Comparing CAR T-cell toxicity grading systems: application of the ASTCT grading system and implications for management

Two anti-CD19 chimeric antigen receptor (CAR) T-cell products are approved by the US Food and Drug Administration (FDA): axicabtagene-ciloleucel (axi-cel; Yescarta, Kite/Gilead) for the treatment of adult relapsed/refractory (R/R) diffuse large B-cell lymphomas (DLBCL), based on results of the ZUMA-1 trial (clinicaltrials.gov #NCT02348216),¹  and tisagenlecleucel (Kymriah, Novartis Pharmaceuticals), both for adult R/R DLBCL (JULIET trial; #NCT02445248)²  and pediatric R/R B-cell acute lymphoblastic leukemias (B-ALL; ELIANA trial; #NCT02435849).³  These therapies have been associated with specific toxicities, including cytokine release syndrome (CRS) and neurotoxicity,^4,5  for which investigators involved in the pivotal clinical trials developed different grading systems. For CRS, they include (1) consensus-based score by Lee et al (referred to herein as Lee),⁶  used in the ZUMA-1 trial¹ ; (2) University of Pennsylvania’s score (referred herein to as Penn),⁷  used in the JULIET and ELIANA trials^2,3 ; (3) National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE), version 5.0 (CTCAEv5.0)⁸ ; (4) the Memorial Sloan Kettering Cancer Center (MSKCC) grading used in trial #NCT01044069⁹ ; and (5) a score developed by the multiinstitutional CAR T-Cell Therapy–Associated Toxicity (CARTOX) working group.¹⁰  In most clinical trials, neurotoxicity has been reported based on the presence of any symptoms of encephalopathy individually graded according to CTCAE, version 4.03 (CTCAEv4.03).¹¹  The CARTOX group proposed a separate grading system for neurotoxicity, the CARTOX-10 test, as a measure of cognitive function.¹⁰  However, these distinct scores differ slightly in variables included and in definitions of severity. Using different grading systems in the postapproval clinical era and in clinical trials not only complicates comparisons between different products, but may also impact clinical management. Recently, the American Society for Transplantation and Cellular Therapy (ASTCT) proposed a consensus grading system that includes a simplified version of the Lee score for CRS and introduces the definition “immune effector cell–associated neurotoxicity syndrome” (ICANS) and a minimally modified version of the CARTOX-10 test, renamed “immune effector cell–associated encephalopathy” (ICE) score.¹²  The ASTCT consensus score is easily calculated at the bedside and has been adopted increasingly in clinical practice. However, to our knowledge, CRS and ICANS have not been reported according to this score. To investigate the ASTCT grading system, we compared it to the other scores in adult patients with B-ALL and DLBCL treated at MSKCC.

Eligible patients included adults with B-ALL from a previously reported phase I trial of 1928z CAR T-cells from 2010 through 2016 (#NCT01044069),⁹  and patients with DLBCL treated with axi-cel or tisagenlecleucel from February 2018 through March 2019. Toxicities were graded according to ASTCT grading and the other scores: Lee, Penn, MSKCC, CTCAEv5.0, and CARTOX, for CRS, and CTCAEv4.03 and CARTOX for ICANS (supplemental Tables 1 and 2). Approval for this retrospective review was obtained from the Institutional Review Board.

Data were retrospectively extracted from electronic medical records. Individual CRS and ICANS signs and symptoms were used to assign final scores for each grading system. The following indicators of CRS were recorded: fever ≥38°C; hypotension requiring >1 fluid bolus, vasopressor administration, minimum blood pressure (BP) values; oxygen requirement for hypoxia/dyspnea; organ damage according to CTCAEv5.0 (tachycardia and/or arrhythmia, reduction in left ventricular ejection fraction, nausea/emesis and/or diarrhea, liver enzymes and/or bilirubin increase, creatinine increase, and coagulation alterations). For ICANS, we recorded ICE score symptoms (orientation, naming, following commands, writing, and attention), lethargy, reduced consciousness, motor findings, seizures, and cerebral edema [including papilledema and cerebrospinal fluid (CSF) opening pressure (OP) evaluation], according to CTCAEv4.03. Given minimal differences between CARTOX-10¹⁰  and ICE,¹²  data were collected using the ICE score for both CARTOX and ASTCT grading. Because the B-ALL cohort was mostly treated before the introduction of CARTOX-10/ICE tests, ICE score data were obtained from a modified Mini-Mental Status Exam (including, among others, all components of the ICE score [orientation, attention span and concentration, naming, command following], except writing), which were performed in all patients with neurological complications. When ICE score data were missing, a range was calculated between the minimum score obtained with the available data and the maximum score appraisable if complete data had been obtained. For example, if a patient had missing data on writing and attention tasks (1 point each) and, based on other tasks, had a total ICE score of 7, an ICE score range of 7 to 9 was assigned. Alternatively, the score was recorded as nonavailable. CRS and ICANS were considered severe for grades ≥3. All toxicity data for the DLBCL cohort was collected prospectively and overseen by an internal consensus group. To enhance accuracy, 2 experts independently graded all toxicities for patients with B-ALL (one each for CRS and ICANS). After review of discrepancies with a third expert, the interinvestigator grading variability was 3% for CRS and 2% for ICANS.

Patients with B-ALL had been treated in an earlier era. As a result, given the significant differences in treatment of toxicities in the 2 cohorts, we selected the patients with DLBCL (treated with axi-cel or tisagenlecleucel) and used recorded signs of CRS/ICANS to predict individual treatment interventions (administration of tocilizumab and/or steroids) that would be made according to published management guidelines: axi-cel and tisagenlecleucel FDA package inserts and CARTOX and NCCN guidelines (supplemental Tables 3 and 4).^9,11-13  Predicted treatment was then paired to ASTCT grades and compared with the actual treatment received at our institution.

We evaluated concordance on rates of CRS and ICANS (yes vs no) across all grading systems. Then, the rate of concordance grade by grade was calculated for CRS/ICANS as the sum of cases for which all grading systems agreed on the same score divided by the total number of CRS/ICANS cases (for example: total patients scored as grade 1 in all CRS grading systems, plus total patients scored as grade 2 in all CRS grading systems, and so on, until grade 5). To evaluate agreement on CRS/ICANS grading between different grading systems, Cohen’s κ coefficient was calculated combining grading systems in pairs and evaluating scores as ordinal variables.

One hundred two patients were included: 53 with B-ALL, who received 1928z CAR T-cells (#NCT01044069),⁹  and 49 with DLBCL, 36 treated with axi-cel and 13 with tisagenlecleucel, based on physician’s preference (Table 1). Clinical data for toxicity grading were available in all patients who developed CRS (n = 85) and ICANS (n = 56). However, ICE score data were complete in only 52% (29 of 56) of patients, including all 23 DLBCL patients and 6 B-ALL patients. An ICE score range was calculated in 25% of patients (14 of 56), whereas in 23% of patients (13 of 56) in whom a range could not be obtained, the grade was determined by other ICANS defining symptoms (mainly seizures). Rates of specific supportive treatment interventions are reported in supplemental Table 6.

Table 2 summarizes the incidence of CRS and ICANS by ASTCT grading. Eighty-two percent (84 of 102) of patients had CRS of any grade, with grade ≥3 CRS in 20% (20 of 102). When analyzed by disease, total incidence of CRS of any grade was 87% (46 of 53) in B-ALL patients, with 28% (15 of 53) of grade ≥3, and 77% (38 of 49) in DLBCL patients, with 10% (5 of 49) of grade ≥3. Within the DLBCL cohort, CRS incidence was 86% (31 of 36) and 54% (7 of 13) for axi-cel and tisagenlecleucel, respectively, and grade ≥3 CRS occurred only in patients treated with axi-cel (14%; 5 of 36). Fifty percent (51 of 102) of the overall population developed ICANS of any grade. ICANS was recorded in 55% (29 of 53) of B-ALL patients, grade ≥3 in 45% of cases (24 of 53), and in 45% (22 of 49) of DLBCL patients, with 24% (12 of 49) grade ≥3. Within the DLBCL cohort, 55% of patients (20 of 36) treated with axi-cel developed ICANS, grade ≥3 in 33% (12 of 36), whereas 15% (2 of 13) of tisagenlecleucel patients experienced ICANS, with no grade ≥3.

The global incidence of CRS across all grading systems was 83% compared with 82% by ASTCT grading (85 vs 84 of 102 patients), leading to overall CRS concordance in 99% of patients. The discordant case was a patient who developed CRS symptoms with a maximum temperature of 37.6°C. Because fever ≥38°C is required for CRS diagnosis, he did not meet ASTCT criteria for CRS; however, he was scored as grade 2 by other grading systems.

When CRS was evaluated by grade, 25% (21 of 85) of patients had the same CRS grade across all grading systems. The major differences were related to grades 1 to 3 and were seen mainly in the Penn grading (Figure 1A). Excluding Penn grades, the overall concordance rate across grading systems increased to 75% (64 of 85). The CRS intergrading systems agreement was generally high (Figure 1C; Cohen’s κ coefficient, ranging from 0.69 to 0.98) except for Penn grading (ranging from 0.08 to 0.12). When comparing ASTCT to Penn grades (Figure 2A), 91% (32 of 35) of patients with grade 1 CRS were upgraded to Penn grade 2 for febrile neutropenia requiring inpatient administration of intravenous antibiotics. Similarly, 93% (27 of 29) of ASTCT grade 2 patients were upgraded to Penn grade 3 for fluid-responsive hypotension (n = 7), low-dose oxygen treatment of hypoxia or dyspnea (n = 10), or the presence of both (n = 10). Conversely, grade 4 rates were mostly overlapping across all scales. When analyzed by disease (B-ALL and DLBCL) and products (axi-cel and tisagenlecleucel in DLBCL), analogous inter-grading systems concordance was observed (Figure 3). When comparing CRS rates by disease, we found a statistically significant difference for B-ALL vs DLBCL, if CRS was graded according to the ASTCT, CARTOX, Lee, and CTCAEv5.0 systems, but not by Penn and MSKCC, further highlighting the variability across grading systems (supplemental Table 5).

Excluding upgrading by Penn, other discrepancies relied on different evaluation of fluid-responsive hypotension, which is grade 1 by MSKCC, as well as CARTOX when systolic BP is >90, whereas it is scored grade 2 by ASTCT, independent of BP values. This led to downgrading of 27% (8 of 29) of ASTCT grade 2 patients, 3 of whom received tocilizumab for hypotension. Use of low- vs high-dose vasopressors did not result in a grading discrepancy. Other differences resulted from CARTOX and Lee inclusion of CRS-related organ damage. Ninety-four percent (80 of 85) of CRS patients had at least 1 organ involved (cardiac, gastrointestinal, hepatic, renal, or coagulation alterations), but only 8 patients (9%) were upgraded by CARTOX and/or Lee, compared with ASTCT, because of organ damage: 2 for reduced left ventricular ejection fraction, 5 for transaminitis, and 1 for coagulation disorders (Figure 2A).

According to ASTCT and CARTOX, the ICANS rate was 50% (51 of 102) in the overall population. In comparison, the rate of ICANS by CTCAEv4.03 was slightly higher at 55% (56 of 102), leading to an overall concordance of 91%. Five patients (4 with B-ALL and 1 with DLBCL) had mild headache, confusion, or slurred speech with recurrent troubles in word-finding, either during or after resolution of CRS. Despite having mild recurring symptoms, scored as grade 1 by CTCAEv4.03, these patients did not meet the criteria for ICANS according to ASTCT and CARTOX, because they performed normally on serial ICE assessments and had no other ICANS-defining features (Figure 2B). No patient experienced progression of neurologic symptoms. Similarly, the ICANS rate by ASTCT was 55% in B-ALL and 45% in DLBCL patients, compared with 62% and 47%, respectively, according to CTCAEv4.03 (Figure 4).

When evaluating ICANS grade by grade, 54% (30 of 56) of patients had the same grade across all grading systems with an intergrading system agreement ranging from 0.40 to 0.71 (Figure 1B-D). Besides cases with mild symptoms, discordance was mainly encountered for grades 3 and 4. According to CARTOX, 21% (21 of 102) of patients had grade 4 ICANS, compared with 5% (5 of 102) by CTCAEv4.03 and ASTCT (Figure 2B). Of these, 67% (n = 14) were upgraded from ASTCT grade 3 because of different scoring of seizure severity. In fact, any generalized and/or convulsive seizure is grade 4 ICANS by CARTOX. Conversely, by ASTCT, seizures (either local, generalized or convulsive) are grade 3, if rapidly resolved, and grade 4, if life-threatening or repetitive with no return to baseline. CTCAEv4.03 scores brief, generalized seizures as grade 2. However, this led to downgrading of ICANS in only 1 patient. Notably, when analyzed by disease, these discordances were seen only in the B-ALL cohort, reflecting the higher incidence of seizures in this population (Figure 4). In fact, of 19 patients who developed seizures, 16 had B-ALL and 3 had DLBCL. Similar to our observation for CRS, when we compared ICANS rates by disease, we found a significant difference for B-ALL vs DLBCL in ICANS graded by CARTOX, but not by ASTCT and CTCAEv4.03 grading systems (supplemental Table 5).

The ASTCT simplifies the evaluation of raised intracranial pressure by not relying on Frissen papilledema grading and CSF OP alone but by basing it on clinical signs/symptoms of elevated intracranial pressure: decerebrate or decorticate posturing, Cushing’s triad, papilledema, or diffuse cerebral edema on neuroimaging. Of 56 patients with ICANS, neuroimaging was available in 53 and fundoscopic examination in 25. CSF OP measurement was available in 18 of 56 patients. No evidence of cerebral edema or papilledema was recorded. Only 5 patients had an OP >20 mm Hg (all in the B-ALL group), 2 of which would be upgraded to grade 4 ICANS according to CARTOX, whereas they were scored grades 2 and 3 by ASTCT criteria. Notably, the patient with grade 2 ICANS by ASTCT presented only with reduced consciousness and an ICE score of 7, but she had a 34 mm Hg OP. Other minor discrepancies were related to 9% (5 of 56) of ICANS patients who were downgraded because CTCAEv4.03 does not include the ICE score (Figure 2B).

We predicted treatment interventions in DLBCL patients according to published management guidelines (Table 3).^10,13-15  According to tisagenlecleucel’s label, which does not recommend tocilizumab for symptoms of grade 2 CRS, tocilizumab would have been given to only 4 patients (with grade 3-4 CRS), whereas 24, 19, and 25 would have received tocilizumab for a CRS grade ≥2 according to axi-cel’s label, CARTOX, and NCCN, respectively. At our center, 58% (22 of 38) of patients with CRS received tocilizumab, mostly for grades 2 to 4 CRS. Steroids would have been given to 2 patients with grade 4 CRS, according to tisagenlecleucel’s label, compared with 4, 5, and 5 patients with grade 3 or 4 CRS, based on axi-cel’s label, CARTOX, and NCCN guidelines, respectively. Similarly, we treated 5 patients with steroids for CRS. For ICANS, steroid administration was consistent across guidelines, with 16 patients treated (5 for grade 2 and 11 for grade 3 or 4 ICANS). We treated 19 patients with steroids, including 2 with grade 1 ICANS in the context of CRS symptoms and 1 with severe headache and trouble with word finding who did not meet ASTCT ICANS criteria (normal ICE score). Conversely, only 2 patients at our institution received tocilizumab for the concurrent presence of ICANS and CRS symptoms, compared with 9, 11, and 11 predicted according to axi-cel’s label, CARTOX, and NCCN, respectively. However, it should be noted that of these 11 patients, 6 had received tocilizumab for CRS before the onset of ICANS.

We finally investigated the impact of management specifically based on grade alone (not on signs/symptoms), by assessing possible treatment changes related to upgrading/downgrading of CRS and ICANS in all patients. For CRS, of 32 patients upgraded to Penn grade 2 for febrile neutropenia, only 4 (12%) received tocilizumab for concurrent ICANS or high fever lasting for >72 hours. According to tisagenlecleucel’s label, none would have received treatment. Conversely, they would all have received tocilizumab, according to axi-cel’s label, CARTOX, and NCCN. Moreover, 18 of 27 (66%) patients upgraded to Penn grade 3 received tocilizumab at our center. Although tisagenlecleucel’s label would not specify any treatment, according to axi-cel’s label, CARTOX, and NCCN, they would all have received both tocilizumab and steroids. The presence of organ damage did not lead to changes in CRS management in the 8 upgraded patients, none of whom received treatment. This management is consistent with tisagenlecleucel’s label, whereas, according to axi-cel’s label, CARTOX, and NCCN, 5 patients would have received tocilizumab and 2 both tocilizumab and steroids.

For ICANS, 2 of 5 patients who did not meet ASTCT criteria received steroids because of persistent symptoms or headache nonresponsive to analgesia. However, even if classified as grade 1 ICANS, they would not have received treatment according to axi-cel’s label, CARTOX. or NCCN. Nine of 14 patients upgraded from grade 3 to grade 4 ICANS by CARTOX because of seizures received steroids (eventually high doses), and the remaining 5 would probably have received it if treated in the current era (B-ALL patients).

CAR T cells present a paradigm shift in the treatment of lymphoid malignancies and, although 2 products are already FDA approved, many others are currently under investigation. However, direct comparisons of the toxicities associated with different products have not been reported to date, in part due to different grading systems.^6,7,10  We retrospectively collected data from 102 adults with B-ALL and DLBCL, and graded CRS and ICANS toxicities according to the available grading systems for comparison with the newly proposed ASTCT consensus grading system.¹²  We found significant differences between grading systems and the potential to undertreat or overtreat CRS and ICANS based on current management guidelines.

The ASTCT grading system proved to be easily applicable, and largely reproduced the incidence of toxicities previously reported in clinical trials, with some notable exceptions. For patients with B-ALL, the ASTCT CRS rate was 87%, with 28% grade ≥3 CRS, confirming results previously reported according to MSKCC criteria of 85% CRS with 26% grade ≥3.⁹  Similarly, the incidence of CRS by ASTCT in the axi-cel DLBCL cohort was comparable to that of ZUMA-1 (graded by Lee), with 86% vs 93% overall CRS and 14% vs 12% grade ≥3, respectively.^1,16  For patients treated with tisagenlecleucel, conversely, whereas the overall CRS rate approximated that in the JULIET trial (54% vs 58%), the rate of grade ≥3 CRS was 0% by ASTCT, compared with 22% in JULIET,¹⁷  highlighting differences with Penn grading. Analyzing ICANS, the overall rate in B-ALL patients was 55%, with 45% grade ≥3, compared with 62% and 42%, respectively, as previously reported by CTCAEv4.03 criteria.^9,18  Similarly, in the DLBCL cohort, 55% patients treated with axi-cel developed ICANS, grade ≥3 in 33%, quite consistent with 64% ICANS and 32% grade ≥3 reported in ZUMA-1 using CTCAEv4.03.^1,16  In our smaller tisagenlecleucel cohort, 3 patients (15%) had ICANS with no severe symptoms, whereas in JULIET, the ICANS rate was 21%, with 12% of grade ≥3.¹⁷ 

Evaluating toxicities grade by grade, we showed that different grading systems score symptoms differently (upgrading/downgrading their severity), resulting in inconsistent final grades across all grading systems. Some of the differences highlighted are relevant for management implications. For CRS, except for Penn grading, the ASTCT and other available scores performed quite similarly, as anticipated, given that the ASTCT, CARTOX, and CTCAEv5.0 grading systems are mostly derived from the Lee criteria. Conversely, according to Penn, most of the grade 1 and 2 toxicities would be upgraded to grade 2 and 3, respectively. When we predicted treatment based on current guidelines, we found that, whereas tisagenlecleucel’s label would not dictate any specific treatment of grade 2 CRS, application of axi-cel’s label, CARTOX, or NCCN guidelines would lead to tocilizumab use in patients with upgraded grade 2 CRS. Similarly, grade 3 CRS may either not be treated, according to tisagenlecleucel’s label or may require both steroids and tocilizumab, according to other guidelines. It should be noted that the treatment guidelines in the package inserts were based on different grading systems and were primarily a reflection of the experience derived from the specific toxicity profiles of single products tested in the pivotal clinical trials. In fact, in trials of tisagenlecleucel, patients rarely received treatment of grade 2 CRS (Penn), and tocilizumab was reserved for patients with grade 3 or 4 CRS.^7,14  Conversely, the other guidelines mostly rely on Lee, CARTOX, or ASTCT grading systems and on experience from patients treated with axi-cel, with recommendations for tocilizumab in grade ≥2 CRS.^{6,10,13,15,19}  As a result, at present, Penn grading should be applied only to patients treated with tisagenlecleucel. For ICANS, global incidence was only 5% lower by ASTCT compared with CTCAEv4.03, consistent with previously reported results.^1,9,16  ICANS grading may not capture mild symptoms like headache or very mild speech alterations without frank dysphasia, leading to slight underestimation of neurotoxicity by ASTCT. This difference, however, did not result in a change in management when treatment was predicted according to current guidelines, endorsing their exclusion by ASTCT criteria. However, this could become relevant in ongoing trials where early treatment is administered to patients with mild symptoms (grade 1) to reduce incidence of severe symptoms.²⁰  Moreover, according to current guidelines, upgrading of brief generalized seizures from grades 3 to 4 by CARTOX would result in use of high-dose steroids, methylprednisolone 1 g/d, whereas lower doses of steroids or benzodiazepines may be sufficiently appropriate. This difference was highlighted mainly in the B-ALL cohort, which had a higher incidence of seizures and therefore of severe ICANS (45%), compared with the DLBCL cohort (33% axi-cel and 0% tisagenlecleucel). Because the B-ALL cohort was treated from 2010 through 2016, this difference may be related to a more conservative use of tocilizumab and steroids, given concerns for reducing the efficacy of CAR T-cells. However, it may also suggest that the severity of toxicities may vary based on different diseases, with leukemia being a higher risk factor than lymphoma, and on different products. Neuroimaging was regularly performed without any pathologic findings, confirming the rare occurrence of cerebral edema. OP assessment led to upgrading of only 2 patients, without related symptoms, endorsing its exclusion from ASTCT grading, especially as other ICANS measures captured the severely affected patients.

In general, after applying the ASTCT grading system to all patients to predict management, irrespective of the product used, we showed that the treatment that patients actually received differed from both specific label recommendations and guidelines (CARTOX and NCCN), underlining that (1) application of different grading systems to nonspecific management guidelines should be discouraged, because it may result in either overtreatment or undertreatment of toxicities; (2) based on clinical experience, some centers have developed internal guidelines that may not completely adhere to the package inserts; and (3) as the management of toxicities is rapidly evolving, treatment recommendations will likely change. This is illustrated, for example, by the discrepancy between the predicted use of both tocilizumab and steroids for concurrent CRS and ICANS, and the actual administration at our center, which highlights 2 challenging clinical scenarios: concomitant onset of CRS and ICANS when both require treatment, and development of ICANS in the context of mild CRS not requiring treatment or that has improved after prior treatment with tocilizumab.

This study has some limitations, including its retrospective nature and the challenges of interpreting reported symptoms and missing information. Moreover, the small sample sizes do not allow further important comparisons between different CAR T-cell products, particularly in the DLBCL population. Prospective registry data collection is highly recommended for future larger toxicity analyses. Moreover, the different time frame of treatment, 2010 through 2016 for B-ALL vs 2018 for DLBCL, introduces an important bias in the evaluation of management, because patients treated less recently were less likely to receive CRS/ICANS treatment because of concerns about impairing CAR T-cell efficacy.

In conclusion, although our data suggest different toxicity profiles of different products when evaluated with a unified grading system, validation in larger series is needed. Nevertheless, the need for CRS- or ICANS-specific treatment must be clarified, as it cannot be inferred from the results of clinical trials where toxicities and treatments were reported in a nonhomogeneous fashion. Furthermore, with an increasing number of centers treating patients for different indications and with different products, the existence of multiple grading systems may generate confusion and possible overtreatment or delayed treatment in recipients of CAR T cells. Even if no conclusion on treatment recommendations can be inferred, our analysis supports the notion that management guidelines should be based on a unified grading system. Therefore, we propose the use of the ASTCT score as a unified grading system applicable to all CAR T-cell products. Future management guidelines, to be paired to the ASTCT grading system, are warranted and should include product-specific indications.

This work was supported by National Institutes of Health (NIH), National Cancer Institute Cancer Center Support Grant P30CA008748. M.P. was supported by an American-Italian Cancer Foundation Post-Doctoral Research Fellowship and by Associazione italiana contro le leucemie-linfomi e mieloma Milano e Provincia ONLUS. M.S.-E. was supported by a Research Institute of Marques de Valdecilla Wenceslao-Lopez-Albo grant (WLA17/03).

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Contribution: M.P. and M.-A.P. designed the study and wrote the manuscript; M.P. collected the data and conducted the analysis; M.L.S., Y.B., M.S.-E., and M.M. participated in data collection; T.J., E.M., B.D.S., K.W., and R.S. participated in data analysis; S.M.D. conducted the statistical analysis; C.B., R.J.B., P.B.D., C.D., E.F.H., S.G., M.L.P., C.S.S., M.S., G.S., and J.H.P. took care of the patients; and all authors reviewed and approved the manuscript.

Conflict-of-interest disclosure: B.D.S. has consulted for Kite/Gilead, Juno/Celgene, and Novartis. C.B. has consulted for and served on advisory boards for Juno Therapeutics. R.J.B. has consulted for Celgene and has consulted for, holds patents, and has received royalties and research funding from Juno Therapeutics. S.G. has consulted for and received research funding from Amgen, Actinium, Celgene, Johnson & Johnson, and Takeda; has consulted for Jazz Pharmaceuticals, Novartis, Kite, and Spectrum Pharmaceuticals; and has received research funding from Miltenyi. M.L.P. has consulted for Noble Insights and Merck & Co, Inc; has served on advisory boards for STRAXIMM, Kite Pharmaceuticals, Pharmacyclics, and Seres Therapeutics; has served on the speakers bureau for Hemedicus; has equity ownership in Seres Therapeutics and Evelo; and holds patents and receives royalties from MSKCC (intellectual property for Juno and Seres Therapeutics). C.S.S. has consulted for Spectrum Pharmaceuticals, Novartis, Genmab, Precision Biosciences, Kite/Gilead, Celgene, and GSK and has consulted for and received research funding from Juno Therapeutics and Sanofi-Genzyme. M.S. has consulted for McKinsey & Company and Angiocrine Bioscience, Inc. G.S. has received research funding from Amgen and Janssen Pharmaceuticals. J.H.P. has consulted for Allogene, Amgen, AstraZeneca, Autolus, GSK, Incyte, Kite Pharma, Novartis, and Takeda. M.-A.P. has served on advisory boards for MolMed, NexImmune, Medigene, and Servier; has received honoraria and served on advisory boards for Abbvie, Bellicum, Bristol-Meyers Squibb, Nektar Therapeutics, Novartis, Omeros, and Takeda; has consulted for and received honoraria from Merck; and has received research funding from Kite/Gilead, Incyte and Miltenyi. The remaining authors declare no competing financial interests.

The current affiliation for T.J. is Division of Hematological Malignancies and Bone Marrow Transplantation, Department of Oncology, Johns Hopkins University, Baltimore, MD.

Correspondence: Miguel-Angel Perales, Adult Bone Marrow Transplantation Service, Department of Medicine, Memorial Sloan Kettering Cancer Center, 1275 York Ave, Box 298, New York, NY 10065; e-mail: peralesm@mskcc.org.