HLH-like syndromes associate with increased risk of death and CD19⁻ relapse after CAR T in pediatric B-ALL

Key Points

• •Adjusting for baseline disease burden, inferior event-free survival and overall survival were observed in those with each HLH-like syndrome.
• •Inferior outcomes in HLH-like syndromes were driven primarily by CD19⁻ relapse events and nonrelapse mortality.

Visual Abstract

Abstract

We previously reported poor long-term outcomes in patients with B-acute lymphoblastic leukemia (B-ALL) who developed hemophagocytic lymphohistiocytosis (HLH)–like toxicities (HLH-LT) after treatment with tisagenlecleucel. Since then, a consensus definition, immune effector cell–associated HLH-like syndrome (IEC-HS), was developed. Here, we report results from a multicenter retrospective study of children and young adults with B-ALL treated with tisagenlecleucel between 2018 and 2023, describing the incidence, diagnostic overlap, clinical phenotypes, and associated long-term outcomes of HLH-like syndromes. Of 154 tisagenlecleucel recipients, 14.9% met IEC-HS criteria, 21.4% met HLH-LT criteria, and 13% were identified as having HLH by treating institutions in the electronic health record; 10% met all definitions. Each HLH-like syndrome was associated with shortened overall survival and event-free survival, adjusting for baseline disease burden (P < .0001). Those with any HLH-like syndrome experienced 1-year cumulative incidence of nonresponse/relapse of 60% (95% confidence interval [CI], 43.1-76.9) vs 28.2% (95% CI, 20.0-36.5), P < .001, and nonrelapse mortality of 11.4% (95% CI, 0.7-22.1) vs 0% (95% CI, 0-0), P < .001, compared with those without. In those with high disease burden, patients with any HLH-like syndrome had a 1-year cumulative incidence of CD19⁻ relapse of 65.2% (95% CI, 44.7-85.8) vs 2% (95% CI, 0-6.4) in those with no HLH-like syndrome, P < .001. Furthermore, adjusting for disease burden, any HLH-like syndrome was independently predictive of CD19⁻ relapse with a hazard ratio of 8.10 (95% CI, 3.6-18.4), P < .001. These findings demonstrate partial definitional overlap of HLH-like syndromes and the prognostic significance of IEC-HS and identify CD19⁻ relapse and nonrelapse mortality as key outcome drivers.

Introduction

Hemophagocytic lymphohistiocytosis (HLH)–like syndromes after CD19-directed and B-cell maturation antigen–directed chimeric antigen receptor (CAR) T-cell therapies have been recognized as infrequent immune-related adverse events after CAR T-cell therapy. These syndromes, characterized by hyperferritinemia, organ dysfunction, and coagulopathy, have been associated with an increased rate of infections, cytopenias, relapse, and death.1, 2, 3, 4, 5 Previous studies demonstrated incidences ranging from 6% to 36% after CAR T-cell therapy,1, 2, 3, 4, 5, 6 with the highest rates reported in those treated with CD22-CAR T cells for B-acute lymphoblastic leukemia (B-ALL) and B-cell maturation antigen–CAR T cells for multiple myeloma. Among children with B-ALL treated with CD19/4-1BB CAR T-cell therapy, HLH-like syndromes are described in ∼15% of patients.^2,4,6 Although these syndromes have been associated with both morbidity and mortality, the lack of a standardized definition has limited the generalizability of previous findings.^4,5,7,8 Since then, an American Society of Transplantation and Cellular Therapy (ASTCT) consensus definition for immune effector cell–associated HLH-like syndrome (IEC-HS) was developed, representing an important step forward for identification of this toxicity.⁹

To date, the IEC-HS consensus definition has not been evaluated in children with B-ALL receiving tisagenlecleucel. However, the poor outcomes in patients with HLH-like syndromes, as previously defined, highlight the need to define their incidence and association with long-term outcomes using the consensus definition. Additionally, a challenge has been noted in distinguishing IEC-HS from cytokine release syndrome (CRS), because several HLH features are commonly observed during CRS. Thresholds for peak ferritin and number of HLH criteria present in patients with HLH-like syndromes have potential clinical utility for distinguishing between uncomplicated CRS and IEC-HS.

This study seeks to (1) establish the incidence of IEC-HS and its overlap with previously used definitions in children and young adults with B-ALL after tisagenlecleucel, (2) characterize HLH features and peak ferritin values in patients with and without HLH-like syndromes, and (3) evaluate long-term outcomes in patients who experience HLH-like syndromes.

Methods

Study design

We conducted a retrospective, multi-institutional study involving 5 institutions through the Pediatric Real World CAR Consortium. All children and young adults with B-ALL who received infusion with commercial tisagenlecleucel between June 2018 and December 2023 and not included in previous HLH-like toxicity (HLH-LT) analysis⁴ were included.

We defined HLH-like syndromes within 30 days of CAR T-cell infusion by IEC-HS, per the ASTCT consensus definition⁹; HLH-LT, as defined previously⁴; and/or institutional HLH, as determined by the treating physician team within the electronic health record (supplemental Table 1). Patients were defined as having “any HLH” if they met any of the 3 HLH-like syndrome definitions, and as having “severe CRS without HLH” if they experienced peak ASTCT CRS grade ≥3 without meeting any HLH-like syndrome definitions.

We evaluated the overlap of HLH-like syndrome definitions with each other and with severe CRS using Venn diagrams and UpSet plots with programs “VennDiagram” and “UpSetR.” We described the proportion of patients with each HLH-like syndrome that had severe CRS.

HLH features

We measured the occurrence of HLH features within 30 days of CAR T-cell infusion (supplemental Table 2). We assessed the proportion of patients with and without each HLH-like syndrome that had each HLH feature. For peak ferritin, 10 000 ng/mL was chosen because it has been identified as highly sensitive and specific for HLH in the pediatric population.¹⁰ Receiver operating characteristic curve (ROC) analysis was performed to assess the accuracy of peak ferritin in distinguishing patients who developed HLH-like syndromes from those who did not. The optimal threshold for discriminating HLH-like syndrome definitions was identified by maximizing the Youden Index.

HLH-like syndrome timing

We describe the median days to onset of HLH-like syndromes from CAR T-cell infusion and detail the timing of HLH-like syndrome onset relative to CRS status (ongoing, resolving, or resolved).

Inflammatory toxicity treatments

We described tocilizumab doses, days of steroid, and alternative inflammatory toxicity treatments in patients with any HLH-like syndrome definition, severe CRS without HLH, and patients without either severe CRS or HLH-like syndromes.

Survival analyses

Overall survival (OS) was defined as days from tisagenlecleucel infusion to death. Event-free survival (EFS) was defined as days from infusion to nonresponse, relapse, or death. Patients were censored at last follow-up. Complete response (CR) was defined as bone marrow minimal residual disease–negative status by flow cytometry without evidence of extramedullary disease at 28 to 35 days after infusion. Those without CR who did not have days to nonresponse recorded were imputed as having nonresponse at day 35 after infusion.

We evaluated OS and EFS for each HLH-like syndrome. Recognizing that HLH-like syndromes largely occur in patients with high baseline disease burden, a well-described risk factor for poor outcomes after CAR T-cell therapy,^11,12 we stratified survival analyses by high disease burden vs no or low disease burden. High disease burden was defined as described previously.¹² No or low disease burden was defined as not meeting the definition of high-disease burden.

To compare outcomes between patients with severe CRS alone and those with HLH-like syndromes, we analyzed OS and EFS stratified by the presence or absence of HLH-like syndromes and severe CRS.

We evaluated OS/EFS in patients with HLH-like syndromes stratified by overlap with CRS (during ongoing CRS, with resolving CRS, or after CRS was resolved).

Multivariable analysis for OS and EFS

To further evaluate the influence of disease burden and each HLH-like syndrome on OS and EFS, bone marrow blast percentage as a continuous variable and HLH-like syndrome definitions were included in Cox proportional hazards models for OS and EFS events. Cox proportional hazards models were plotted using forest plots using package “forestmodel.”

Statistical analysis

We performed comparisons between categorical variables using the Fisher exact test, and between continuous variables with Kruskal-Wallis and Wilcoxon rank-sum tests. We conducted survival analyses using the Kaplan-Meier method with comparisons via the log-rank test. All data processing was conducted using R Studio version 2024.04.0 + 735. For all statistical analyses, significance was defined as P < .05.

Cumulative incidence for relapse immunophenotype

Time-to-event analyses for relapse immunophenotypes were conducted within a competing risks framework. Patients who achieved a CR with a known immunophenotype at relapse were included. For each patient, follow-up time was measured from CAR T-cell infusion to the earliest of (1) immunophenotype-specific relapse, (2) death without previous relapse or nonresponse, (3) relapse or nonresponse of a different immunophenotype, or (4) last clinical follow-up. CD19⁺ and CD19⁻ relapses were modeled as distinct failure types. Myeloid lineage switch was not counted as CD19⁻ or CD19⁺ relapse. When estimating the cumulative incidence of CD19⁺ relapse, the competing events were death without previous relapse and any relapse that was not CD19⁺. Similarly, when estimating the cumulative incidence of CD19⁻ relapse, competing events were defined as death without previous relapse and any relapse that was not CD19⁻. Patients who remained alive and relapse-free at last follow-up were censored at that time. Cumulative incidence functions for CD19⁺ and CD19⁻ relapse were estimated using the method of Kalbfleisch and Prentice for competing risks. Group differences in the cumulative incidence of each relapse immunophenotype (eg, by baseline characteristics or risk strata) were assessed using the Gray test. Subgroup analyses for CD19⁻ relapse were conducted stratified by disease burden (high vs no-to-low). Considering aforementioned competing events, Cox proportional hazards models were created for CD19⁻ relapse using age, baseline bone marrow disease burden, and HLH-like syndrome status as predictors. Variables significant (P < .05) as predictors of CD19⁻ relapse were included in the multivariable model.

Cumulative incidence for NRM

Time-to-event analyses for nonrelapse mortality (NRM) by HLH-like syndrome status were conducted within a competing risks framework. For each patient, follow-up time was measured from CAR T-cell infusion to the earliest of (1) death without previous disease progression, (2) relapse or refractory disease, (3) hematopoietic stem cell transplantation (HSCT), or (4) last clinical follow-up. NRM was defined as death occurring in the absence of relapse, refractory disease, or post–CAR T-cell HSCT. Cumulative incidence functions for NRM were estimated using the Kalbfleisch and Prentice method for competing risks. Group differences in NRM by HLH status were evaluated using the Gray test.

This study received institutional review board approval with waiver of informed consent.

Results

Patient population

A total of 154 children and young adults with B-ALL treated with tisagenlecleucel were included (Table 1). Median age was 12 years (interquartile range [IQR], 7-16.75). Twenty-two (14.3%) received previous HSCT, a median of 16.2 months before tisagenlecleucel. Median bone marrow disease burden was 1.36% (IQR, 0%-30%), with bone marrow aspirate/biopsy assessments obtained a median of 14 days (IQR, 10-20) before CAR T-cell infusion. Preinfusion extramedullary disease and peripheral lymphoblasts were each present in 30 (20%) patients.

Table 1.

Demographic information and baseline characteristics of the cohort

	Overall (N = 154)
Age at CAR T-cell infusion, median (IQR), y	12.0 (7-16.5)
Sex
Female, n (%)	64 (41.6)
Male, n (%)	90 (58.4)
Previous HSCT
No, n (%)	132 (85.7)
Yes, n (%)	22 (14.3)
Time from previous HSCT, median (IQR), mo	16.2 (14-21.3)
Remission status at CAR T-cell therapy
CR1, n (%)	34 (22.1)
CR2, n (%)	35 (22.7)
CR3, n (%)	8 (5.2)
CR4, n (%)	1 (0.6)
Induction Failure, n (%)	21 (13.6)
Relapse 1, n (%)	28 (18.2)
Relapse 2, n (%)	17 (11.0)
Relapse 3, n (%)	2 (1.3)
Missing, n (%)	8 (5.2)
Baseline bone marrow disease burden, median (IQR), %	1.36 (0-30)
Missing, n (%)	1 (0.6)
Time between bone marrow aspirate and CAR T-cell infusion, median (IQR), d	14.0 (10-19.6)
Extramedullary disease before CAR T-cell therapy
No, n (%)	123 (79.9)
Yes, n (%)	30 (19.5)
Missing, n (%)	1 (0.6)
Peripheral lymphoblasts before CAR T-cell therapy
No, n (%)	124 (80.5)
Yes, n (%)	30 (19.5)

Incidence of HLH toxicities and definition overlap

Twenty-three (14.9%) patients met criteria for IEC-HS, 33 (21.4%) met criteria for HLH-LT, and 20 (13.0%) had HLH by institution determination (Figure 1A-B). Thirty-five (22.7%) met any of the HLH criteria, 21 (13.6%) met criteria for HLH-LT and IEC-HS, and 15 (9.7%) met all 3 criteria. There were no patients identified by institutional determination who were not also identified by IEC-HS and/or HLH-LT criteria.

Figure 1.

HLH-like syndromes overlap with each other and with ongoing CRS. (A)Venn diagram of HLH-like syndrome definitions. (B) UpSet plot of HLH-like syndrome definitions with severe CRS. (C) Bar plots demonstrating the proportion of patients that experienced each clinical HLH feature who did not have (blue) or had (yellow) each condition (HLH-like syndrome or severe CRS). (D) A ROC of peak ferritin for the outcome of any HLH-like syndrome. (E) The proportion of patients with any HLH-like syndrome that developed onset of HLH-like syndrome with ongoing CRS, with resolving CRS, or after resolved CRS.

Severe CRS occurred in 34 patients (22%) of whom 11 (32%) had no HLH-like syndrome and 23 (68%) met criteria for an HLH-like syndrome. Severe CRS was common across all HLH-like syndromes, 61% with IEC-HS, 64% with HLH-LT, 70% with institutional HLH, and 66% with any HLH experienced severe CRS (supplemental Figure 1A-B). These findings indicate that HLH-like syndromes arise in patients with and without severe CRS, although most patients with HLH-like syndromes also had severe CRS.

Clinical features of HLH across definitions

All patients meeting criteria for any HLH-like syndrome definition had a peak ferritin of ≥10 000 ng/mL (Figure 1C). Other HLH features present in most patients with HLH-like syndromes included cytopenias (new onset, worsening, or refractory), aspartate aminotransferase and alanine aminotransferase elevations, hypoxia or pulmonary edema, cytopenias, renal insufficiency, hypertriglyceridemia, and hypofibrinogenemia. lactate dehydrogenase and prothrombin time/international normalized ratio were elevated in most of those with HLH-like syndrome definitions but also in those not meeting HLH-like syndrome definitions. Hemophagocytosis and splenomegaly were reported in <25% of those with any HLH-like syndrome definition. The median number (IQR) of measured HLH criteria for IEC-HS, HLH-LT, and institutional HLH were 8 (7-9), 8 (7-10), and 8.5 (7.75-9.25), respectively (supplemental Table 3A). Expectedly, patients with any HLH-like syndrome definition had a greater number of HLH criteria than those with severe CRS alone (median [IQR], 8 [7-9.5]) vs 2 [1.5-4], P < .0001). This was also true comparing patients with severe CRS and HLH vs severe CRS with no HLH (median [IQR], 8 [7-10] vs 2 [1.5-4], P = .0001; supplemental Table 3B). Peak ferritin in the group with any HLH was higher than in those with severe CRS without HLH or those without HLH-like syndrome or severe CRS (median [IQR], 74 600 ng/mL [37 200-100 001] vs 3670 ng/mL [2847-5167.5] vs 1276 [515-2916]; P < .0001; supplemental Figure 2A). ROC analysis of peak ferritin for any HLH identified an optimal threshold of 10 960 ng/mL (area under the ROC, 0.992; 95% confidence interval [CI], 0.982-1.0; Figure 1D). At this threshold, the sensitivity was 1, and specificity was 0.921 for any HLH. Area under the ROC, sensitivity, and specificity are shown for each HLH-like syndrome definition in supplemental Figure 2B-D. Comparing peak ferritin and number of HLH criteria between groups with IEC-HS with or without HLH-LT (n = 23) vs HLH-LT alone (n = 12), we found no significant differences (supplemental Figure 2E-F).

Timing of HLH-like syndrome onset across definitions

The median days to meeting criteria for IEC-HS and HLH-LT from CAR T-cell infusion were 8 days (IQR, 7-11) and 8 days (IQR, 5-10), respectively (supplemental Table 3A). HLH-like syndrome onset occurred with ongoing CRS, resolving CRS, and resolved CRS 65.7%, 28.6%, and 5.7% of the time, respectively (Figure 1E). For patients with IEC-HS specifically, 52.2%, 39.1%, and 8.6% had onset of IEC-HS with ongoing, resolving, and resolved CRS, respectively (supplemental Table 4). These findings suggest HLH-like syndromes typically develop in the second postinfusion week, most often with ongoing CRS, in patients with B-ALL treated with tisagenlecleucel.

CAR T-cell toxicity–directed anti-inflammatory treatments

Patients with any HLH received more doses of tocilizumab than patients with severe CRS alone or without severe CRS or HLH-like syndrome (median, 3 [IQR, 3-5], vs 2 [IQR, 1-4] vs 0 [IQR, 0-1]; P < .0001; supplemental Table 5) Similarly, those with any HLH received more days of steroids (9 [IQR, 3-12.5] vs 0 [IQR, 0-6.5] vs 0 [IQR, 0-0]; P < .0001.) Use of alternative toxicity-directed treatments was also more common in the group with any HLH-like syndrome. Anakinra was used in 60.0% of patients meeting criteria for any HLH definition, vs 45.5% in patients with severe CRS alone, and 7.4% in those without any severe CRS or any HLH-like syndrome (P < .0001). The use of ruxolitinib, emapalumab, and siltuximab, or combinations thereof occurred exclusively in patients with any HLH-like syndrome. Notably, only 14.3% of patients with any HLH-like syndrome received no alternative toxicity-directed treatment vs 54.5% with severe CRS alone, and 92.6% without severe CRS or any HLH-like syndrome (P < .0001). No patients in any group received treatment with dasatinib, etoposide, or cyclophosphamide.

Clinical outcomes

HLH-like syndrome definitions

With a median follow-up of 1156 days (95% CI, 1040-1304) in the overall cohort, survival analyses for HLH-like syndromes revealed median OS (95% CI) for IEC-HS, HLH-LT, and Institutional HLH of 186 days (122-527), 287 days (188-NA), and 234 days (142-NA), inferior to respective groups without IEC-HS, HLH-LT, or Institutional HLH, which did not reach median OS, P < .0001 for all (supplemental Figure 3). Similarly, median EFS (95% CI) for IEC-HS, HLH-LT, and Institutional HLH were 63 days (35-160), 114 days (76-187), and 92.5 days (51-211), inferior to groups without respective HLH-like syndrome definitions, P < .0001 for all. These results indicate substantially inferior OS and EFS for groups meeting criteria for each of the HLH-like syndrome definitions. We additionally evaluated OS and EFS in groups with IEC-HS alone (n = 2), IEC-HS and HLH-LT (n = 21), and those with HLH-LT alone (n = 12); Groups with IEC-HS ± HLH-LT experienced inferior outcomes than those with HLH-LT alone (supplemental Figure 4A).

HLH-like syndromes and disease burden

The baseline characteristics of each HLH-like syndrome, stratified by disease burden, are shown in supplemental Table 6. Those with high disease burden and each HLH-like syndrome experienced significantly shorter median OS compared with those with high disease burden without the corresponding HLH-like syndrome, P < .0001 (Figure 2). Similarly, median EFS was inferior in groups with each HLH-like syndrome and high disease burden compared to those with high disease burden without HLH-like syndromes, P < .0001. The median (IQR) EFS for IEC-HS, HLH-LT, and institutional HLH were 60 days (35-142), 110 days (63-175), and 78 days (51-211), respectively. These results indicate that patients with high disease burden who experienced HLH-like syndromes experienced inferior OS/EFS, while those with high disease burden without HLH-like syndromes had comparable outcomes to groups with no-low disease burden and no HLH-like syndromes. There were too few patients (n ≤ 3) with low or no disease burden and HLH-like definitions to draw conclusions about the outcomes of these groups.

Figure 2.

In patients with high disease burden, those with HLH-like syndromes experience significantly inferior OS and EFS relative to those without. OS (left) and EFS (right) for IEC-HS (A), HLH-LT (B), and institutional HLH (C), with high disease burden or no/low disease burden, plotted via the Kaplan-Meier method. Comparisons via log-rank test with significance defined as P < .05. Inst HLH, institutional HLH.

Multivariable Cox proportional hazards models including bone marrow disease burden and HLH-like syndromes demonstrated HLH-like syndromes were independently predictive of OS and EFS (Figure 3). For each model, HLH-like syndromes and disease burden were independently predictive of OS and EFS events. IEC-HS carried the highest hazard ratio (HR) for OS and EFS, with HR of 6.06 (95% CI, 3.03-12.09) for OS and 5.64 (95% CI, 3.21-9.92) for EFS. These results illustrate that all HLH-like syndromes were independently predictive of EFS and OS events along with baseline bone marrow disease burden.

Figure 3.

HLH-like syndromes and baseline disease burden are independently predictive of OS and EFS. Forest plots demonstrating HRs and 95% CIs for multivariable Cox proportional hazards models including bone marrow blast percentage as a continuous variable (by 10% increase) and HLH-like syndromes for OS (A) and EFS (B). Significance defined as P < .05.

HLH-like syndromes and severe CRS

Stratifying patients by severe CRS status and each HLH-like syndrome definition demonstrated inferior OS and EFS for groups with HLH-like syndromes regardless of severe CRS status, P < .0001 for all (Figure 4). Groups with severe CRS without respective HLH-like syndromes had similar outcomes to those in groups with no CRS to low-grade CRS without HLH-like syndromes.

Figure 4.

HLH-like syndromes are associated with inferior OS and EFS compared with those with severe CRS or no CRS to low-grade CRS without HLH-like syndromes. OS (left) and EFS (right) for IEC-HS (A), HLH-LT (B), and institutional HLH (C), stratified by severe CRS vs no/low CRS, and plotted via the Kaplan-Meier method. Comparisons via log-rank test with significance defined as P < .05.

Timing of HLH-like syndrome onset relative to CRS

Survival analyses among patients with any HLH-like syndrome definition, stratified by timing of onset during ongoing CRS, with resolving CRS, or after resolved CRS showed no significant differences in OS or EFS (supplemental Figure 4B). Similarly, no significant difference in OS or EFS was observed for patients with IEC-HS, when stratified by timing of onset during ongoing, resolving, or resolved CRS (supplemental Figure 4C).

NRM

Of the entire 154 patient cohort, 4 NRM events occurred due to CRS (n = 2), HLH (n = 1), and infection (n = 1) with a 1-year cumulative incidence of 2.6% (95% CI, 0.0-0.5). The cumulative incidence of 1-year NRM in patients with any HLH-like syndrome was 11.4% (95% CI, 0.7-22.1), with no new NRM events occurring by 2 years (Figure 5A). In contrast, no patients without an HLH-like syndrome experienced NRM by 2 years after infusion, 0% (95% CI, 0-0), P < .01. At 1 year, the cumulative incidence of NRM was significantly higher for IEC-HS, HLH-LT, and institutional HLH than those without (supplemental Figure 5). Although NRM was infrequent, it occurred exclusively in patients with HLH-like syndromes.

Figure 5.

NRM and nonresponse/relapse are more frequent in patients with any HLH-like syndromes. Cumulative incidence curves for the first event of nonresponse/relapse, NRM, or posttisagenlecleucel HSCT∗ stratified by HLH-like syndrome status. (A) NRM; (B) nonresponse relapse; (C) cumulative incidence function of events at 1 year and 2 years after tisagenlecleucel with 95% CI. Comparisons via the Gray test with significance of P < .05. ∗HSCT is suppressed on CIF plot. CIF, cumulative incidence functions.

Relapse immunophenotype

The overall 1-year cumulative incidence of nonresponse or relapse was 35.4% (95% CI, 27.9-42.9). Patients with any HLH-like syndrome had a markedly higher 1-year cumulative incidence of nonresponse or relapse than those without, 60% (95% CI, 43.1-76.9) vs 28.2% (95% CI), P < .001; Figure 5B-C). Twenty patients were excluded from the relapse immunophenotype analysis (4 patients who died before initial disease evaluations, 14 patients with nonresponse, and 2 patients with unknown relapse immunophenotype), leaving 134 evaluable patients. Among these, the overall relapse rate was 36.5% (49/134 patients), with a rate of 70.4% (19/27) in patients with any HLH vs 30/107 (28.1%) without any HLH-like syndrome. The median times to relapse were 114 days (IQR, 77-176) in patients with any HLH-like syndrome vs 234 days (IQR, 142-476) in patients without HLH-like syndromes. Of patients who experienced disease relapse, patients with HLH-like syndromes had a significantly higher proportion of CD19⁻ relapse than those without HLH (84% vs 33%, P < .01; Figure 6A). The cumulative incidence of CD19⁺ and CD19⁻ relapses was assessed by the presence and absence of any HLH-like syndrome (Figure 6B) and by each HLH-like syndrome definition (supplemental Figure 6). CD19⁻ relapse occurred significantly more frequently in patients with each HLH-like syndrome compared with groups without respective HLH-like syndromes (P < .001 for all). Because patients with HLH-like syndromes were enriched for high disease burden, a known risk factor for CD19⁻ relapse, relapse immunophenotype was further analyzed by disease burden status (Figure 6C). Among patients with high disease burden, those with any HLH-like syndrome had a 1-year cumulative incidence of CD19⁻ relapse of 65.2% (95% CI, 44.7-85.8), compared with 2.2% (95% CI, 0-6.4) in those without any HLH-like syndrome, P < .001. Similar patterns were observed across each HLH-like syndrome definition, P < .001 for each (supplemental Figure 7).

Figure 6.

CD19⁻ relapses occur frequently in patients affected by HLH-like syndromes and account for most CD19⁻ relapses in high disease burden patients. (A) Bar plots of relapse immunophenotype in patients with and without any HLH-like syndrome who achieved CR and subsequently experienced disease relapse. (B) Cumulative incidence of CD19⁺ relapse (left panel) and CD19⁻ relapse (right panel) in patients with and without any HLH-like syndrome. (C) Cumulative incidence of CD19⁻ relapses in patients with high baseline disease burden (left panel) or no to low baseline disease burden (right panel). Comparisons via the Gray test with significance of P < .05. (D) Forest plot of a multivariable Cox proportional hazards model evaluating the association between disease burden, analyzed as a continuous variable with HR reported per 10% increase, and any HLH status with the risk of CD19⁻ relapse.

In univariate analysis for CD19⁻ relapse, disease burden and any HLH-like syndrome status were associated with CD19⁻ relapse. Age was not significant as a continuous or categorical variable. Disease burden and occurrence of any HLH-like syndrome were evaluated in a multivariable Cox proportional hazards model, identifying any HLH-like syndrome as an independent predictor of CD19⁻ relapse with a HR of 8.1 (95% CI, 3.6-18.4), adjusting for disease burden (Figure 6D). These findings were replicated for each HLH-like syndrome (supplemental Figure 8). Together, these results indicated HLH-like syndromes after CAR T-cell therapy were strongly and independently associated with CD19⁻ relapse events.

Discussion

In this study, we found that 23% of our cohort met criteria for any of the HLH-like syndrome definitions and 14.9% met criteria using the IEC-HS definition. This incidence was comparable with the ∼15% incidence after CD19⁻/4-1BB CAR T-cell therapy in B-ALL reported previously.^2,4,6 HLH-like syndrome definitions overlapped in this cohort and were not clearly distinguishable from each other by the number of HLH features or peak ferritin values. Those with HLH-like syndromes were, however, differentiated from those with severe CRS in 3 ways: (1) peak ferritin was uniformly ≥10 000 ng/mL in HLH-like syndromes compared with 10% of those with severe CRS alone; (2) the number of HLH features was significantly higher in those with each HLH-like syndrome than those with severe CRS alone; and (3) OS and EFS were markedly reduced in groups with HLH-like syndromes, but not in severe CRS alone.

HLH-like syndrome onset overlapped with ongoing CRS in about two-thirds and with resolving or resolved CRS in the remaining third. The timing of HLH-like syndromes in this tisagenlecleucel-treated cohort is distinct from that reported after CD22-directed CAR T-cell products for B-ALL, which occurred a median of 2 weeks after infusion.^6,8,13 Because tisagenlecleucel has a shorter time to peak expansion (10 days) than CD22-directed CAR T-cell products (14-21 days after infusion), we speculate earlier peak CAR expansion in tisagenlecleucel-treated patients may explain the earlier onset of HLH-like manifestations observed in this cohort.^8,14,15 Notably, the timing of HLH-like syndrome onset relative to CRS did not influence outcomes, which were poor regardless of onset with ongoing, resolving, or resolved CRS. The co-occurrence of HLH manifestations with ongoing CRS in most patients with HLH-like toxicities does indicate the possibility that some patients had refractory CRS that evolved to meet criteria for HLH-like syndromes. The definitions used, IEC-HS and HLH-LT, were developed with the intent of distinguishing HLH-like pathological inflammation from CRS, which is defined by fever, hypotension, and hypoxia, although we acknowledge that these syndromes occur along a spectrum. We speculate that CD19-directed CAR T-cell–treated patients who have heightened levels of inflammation from CRS are most at risk for subsequent immune-mediated HLH-like manifestations, as they are for immune effector cell–associated neurotoxicity syndrome^14,16,17 and prolonged severe neutropenia.¹⁸

The prognostic significance of HLH-like syndromes, previously demonstrated in a cohort of 185 children and young adults with B-ALL,⁴ was validated in this cohort. In patients with high disease burden, the median OS and EFS were substantially reduced in those meeting each HLH-like syndrome definition relative to those without HLH-like syndromes. Multivariable analyses identified each HLH-like syndrome as independently predictive of OS and EFS events, with IEC-HS having the highest HRs for both OS and EFS. Furthermore, we found patients who experienced IEC-HS with or without HLH-LT had inferior outcomes than those with HLH-LT alone, providing additional support for use of the IEC-HS consensus definition in the pediatric B-ALL population undergoing treatment with tisagenlecleucel.

The undesirable outcomes demonstrated for the HLH-like syndrome groups were driven by increased rates of relapse or nonresponse, whereas increased rates of NRM represented an important contributor. Relapse or nonresponse in patients with any HLH-like syndrome occurred at approximately double the rate observed in those without HLH-like syndromes. A key finding of our study was the high proportion of CD19⁻ disease at relapse (84%) in patients with any HLH-like syndrome, consistent with our previously published work on HLH-LT.⁴ However, we additionally show that patients with high disease burden without any HLH-like syndrome had markedly lower 1-year cumulative incidence of CD19⁻ relapse than those with high disease and any HLH-like syndrome. Disease burden and age have been previously identified as key risk factors for CD19⁻ relapse in B-ALL.¹⁹ Our multivariable analyses suggest HLH-like pathophysiology is the predominant factor associated with CD19-antigen escape.

This important finding raises critical questions into the mechanisms of antigen escape in patients who develop HLH-like syndromes after CAR T-cell therapy. Based on the findings of extreme hyperferritinemia observed in this study, and greater elevation of proinflammatory cytokines (interferon gamma [IFN-γ], interleukin-1 [IL-1], IL-4, IL-6, IL-13, IL-18, macrophage inflammatory protein 1α) relative to CRS shown previously,⁶ we speculate that excessive proinflammatory cytokines may play a role in leukemia resistance to CAR T cells, as has been shown in CAR T-cell therapy for acute myeloid leukemia.²⁰ The selection pressure exerted by CAR T cells, combined with a highly inflamed environment, may drive antigen loss. Conversely, preexisting CD19⁻ leukemia populations may be present before infusion²¹ and contribute to excessive inflammation after infusion. Genetic factors predisposing to HLH-like manifestations (eg, STXBP2, RAB27A) have been observed to be enriched in patients with severe CAR T-cell toxicities, and represent possible contributors here as well.^22,23 Furthermore, given the increased expression of inhibitory ligands induced by IFN-γ,²⁴ a key cytokine in HLH that is also highly elevated in children and young adults with HLH-like syndromes after CAR T-cell therapy,⁶ we speculate leukemia inhibitory ligand expression could be contributing to slowed leukemia killing, allowing for leukemic adaptation in those with HLH-like syndromes. Finally, groups with HLH required more days of steroids, tocilizumab doses, and alternative toxicity-directed treatments than groups without HLH, which may have affected leukemia clearance. Each of these hypotheses requires testing in future investigations.

This study is limited by its retrospective nature, including retrospective application of HLH-like syndrome definitions. We have attempted to balance the retrospective application of IEC-HS and HLH-LT criteria by evaluating for institutional criteria of HLH and have captured granular data on individual HLH features among all patients. This approach allowed us to comprehensively characterize HLH-related clinical features and peak ferritin levels, which is a strength of this study. The potential overlap of refractory CRS with IEC-HS, particularly in the 12 patients identified as having IEC-HS onset with ongoing CRS, raises the possibility of patients with refractory CRS being classified as IEC-HS. Limited data on refractoriness to CRS therapy were available to assess this overlap and warrants further study in future analyses. These findings are not meant to suggest that CRS should not be treated as CRS, if ongoing. Finally, the relatively small number of relapse events, particularly CD19⁻ relapse, combined with multiple subgroup and comparative analyses, increases the risk of type I error and instability of subgroup estimates. Accordingly, these findings should be interpreted with caution and considered hypothesis-generating, warranting validation in larger prospective cohorts.

The findings in this study are important because they highlight that IEC-HS occurs in ∼1 in 6 children and young adults after tisagenlecleucel and associates with poor outcomes driven by high rates of CD19⁻ relapse and NRM. There was diagnostic overlap of IEC-HS with HLH-LT and institutionally identified HLH, although patients identified as having IEC-HS were found to experience significantly inferior outcomes compared with those with HLH-LT alone. There remains a critical need for investigations to further characterize IEC-HS and improved treatment strategies. Because of the striking rate of CD19⁻ relapse, we speculate that post–CAR T-cell consolidative therapies may have utility in improving outcomes for patients with B-ALL who experience IEC-HS after CD19-CAR T-cell therapy and can tolerate subsequent therapy, although this requires prospective study. Inflammation mitigation strategies such as preemptive use of anakinra (initial therapy recommended for IEC-HS management),⁹ or IFN-γ–targeting approaches, offer potential mitigation strategies for patients with high risk. Despite greater use of steroids, tocilizumab, and alternative treatment strategies in this group, outcomes remained poor, illustrating a need for prospective investigations of IEC-directed treatment strategies to guide optimal management. Finally, predictive models for the development of IEC-HS can inform which patients are at greatest risk and are currently being validated by our group.

Conflict-of-interest disclosure: K.O.M. served on an advisory board for Autolus and Electra Therapeutics. C.L.M. is a founder, equity holder, and consultant for Link Cell Therapies; has received consulting fees from Cargo, Ensoma, AstraZeneca, Immatics, Red Tree Venture Capital, Grace Science, Kite Pharma, and Nektar; and has received research funding from Tune Therapeutics. The remaining authors declare no competing financial interests.