Real-world comparison of lisocabtagene maraleucel and axicabtagene ciloleucel in large B-cell lymphoma: an inverse probability of treatment weighting analysis with 3-year follow-up

Abstract

Lisocabtagene maraleucel (liso-cel) and axicabtagene ciloleucel (axi-cel) are Food and Drug Administration- and European Medicines Agency-approved chimeric antigen receptor (CAR) T-cell therapies for relapsed/refractory large B-cell lymphoma (LBCL). However, comparative real-world analyses of their efficacy and toxicity with extended follow-up are lacking. We conducted a retrospective study of 160 LBCL patients treated at the Fred Hutchinson Cancer Center with commercial liso- cel or axi-cel per standard of care. Using inverse probability of treatment weighting (IPTW) to mitigate treatment allocation bias and multivariable adjustments to minimize other sources of confounding, we assessed the impact of CAR T-cell product type on outcomes. Axi-cel was associated with significantly higher rates of cytokine release syndrome (CRS; grade [G]1+: adjusted odds ratio [aOR] =4.27; P=0.004; G2+: aOR=2.88; P=0.006), immune effector cell-associated neurotoxicity syndrome (ICANS; G1+: aOR=2.10; P=0.048), and immune effector cell-associated hematotoxicity (ICAHT; G1+: aOR=8.09; P<0.001; G2+: aOR=3.86; P=0.001). Axi-cel was also associated with more frequent use of supportive care measures, such as tocilizumab (aOR=2.50; P=0.017), dexamethasone (aOR=2.77; P=0.007), and cefepime (aOR=3.37; P=0.001). We could not confirm statistically significant differences in the response rates and survival outcomes after liso-cel versus axi-cel (complete response: aOR=1.12; P=0.8; overall survival: adjusted hazard ratio [aHR] =1.34; P=0.3; progression-free survival: aHR=0.97; P=0.9; duration of response: aHR=0.89; P=0.7; cumulative incidence of relapse: aHR=0.92; P=0.8). In summary, although axicel was associated with greater toxicity requiring more intensive management, the response rates and survival outcomes were comparable between axi-cel and liso-cel.

Introduction

CD19 chimeric antigen receptor (CAR) T-cell therapy has transformed the management of relapsed/refractory (R/R) large B-cell lymphoma (LBCL), offering the potential to cure 30-40% of treated patients.¹ Currently approved products include axicabtagene ciloleucel (axi-cel), tisagenlecleucel (tisa-cel), and lisocabtagene maraleucel (liso-cel). Axi-cel and liso-cel are approved by the Food and Drug Administration (FDA) and the European Medicines Agency (EMA) as second-line treatments for high-risk LBCL patients with primary refractory disease or relapse/progression within 12 months, having demonstrated superiority over standard-of-care (SOC) therapy.^2-9 In contrast, the BELINDA study did not confirm a similar benefit for tisa-cel in this setting.¹⁰ Retrospective adjusted comparisons have revealed differences in efficacy and toxicity profiles among these therapies. Axi-cel showed superior anti-tumor activity compared to tisa-cel, albeit with greater toxicity.^11,12 The TRANSCEND NHL 001 trial, which led to the initial approval of liso-cel, indicated a more favorable toxicity profile, a finding which was re-demonstrated in a non-transplant eligible population.^13-15 A matching-adjusted indirect treatment comparison (MAIC) suggested comparable efficacy between liso-cel and axi-cel, with a better safety profile for liso-cel. However, the analysis was restricted to clinical trial patients, who generally have better outcomes than non-trial patients,¹⁶ and lacked patient-level data for axi-cel.¹⁷ A more recent real-world comparison, though limited by a small sample size of 87 patients and 11 months of follow-up, suggested similar outcomes between these products, with a propensity score matching analysis indicating superior progression-free survival (PFS) for axi-cel.¹⁸

To address the limitations of prior studies and provide a real-world, patient-level comparison of liso-cel and axicel, we utilized inverse probability of treatment weighting (IPTW). This statistical approach balances baseline characteristics between treatment groups to mitigate treatment allocation bias, enabling an unbiased assessment of the independent impact of CAR T-cell product type on outcomes in LBCL patients treated with liso-cel and axi-cel at our center per SOC.

Methods

Patients

The study was approved by Fred Hutchinson Cancer Center’s (FHCC) Institutional Review Board based on federal regulations and the ethical standards of the FHCC Human Research Protection Program. Informed consent was obtained for the collection of medical data. All adult patients (age ≥18 years) with LBCL who received liso-cel or axi-cel at FHCC through May 2024 were included, with data collected through November 15, 2024.

Definitions

Definitions of LBCL subtypes, treatment response, adverse event grading, and other clinical criteria are detailed in the Online Supplementary Appendix.

Lymphodepletion chemotherapy

Patients received lymphodepletion chemotherapy with fludarabine and cyclophosphamide per product-specific FDA package inserts.^4,5

Supportive care

Packed red blood cells and platelets were transfused for hematocrit <26% and platelet count <11×10³/VL, respectively, or when clinically indicated. Broad-spectrum intravenous antibiotics were initiated for neutropenic fever per institutional guidelines. CRS and ICANS management followed product-specific risk evaluation and mitigation strategy and institutional guidelines.

Statistical analysis

All statistical analyses were conducted using R Statistical Software (Version 4.4.1; R Core Team 2024).¹⁹ Continuous variables were compared with the Wilcoxon rank sum test, and categorical variables with Pearson’s X² test or Fisher’s exact test. PFS was defined as time to death, relapse, or progressive disease (PD). Duration of response (DOR) was time to relapse, PD, or death for patients achieving CR or PR after CAR T-cell therapy. Overall survival (OS), PFS, and DOR were estimated using the Kaplan-Meier (KM) method. The cumulative incidence of relapse or progression was computed with tidycmprsk::cuminc() (Version 1.0.0), treating non-relapse mortality as a competing event. Median follow-up was determined using the reverse KM method. Binomial outcomes were assessed using univariate (UV) and multivariable (MV) logistic regression, and time-to-event outcomes with Cox regression. Logistic regression for CR was restricted to response-evaluable patients.

Inverse probability of treatment weighting

IPTW was employed to mitigate treatment allocation bias, while MV analysis addressed additional confounders. IPTW uses the inverse of the propensity score to create a weighted pseudo-population where covariates are balanced, enabling estimation of treatment effects as if treatments were randomly assigned. Unlike propensity score matching, IPTW retains all participants, preserving statistical power and providing an unbiased estimate of the average treatment effect.^20-22 Further details on the IPTW rationale and MV regression models are provided in the Online Supplementary Appendix.

Covariates for IPTW were selected based on subject-matter expertise regarding known and measurable confounders as described previously.¹¹,¹²,²³,²⁴ The following variables were included for IPTW: pre-leukapheresis lactate dehydrogenase (LDH; U/L; continuous), pre-leukapheresis absolute lymphocyte count (ALC; x10³/μL; continuous), age (years; continuous), prior autologous hematopoietic cell transplantation (HCT; binary), use of CAR T-cell therapy in the second line setting (binary), Eastern Cooperative Oncology Group (ECOG) performance status ≥2 (binary), transformed LBCL versus other (binary), largest lesion diameter as measured on pre-infusion positron-emission tomography/computed tomography (PET/CT) (cm; continuous), presence of non-CNS extranodal disease (binary), presence of central nervous system (CNS) involvement (binary), HCT-specific comorbidity index (HCT-CI) score (continuous), and bridging response category (no bridging vs. complete reposnse/ partial remission [CR/PR] after bridging vs. stable disease/ disease progression [SD/PD] after bridging vs. bridging with unknown response).

Results

Patient characteristics prior to inverse probability of patient weighing

Between January 2018 and May 2024, 160 LBCL patients were treated with liso-cel (N=58) or axi-cel (N=102) at our center per SOC. Unweighted and IPTW patient characteristics are shown in Table 1. An out-of-specification product was administered to 17 (11%) patients. Patients who received liso-cel were older (median age: 66 vs. 61 years; P=0.003) and experienced a longer median time interval from apheresis to CAR T-cell infusion (median time: 35 vs. 27 days; P<0.001; Online Supplementary Figure S1). There was numerically lower disease bulk in patients given liso-cel compared to axi-cel (median largest lesion: 3.1 vs. 3.5 cm; P=0.2). The distribution of bridging response categories differed for liso-cel versus axi-cel: no bridging, 41% versus 65%; CR/PR post-bridging, 26% versus 9.8%; and SD/PD post-bridging, 26% versus 25% (P<0.001). ECOG performance status and HCT-CI scores were similar between products. Prophylactic dexamethasone was used in one (1.7%) liso-cel and six (5.9%) axi-cel patients.

Inverse probability of patient weighing

For clarity, all summary statistics and estimates in the following sections refer to the pseudo-population generated by IPTW, unless otherwise specified.

The sample size of the pseudo-population after IPTW was 155 patients (liso-cel, N=55; axi-cel, N=100). IPTW resulted in good balance with an appropriate distribution of weights (Online Supplemental Figure S2). After matching the co-variates, all absolute standardized mean differences were <0.2 indicating adequate balance.²⁵ The variance ratios of the continuous variables were 1.46 for age, 0.50 for largest lesion diameter, 9.1 for pre-leukapheresis LDH, 0.18 for pre-leukapheresis ALC, and 1.4 for HCT-CI.

Table 1.Inverse probability of treatment weighting and unweighted baseline characteristics of patients who received liso-cel and axi-cel.

Patient characteristics following IPTW are shown in Table 1, where patient numbers are rounded to the nearest integer. The median age was 63 years (IQR, 53-68) and 66% (N=102) were male. A minority received CAR T-cell therapy as second line (9.7%, N=15) or had undergone prior autologous HCT (17%, N=27). A significant proportion of patients did not meet the eligibility criteria for the TRANSCEND-NHL-001 (32%, N=50) or ZUMA-1 (46%, N=71) clinical trials. Most patients had DLBCL (67%, N=103) or transformed DLBCL (23%, N=36). Other disease categories included high-grade B-cell lymphoma (5.7%, N=9), primary mediastinal large B-cell lymphoma (1.3%, N=2), and T-cell/histiocyte-rich LBCL (3.0%, N=5).

Toxicity

All patients were evaluable for toxicity assessment. IPTW and unweighted rates of CRS, ICANS, and ICAHT are depicted in Figure 1. For liso-cel versus axi-cel, unadjusted IPTW toxicities were as follows: any-grade CRS, 68% versus 88% (P=0.011); grade 2+ CRS, 25 versus 50% (P=0.008); grade 3+ CRS, 3.1% versus 7.5% (P=0.4); any-grade ICANS, 35% versus 55% (P=0.041); grade 2+ ICANS, 28% versus 42% (P=0.12); grade 3+ ICANS, 19% versus 20% (P>0.9); any-grade ICAHT, 63% versus 92% (P<0.001); grade 2+ ICAHT, 19% versus 49% (P=0.001); and grade 3+ ICAHT, 7.3% versus 19% (P=0.049). Fewer liso-cel patients received tocilizumab (32% vs. 55%; P=0.017) and non-prophylactic dexamethasone (38% vs. 64%; P=0.007). The proportion of liso-cel and axi-cel patients who received granulocyte colony-stimulating factor (G-CSF) was similar (72% vs. 81%; P=0.3), though liso-cel patients received fewer median doses (2 vs. 5; P=0.006). None received peg-G-CSF. IPTW and unweighted peak inflammatory markers are shown in Online Supplementary Table S1. We measured lower median peak serum CRP (71 vs. 114 mg/dL; P=0.002), D-dimer (1.3 vs. 2.4 mg/L; P=0.011), ferritin (666 vs. 1,081 ng/mL; P=0.009), IL-6 (55 vs. 196 ng/ mL; P=0.030), and ALT (32 vs. 51 U/L; P=0.020) after liso-cel treatment compared to axi-cel.

Post-infusion median nadir cytopenias were less severe after liso-cel: ANC, 0.34 versus 0.04x10³/μL (P<0.001); hemoglobin, 8.80 versus 8.20 g/dL (P=0.015); and platelets, 84 versus 35x10³/μL (P=0.002). Fewer patients developed severe neutropenia after liso-cel, 63% vs. 92% (P<0.001), and the median duration of neutropenia was shorter, 2.0 versus 6.0 days (P<0.001). Unweighted longitudinal ANC and platelet count data are shown in Online Supplementary Figure S3. At day +30, the median platelet count was higher for liso-cel compared to axi-cel (101 vs. 63x10³/μL; P=0.045), but otherwise median laboratory values were comparable: ANC, 1.56 versus 1.37x10³/μL (P=0.2); and hemoglobin, 10.60 versus 10.20 g/dL (P=0.11).

The incidence of infectious complications (Online Supplementary Table S2) after liso-cel versus axi-cel was as follows: bacteremia, 3.6% versus 8.3% (P=0.3); Cytomegalo virus (CMV) viremia, 32% versus 28% (P=0.6); and CMV level ≥150 IU/mL (threshold requiring anti-viral therapy per our standard practice), 7.5% versus 5.8% (P=0.7). Fewer liso-cel patients received cefepime (43% vs. 71%; P=0.003).

While axi-cel patients were required to remain inpatient for at least 7 days following infusion, liso-cel had no mandatory inpatient stay and was administered in an outpatient setting for 39% of patients (N=21). Compared to axi-cel, liso-cel patients experienced significantly shorter total inpatient duration (5 vs. 14 days; P<0.001), and a smaller proportion required 2+ admissions (5.1% vs. 25%; P<0.001). Both metrics were assessed from the day of infusion through day +30. Of note, 14 patients (liso-cel: 11; axi-cel: 3) were in CR at the time of lymphodepletion chemotherapy, and none experienced grade 3+ CRS, ICANS, or early ICAHT. The incidence of any-grade toxicity was as follows for liso-cel compared to axi-cel: CRS, 50% versus 46%; ICANS 12% versus 24%; and early ICAHT, 39% versus 46%.

Response rates

IPTW and unweighted responses to CAR T-cell therapy in response-evaluable patients are shown in Table 2. Among response-evaluable patients (N=141), we observed comparable best response rates with liso-cel versus axi-cel: ORR, 76% versus 78% (P=0.8); and CR, 46% versus 49% (P=0.8). The proportions of patients who converted from an initial PR to a CR were similar across CAR T-cell products (li-so-cel vs. axi-cel: 11% vs. 7.0%; P=0.4). Among responders, the median time to best response was similar (liso-cel vs. axi-cel: 28 vs. 29 days; P=0.12).

Among patients who received bridging therapy and were evaluable for response (N=62), the best response rates were comparable for liso-cel versus axi-cel: ORR, 73% versus 74% (P>0.9); and CR, 34% versus 40% (P=0.7).

Long-term outcomes

Kaplan-Meier plots depicting OS, PFS and DOR, and cumulative incidence plots depicting relapse and non-relapse mortality (NRM), are shown in Figure 2. The median follow-up was 35.3 months (liso-cel, 29.8; axi-cel, 53.8). The median OS, PFS, and DOR were 41.1 months (95% CI: 13.6-not reached), 10.7 months (95% CI: 5.7-29.3), and 28.4 months (95% CI: 15.4-not reached), respectively. Median OS (liso-cel, not reached; axi-cel, 16.3 months; P=0.12), PFS (li-so-cel, 11.9; axi-cel, 10.7 months; P=0.89), DOR (liso-cel, 21.4 vs. 38.4 months; P=0.78), and 3-year cumulative incidence of relapse (53.2% vs. 46.2%; P=0.69) were not significantly different between products. Unweighted survival plots are shown in Online Supplementary Figure S4. We next evaluated outcomes in high-risk patients specifically. For liso-cel versus axi-cel, the OS and PFS were assessed among patients with bulky disease (largest lesion diameter ≥5 cm; OS: 6.8 vs. 7.2 months; P=0.91; PFS: 4.7 vs. 3.0 months; P=0.91), high LDH (≥210 U/L; OS: 13.6 vs. 12.1 months; P=0.62; PFS: 7.2 vs. 5.7 months; P=0.84), extranodal disease (OS: not reached vs. 16.3 months; P=0.4; PFS: 6.8 vs. 5.7; P=0.83), or who received bridging therapy (OS: 12.0 vs. 9.5 months; P=0.83; PFS: 4.0 vs. 5.7; P=0.22), as depicted in Figures 3 and 4, respectively.

Figure 1.Comparative analysis of toxicities between CAR T-cell product types. (A) Inverse probability of treatment weighting (IPTW) and (B) unweighted comparisons of cytokine release syndrome (CRS), immune effector cell-associated neurotoxicity syndrome (ICANS), and early immune effector cell-associated hematotoxicity (ICAHT). Weighted patient numbers have been rounded to the nearest integer. P values for comparisons of toxicities were derived from the χ² test or Fisher’s exact test (with Rao and Scott’s second-order correction for IPTW comparisons). CAR: chimeric antigen receptor; axi-cel: axicabtagene ciloleucel; liso-cel: lisocabtagene maraleucel.

Trial-eligible patients demonstrated superior OS (ZUMA-1: not reached vs. 12.0 months; P=0.003; TRANSCEND: 51.1 vs. 16.0 months; P=0.051) and PFS (ZUMA-1: 22.6 vs. 5.1 months; P=0.019; TRANSCEND: 15.9 vs. 5.3 months; P=0.15), as shown in Online Supplementary Figures S5 and S6, respectively. Unweighted comparisons are shown in Online Supplementary Figures S7 and S8, respectively.

Among patients not in a CR at time of lymphodepletion chemotherapy (liso-cel: 44 patients; axi-cel: 97 patients), the median duration of follow-up was 38.0 months (liso-cel: 26.5 months; axi-cel 53.8 months). No significant differences were observed between liso-cel and axi-cel in terms of median OS (not reached vs. 24.3 months; P=0.57), PFS (5.1 vs. 10.7 months; P=0.38), or DOR (11 vs. 38.4 months; P=0.32). Additionally, the 1-year cumulative incidence of relapse was comparable between liso-cel and axi-cel (52.8% vs. 43.3%; P=0.28), as shown in Online Supplementary Figure S9.

Among the 14 patients in a CR at the time of lymphodepletion chemotherapy (liso-cel: 11; axi-cel: 3), one axi-cel patient experienced an early relapse at 27 days. Excluding this case, the 1-year PFS was 88% for liso-cel and 100% for axi-cel, while 1-year OS was excellent for both groups at 100%.

Regression models estimating the effect of the CAR T-cell product type on outcomes

Efficacy

In IPTW UV and MV analyses (Table 3), we did not observe an independent impact of the product type (reference: liso-cel) on CR (adjusted OR [aOR] for axi-cel: 1.12, 95% CI: 0.52-2.40; P=0.8), DOR (aHR for axi-cel: 0.89, 95% CI: 0.48-1.65; P=0.7), relapse (aHR for axi-cel: 0.92, 95% CI: 0.55-1.54; P=0.8), PFS (aHR for axi-cel: 0.97, 95% CI: 0.60-1.56; P=0.9), or OS (aHR for axi-cel: 1.34, 95% CI 0.73-2.47; P=0.3). Similar findings were noted in unweighted analyses (Online Supplementary Table S3).

In IPTW MV analysis, increasing pre-leukapheresis LDH remained independently associated with lower odds of a CR, and increased hazard of PFS and OS.

Among patients not in a CR at time of lymphodepletion chemotherapy, we did not observe an independent impact of product type (reference: liso-cel) on PFS (aOR for axicel: 0.80, 95% CI: 0.47-1.36; P=0.4), DOR (aOR for axi-cel: 0.76, 95% CI: 0.40-1.46; P=0.4), or OS (aOR for axi-cel: 1.13, 95% CI: 0.61-2.10; P=0.7).

Toxicity

In all models, axi-cel (reference: liso-cel) was associated with higher odds of any-grade CRS, ICANS, and early ICAHT; and grade 2+ CRS and early ICAHT. In IPTW MV analysis, the aOR for axi-cel for any-grade CRS was 4.27 (95% CI: 1.59-11.4; P=0.004), for grade 2+ CRS was 2.88 (95% CI: 1.36-6.14; P=0.006), for any-grade ICANS was 2.10 (95% CI: 1.01-4.40; P=0.048), for any-grade early ICAHT was 8.09 (95% CI: 2.87-22.8; P<0.001), and for grade 2+ early ICAHT was 3.86 (95% CI: 1.72-8.65; P=0.001). In IPTW MV analysis, we did not detect significantly higher odds of grade 3+ toxicities (CRS: aOR=1.50, 95% CI: 0.24-9.30; P=0.7; ICANS: aOR=0.97, 95% CI: 0.39-2.38; P>=0.9; early ICAHT: aOR=3.46, 95% CI: 0.96-12.5; P=0.057). In IPTW MV analysis, increasing largest lesion size remained associated with higher odds of any-grade CRS and any-grade/grade 2+/grade 3+ ICANS; and increasing pre-leukapheresis LDH was associated with higher odds any-grade/grade 3+ CRS, any-grade ICANS, and grade 2+ ICAHT.

Table 2.Inverse probability of treatment weighting and unweighted response assessments among response-evaluable patients.

Additionally, in IPTW MV analysis, axi-cel (reference: liso-cel) was associated with higher odds of receiving tocilizumab (aOR=2.50, 95% CI: 1.17-5.34; P=0.017), dexamethasone (aOR=2.77, 95% CI: 1.31-5.83; P=0.007), and cefepime (aOR=3.37, 95% CI: 1.62-7.01; P=0.001).

Sensitivity analyses

Sensitivity analysis #1: contemporaneous cohort

To evaluate the impact of CAR T-cell product type on outcomes in a contemporaneous cohort, we analyzed patients treated between April 2021 and May 2024, after the introduction of liso-cel at our institution. During this period, 58 patients received liso-cel and 34 received axicel. Absolute standardized mean differences are shown in Online Supplementary Figure S10.

IPTW MV regression analysis (Figure 5) could not confirm differences in efficacy outcomes between liso-cel and axi-cel (reference: liso-cel) for CR (aOR=1.01, 95% CI: 0.39-2.58; P>0.9), DOR (aHR=1.0, 95% CI: 0.50-2.02; P>0.9), PFS (aHR=1.02, 95% CI: 0.57-1.83; P>0.9), or OS (aHR=1.31, 95% CI: 0.60-2.82; P=0.5). However, axi-cel was associated with higher odds of grade 1+ CRS (aOR=18, 95% CI: 2.57-126; P=0.022) and grade 2+ CRS (aOR=2.98, 95% CI: 1.17-7.58; P=0.022). In UV analysis, axi-cel was associated with grade 1+ ICANS (OR=2.47, 95% CI: 1.03-5.93; P=0.042), although this did not reach statistical significance in MV analysis (aOR=2.53, 95% CI: 0.90-7.12; P=0.078).

Figure 2.Analysis of survival outcomes and non-relapse mortality. Inverse probability of treatment weighting Kaplan-Meier plots, stratified by product type, showing (A) overall survival, (B) progression-free survival, (C) duration of response, and (D) cumulative incidence of relapse, and non-relapse mortality (NRM). Median survival times were analyzed using weighted univariate Cox regression, and the 1-year cumulative incidence of relapse was assessed with Gray’s test. axi-cel: axicabtagene ciloleucel; liso-cel: lisocabtagene maraleucel; NA: not applicable.

No significant differences were observed between product types for early ICAHT outcomes. Notably, axi-cel patients treated during this period had significantly higher G-CSF use compared to their earlier counterparts (98% vs. 72%; P<0.001; median 8 vs. 3 doses; P<0.001) and compared to liso-cel patients (98% vs. 72%; P<0.001; median 8 vs. 2 doses; P<0.001). These differences, shown in Online Supplementary Figure S11, may have contributed to ICAHT outcomes.

Figure 3.Kaplan-Meier analysis of overall survival in patient subgroups. Inverse probability of treatment weighting Kaplan-Meier plots depicting overall survival of patients with (A) bulky disease (largest lesion diameter ≥5 cm), (B) high lactate dehydrogenase (LDH) (≥210 U/L), (C) extranodal involvement, and (D) who received bridging therapy. P values were derived from weighted univariate Cox regressio to axi-cel: axicabtagene ciloleucel; liso-cel: lisocabtagene maraleucel.

Table 3.Summary of inverse probability of treatment weighting univariate and multivariable regression analyses.

Sensitivity analysis #2: baseline inflammatory markers

To evaluate the influence of baseline inflammatory markers, IPTW was performed using day 0 ferritin and CRP (available for 147 and 146 patients, respectively). Absolute standardized mean differences are shown in Online Supplementary Figure S10.

As summarized in Figure 5, no significant differences were found between liso-cel and axi-cel (reference: liso-cel) for CR (aOR=1.51, 95% CI: 0.65-3.47), DOR (aHR=1.02, 95% CI: 0.52-2.02; P>0.9), PFS (aHR=1.06, 95% CI 0.62-1.80; P=0.8), or OS (aHR 1.72, 95% CI: 0.84-3.51; P=0.14). Axi-cel was associated with higher odds of grade 1+ CRS (aOR=4.41, 95% CI: 1.39-13.9; P=0.012), grade 2+ CRS (aOR=3.15, 95% CI: 1.40-7.07; P=0.005), any-grade ICANS (aOR=2.75, 95% CI: 1.19-6.34; P=0.018), grade 1 eICAHT (aOR=7.05, 95% CI: 2.28-21.8; P<0.001), and grade 2 eICAHT (aOR=4.27, 95% CI: 1.74-10.5; P=0.002).

Sensitivity analysis #3: disease activity (maximum standardized uptake value)

To examine the role of disease activity, we performed IP-TW incorporating the maximum tandardized uptake value (SUV) from pre-CAR-T PET/CT scans from 153 available patients. Absolute standardized mean differences are shown in Online Supplementary Figure S10.

As shown in Figure 5, showed no significant differences were observed between liso-cel and axi-cel (reference: liso-cel) for CR (aOR=1.66, 95% CI: 0.78-3.63; P=0.2), DOR (aHR=0.84, 95% CI: 0.46-1.55; P=0.6), PFS (aHR=0.83, 95% CI: 0.51-1.35; P=0.5), or OS (aHR=1.32, 95% CI: 0.72-2.43; P=0.4). Axi-cel was associated with higher odds of grade 1+ CRS (aOR=7.08, 95% CI: 2.24-22.4; P<0.001), grade 2+ CRS (aOR=3.33, 95% CI: 1.51-7.34; P=0.003), grade 1 eICAHT (aOR=6.15, 95% CI: 2.06-18.4; P=0.001), and grade 2 eICAHT (aOR=2.72, 95% CI: 1.23-6.01; P=0.013). No significant association between axi-cel and any-grade ICANS was observed in UV (OR=1.64, 95% CI: 0.85-3.14; P=0.14) or MV (aOR=1.78, 95% CI: 0.84-3.75; P=0.13) models, though the effect sizes were similar to the other models.

Discussion

CD19 CAR T-cell therapies have revolutionized the management of patients with high risk or R/R LBCL, though their use remains limited by significant toxicities. Retrospective analyses suggest axi-cel may offer stronger anti-tumor activity than tisa-cel, albeit with increased toxicity.^11,12 Liso-cel, more recently approved by the FDA and EMA for high-risk or R/R LBCL, has demonstrated high efficacy with lower rates of severe CRS and ICANS in single-arm studies, making it an appealing option for older or frail patients.^13,14,26 However, real-world data comparing these therapies are limited. A prior study of 87 patients (50 treated with axi-cel and 27 with liso-cel) with 11 months of follow-up suggested superior PFS with axi-cel but did not adjust for key confounders such as response to bridging therapy or the presence of extranodal disease, particularly CNS involvement.^18,23,24 Using a larger cohort of 160 patients and nearly 3 years of follow-up, our analysis employed IPTW to account for all known factors impacting treatment allocation and survival outcomes, enabling a more robust real-world comparison of the two CAR T-cell products.

Baseline differences in patient characteristics reflected treatment allocation bias. Liso-cel patients were older (66 vs. 61 years) and more likely to receive CAR T-cell treatment as second-line therapy, while axi-cel patients had greater tumor burden and were less likely to respond to bridging therapy. These trends likely reflect a preference for liso-cel in older or frail patients, axi-cel being approved for a longer duration in the third-line setting, and a tendency to reserve axi-cel for patients with greater disease burden. To mitigate these biases, IPTW was applied to balance key confounders influencing treatment allocation.

We estimated comparable anti-tumor efficacy between the two CAR T-cell products. The ORR were nearly identical (liso-cel vs. axi-cel: 76% vs. 78%) among response-evaluable patients, while CR rates were similar (liso-cel vs. axi-cel: 46% vs. 49%). These CR rates were lower than those reported in clinical trials for liso-cel (TRANSCEND-NHL-001: 53%; TRANSFORM: 66%) and axi-cel (ZUMA-1: 54%; ZUMA-7: 65%).^2,3,14,27 Median OS and PFS showed no significant differences for liso-cel versus axi-cel, with OS favoring liso-cel (not reached vs. 16.3 months) and PFS showing minimal variation (11.9 vs. 10.7 months). Although liso-cel demonstrated a numerically shorter DOR compared to axi-cel (21.4 vs. 38.4 months) - particularly among patients not in CR at the time of lymphodepletion (PFS: 5.1 vs. 10.7 months; DOR: 11 vs. 38.4 months) - these differences did not reach statistical significance. These findings suggest that both liso-cel and axi-cel are excellent options for high-risk and/ or R/R LBCL patients. Notably, our IPTW-adjusted estimates closely aligned with a MAIC analysis of clinical trial patients, with nearly identical HR (95% CI) for OS (IPTW aHR vs. MAIC HR=1.34 [95% CI: 0.73-2.47] vs. 1.23 [95% CI: 0.67-2.27]) and PFS (0.97 [95% CI: 0.60-1.56] vs. 0.93 [95% CI: 0.58-1.48]).¹⁷ This reinforces the robustness of our real-world findings. Safety is a critical factor in selecting CD19 CAR T-cell therapy, particularly for older or frail patients. In our analyses, axi-cel was associated with higher odds of any-grade and grade 2+ CRS, as well as any-grade ICANS, in both IPTW and unweighted MV regression models. These findings correlated with elevated peak inflammatory markers (i.e., CRP, ferritin, IL-6, D-dimer, ALT) observed in axi-cel patients. Moreover, a greater proportion of axi-cel patients required tocilizumab (55% vs. 32%) and dexamethasone (64% vs. 38%), reflecting a higher need for toxicity management. These results align with prior adjusted comparisons of axi-cel and liso-cel, which have also demonstrated significantly higher rates of CRS and neurologic toxicity in axi-cel recipients.^17,18 However, it is important to note that we did not observe significant differences in rates of grade 3+ CRS or ICANS - events that have the most substantial impact on morbidity and mortality - and rates of NRM were comparable.

Figure 4.Kaplan-Meier analysis of progression-free survival in patient subgroups. Inverse probability of treatment weighting Kaplan-Meier plots depicting progression-free survival of patients with (A) bulky disease (largest lesion diameter ≥5 cm), (B) high lactate dehydrogenase (LDH) (≥210 U/L), (C) extranodal involvement, and (D) who received bridging therapy. P values were derived from weighted univariate Cox regression. axi-cel: axicabtagene ciloleucel; liso-cel: lisocabtagene maraleucel.

Figure 5.Sensitivity analyses of response, survival, and toxicity outcomes. Sensitivity analyses evaluating: (A) complete response (CR), (B) duration of response (DOR), (C) relapse rate, (D) progression-free survival (PFS), (E) overall survival (OS), (F) cytokine release syndrome (CRS) grade 1 (G1) or higher, (G) CRS grade 2 or higher, (H) immune effector cell-associated neurotoxicity syndrome (ICANS) G1 or higher, (I) early immune effector cell-associated hematotoxicity (ICAHT) G1 or higher, and (J) early ICAHT G2 or higher. Three inverse probability of treatment weighting models were compared: the original model, Sensitivity #1 (restricted to patients treated from April 2021 onwards), Sensitivity #2 (incorporating day 0 ferritin and CRP values), and Sensitivity #3 (incorporating the maximum standardized uptake value from pre-CAR T positron emission tomography/computed tomography scans). Adjusted hazard ratios (HR) or odds ratios (OR) are displayed with 95% confidence intervals. Statistical significance is denoted as *P<0.05; **P<0.01; ***P<0.001. CAR: chimeric antigen receptor. axi-cel: axicabtagene ciloleucel; liso-cel: lisocabtagene maraleucel.

We additionally observed more severe hematologic toxicity following axi-cel, with higher odds of any-grade and grade 2+ early ICAHT in both IPTW and unweighted MV regression analyses. Although a significantly higher proportion of axi-cel patients experienced grade 3+ early ICAHT, this finding was not statistically confirmed in our regression models. Several factors likely contribute to these observations. First, the higher cyclophosphamide dose used in the lymphodepletion regimen for axi-cel (500 mg/m²/day for 3 days) compared to liso-cel (300 mg/m²/day for 3 days) may play a significant role. Second, more pronounced systemic inflammation during CRS, as shown in our prior research,²⁸ likely exacerbates hematologic toxicity. Finally, the CD28 co-stimulatory domain in axi-cel, known to induce greater toxicity compared to the 4-1BB domain used in liso-cel,²⁹ may further explain the differences in toxicity profiles.

This study has several limitations that should be considered when interpreting the findings. First, its retrospective design is inherently vulnerable to treatment allocation bias. To mitigate this, we applied IPTW and utilized MV models to reduce bias and improve the accuracy of our estimates. Nevertheless, residual confounding may still exist due to unmeasured factors such as physician preference, frailty, and total metabolic tumor volume. Sensitivity analyses incorporating baseline inflammatory markers (ferritin and CRP) and maximum SUV supported the findings of the primary analysis. However, the absence of standardized frailty assessments may have led to an underestimation of frailty in liso-cel recipients, despite no significant differences in HCT-CI or ECOG performance status scores between groups. Second, the earlier approval of axi-cel resulted in longer follow-up for axi-cel patients and a greater proportion treated in an earlier era, which could have influenced outcomes. Third, this was a single-center study, which may limit the generalizability of the findings to other centers. While a multi-center study could enhance statistical power and broaden applicability, it would likely introduce additional sources of confounding due to variations in practice patterns, patient populations, and data collection methods. Furthermore, logistical challenges, such as obtaining longitudinal laboratory data across multiple sites, may hinder the feasibility of achieving the same level of detail. Finally, while we could not detect an independent impact of CAR T-cell product type on efficacy, the relatively broad 95% CI of our IPTW estimates suggest the study may have been underpowered to detect small effect sizes. Larger studies with more comprehensive follow-up, narrower CI, and the inclusion of multi-center data will be essential to confirm these findings and provide greater generalizability.

To address the potential bias introduced by treatment era, we conducted a sensitivity analysis restricted to patients treated contemporaneously with liso-cel or axi-cel between April 2021 and May 2024. This analysis confirmed no significant differences in survival outcomes between the two products, supporting the conclusions of our primary analysis. Additionally, axi-cel remained associated with higher rates of CRS and ICANS. Importantly, no significant association was observed between product type and early ICAHT in the sensitivity analysis. However, it is noteworthy that axi-cel patients treated during this period received significantly more G-CSF compared to both liso-cel patients and axi-cel patients treated in earlier years. This increased G-CSF use among contemporaneous axi-cel patients may have masked an intrinsic product-related difference in the risk of early ICAHT.

In summary, our analysis, which applied IPTW to address treatment allocation bias and incorporated MV adjustments for additional confounders, found no evidence that CD19 CAR T-cell product type independently influenced efficacy outcomes in a relatively large cohort of LBCL patients treated with liso-cel or axi-cel in a real-world, non-trial setting with extended follow-up. However, liso-cel demonstrated a significantly lower incidence and severity of CRS and early ICAHT, as well as a reduced incidence of ICANS, highlighting its more favorable safety profile compared to axi-cel. Notably, rates of NRM were similar between the two products, highlighting the need for personalized treatment decisions. Our analysis demonstrates liso-cel’s more favorable safety profile, which should be a key consideration when choosing between liso-cel and axi-cel. However, further data are needed to confirm whether axi-cel and liso-cel achieve comparable long-term outcomes. The lower incidence and severity of CRS and early ICAHT, along with reduced rates of ICANS, position liso-cel as a preferred option for older, frail, or hematologically compromised patients. Additionally, the shorter hospital stays associated with liso-cel treatment enable outpatient management, significantly alleviating resource demands – an especially critical factor in resource-constrained environments. For high-risk subgroups, treatment decisions should be guided by patient-specific factors, such as tolerance for toxicity and logistical considerations, including the shorter vein-to-vein time offered by axi-cel. Ultimately, the choice of therapy must be individualized, taking into account patient performance status, treatment objectives, and institutional protocols.

Footnotes

• Received November 19, 2024
• Accepted March 7, 2025

Correspondence

A.J. Portuguese aportugu@fredhutch.org

Disclosures

AJP reports consultancy/honoraria from Capvision and Karyopharm Therapeutics. AVH has received research funding from Juno Therapeutics and Nektar Therapeutics; and has received honoraria from Bristol Myers Squibb. AKG has received research funding from Merck, BMS, Gilead Sciences, Seagen, Teva, Pfizer, Janssen Oncology, Millennium, IgM, I-Mab, Takeda, BeiGene, and AstraZeneca; and reports consultancy/honoraria from Pfizer, Seagen, Janssen Oncology, SciTek, Compliment Corporation, Millennium, Gilead Sciences, Nurix, Cellectar, Kite/Gilead, Morphosys/Incyte, I-Mab, TG Therapeutics, Pfizer, ADC therapeutics, Amgen, Actinium Pharmaceuticals, Takeda, Epizyme, and Merck. RDC has received research funding from Amgen, Kite/Gilead, Incyte, Merck, Pfizer, Servier, and Vanda Pharmaceuticals; consultancy/honoraria from Amgen, Autolus, Jazz, Kite/Gilead, and Pfizer; discloses membership on a board or advisory committee for Autolus and PeproMene Bio; and discloses that spouse was employed by and owned stock in Seagen. BGT reports research funding from Mustang Bio and Juno Therapeutics; patents/ royalties from Mustang Bio; and consultancy for Mustang Bio and Proteios Technology. DGM has received research funding from Kite, Juno Therapeutics, Celgene, Legend Biotech, and BMS; honoraria from BMS, Caribou Biosciences, Inc. Celgene, Genentech, Incyte, Juno Therapeutics, Kite, Lily, Mustang Bio, Novartis, and Umoja; and reports membership on a board or advisory committee for A2 Biotherapeutics, Navan Technologies, Chimeric Therapeutics, Genentech, BMS, ImmPACT Bio, Gilead Sciences, Interius, and BMS.

Contributions

AJP interpreted data and drafted the manuscript. JG conceived of the study, edited the manuscript, and provided critical oversight. JH abstracted chart data. ECL computed early ICAHT scores. QVW performed statistical analysis and provided statistical oversight. MT, AA, AVH, ELK, LI, CP, AKG, MS, BGT, FM, AGC, FO, RDC, RSB, YJ and DGM reviewed and edited the manuscript.

Funding

AJP received support from the Kuni Foundation (Imagination Grant) and the National Heart, Lung, and Blood Institute (T32 HL007093). JJH received support from the National Cancer Institute (5T32CA009515-39). ECL received support from the National Heart, Lung, and Blood Institute (T32 HL007093). JG received support from Swim Across America and the National Cancer Institute (P30 CA015704-45).

Acknowledgments

The authors thank Gary Schoch, Tim Monahan, and Chris Davis for maintaining the FHCC Gateway Database and aiding data retrieval.