Toxicity was considered as events within 30 days of CAR-T infusion or during index-admission. Individual CRS components were assessed according to consensus (Penn) criteria.20 Specifically, fever, coagulopathy, hypoxia, hypotension, and end organ toxicity (eg, renal failure) were included as criteria for incident or worsening CRS (online supplemental table 1). Fever was defined as oral temperature >38°C, and hypoxia was considered oxygen saturation <90% on room air or need for supplemental oxygen, while hypotension was considered as minimum systolic blood pressure (SBP) of <105 mm Hg. We selected <105 mm Hg due to the desire to capture even those with probable relative hypotension. Neurotoxicity grades included encephalopathy, seizure, dysphasia, tremor, headache, confusion, depressed level of consciousness, and cerebral edema.20 Furthermore, CVD events were defined as myocarditis, congestive heart failure, stroke, myocardial infarction, symptomatic arrhythmia (eg, atrial fibrillation ventricular tachycardia), or cardiac death.12 13

Available clinical laboratory measures, including c-reactive protein (CRP), ferritin, complete blood counts with differential, troponin, alanine aminotransferase, and lactate dehydrogenase (LDH) were also considered. These were widely available and routinely measured among treated patients, including those with concern for toxicity.23 24 The maximum value was included where multiple measures were obtained.

The primary outcome was the development of cancer disease progression or death (ie, PFS) after CAR-T initiation. The secondary outcome was the occurrence of partial or complete response at 1 year, and overall survival. Follow-up began from the time of CAR-T initiation. Toxicity events were graded using CTCAE V.5, and Lugano consensus criteria. Moreover, we assessed the incidence and outcomes CRS, by CAR-T product employed (axicabtagene ciloleucel (axi-cel), and tisagenlecleucel (tisa-cel), respectively). Axicabtagene ciloleucel and brexucabtagene autoleucel were considered together as axi-cel, as the two products are clinically identical, with some difference in T-cell enrichment/processing.

Descriptive statistics were used to summarize patient characteristics, using mean±SD or median (IQR) for continuous variables and frequency counts with percentages for categorical variables. Time-to-event analysis methods were used to summarize PFS and evaluate associations with these outcomes. Patients without progression or death were considered censored at the last follow-up date. PFS was estimated and displayed by Kaplan-Meier curves. Cox proportional hazards models were used to assess association of patient factors with PFS. Univariable models were fit using maximum CRS grade as the independent variable, followed by using individual CRS components, neurotoxicity, and patient laboratory values. Multivariable Cox proportional hazards models were then fit using potential risk factors identified in the univariable models. To avoid overfitting, we considered the number of PFS events when evaluating how many independent variables to include in the models. Log-rank tests and Kaplan-Meier curves were also used to compare PFS by product type received. Furthermore, χ² tests were used to assess differences in CRS grade, and its’ individual components (eg, hypotension, etc.) between patients with complete (disease) response at 3, 6, and/or 12 months, versus those not seeing complete response. Multivariable logistic regression was used to account for potential confounders in 12-month disease response. To further delineate the impacts of toxicity, we performed additional landmark analysis of time to PFS, beginning at 30 days after CART initiation. This analysis included only those patients who survived to 30 days without disease progression, stratified by toxicity grade.

In order to understand the potential contribution of specific components of CRS in association with clinical outcomes, individual components of CRS, as well as the presence of neurotoxicity, and cardiotoxicity, respectively, were considered. Blood pressure variables were included alternatively as hypotension (SBP <80 mm Hg) and percent SBP decrease ranges; these were not included together in any model so as to avoid multicollinearity issues. A univariable model for overall survival was also fit with maximum CRS grade as the independent variable. We also assessed the effects of the antitoxicity interventions (tocilizumab and steroids, respectively, at grade ≤2 toxicity) on the maximum grade observed. All statistical tests were two-sided and evaluated at the α=0.05 type-I error rate, with no adjustments for multiple comparisons or evaluation of multiple cut points. All analyses were performed with SAS Software V.9.4.

Following CAR-T infusion, 88.9% (80) of patients developed CRS (figure 1), with a median time to CRS onset of 2.5 days (IQR: 1–5 days; range: 0–9 days). This included 98.3% (57/58) with CRS after axi-cel, and 71.9% (23/32) after tisa-cel infusion. Of those patients who developed CRS, 38.7% (31) saw a maximum severity of grade 2, and 16.3% (13) reached at least grade 3; with hypotension (87.8% (79)), followed by fever (86.7% (78)) being the most common manifestations (online supplemental figure 1; table 2); 52.2% patients experienced a minimum SBPs of ≤90 mm Hg. There were no specific clinical or disease-related factors associated with the development of CRS. Neurotoxicity occurred in 41 (45.6%) patients, including 34 (37.8%) reaching at least grade 2 (online supplemental figure 2). Furthermore, all patients with neurotoxicity experienced CRS within the same admission. In addition, 17 (18.9%) patients developed a CVD event within days of infusion, including 11 (12.2%) with arrhythmias, 2 (2.2%) with myocarditis, 1 (1.1%) heart failure, and 3 (3.3%) with other CVD events (online supplemental figure 2). Among those with cardiac events, there was no difference in prevalence of concurrent or preceding high-grade (≥3) CRS (p=0.3).

The use of axi-cel was associated with a higher incidence of CRS (98.3% vs 71.9%, p<0.01; online supplemental table 2). The median time to CRS onset differed by product (3.0 vs 3.4 days for axi-cel and tisa-cel, respectively; p=0.45), as did maximum CRS grade observed (1.93±1.1 vs . 1.15±1.08 for axi-cel and tisa-cel, respectively; p=0.005). Neurotoxicity was more common after axi-cel (63.6% vs 20.3%, p=0.001). There was no difference in cardiotoxicity development, by CAR-T product. There was also no difference in PFS or mortality, by product.

Over a median follow-up of 16.5 months (range: 1–54 months), the median PFS was 6.0 months (95% CI: 3.3 to 8.8 months), and the 25th percentile for overall survival was 9.6 months (95% CI: 5.1 to 14.9 months).Fifty-four (60.0%) patients experienced disease progression or death at 12 months (online supplemental table 3). Complete remissions was attained in 42 (4672%) patients at 3 months postinfusion, while nine (10.0%) had a partial remission, and 38 (42.2%) had progressive (unresponsive) disease; one (1.1%) had stable disease. At 12 months, complete remission remained in 30 (33.3%), with no partial remissions; and 54 (60.0%) had progressive disease. Overall survival among all patients was 90.0% at 3 months, 80.0% at 6 months, and 72.2% at 12 months, with corresponding PFS of 57.8% at 3 months, 47.8% at 6 months, and 40.0% at 12 months. Response rates were similar to the ZUMA-11 and JULIET2 trials.

When stratified by CRS-status, maintenance of PFS at 3 months was observed among 70.0% of those with grades 0–1 CRS, and 55.0% with grade 3–4, and 80.0% in those with grade 2 CRS. This pattern remained through 12 months, with 36.0% maintaining PFS with grade 0–1, 33.0% with grade 3–4, and 54.0% with grade 2, respectively (figure 2). Similarly, complete disease response at 1 year was highest among those with moderate (grade 2) CRS (figure 3). These relationships remained even after accounting for other factors linked with disease and therapeutic outcomes (HR: 0.52; 95% CI: 0.19 to 0.88; p=0.03 in landmark analysis; table 3, online supplemental figure 3 and online supplemental table 4). Furthermore, similar differences in overall survival by CRS status were also observed, with moderate CRS seeing improved survival (HR: 0.43; 95% CI: 0.20 to 0.94; p=0.03; online supplemental figure 4 and online supplemental table 4).

In univariable analysis of CRS components, hypoxia (HR: 1.88; 95% CI: 1.02 to 3.47; p=0.044), coagulopathy (HR: 11.62; 95% CI: 4.26 to 31.67; p<0.01), organ toxicity (HR: 3.99; 95% CI: 1.92 to 8.33; p<0.01), and profound hypotension (minimum SBP <80 mm Hg; HR: 2.07; 95% CI: 1.11 to 3.87; p=0.02) were linked with worse PFS (online supplemental table 3). Yet, in a multivariable model, only coagulopathy remained associated with worse PFS (HR: 4.72; 95% CI: 1.22 to 18.23; p=0.024; table 3). On stratification by temperature, there were no differences in outcomes. However, on stratification by the degree of SBP reduction, those with a minimum SBP of 80–104 mm Hg saw PFS at 3 and 12 months of 74.0% and 50.0%, respectively, compared with 70.0% at 3 months and 11.0% at 12 months in those with minimum SBPs of ≥105 or the 50.0% at 3 months and 18.0% at 12 months in those with SBPs<80 mm Hg; online supplemental figure 5. Furthermore, by the degree of SBP change postinfusion (≤80, 81–104, or ≥105 mm Hg SBP after CAR-T treatment blood pressure, and absence of active infection), SBP of ≤80 or ≥105 mm Hg was associated with higher PFS, which remained in multivariable analysis (HR: 2.25; 95% CI: 1.29 to 3.91; p<0.01; online supplemental table 5). Relative percent fall (≤10%, >10%–40%, or >40% decline in SBP from pretreatment blood pressure, and absence of active infection) in SBP of ≤10% or >40% was associated with higher PFS on univariate analysis, but not in multivariable analysis. There was no association between the development of other cardiovascular or neurologic events and PFS.

Where measured, the presence of CRP (≥50 mg/L), ferritin (≥500 ng/mL), and LDH (>280 units/L) elevation, respectively, was associated with lower rates of PFS (online supplemental figure 5). However, following multivariable adjustment, outside of ferritin, no associations with outcomes were observed (online supplemental table 4). Similarly, there was no association between prothrombin time, d-dimer, or alkaline phosphatase and disease outcomes after CAR-T infusion. Furthermore, baseline CRP was known in the majority (72.2%) of patients, with an average delta (Δ) peak increase in CRP of 24.0 mg/L, at a median of 3 days post-CART infusion. Among those with available measures, there was a correlation between Δ CRP and maximum CRS grade (r²=0.28, p=0.02).

The effect of CRS grading system on PFS was assessed based on the three primary CRS grading criteria: Penn, Transplantation and Cellular Therapy(TCT), and Lee, respectively. The variance across the three scoring scales was small. The proportion of patients with grade 2 CRS was 34.4% by Penn criteria, 34.4% by Lee criteria, and 38.9% by TCT criteria. There was no difference in the relationship between grading system and PFS (p=0.9995, online supplemental figure 6). Furthermore, on multivariate analysis, a similar degree of significance was maintened between the relationship of b moderate CRS and PFS outcomes, by Penn, TCT, and Lee criteria (p=0.067, 0.067, and 0.033, respectively; online supplemental table 6); with axicel showing stronger associations (HR: 0.40; 95% CI: 0.18 to 0.87; p=0.02; online supplemental tables 7 and 8).

Tocilizumab, an interleukin-6 inhibitor, was used in 36 (40.0%) patients due to worsened CRS, with a mean CRS grade of 1.87±0.63 (median grade of 2) at the time of initiation. There was no difference in mortality (p=0.91), CRS worsening, cardiotoxic or neurotoxic events, by tocilizumab-status (p=0.99). Furthermore, steroids were used in 39 (43.3%) patients due to neurotoxicity (and three patients received steroids for CRS), with a mean toxicity grade at the time of initiation of 2.17±0.71 (median grade 2). Similarly, there was no difference in mortality (p=0.538), including in those with concurrent CRS (p=0.573); or the likelihood of neurotoxicity worsening (p=0.594), or cardiotoxic events (p=0.99). Furthermore, there was no difference in PFS or disease response based on the presence or absence of tocilizumab or steroid use, respectively (online supplemental table 4).

Several limitations should be acknowledged. Given the retrospective nature of the study, available follow-up within the cohort was not uniform. Also, we could not exclude information bias. Uniform cardiac, neurologic, and correlative laboratory testing protocols were not in place near the initiation of programmatic CAR-T-based treatments, as many patients were in early phase clinical trials. Similarly, timely tocilizumab and steroid application was at the discretion of the treating clinicians and thus varied across time. CART management strategies shifted over time. Notably, the proportion of patients receiving tisa-cel increased after FDA approval in 2018. Furthermore, tocilizumab was utilized more frequently in practice than was initially outlined in the prior JULIET and ZUMA-1 trials.1 2 Despite these factors, the data represent real-world practice, and are thus generalizable to this population. Although we did not observed mitigation in maximum toxicity grade by intervention antitoxicity therapies, it is not known if treatment in combination with supportive therapies (eg, fluids), may have shortened toxicity durations,28 or prevented some patients from reaching high grades or death. We did not include children. Relationships with axi-cel were more robust. Keppra was not commonly used for seizure prophylaxis. The population consisted primarily of DLBCL patients, and it is unclear if the same relationships would be observed among non-lymphoma patients. The current study focused on relapsed and refractory lymphoma patients treated with axi-cel and tisa-cel, respectively, and did not include data from more recently available CAR-T products or other disease populations. Pooled product data were employed for most analyses, as sample size limited further product subgroup analyses. Yet, there was no difference in PFS by CAR-T product used. However, given the desire to best reflect commonly used CAR-Ts in contemporary clinical practice, we focused on these high-profile and more available therapies. Additionally, given that the risk of non-CRS toxicities (namely cardiotoxicity) was not known during the early experience with CAR-T therapy, several adverse events may have gone uncaptured despite extensive search.