57.8%	<0.01*	57.8%	57.8%	1.00
Male	61.3%	66.0%	0.47	66.0%	66.0%	1.00
HGBCL	14.2%	12.9%	0.88	12.9%	12.9%	1.00
DLBCL	66.0%	51.2%	<0.05*	51.2%	51.2%	1.00
ECOG PS of 0	57.5%	40.9%	<0.01*	40.9%	40.9%	1.00
40% ≤ LVEF < 50%	1.9%	4.8%	0.37	4.8%	4.8%	1.00
SPD (≥50 cm^2)	22.6%	28.3%	0.31	28.3%	28.3%	1.00
LDH (≥500 U/L)	17.0%	22.3%	0.32	22.3%	22.3%	1.00
Prior SCT	47.2%	34.0%	<0.05*	34.0%	34.0%	1.00
Prior lines of therapy ≥3	49.1%	51.7%	0.75	51.7%	51.7%	1.00
Received bridging chemotherapy	89.6%	58.6%	<0.01*	58.6%	58.6%	1.00
Refractory	73.6%	66.8%	0.25	66.8%	66.8%	1.00
Never achieved CR with prior therapy	44.3%	44.2%	1.00	44.2%	44.2%	1.00

CR: complete response; DLBCL: diffuse large B-cell lymphoma; ECOG PS: Eastern Cooperative Oncology Group Performance Status; EES: efficacy-evaluable set; HGBCL: high-grade B-cell lymphoma; LDH: lactate dehydrogenase; LVEF: left ventricular ejection fraction; SCT: stem cell transplantation; SPD: sum of product diameter.

*p < 0.05.

Comparison of efficacy

After adjusting for differences in baseline characteristics, the JULIET patients had a lower but not statistically different ORR compared to the TRANSCEND EES (62.9 and 72.7%, respectively; rate difference: −9.7%, p = 0.07) and a similar CRR (47.7 and 53.1%; −5.4%, p = 0.29) (Table 2). PFS (HR [95% confidence interval (CI)]: 1.16 [0.64, 2.09], p = 0.62) and OS (1.12 [0.62, 2.05], p = 0.71) were comparable for the two trial cohorts following adjustment (Table 3).

Table 2. Comparison of ORR and CR between JULIET patients and the TRANSCEND EES.

	Before matching				After matching
	JULIET, main cohort^a (N = 91) [A]	TRANSCEND EES (N = 256) [B]	Response difference (95% CI) [A] − [B]	p	JULIET, main cohort^a (N = 91; effective sample size = 27) [C]	TRANSCEND EES (N = 256) [D]	Response difference (95% CI) [C] − [D]	p
Best overall response
ORR (CR + PR)	57.1%	72.7%	−15.5% (−27.1, −3.9%)	<0.01*	62.9%	72.7%	−9.7% (−20.0, 0.6%)	0.07
CRR	42.9%	53.1%	−10.3% (−22.2, 1.6%)	0.09	47.7%	53.1%	−5.4% (−15.5, 4.7%)	0.29

CI: confidence interval; CRR: complete response rate; EES: efficacy-evaluable set; ORR: overall response rate; PR: partial response.

*p < 0.05.

^aORR and CRR were assessed only in the JULIET Main Cohort to be consistent with the main analyses per the JULIET trial protocol, which only included patients treated with tisa-cel from the United States manufacturing facility in Morris Plains, New Jersey.

Table 3. Comparison of OS and PFS between JULIET patients and the TRANSCEND EES, before and after matching.

	Before matching			After matching
	JULIET (N = 106)	TRANSCEND EES (N = 256)	p	JULIET (N = 106; effective sample size = 29)	TRANSCEND EES (N = 256)	p
OS
OS from infusion
Median, months (95% CI)	11.1 (6.6, 28.5)	21.1 (13.3, NR)		20.7 (6.5, NR)	21.1 (13.3, NR)
JULIET vs. TRANSCEND HR (95% CI)	1.29 (0.95, 1.76)		0.10	1.12 (0.62, 2.05)		0.71
OS rate at month 6, % (95% CI)	61.7% (51.6, 70.3%)	74.7% (68.9, 79.6%)		65.3% (53.1, 80.4%)	74.7% (68.9, 79.6%)
OS rate at month 12, % (95% CI)	48.5% (38.5, 57.8%)	57.9% (51.3, 63.8%)		55.1% (42.5, 71.4%)	57.9% (51.3, 63.8%)
OS rate at month 24, % (95% CI)	41.1% (31.4, 50.5%)	43.8% (35.2, 52.1%)		45.6% (33.2, 62.6%)	43.8% (35.2, 52.1%)
PFS
PFS from infusion
Median, months (95% CI)	3.0 (2.6, 6.2)	6.8 (3.3, 14.1)		9.0 (2.5, NR)	6.8 (3.3, 14.1)
JULIET vs. TRANSCEND HR (95% CI)	1.34 (0.99, 1.81)		0.06	1.16 (0.64, 2.09)		0.63
PFS rate at month 6, % (95% CI)	41.1% (31.1, 50.8%)	51.4% (44.6, 57.7%)		50.6% (37.6, 68.1)	51.4% (44.6, 57.7%)
PFS rate at month 12, % (95% CI)	36.3% (26.6, 46.0%)	44.1% (37.3, 50.7%)		47.4% (34.4, 65.3)	44.1% (37.3, 50.7%)
PFS rate at month 24, % (95% CI)	33.5% (23.9, 43.3%)	42.1% (35.0, 48.9%)		38.2% (25.7, 56.8)	42.1% (35.0, 48.9%)

CI: confidence interval; EES: efficacy-evaluable set; HR: hazard ratio; NR: not reached; OS: overall survival; PFS: progression-free survival.

Figure 1(a,b) illustrate the KM curves for PFS and OS, respectively, for JULIET patients and the TRANSCEND EES after matching. The estimated 2-year OS and 2-year PFS rates were 45.6 and 38.2%, respectively, in the tisa-cel arm vs. 43.8 and 42.1% in the liso-cel arm.

Figure 1. Kaplan-Meier curves of (a) PFS and (b) OS for JULIET patients and the TRANSCEND EES, after matching. CI: confidence interval; EES: efficacy-evaluable set; ESS: effective sample size; HR: hazard ratio; liso-cel: lisocabtagene maraleucel; OS: overall survival; PFS: progression-free survival; tisa-cel: tisagenlecleucel.

Scenario and sensitivity analyses

In the scenario analysis, the ORR was significantly higher in the TRANSCEND PAS compared with patients in JULIET (rate difference [95% CI]: −13.5% [−25.6, −1.5%]; p < 0.05) (Table 4). However, no significant difference was observed in the CRR (rate difference [95% CI]: −7.7% [−20.1, 4.8%]; p = 0.23) after adjusting for baseline characteristic differences. After matching, the median PFS and OS of JULIET patients overlapped with the CIs of those reported for the TRANSCEND PAS (PFS: 6.0 [95% CI: 1.9, not reached (NR)] vs. 9.0 [3.1, NR] months; OS: 20.7 [3.9, NR] vs. 19.9 [10.4, NR] months).

Table 4. Comparison of ORR, CR, PFS, and OS between JULIET patients and the TRANSCEND PAS, after matching.

	After matching
	JULIET (N = 106; effective sample size = 26^a)	TRANSCEND PAS (N = 133)	p
JULIET vs. TRANSCEND PAS
ORR difference (95% CI)	−13.5% (−25.6, −1.5%)		<0.05*
CRR difference (95% CI)	−7.7% (−20.1, 4.8%)		0.23
PFS
Median, months (95% CI)	6.0 (1.9, NR)	9.0 (3.1, NR)
PFS rate at month 6, %	48.6%	53.2%
OS
Median, months (95% CI)	20.7 (3.9, NR)	19.9 (10.4, NR)
OS rate at month 6, %	63.6%	72.9%

CI: confidence interval; CRR: complete response rate; NR: not reached; ORR: overall response rate; OS: overall survival; PAS: primary analysis set; PFS: progression-free survival.

*p < 0.05.

^aORR and CRR analyses were conducted among the main cohort of 91 JULIET patients, with an effective sample size of 25.

The findings from the two sensitivity analyses were consistent with those of the main analysis (Table 5), with no significant differences observed between JULIET patients and the TRANSCEND EES in any outcomes. In the sensitivity analysis of JULIET patients with fludarabine-based LDC, the estimated 2-year OS and 2-year PFS were 53.3 and 49.3% in the tisa-cel arm vs. 43.8 and 42.1% in the liso-cel arm. In the sensitivity analysis only adjusting for variables that significantly differed at baseline, the estimated 2-year OS and PFS were 43.6 and 38.0%, respectively, in the tisa-cel arm vs. 43.8 and 42.1% in the liso-cel arm.

Table 5. Comparison of ORR, CR, PFS, and OS from the sensitivity analyses, after matching.

	Sensitivity analysis 1: restricting to JULIET patients with fludarabine-based LDC			Sensitivity analysis 2: reduced list of matching variables that are significantly different
	JULIET (N = 84; effective sample size = 26^a)	TRANSCEND EES (N = 256)	p	JULIET (N = 106; effective sample size = 32^b)	TRANSCEND EES (N = 256)	p
JULIET vs. TRANSCEND EES
ORR difference, % (95% CI)	3.8% (−4.9, 12.4%)		0.40	−11.6% (−25.1, 1.8%)		0.09
CRR difference, % (95% CI)	2.9% (−9.1, 14.9%)		0.64	−3.0% (−17.1, 11.1%)		0.67
PFS
Median, months	34.2 (4.2, NR)	6.8 (3.3, 14.1)		9.0 (2.5, NR)	6.8 (3.3, 14.1)
JULIET vs. TRANSCEND HR	0.73 (0.43, 1.25)		0.26	1.13 (0.65, 1.98)		0.67
PFS rate at month 12, % (95% CI)	60.8% (46.4, 79.7%)	44.1% (37.3, 50.7%)		47.3% (35.4, 63.3%)	44.1% (37.3, 50.7%)
PFS rate at month 24, % (95% CI)	49.3% (34.7, 70.0%)	42.1% (35.0, 48.9%)		38.0% (26.5, 54.7%)	42.1% (35.0, 48.9%)
OS
Median, months	NR (17.9, NR)	21.1% (13.3, NR)		20.7 (6.7, NR)	21.1 (13.3, NR)
JULIET vs. TRANSCEND HR	0.80 (0.47, 1.38)		0.42	1.16 (0.67, 2.04)		0.59
OS rate at month 12, % (95% CI)	64.5% (50.9, 81.7%)	57.9% (51.3, 63.8%)		54.7% (43.0, 69.7%)	57.9% (51.3, 63.8%)
OS rate at month 24, % (95% CI)	53.3% (39.4, 72.1%)	43.8% (35.2, 52.1%)		43.6% (32.1, 59.1%)	43.8% (35.2, 52.1%)

CI: confidence interval; CRR: complete response rate; EES: efficacy-evaluable set; HR: hazard ratio; LDC: lymphodepleting chemotherapy; NR: not reached; ORR: overall response rate; OS: overall survival; PFS: progression-free survival.

^aIn sensitivity analysis 1, ORR and CRR analyses were conducted among the main cohort of 73 JULIET patients, with an effective sample size of 23.

^bIn sensitivity analysis 2, ORR and CRR analyses were conducted among the main cohort of 91 JULIET patients, with an effective sample size of 30.

Discussion

This indirect comparison between the two pivotal phase II trials of tisa-cel and liso-cel in r/r LBCL provides insight into differences between the trials and an understanding of their relative efficacy among adults with r/r LBCL, which can be informative for patients and physicians in treatment decision-making. Specifically, the MAIC results demonstrate that, after adjustment for baseline differences in the trial populations, the two CAR T-cell therapies were associated with similar efficacy. The results in terms of CRR, OS, and PFS were consistent across all scenario and sensitivity analyses. The only significant difference in efficacy between the two trials was the ORR, which was higher in the TRANSCEND PAS than in the respectively matched JULIET patients (74.4 vs. 60.9%, p < 0.05), but this difference was not observed when comparing JULIET patients with TRANSCEND EES patients.

Because all three available CAR T-cell therapies were evaluated in single-arm clinical trials, it is not feasible to directly compare them with the widely used network meta-analysis approach which requires a common treatment as an anchor to connect different studies to a network. In a situation where IPD is available for one trial and aggregated data is available for another trial, MAIC is commonly used for pair-wise comparison. As with any indirect treatment comparison, heterogeneity across trials needs to be carefully assessed to understand its potential impact on the validity of the comparative results from a MAIC. Due to the substantial differences in the patient populations and trial designs of JULIET (tisa-cel) and ZUMA-1 (axi-cel), especially with regard to the use of bridging treatment (90% in JULIET vs. not allowed in ZUMA-1), LDC regimens, and timing of leukapheresis, Zhang et al. concluded that indirect treatment comparisons of tisa-cel and axi-cel using data from these trials could not lead to conclusive results [22,23].

The differences in the rates of bridging treatment were less dramatic between JULIET and TRANSCEND because both trials allowed its use. To adjust for differences between the JULIET and TRANSCEND populations, the main analysis adjusted for 13 baseline factors including demographics, baseline disease characteristics, prior treatments, and response to prior treatments. With these adjustments, the main analysis led to a reduction of 73% in the effective sample size for the JULIET population. This large reduction in the effective sample size is likely a reflection of the substantial differences between the JULIET and TRANSCEND trial populations. Such a reduction, however, is anticipated in MAIC studies. The National Institute for Health and Care Excellence Decision Support Unit reported that the median reduction in effective sample size in published MAIC studies was 74.2% and ranged from 7.9 to 94.1% [29]. To increase the effective sample size, a sensitivity analysis was conducted to adjust for factors that significantly differed between the two trial populations, including age <65 years, DLBCL histology, ECOG PS of 0, prior SCT, and the use of bridging treatment. The sensitivity analysis only slightly increased the effective sample size (from a reduction of 73 to 70%), and the results were consistent with the main analysis.

The current study only compared the CAR T-cell infused trial populations and did not compare enrolled JULIET with TRANSCEND patients due to differences in enrollment strategies. CAR T-cell therapy requires the engineering of patients’ T-cells, collected via leukapheresis and then genetically modified during the manufacturing process before preparation for infusion. The manufacturing process starts after the receipt of T-cells by the processing labs. In JULIET, leukapheresis and enrollment were performed regardless of manufacturing slot availability and the cells were cryopreserved, to allow flexibility for the timing of apheresis and manufacturing. However, in TRANSCEND, leukapheresis, and enrollment were not allowed until a manufacturing slot became available. Cells were not cryopreserved, and fresh cells were delivered by courier directly for manufacture. This difference creates a potential source of bias when conducting indirect treatment comparisons using enrolled populations [23]. A trial allowing flexible enrollment criteria (i.e. without the requirement to wait for a manufacturing slot) could enroll patients who might not otherwise be considered. Because r/r LBCL is associated with high mortality risk, patients might be able to enroll in JULIET but not survive to the time of enrollment availability for TRANSCEND. It is impossible to estimate how TRANSCEND patients would be enrolled under the criteria of JULIET or vice versa.

A MAIC by Maloney et al. also indirectly compared the associated efficacy and safety outcomes of tisa-cel and liso-cel using aggregate data extracted from the JULIET trial publication and IPD from TRANSCEND [31] and concluded that liso-cel was associated with better efficacy compared with tisa-cel with regards to ORR, CRR, PFS, and OS. However, the difference in results may be a measure of study design, as the latter study adjusted for fewer factors; namely, disease histology, ECOG PS, central nervous system (CNS) involvement, prior SCT, and refractory status to last therapy. The current study adjusted for additional factors known to be associated with differences in efficacy outcomes (e.g. age, sex, LDH, number of prior lines of therapy, never achieved CR with prior therapy, SPD, LVEF, and use of bridging chemotherapy).

Conversely, safety was compared in Maloney et al., which was not compared in the current study recognizing the advancements in CAR T-cell therapy clinical care and management of CRS and neurotoxicity since the conduct of JULIET [16,32–34]. Tocilizumab is now commonly used for lower grades of CRS, which can reduce the likelihood of progression to high-grade CRS [32,34,35]. The majority of tocilizumab was initiated for grade 1 or 2 CRS in TRANSCEND, however, it was reserved for grade 3 or 4 CRS in JULIET [34]. The evolution of AE management is further underscored by the lower CRS rate (1–4% for grade 3–5 CRS [36,37]) reported with tisa-cel in the real world vs. that observed in JULIET [38]. Finally, while TRANSCEND used the Lee grading system for CRS, JULIET used the Penn grading system, which upgrades CRS as demonstrated by a blinded independent analysis of the JULIET database [34]. Such differences in AE grading and management cannot be adjusted for using MAIC.

This study is subject to several limitations, some of which are inherent to indirect treatment comparisons, recognizing that the data obtained are not derived from a comparative, randomized controlled trial. First, it was not possible to adjust for all baseline characteristics of interest. Only observed baseline factors consistently reported in both studies were included in the adjustment for the MAIC. Baseline variables, such as International Prognostic Index, disease stage, and bulky disease could not be matched due to limited or incomplete information in the TRANSCEND publication. Second, some differences in the trial design of JULIET and TRANSCEND could not be adjusted for. Retreatment was allowed in TRANSCEND, but not in JULIET, and 16 patients in the liso-cel treated set (N = 269) who achieved a CR and subsequently relapsed received retreatment. The difference in retreatment might have an impact on the OS outcome between the two trials. In addition, TRANSCEND permitted enrollment of patients with prior allo-SCT, secondary CNS involvement, PMBCL, FL grade 3B, and ECOG PS of 2 (1.5% of TRANSCEND population), while JULIET excluded all such patients. Due to the lack of access to TRANSCEND IPD, the current study was not able to account for these factors. The scenario analysis of JULIET vs. the TRANSCEND PAS partially addressed this limitation as the PAS population excluded patients with ECOG PS of 2, prior allo-SCT, PMBCL, FL grade 3B, or transformation from indolent lymphoma other than FL. However, baseline characteristics were not reported for the PAS in Abramson et al. [14], and the current analyses assumed that they had the same characteristics as that of the dose-level 2 population. Third, per the FDA label of liso-cel, a broader range of dosages is considered therapeutic (i.e. 50–110 × 10⁶ CAR T-cells) [19], which is a subset of the three doses tested in TRANSCEND: 50 × 10⁶, 100 × 10⁶, and 150 × 10⁶ CAR T-cells (doses 1, 2, and 3, respectively) [14]. As there are limited publicly available baseline data for the label dose, it was not possible to conduct a MAIC to compare the tisa-cel vs. liso-cel label doses. Lastly, differences in the two trial populations based on the matched prognostic factors resulted in a low effective sample size after matching and wide ranges of CIs in some analyses, which may have resulted in bias if some weighted patients were overly influential in the analysis.

Conclusions

The MAIC results indicate that OS, PFS, or CRR are comparable between tisa-cel and liso-cel in r/r LBCL. Future studies using either IPD from both JULIET and TRANSCEND or real-world data will be important for comparisons of the efficacy of tisa-cel and liso-cel for the treatment of r/r LBCL.

Ethical approval

This study used existing, de-identified data from clinical trials. Thus, no institutional board review was required.

Author contributions

The contributions of the authors are as follows: conception and design (all authors); acquisition of data (Jie Zhang); analysis and interpretation of data (all authors); drafting and revision of the manuscript (all authors). All authors have given their approval for the final version to be published and agree to be accountable for all aspects of the work.

Acknowledgments

Medical writing assistance was provided by Shelley Batts, Ph.D., an employee of Analysis Group, Inc. Support for this assistance was provided by Novartis.

Disclosure statement

Jie Zhang, Marcela Martinez-Prieto, Vamsi Bollu, David Kuzan, and Abhijit Agarwal are employees of Novartis and own stock/options. Hongbo Yang and Wenxi Tang are employees of Analysis Group, Inc., which has received consulting fees from Novartis. Stephen Schuster has received research funding from Acerta, Abbvie, Adaptive Biotechnologies, Celgene/Juno, DTRM, Genentech/Roche, Gilead, Incyte, Merck, Novartis, Pharmacyclics, and TG Therapeutics, serves on steering committees for Celgene, Nordic Nanovector, Novartis, and Pfizer, has received honoraria from Abbvie, Acerta, Alimera Sciences, BeiGene, AstraZeneca, Celgene, Juno Therapeutics, Genentech/Roche, Loxo Oncology, Nordic Nanovector, Novartis, Pfizer, and Tessa Therapeutics, and has a patent for Combination Therapies of CAR and PD-1 Inhibitors (royalties to Novartis). Richard Maziarz is an advisor or consultant for AlloVir, Artiva, CRISPR Therapeutics, CytoDyn, Incyte, and Novartis; reports honoraria from Bristol-Myers Squibb/Celgene, Incyte, Intellia, and Kite; research support from BMS, Allovir, and Novartis; participation in a data and safety monitoring board for Athersys, Vor Pharma, and Novartis; and a patent with Athersys. Marie José Kersten has received research support from Kite/Gilead and Roche, and has received honoraria for speaking or advisory boards or travel support from Novartis, Kite/Gilead, BMS/Celgene, Roche, MSD, Amgen, Janssen/Cilag, and Miltenyi Biotech.

Data availability statement

Data sharing is not applicable to this study.

Additional information

Funding

Support for this study was provided by Novartis.