Impact of CD19 CAR T-cell product type on outcomes in relapsed or refractory aggressive B-NHL

CD19-targeted chimeric antigen receptor-engineered (CD19 CAR) T cells are novel therapies showing great promise for patients with relapsed or refractory (R/R) aggressive B-cell non-Hodgkin lymphoma (B-NHL). To date, 4 CD19 CAR T-cell products for R/R aggressive B-NHL are already US Food and Drug Administration (FDA)-approved and commercially available: axicabtagene ciloleucel (axicel), tisagenlecleucel (tisacel), brexucabtagene autoleucel, and lisocabtagene maraleucel. Overall, clinical trial^1-6 and “real-world”^7,8 data showed promising efficacy in R/R B-NHL patients across all CAR T-cell products compared with historical reports,⁹ with overall response rates and complete response (CR) rates ranging from 51% to 93% and 40% to 64%, respectively. Tempering their high efficacy, CD19 CAR T-cell therapy can be associated with severe, potentially life-threatening toxicities,^10-12 such as cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS). Published data suggest important variations in the toxicity and efficacy rates across CAR T-cell products, but these studies were limited by the absence of comparative groups and did not specifically adjust estimates for key confounders. Using retrospective data from 129 R/R aggressive B-NHL patients treated with the FDA-approved CAR T-cell products axicel or tisacel, or the investigational product JCAR014, we aimed to estimate the independent impact of the CAR T-cell product type on toxicity and efficacy.

We performed a retrospective analysis of R/R aggressive B-NHL patients treated at our institutions with cyclophosphamide and fludarabine lymphodepletion followed by the infusion of CD19 CAR T cells. Tisacel and axicel patients were treated per standard of care using FDA-approved, commercially available products. The median tisacel dose was 2.2 × 10⁸ CAR T cells (interquartile range [IQR], 1.4-2.85; range, 0.31-4). The exact axicel dose was not measured; the target axicel dose per the FDA label was 2 × 10⁶ CAR-positive viable T cells/kg, with a maximum of 2 × 10⁸ CAR-positive viable T cells. Investigational, defined CD4⁺/CD8⁺-composition CD19 CAR T-cells (JCAR014) were administered in all patients at the dose of 2 × 10⁶/kg on a phase 1/2 clinical trial (NCT01865617) as previously described.^4,5 All JCAR014 patients included in our dataset were enrolled and apheresed between March 2015 and January 2017, before the approval by the FDA of axicel (October 2017) and tisacel (May 2018) for patients with R/R large B-cell lymphoma. Twenty-nine of 30 JCAR014 patients were infused before FDA approval of axicel and tisacel. One JCAR014 patient was apheresed in November 2016 and infused in February 2018. Details regarding inclusion and exclusion criteria for screening and JCAR014 treatment are detailed in the supplemental Material, available on the Blood Web site. History of prior transplant and bridging therapy were allowed in the study, and there was no absolute lymphocyte count (ALC)-based or T-cell function-based criteria to limit study inclusion. Since we had previously demonstrated a significant impact of the CAR T-cell dose and cyclophosphamide and fludarabine lymphodepletion on outcomes,^5,13 JCAR014 patients treated with a CAR T-cell dose of 2 × 10⁵ or 2 × 10⁷/kg and JCAR014 patients who received lymphodepletion other than cyclophosphamide and fludarabine were excluded from our analysis, as detailed in the study flowchart (Figure 1). This study was conducted with the approval of the Fred Hutchinson Cancer Research Center Institutional Review Board and in accordance with the Declaration of Helsinki.

In axicel and tisacel patients, peak CRS and ICANS severity after CD19 CAR T-cell therapy were graded retrospectively using chart review according to the ASTCT (American Society of Transplant and Cellular Therapy) consensus grading system¹⁴ and the CTCAE (Common Terminology Criteria for Adverse Events) 4.03 criteria, respectively. In patients receiving investigational JCAR014, peak CRS and ICANS severity were initially graded prospectively per protocol using the CTCAE 4.03 criteria; peak CRS severity was then retrospectively regraded using the 2021 ASTCT criteria to apply the same criteria across all product types. CRS and ICANS clinical management were conducted according to the Yescarta¹⁵ and Kymriah¹⁶ Risk Evaluation and Mitigation Strategy programs and according to established institutional guidelines. The protocol-recommended CRS and ICANS management for JCAR014 patients is shown in the supplemental Material. Severe cytopenia and infections were defined as grade ≥3 per CTCAE 4.03 criteria.

Response to CAR T-cell therapy was defined as the best response within 3 months after CAR T-cell infusion using Lugano 2014 positron emission tomography-computed tomography (PET-CT)–based response criteria or CT-based response criteria¹⁷ when a PET was not performed.

Comparisons between groups of continuous variables were made using the Kruskal Wallis test and of categorical variables using Pearson's χ-squared test or Fisher’s exact test. All P values are 2-sided and were not corrected for multiplicity. We hypothesized several preleukapheresis variables would confound the effect of the CAR T-cell product type on both toxicity and efficacy. Based on past literature and subject-matter knowledge, we prespecified variables suspected to impact both the choice of the CAR T-cell product and outcomes and/or known to independently impact outcomes. The hypothesized causal relationships between variables were built into a directed diacyclic graph (DAG) (supplemental Figure 1). DAGs are an established framework to address confounding using observational nonrandomized data.^18-21 Using the R package Dagitty²² to analyze our DAG, the following predictors were selected as the minimal adjustment set of covariates to estimate the direct effect of the CAR T-cell product type on outcomes: CAR T-cell product type, lactate dehydrogenase (LDH), largest lesion diameter, age, and hematopoietic cell transplantation-specific comorbidity index (HCT-CI). While not considered a confounder, we also included the preleukapheresis ALC. Next, we applied multivariable proportional odds logistic regression to model peak CRS (grade 0, grade 1, grade 2-4) and ICANS grade (grade 0, grade 1, grade 2, grade 3-4), and logistic regression to model day-28 response (complete or partial response vs stable or progressive disease) including the predictors in the minimal sufficient adjustment set, as described above. Missing preleukapheresis largest lesion diameter data (n = 24) were imputed using predictive mean matching with 10 imputations, using the rms::aregImpute R function. Multivariable models were fitted to all imputed datasets using the rms::fit.mult.impute() R function. For continuous variables, all reported odds ratios were rescaled to reflect an increase from the first (25% percentile) to the third quartile (75% percentile), unless stated otherwise. We used the rms::calibrate R function for internal validation of model calibration, which generates bias-corrected smooth calibration curves using bootstrapping (200 repetitions). All statistical analyses were performed using RStudio software (version 1.2.5033, RStudio, Boston, MA). The list of R packages used is available at the end of the supplemental Material. A Reproducible Data Supplement, including the R code to replicate our analyses and including tables, figures, model outputs, analysis of covariance tables, and calibration curves, is available at https://drjgauthier.github.io/lbclcartcomparison/. Datasets are available upon request to the corresponding author.

Axicel, tisacel, and JCAR014 were administered in 68 (53%), 31 (24%), and 30 (23%) patients, respectively. Most patient and disease characteristics before leukapheresis were comparable across the 3 CAR T-cell products; specifically, we could not demonstrate statistically significant differences in HCT-CI, LDH, Ann Arbor stage, largest lesion diameter, or extranodal disease (Table 1). We noted a higher proportion of patients with non-DLBCL large B-cell histologies in the JCAR014 group (23%) compared with axicel (6%) and tisacel (3%). Lymphoma histologies for JCAR014 patients are detailed in supplemental Table 1. We observed a higher percentage of patients receiving bridging chemotherapy after leukapheresis in those treated with axicel and tisacel than JCAR014 (P < .001). The number of prior therapies was higher in JCAR014 patients (P = .004), as expected in these patients treated on a phase 1/2 clinical trial before commercial availability of CD19 CAR T-cell products. There was a statistical trend toward older age in tisacel patients (median 64, compared with 62 and 60 in axicel and JCAR014 patients, respectively; P = .08). The median time from leukapheresis to CAR T-cell infusion in axicel, tisacel, and JCAR014 patients was 27, 40, and 17 days, respectively (P < .001) (supplemental Table 2).

Proportions of patients categorized by CRS and ICANS grade and by CAR T-cell product type are shown in Figure 2 and Table 2. We observed higher rates of grade ≥1 CRS in axicel (87%; 95% CI, 76-94) and tisacel (71%; 95% CI, 52-86) patients compared with JCAR014 (47%; 95% CI, 28-66; P < .001) and grade ≥2 CRS with axicel compared with tisacel and JCAR014 (P = .03). We observed significantly higher rates of grade ≥1 (62%; 95% CI, 49-73; P < .001), grade ≥2 (51%; 95% CI, 39-64; P = .002), and grade ≥3 ICANS (29%; 95% CI, 19-42; P = .04) in axicel patients compared with those who received tisacel and JCAR014. A higher proportion of patients requiring tocilizumab was seen in axicel (49%) and tisacel (42%) patients compared with JCAR014 (7%; P < .001), and higher proportion of patients requiring corticosteroids with axicel (60%) compared with tisacel (16%) and JCAR014 (10%; P < .001).

Rates of severe infection occurring within 30 days after axicel, tisacel, or JCAR014 were 9%, 3%, and 0%, respectively (P = .20). We observed higher rates of severe neutropenia, thrombocytopenia, and anemia at day 30 in axicel and tisacel patients compared with JCAR014 patients (supplemental Table 3). Early death before disease restaging at day 30 after CAR T-cell infusion was reported in 3 axicel patients (ventricular tachycardia in the context of severe ICANS, ischemic stroke, and neutropenic colitis, n = 1; cytomegalovirus pneumonia complicated with diffuse alveolar hemorrhage and multiorgan failure, n = 1; Escherichia coli septic shock, n = 1).

The best overall response rates after axicel, tisacel, and JCAR014 were 75% (95% CI, 63-85), 58% (95% CI, 39-75), and 57% (95% CI, 37-75), respectively (P = .10). The complete response rates after axicel, tisacel, and JCAR014 were 55% (95% CI, 43-68), 32% (95% CI, 17-51), and 40% (95% CI, 23-59), respectively (P = .08).

Nonadjusted and adjusted odds ratios (ORs) suggested JCAR014 was independently associated with significantly lower CRS severity compared with axicel (adjusted odds ratio [aOR], 0.19; 95% CI, 0.08-0.46; P < .001). The aOR was also compatible with reduced CRS severity with tisacel compared with axicel (aOR, 0.47; 95% CI, 0.21-1.06; P = .07), although we could not exclude a null effect at the 0.05 level. The impact of age on CRS severity was undetermined (aOR per 10-year increase, 0.86; 95% CI, 0.65-1.10; P = .28). Both tisacel and JCAR014 were associated with lower ICANS severity compared with axicel (aORs, 0.17 and 0.17; 95% CI, 0.06-0.48 and 0.06-0.47; P < .001 and P < .001, respectively). The impact of age on ICANS risk was also undetermined (aOR per 10-year increase, 1.30; 95% CI, 0.92-1.70; P = .15).

Unadjusted and adjusted OR for CR suggested lower antitumor efficacy with tisacel (aOR, 0.23; 95% CI, 0.06-0.78; P = .02) and JCAR014 (aOR, 0.21; 95% CI, 0.06-0.73; P = .01) compared with axicel, as shown in Table 3, using a multivariable model not including interaction terms. The preleukapheresis LDH (aOR, 0.22; 95% CI, 0.10-0.51; P < .001), largest lesion diameter (aOR, 0.29; 95% CI, 0.09-0.87; P = .03), and ALC (aOR, 2.41; 95% CI, 1.22-4.76; P = .01) were also strongly and independently associated with the odds of CR (Figure 3). Since a higher proportion of tisacel patients underwent CT-based restaging alone (39%) compared with axicel (9%) and JCAR014 (10%) patients, we performed sensitivity analyses modeling CR by either CT- or PET-based criteria to assess the impact of imaging type on our estimates. As shown in Table 3, aORs still suggested JCAR014 was associated with lower efficacy (aOR for CT-based CR, 0.25; 95% CI, 0.06-1.06; P = .06; aOR for PET-based CR, 0.30; 95% CI, 0.09-1.00; P = .05). In contrast, the impact of tisacel compared with axicel was undetermined (aOR for CT-based CR, 0.78; 95% CI, 0.24-2.52; P = .68; aOR for PET-based CR, 0.42; 95% CI, 0.11-1.54; P = .19). Additional sensitivity analyses are available in the supplemental Material (supplemental Table 5, addition of bridging therapy to the set of covariates; supplemental Table 6, exclusion of JCAR014 patients; supplemental Table 7, inclusion of prelymphodepletion instead of preleukapheresis ALC).

We hypothesized that 4-1BB costimulated products (tisacel and JCAR014) might be differentially affected by tumor burden and T-cell function-altering factors compared with the CD28-costimulated product axicel. Hence, we tested for the presence of covariates modifying the effect of the CAR T-cell product type on outcomes (ie, interaction effects). To maximize statistical power, and since our models would not fit with the 3-level CAR T-cell product type variable due to the limited sample size, the tisacel and JCAR014 categories were pooled into a single category to create a 2-level categorical variable (axicel, tisacel/JCAR014). Multivariable models for CRS grade, ICANS grade, and complete response were fitted, including the following interaction terms: CAR T-cell product*preleukapheresis LDH, CAR T-cell product*preleukapheresis largest lesion diameter, CAR T-cell product*age, and CAR T-cell product*HCT-CI.

We could not detect significant associations between the interaction terms listed above and CRS (supplemental Table 6) or ICANS (supplemental Table 7). In contrast, our multivariable modeling suggested an interaction effect between the preleukapheresis ALC (P = .04) and the CAR T-cell product type on the odds of achieving a CR (supplemental Table 8) with a stronger association between ALC and efficacy in axicel patients compared with tisacel/JCAR014 patients. The probabilities of CR as a function of the CAR T-cell product type and preleukapheresis ALC are shown in Figure 4.

Single-arm clinical trial data of CD19 CAR T-cell therapy have suggested important variations in the toxicity and efficacy rates across distinct CAR T-cell products. Using data from patients treated with an FDA-approved, commercially available CAR T-cell product (axicel or tisacel) or the investigational product JCAR014, we aimed to estimate the independent impact of the CAR T-cell product type on outcomes. To minimize treatment allocation bias, we used an established causal framework based on a DAG to adjust for the key patient and disease characteristics that might have influenced both the choice of CAR T-cell product and outcomes after CAR T-cell therapy.

We found that axicel was independently associated with higher ICANS severity compared with tisacel and JCAR014 and with higher CRS severity compared with JCAR014. Although retrospective, nonblinded grading of toxicities could have introduced bias, we also observed longer CRS duration and higher corticosteroid use after axicel than tisacel and JCAR014 (Table 2).

As previously described by our group and others, the preleukapheresis LDH^4,8 and ALC⁷ were strong predictors of CR after CD19 CAR T-cell therapy. Furthermore, we detected an interaction effect between the preleukapheresis ALC and the CAR T-cell product type, implying the efficacy of distinct CD19 CAR T-cell products may be differentially affected by factors impacting T-cell function. While higher CR rates were observed and predicted after axicel in preleukapheresis ALC_high patients, lower and comparable efficacy was observed and predicted in ALC_low patients across all CAR T-cell products (Figure 4). More research is needed to characterize the mechanisms responsible for this differential impact.

An important limitation of our study is that differences in response assessments may have influenced the observed response rates in our dataset. Specifically, CT-only assessments were performed more frequently in tisacel patients (39%) due to distinct insurance policies between institutions and states for PET imaging coverage, compared with axicel (9%) and JCAR014 (10%) patients. When considering PET criteria or CT criteria alone, the impact of tisacel on the odds of CR compared with axicel became undetermined (Table 3). While this could be related to low power (due to the reduced sample size), this also highlights the need for homogenous response assessments across CAR T-cell products.

In this study, we chose to compare clinical trial patients to real-world patients. It is generally accepted that clinical trial data suffer from the selection of patients with less aggressive disease. In contrast, we found that patients treated with investigational JCAR014 had been more heavily pretreated and that response rates were comparable or inferior to real-world patients.

It is also possible that other confounding factors related to tumor aggressiveness not captured in our dataset may have confounded our estimates. We tried to minimize this bias by adjusting for a measure of both tumor kinetics (preleukapheresis LDH) and bulk (largest lesion diameter before leukapheresis). Recent publications have suggested the total metabolic tumor volume may be a better predictor of outcomes after CD19 CAR T-cell therapy over LDH or tumor bulk.^23,24 Unfortunately, the total metabolic tumor volume is not routinely measured at our institutions, and the data were not available to be included in our models. Last, unmeasured confounders, such as physician and patient preferences and insurance coverage, not modeled in our DAG, might have also generated allocation bias. A randomized clinical trial would address this limitation but is not currently forthcoming as it would most likely require the support of nonindustry sponsors such as the National Clinical Trial Network of the National Cancer Institute.

Given significant differences in the manufacturing process between JCAR014 and JCAR017 (JCAR014, antigen presenting cell-based ex vivo stimulation with an infusion of a fresh product containing 2 × 10⁶CAR T cells/kg; JCAR017, cytokine-based ex vivo stimulation with an infusion of a cryopreserved product containing a total dose of 50 to 110 × 10⁶ CAR T cells), our findings should not be extrapolated to compare outcomes between lisocel and other commercially approved CAR T-cell products. A recently published matching-adjusted indirect treatment comparison of lisocel vs axicel suggested comparable efficacy and a more favorable safety profile with lisocel.²⁵ 

In conclusion, our analyses show a strong and independent impact of the CAR T-cell product type on toxicity severity and antitumor efficacy. Antitumor efficacy was comparable across all 3 products in preleukapheresis ALC_low patients. Higher CR probabilities were observed and predicted after axicel in ALC_high patients compared with tisacel and JCAR014. Further studies are needed to analyze cohorts of patients including other FDA-approved CAR T-cell products (eg, lisocel, brexucabtagene autoleucel) to confirm our findings, pursue comparative studies of health resource utilization, and further optimize patient and CAR T-cell product selection in patients with R/R B-NHL.

The authors thank both the Fred Hutchinson Cancer Research Center Cell Processing Facility and Seattle Cancer Care Alliance Cell Therapy Laboratory, the staff of the Program in Immunology and Seattle Cancer Care Alliance Immunotherapy Clinic, and the Oregon Health and Science University Cell Therapy Clinical, Administrative, and Laboratory Programs.

J.G. thanks the National Institutes of Health/National Cancer Institute Cancer Center Support Grant (P30 CA015704-45).

Contribution: J.G. and R.T.M. conceptualized and designed the study; J.G., N.G., A.V.H., C.J.T., D.G.M., S.W., J.M., and J.A.H. collected and assembled data; J.G., C.J.T., D.G.M., and R.T.M. analyzed and interpreted data; and all authors wrote and approved the final manuscript and are accountable for all aspects of the study.

Conflict-of-interest disclosure: J.G. reports honoraria from Larvol, Eusapharma, JMP, Multerra Bio; research support from Juno Therapeutics, a Bristol Myers Squibb company, and Sobi; and participation in an advisory board for JNJ/Legend Biotech. A.V.H. reports honoraria from Bristol Myers Squibb and Novartis and research support from Juno Therapeutics, a Bristol Myers Squibb company. A. Chen has participated in advisory board meetings for Morphosys, Mesoblast, and Fate Therapeutics. J.A.H. has served as a consultant for Allogene and Gilead, with research support from Gilead. S.N. is employed by Novartis. V.A.C. has received research funding from AstraZeneca. D.G.M. has received research funding paid to his institution and honoraria from Juno Therapeutics, a BMS Company, Celgene, a BMS Company, and Kite Pharma; has participated in ad hoc advisory board meetings and received honoraria from Amgen, BMS, Genentech, Gilead Sciences, Incyte, Janssen, Legend Biotech, Mustang Bio, MorphoSys, Novartis, Pharmacyclics, and Umoja; has rights to royalties from Fred Hutch for patents licensed to Juno/BMS; and is a member of the A2 Biotherapeutics Scientific Advisory Board with stock options and compensation and is a member of the Navan Technologies Scientific Advisory Board with stock options and compensation. C.J.T. receives research funding from Juno Therapeutics/BMS and Nektar Therapeutics; serves on Scientific Advisory Boards for Precision Biosciences, Eureka Therapeutics, Caribou Biosciences, T-CURX, Myeloid Therapeutics, ArsenalBio, and Century Therapeutics; has stock/options in Precision Biosciences, Eureka Therapeutics, Caribou Biosciences, Myeloid Therapeutics, and ArsenalBio; has the right to receive royalties from Fred Hutch as an inventor on patents related to CAR T-cell therapy that are licensed to Juno Therapeutics/BMS; and served on ad hoc advisory boards (last 12 months) for GlaxoSmithKline. M.L.S. is a member of consulting, advisory boards, steering committees, or data safety monitoring committees: Abbvie, Genentech, AstraZeneca, Sound Biologics, Pharmacyclics, Beigene, Bristol Myers Squibb, Morphosys/Incyte, TG Therapeutics, Innate Pharma, Kite Pharma, Adaptive Biotechnologies, Epizyme, Eli Lilly, Adaptimmune, Mustang Bio, Regeneron, Merck, Fate Therapeutics, MEI Pharma, and Atara Biotherapeutics; Research funding: Mustang Bio, Celgene, Bristol Myers Squibb, Pharmacyclics, Gilead, Genentech, AbbVie, TG Therapeutics, Beigene, AstraZeneca, Sunesis, Atara Biotherapeutics, Genmab, and Morphosys/Incyte. R.T.M. is an advisor or consultant for AlloVir, Artiva, CRISPR Therapeutics, CytoDyn, Incyte, and Novartis; reports honoraria from Bristol-Myers Squibb/Celgene, Incyte, Intellia, Kite, Omeros, Orca BioSystems, and PACT Pharma; research support from BMS/Celgene/Juno and Novartis; participation in a data and safety monitoring board for Athersys, Novartis, and Vor Pharmaceutical; and holds a patent with Athersys. The remaining authors declare no competing financial interests.

Correspondence: Jordan Gauthier, Clinical Research Division, Fred Hutchinson Cancer Center, 1100 Fairview Ave N, Seattle, WA 98109; e-mail: jgauthier@fredhutch.org.