EASIX and m-EASIX predict CRS and ICANS in pediatric and AYA patients after CD19-CAR T-cell therapy

CD19-directed chimeric antigen receptor (CD19-CAR) T-cell therapy has resulted in tremendous disease response outcomes in pediatric and adolescent and young adult (AYA) patients with relapsed/refractory (R/R) B-cell acute lymphoblastic leukemia (B-ALL).^1-4 Complications of this therapy are characterized by hyperinflammation and endothelial damage, including cytokine release syndrome (CRS), immune effector cell–associated neurotoxicity syndrome (ICANS), and immune effector cell–associated hemophagocytic lymphohistiocytosis syndrome.^5-7 Although these high-risk toxicities can culminate in multiorgan dysfunction and even death, they can be treated with cytokine-directed therapies and/or corticosteroids.^8,9 Therefore, early recognition of and intervention for these toxicities may mitigate their morbidity and mortality.^10,11 Individual risk factors for severe CRS and ICANS have been identified, primarily relating to higher levels of leukemic disease burden before CD19-CAR T-cell therapy, in addition to CAR T-cell dose, activation, and expansion.^3,12-14 However, comprehensive predictive tools are necessary for early detection of these complications to further guide timely interventions to improve patient outcomes.

The Endothelial Activation and Stress Index (EASIX) score combines lactate dehydrogenase (LDH), creatinine, and platelet (PLT) count (LDH × creatinine/PLT) and has been used to predict endothelial complications associated with hematopoietic cell transplant (HCT).¹⁵ A higher pre-HCT EASIX score has been correlated with higher risk of multiple transplant complications, including graft-versus-host-disease, fluid overload, and sepsis.^15-17 A derivative score, the modified EASIX (m-EASIX), substitutes C-reactive protein (CRP) for creatinine and has also shown to correlate with posttransplant engraftment syndrome, febrile neutropenia, veno-occlusive disease, and transplant-associated microangiopathy.¹⁸ Furthermore, in multiple adult patient populations treated with CD19-CAR T-cell therapy, retrospective calculation of EASIX and m-EASIX scores has shown that higher scores are predictive of the development of CRS and ICANS.^7,17,19-28 Importantly, these studies included different CAR products and varied incidences of CRS and ICANS and, thus, resulted in varied predictive thresholds of EASIX and m-EASIX.^{17,19,20,24-30} This suggests that there is likely an effect of patient and/or treatment variables on score performance.

The EASIX and m-EASIX scores have not yet been applied to a pediatric and AYA patient cohort treated with CAR T-cell therapy, and therefore, the potential role of using these models is unknown in this patient population. The pediatric and AYA treatment setting often differs from that of adults in terms of primary cancer type (leukemia vs lymphoma), prior treatment modalities, disease burden before CAR T-cell treatment, and baseline organ function status. Additionally, children may receive different lymphodepletion regimens and/or CAR T-cell products, which, in turn, can further affect the incidence and severity of CAR T-cell–related complications.^1,31,32 Therefore, predictive tools such as the EASIX and m-EASIX scores need to be evaluated in this specific population to better understand their utility in predicting complications after CD19-CAR T-cell therapy.

Here, we report on the retrospective application of the EASIX and m-EASIX scoring systems to a pediatric and AYA patient cohort, treated with CD19-CAR T cells at 2 institutions. We evaluate EASIX and m-EASIX scores before and after treatment, including their association with and predictive utility for the subsequent development of CRS and ICANS.

This retrospective study included pediatric and AYA patients with R/R CD19⁺ hematologic malignancies treated with lymphodepleting chemotherapy (fludarabine/cyclophosphamide) followed by CD19-CAR T-cell therapy (tisagenlecleucel or an investigational product [NCT03573700]⁴; autologous T cells expressing CD19.41BBz CARs). Data from first infusions of CAR T-cell products were included in the cohort. Treatment occurred at either St. Jude Children’s Research Hospital (SJCRH) or Johns Hopkins Hospital (JHH), between August 2018 and April 2023. Written informed consent/assent was obtained from all patients and/or legal guardians to receive treatment, per institutional guidelines and the Declaration of Helsinki. Approval for this retrospective study was obtained from the institutional review boards at both SJCRH and JHH.

Demographic, clinical, laboratory, and treatment-related data were collected from both a prospective clinical database and retrospective medical record review. Disease status at the time of patient referral to CAR T-cell therapy was recorded, with complete response requiring no detectable minimal residual disease (MRD) evaluated using flow cytometry (limit of detection, 0.01%) when available for a given patient. Pre–CAR T-cell disease burden was recorded using MRD from the most recent bone marrow evaluation before treatment with CAR T-cell therapy; morphologic blast percent was used for 5 patients without available MRD testing. Within the first 35 days of infusion, we captured CRS and ICANS occurrence and grading based on the American Society for Transplantation and Cellular Therapy consensus guidelines.³³ Before adoption of the American Society for Transplantation and Cellular Therapy consensus grading, neurotoxicity was graded using the common terminology adverse event criteria, in conjunction with a protocol defined grading scheme to determine an overall toxicity grade. Severity of CRS or ICANS was further categorized as none, mild (grade 1-2), or severe (grade 3-4).

Laboratory data (creatinine [mg/dL], LDH [units per L], CRP [mg/dL], and PLT count [10³/μL]) were collected at multiple time points to calculate EASIX (LDH × creatinine/PLT) and m-EASIX (LDH × CRP/PLT) scores, including before lymphodepletion (day –5 ± 2 days), on the day of CAR T-cell infusion (day 0 ± 1 day), and on day +3 (±1 day) after infusion. If multiple laboratory values were present on a given day, the highest value was taken for LDH, creatinine, and CRP and the lowest value for PLTs. To account for iatrogenic elevations in PLT counts due to transfusion support, we substituted a value of 20×10³/μL in instances when a PLT transfusion had been given within the prior 72 hours.

Data were summarized using descriptive statistics, counts, proportions, and medians with ranges. We analyzed the association and predictive performance of EASIX and m-EASIX scores with incidences of CRS and ICANS as outcomes. All patients are included in the analyses using day –5 and day 0 scores; patients who experienced CRS or ICANS on or before day +3 were excluded from day +3 analyses. Both EASIX and m-EASIX were transformed with the base-2 logarithm in the association and prediction analyses for consistency with previous reporting. We imputed scores of 0 with half of the minimum of nonzero values for each predictor of interest before transforming.

Univariate logistic regression was conducted to explore the association between scores and complications at the specified time points. We report the estimated odds ratios (ORs) with 95% likelihood ratio-based confidence intervals and computed likelihood ratio test P values. To assess the individual components of EASIX and m-EASIX, we repeated the univariate logistic regression analysis, replacing the score covariate by each component. For assessing predictive performance, we constructed receiver operating characteristic (ROC) curves from the univariate logistic regression models of scores and calculated the respective area under the curves (AUCs) to summarize performance. To internally validate and address potential optimism bias from overfitting,³⁴ we applied Harrell bias correction method with 200 bootstrap replications to adjust the naïve AUCs.³⁵ All analyses were conducted using R 4.2.2. Analyses are intended to be exploratory, and no adjustment was made for multiple comparisons.

Our cohort consisted of 76 pediatric and AYA patients with R/R B-ALL (SJCRH, n = 51; JHH, n = 25). See Table 1 for patient, disease, and treatment characteristics. Twenty-two patients received an investigational CD19-CAR T-cell product (NCT03573700); the remainder of the cohort received tisagenlecleucel. All patients received lymphodepleting chemotherapy before cellular infusion, consisting of fludarabine (total dose, 75 mg/m² for NCT03573700; 120 mg/m² for tisagenlecleucel) and cyclophosphamide (total dose, 900 mg/m² for NCT03573700; 1000 mg/m² for tisagenlecleucel). Before treatment, 62 patients had detectable marrow disease with 32 patients (42%) having >5% MRD in the marrow. The cohort was heavily pretreated, including 28 patients with blinatumomab and/or inotuzumab and 13 patients with prior allogeneic HCT.

After CD19-CAR T-cell infusion, 47 (61.8%) and 17 patients (22.4%) developed CRS and ICANS, respectively, of whom 14 (18.4%) developed both CRS and ICANS. Most patients experienced onset of CRS (n = 31) and ICANS (n = 7) within 5 days of infusion (supplemental Table 1). Severe CRS or ICANS occurred in a subset of patients (CRS, n = 13; ICANS, n = 7; supplemental Table 2). Treatment interventions for CRS and ICANS were guided by institutional guidelines. One patient died within the study period, after experiencing both severe CRS and ICANS.

EASIX and m-EASIX were calculated over time (day –5, day 0, and day 3), with results displayed by the severity (Figure 1) and presence of CRS or ICANS (supplemental Figure 1). Among those patients who developed severe CRS, the median EASIX scores were consistently higher at all time points than patients with mild or no CRS (severe CRS: median log score 2.15, 0.97, and 1.36 for days –5, 0, and +3, respectively; P < .05, days –5 and 0). Comparatively, the median EASIX scores among patients with either mild or no CRS were similar to each other and remained relatively stable over time (no CRS: –1.18, –0.71, and –0.73; mild CRS: –0.56, –0.26, and –0.27 for days –5, 0, and +3, respectively). Notably, the dispersion of EASIX score decreased over time, particularly among those with severe CRS (Figure 1A). Regarding m-EASIX, scores were consistently higher for patients who developed severe CRS (4.52, 3.65, and 5.45 for days –5, 0, and +3, respectively; P < .05) than those with mild or no CRS. However, the difference between patients experiencing mild CRS and no CRS was larger when using m-EASIX scoring rather than EASIX scoring (no CRS: –3.08, –1.21, and –1.44; mild CRS: –1.41, 0.76, and 0.12 for days −5, 0, and +3, respectively; Figure 1B). Additionally, when dividing patients into 2 groups by the presence of CRS or not, those patients who went on to develop CRS displayed higher median EASIX and m-EASIX scores across all time points (supplemental Figure 1).

For patients who developed ICANS, the trend of EASIX scores over time showed a slightly different pattern than what was seen with CRS. Patients with mild and severe ICANS had more similar EASIX scores (mild ICANS: 1.44, 1.50, and 1.68; severe ICANS: 0.14, 0.81, and 0.36 for days –5, 0, and +3, respectively). Furthermore, EASIX scores among those patients with mild or severe ICANS were consistently higher than patients who never developed ICANS (–1.05, –0.47, and –0.53 for days –5, 0, and +3, respectively; Figure 1C). This was further demonstrated when comparing median EASIX scores between patients with any grade ICANS vs no ICANS (supplemental Figure 1). Regarding m-EASIX, there was a noticeable rise in median values only for patients with severe ICANS (2.16, 3.59, and 5.45 for days –5, 0, and +3, respectively) compared with patients with no and mild ICANS (no ICANS: –2.18, –0.11, and –0.62; mild ICANS: –1.39, 1.74, and 0.18 for days –5, 0, and +3, respectively; P < .05, day +3; Figure 1D).

In univariate analyses, at all time points evaluated, the odds of developing CRS or ICANS were higher with higher EASIX and m-EASIX scores (Figure 2). With both EASIX and m-EASIX scoring, the estimated ORs increased over time for the outcomes of severe CRS, any grade ICANS, and severe ICANS. However, for any grade CRS, there was an increase in m-EASIX–estimated ORs over time, which was not seen with EASIX scores. Additionally, across all time points, ORs for CRS/ICANS were estimated with more precision for m-EASIX than EASIX, as demonstrated by smaller confidence intervals.

We next sought to evaluate the individual components of the EASIX and m-EASIX scores, in relation to both CRS and ICANS over time (Figure 3A-H). There was a clear trend for median LDH (increased) and PLTs (decreased) at each time point among CRS severity subgroups (P <.05, days 0 and +3 [LDH] and days –5, 0 and +3 [PLT]; Figure 3A-B). For CRP, there was relative stability over time within each subgroup, with those with severe CRS having higher CRP values across time points (Figure 3C). However, for creatinine, values were similar among CRS subgroups at both the day –5 and day 0 time points, with a downward trend at day +3 among those patients with severe CRS (Figure 3D). Patterns are less apparent when using groupings by ICANS severity (Figure 3E-H). It is notable that, across time points, the median PLT levels are lower among those patients who subsequently developed any severity of ICANS (Figure 3F). Additionally, CRP levels are higher and increase over time among those who develop severe ICANS compared with other subgroups (Figure 3G).

To evaluate the performances of EASIX and m-EASIX scores as predictors of toxicity occurrence and severity, ROC curve analysis was performed, with AUC as the measure of prediction performance (Figure 4). Both scores performed better at predicting severe CRS than any CRS, with m-EASIX having consistently higher AUC than EASIX at each evaluated time point (Figure 4A-B). Furthermore, although performance was stable across time with the EASIX score, among patients who developed severe CRS, m-EASIX demonstrated the best performance at day +3 (AUC, 90.6%) compared with days –5 and 0 (AUCs, 80.6% and 79.8%, respectively; Figure 4B). Prediction performance for ICANS was more modest. EASIX outperformed m-EASIX at predicting any grade ICANS (AUCs on days –5, 0, and +3, consecutively, for EASIX, 73.8%, 70.5%, and 74.2%; for m-EASIX, 59.7%, 64.8%, and 71.4%; Figure 4C). Of note, the predictability of m-EASIX for any ICANS improved over time (Figure 4C). The most robust prediction for ICANS was seen with m-EASIX predicting severe ICANS at day +3 (AUC, 82.2%; Figure 4D). AUCs were similar with and without optimism bias adjustment (data not shown). Additionally, subgroup ROC curve analysis was performed using the cohort treated with tisagenlecleucel. In this analysis, a similar predictive pattern was seen compared with the whole cohort, with the exception of severe ICANS in which both EASIX and m-EASIX had lower predictive power, likely due to even smaller number of events (data not shown).

Multiple studies have looked into EASIX and its derivatives as predictive models for CRS and ICANS in adult patient populations treated with CAR T-cell therapy.^{7,17,19,20,24-29,36,37} However, these data do not exist for a pediatric and AYA cohort. In this study, we evaluated the utility of EASIX and m-EASIX scores as tools to predict the development of CRS and ICANS in a population of pediatric and AYA patients with R/R CD19⁺ hematologic malignancies, treated with CD19-CAR T cells at 2 institutions. Our results show that both scores have utility in the prediction of subsequent development of CRS and/or ICANS, particularly severe toxicity.

CRS and ICANS are potentially life-threatening toxicities of CAR T-cell therapy, and anticipating their development is integral in parent/patient counseling and medical management.^38,39 Although some pretreatment risk factors for development of toxicity have been identified, including higher disease burden,^3,14 there is still largely a lack of predictive capability. Furthermore, although postinfusion interventions such as preemptive tocilizumab have been shown to decrease the chance of developing severe CRS,^10,11 such measures are costly, have potential side effects, and are not necessary in all patients. Additionally, identifying patients at risk of severe CRS/ICANS may inform upon decisions related to the location of administration of CAR T cells (inpatient vs outpatient) and the required duration of observation after infusion.^40,41 Therefore, the ability to anticipate severe toxicities can still have major clinical, financial, and psychological implications to the patient and/or treating team, and continued investigation into methodologies to predict the development of high-risk toxicities are needed.

In adult patients with non-Hodgkin lymphoma treated with axicabtagene ciloleucel, investigators found that the EASIX score calculated before lymphodepletion was predictive of subsequent development of grade 2 to 4 CRS (hazard ratio, 2.3; P = .001) and ICANS (hazard ratio, 2.3; P = .001),²⁵ and EASIX combined with ferritin or ferritin and CRP was predictive of grade 2 to 4 CRS.²⁰ In adult patients with B-ALL and diffuse large B-cell lymphoma, treated with multiple CAR products, m-EASIX was predictive of high-grade CRS at prelymphodepletion and days –1 and +1 peri-infusion (AUCs, 80.4%, 73%, and 75.4%, respectively) and of high-grade ICANS at day +3 (AUC, 73%).^19,24 Other efforts evaluating novel yet more challenging quantifiable combinatorial models for post-CAR toxicity prediction have also been reported. For example, a nomogram was developed to predict severe CRS in patients with multiple myeloma treated with B-cell maturation antigen-directed CAR T cells, by measuring CX3CL1, granzyme B, interleukin-4, interleukin-6, and platelet-derived growth factor AA at days 1 to 2 after infusion, with a bias-corrected AUC of 0.876 (95% confidence interval, 0.871-0.882).³⁶ Analyzing multiple cytokines early after infusion and before the development of grade 4 CRS indicated that an monocyte chemotactic protein 1 >1343.5 pg/mL in febrile patients enhanced the identification of patients who later developed grade 4 CRS.⁷ In a combined pediatric and adult population, grade 4 to 5 CRS was predicted by measuring interferon gamma, sgp130, and sIL1RA (AUC, 0.93 [95% confidence interval, 0.86-1.0]).³⁷ However, in comparison with these algorithms that rely on measuring of cytokines that are not routinely available in a timely manner, EASIX and its derivatives are more appealing, because they are more easily calculated using laboratory data that are more widely available in real time.

In our work using data from pediatric and AYA patients with R/R CD19⁺ hematologic malignancies treated with CD19-CAR T cells, we found that the predictive performance of m-EASIX was greater than EASIX, except for predicting any grade ICANS. We also performed component analysis to investigate the potential causes for differences in prediction performance between EASIX and m-EASIX. Our component analysis showed trends between LDH (increasing), PLT (decreasing), and CRP (increasing) values with CRS occurrence and severity. However, this was not seen with ICANS. Creatinine values, on the contrary, did not display clear patterns across outcomes, possibly relating to different renal profile in children than adults.⁴² CRP has come to be a well-known predictive measure for CRS and ICANS,⁴³ and it showed a positive trend with severity in our population. Therefore, omission of creatinine and addition of CRP likely contributed to the superiority of m-EASIX over EASIX in our population. These data may also be representative of different etiologies of CRS and ICANS. However, it is important to recognize that results could be affected by the small number of ICANS events in this cohort. Larger cohorts are needed to study this further, in addition to studying other markers that could be more specific to ICANS.^5,44 Interestingly, almost all patients who developed ICANS in our cohort had already developed CRS, which could in the future serve as a predictive factor itself.⁴⁵ 

As we consider future work in developing predictive models for severe side effects after CAR T-cell therapy, it is critical to recognize that historical data sets may not be representative of current practice. This may include changes in treatment indications, types of CAR T-cell therapy available, and available interventions. For example, although the incidence of severe CRS and ICANS did not vary over the study time period in this data set, the increasing use of preemptive or prophylactic treatment strategies will likely alter the patterns of adverse events over time, which, in turn, will affect predictive model development. Similarly, the role of disease burden is important to consider, particularly because patients are treated with lower amounts of disease. Furthermore, the CAR T-cell products used in this work are similar and therefore had expected similar side effects profiles. We evaluated this in a subgroup analysis, in which we found no differences in our conclusions based on the product used. However, different CAR constructs may substantially affect the incidence and severity of adverse events after treatment and need to be considered in future models.

Our findings provide, to our knowledge, the first evidence of the predictive performance of EASIX and m-EASIX scores for the development of CRS and ICANS in a pediatric and AYA population treated with CD19-CAR T-cell therapy. Importantly, our data align strongly with prior published adult data, despite differences in disease types, as well as patient and treatment characteristics. This work supports further prospective evaluation of these models in larger patient cohorts. This is necessary to further understand how these prediction tools may improve real-world decision-making, such as implementing cutoff values or decision curve analyses that inform upon treatment decisions. Furthermore, as we are increasingly recognizing the spectrum of inflammatory complications associated with CAR T-cell therapy, including immune effector cell–associated hemophagocytic lymphohistiocytosis syndrome,⁶ and prolonged cytopenias,^46,47 it will be essential to include such outcomes in future prediction studies.

This study was supported by National Institutes of Health (NIH)/National Cancer Institute grant P30CA021765 and the American Lebanese Syrian Associated Charities.

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

Contribution: D.Z. and A.C.T. conceived the study; D.K., Y.L., D.Z., A.C.T., C.L.B., and M.Z. contributed to data collection; S.S. and Y.B. contributed to statistical analysis; A.C.T., S.N., S.S., and D.Z. analyzed and interpreted data and contributed to manuscript writing; and all authors reviewed and edited the manuscript.

Conflict-of-interest disclosure: S.G. is a member of the scientific advisory board of Be Biopharma and CARGO; is a member of the data and safety monitoring board of Immatics; and has received honoraria from Tessa Therapeutics within the last year. S.G. and R.E. are coinventors on patent applications in the fields of cell or gene therapy for cancer. C.L.B. has been awarded and has pending patent applications describing the use of engineered T cells as therapeutics, and has received research support from Merck Sharp & Dohme Inc, Bristol Myers Squibb, and Kiadis Pharma. The remaining authors declare no competing financial interests.

The current affiliation for D.Z. is Department of Pediatric Oncology, King Hussein Cancer Center, Amman, Jordan.

Correspondence: Aimee C. Talleur, Department of Bone Marrow Transplantation & Cellular Therapy, St. Jude Children’s Research Hospital, 262 Danny Thomas Pl, Memphis, TN 38115; email: aimee.talleur@stjude.org.