Development of ALL-Hematotox: predicting post-CAR T-cell hematotoxicity in B-cell acute lymphoblastic leukemia

Immune effector cell–associated hematotoxicity (ICAHT) is increasingly recognized as a major complication of B-cell targeted chimeric antigen receptor (CAR) T-cell therapy.^1-3 Although ICAHT incidence and severity is documented in large B-cell lymphoma (LBCL), mantle cell lymphoma (MCL), and multiple myeloma (MM),⁴ ICAHT has not been described in B-cell acute lymphoblastic leukemia (B-ALL). Importantly, because of infections and morbidity associated with prolonged cytopenias, predicting hematotoxicity early on may help guide post-CAR T-cell management. Accordingly, CAR-HEMATOTOX (CAR-HT), a model to predict post-CAR T-cell hematotoxicity, was developed and validated in adults with relapsed/refractory (r/r) LBCL, MM, and MCL.^5-7 This model uses baseline hematopoietic reserve and inflammatory markers before lymphodepleting chemotherapy to stratify patients into low versus high risk for developing severe prolonged neutropenia, defined as an absolute neutrophil count (ANC) <500/μL for ≥14 days (eg, ≥ICAHT grade 3).¹ However, applicability of CAR-HT in patients with r/r B-ALL is unknown.

One unique concern in B-ALL is related to baseline hematopoietic reserve and how recovery may be impacted by disease. As a bone marrow (BM) infiltrative process, B-ALL causes cytopenias and impaired hematopoiesis.⁸ In contrast, only 19% to 36%^9,10 of patients with LBCL are expected to have BM involvement. Similarly in MM, BM involvement is often variable with retention of hematopoiesis and pancytopenia generally only seen in advanced cases.⁶ Importantly, high BM disease burden has been associated with longer cytopenias following CAR T cells across multiple B-cell malignancies,^4,11 and post-CAR T-cell cytopenias are exceedingly common in B-ALL. Indeed, in 1 report, 41% and 53% of patients experienced grade 3 to 4 thrombocytopenia and grade 3 to 4 neutropenia, respectively—neither of which resolved by day 28.¹² 

As cytopenias post-CAR T-cell therapy may impact long-term outcomes, we sought to first evaluate early ICAHT in B-ALL and subsequently assess utility of CAR-HT in this population, with a plan to refine the CAR-HT model as needed to account for disease-specific differences.

We performed an institutional review board–approved retrospective analysis (NCT03827343) of children, adolescents, and young adults (AYA) with r/r B-ALL treated with CAR T cells on 1 of 4 phase 1 clinical trials targeting CD19 and/or CD22 (NCT01593696, NCT02315612, NCT0344839, or NCT05442515) at the National Institutes of Health, National Cancer Institute (NCI) between July 2012 and June 2023. The primary objectives were to apply early ICAHT grading and assess the applicability of CAR-HT in B-ALL for predicting severe prolonged neutropenia, from CAR T-cell infusion (day 0 [D0]) until day +30 (D30). Our shorter follow-up differs from the original CAR-HT⁵ in which 60 days of monitoring was available, in part because of referral practices and/or use of consolidative hematopoietic stem cell transplant (HSCT) before D+60.

First, we characterized the NCI B-ALL cohort to establish baseline and postinfusion cytopenias and pattern of count recovery. Common Terminology Criteria for Adverse Events (CTCAE) was used to define grade 3 and 4 neutropenia and thrombocytopenia. Neutrophil recovery was defined as achieving an ANC ≥500/μL for 3 consecutive days, as per Center for International Blood and Marrow Transplant Research (CIBMTR) post-HSCT engraftment definitions. Early ICAHT grading was based on Rejeski et al¹ (supplemental Table 1 [available on the Blood website]). Additional variables, including baseline BM disease (using the most proximal pre-lymphodepletion [LD] evaluation [generally within 14 days]), were also collected.

Next, the per patient CAR-HT score was calculated using the necessary variables (hemoglobin, platelets, ANC, C-reactive protein [CRP], and ferritin) from the original scoring.⁵ Laboratory values were chosen from LD initiation or up to 3 days prior. Risk stratification toward low risk (LR; score 0-1) or high risk (HR; score ≥2) was based on the original CAR-HT scoring. Patients were excluded from CAR-HT analysis if missing laboratory values precluded calculating CAR-HT risk assignment. Accordingly, patients meeting the threshold for HR (CAR-HT ≥2) were kept as HR, as a missing value (eg, ferritin or CRP) would only increase their total score without risk of recategorization. In contrast, we excluded those with a CAR-HT score <2 where a missing value could inform HR categorization (ie, increase it to HR threshold).

Considering potential limitations to CAR-HT, we also proposed to refine this model if needed to improve its use in a B-ALL population. For this purpose, the NCI cohort served as the training cohort, whereas the adapted score was validated in 2 independent external cohorts of patients with B-ALL from Memorial Sloan Kettering (MSK) and Seattle Children’s Hospital (SCH). Importantly, all patients received LD with cyclophosphamide or a combination of cyclophosphamide and fludarabine with comparable dosing across trials and centers (supplemental Appendix).

Exploratory objectives included assessing the application of a modified CAR-HT score with clinical CAR T-cell outcomes, including incidence of cytokine release syndrome (CRS), CRS severity, neurotoxicity, BM complete response (CR), minimal residual disease (MRD) negative CR, overall survival (OS), and event-free survival (EFS).

CRS grading was per the American Society for Transplantation and Cellular Therapy (ASTCT) guidelines¹³ for the NCI cohort. Incidence of any grade neurotoxicity was defined by either ASTCT or CTCAE criteria. For NCI and MSK, disease response was assessed with BM morphology and MRD assessment by flow cytometry at day 28 (±4 days) and day +21 at SCH, or earlier in those with rapid disease progression before day 28. High-disease burden was defined by ≥5% marrow involvement based on recent efforts supporting this designation as a critical threshold used in distinguishing post-CAR T-cell outcomes.^14,15 Consistent with the CAR-HT model, patients with B-ALL with early death or progressive disease that necessitated intensive chemotherapy impacting cytopenias before D30 were excluded from analysis in the training cohort (n = 3). A TRIPOD (transparent reporting of studies on prediction models for individual prognosis or diagnosis) checklist for predictive model development and validation¹⁶ is included in the supplemental Appendix.

For all statistical analysis, a nonparametric test with a P < .05 was used to determine statistical significance. The Fisher exact test was used to compare categorical variables. Continuous variables were assessed using a Spearman correlation coefficient (r). A Mann-Whitney U test was performed to compare a continuous variable between 2 groups. A receiver operating characteristic (ROC) curve was used to assess continuous variables and association with severe prolonged neutropenia. The Youden J statistic (sensitivity + specificity −1) based on the ROC curves was used to determine the optimal threshold for risk assignment. Kaplan-Meier curves were generated to characterize survival outcomes (OS, EFS) starting from time of CAR T-cell infusion. Events were determined by earliest date of no response, relapse, or death. Patients without an event were censored at their date of last follow-up. Log-rank P values were calculated in Kaplan-Meier survival analyses for group comparisons. All statistical analyses were performed using GraphPad Prism version 9.3.1.

Across 156 patients with r/r B-ALL treated on an NCI CAR T-cell trial, the median age was 16 years (range, 4-39 years), and 108 (69%) were male. The median number of prior lines of therapy was 5 (range, 1-14), and 88 (56%) had relapsed following a prior allogeneic HSCT. Separated by CAR T-cell construct, 49 (31%) received a CD19/28z CAR T-cell product, 74 (47%) received a CD22/41BB CAR T-cell product, and 33 (22%) received a CD19/CD22 dual-targeted CAR T-cell product. The baseline median BM disease in patients with BM involvement (n = 151) was 27% (interquartile range, 1%-81%), which were performed at a median of 12 days (interquartile range, 4.25-18 days) pre-LD. This represented 36% with an M1 marrow (<5% blasts); 13% with an M2 marrow (5%-25% blasts); and 51% with an M3 marrow (>25% blasts). One hundred (65%) patients were considered as having high disease burden. Overall CR rate was 73%. Grade 3 or higher CRS was seen in 19% (Table 1).

The median baseline ANC was 1370/μL (95% confidence interval [CI], 1020/μL-1760/μL; Figure 1A), and median baseline platelet count was 124 × 10³/μL (95% CI, 92-141 × 10³/μL; Figure 1B). Collectively, 35 (22%) and 14 (9%) of all patients had a grade 4 ANC (<500/μL) and platelet count (<25 × 10³/μL) at baseline, respectively. The median duration of severe neutropenia (ANC <500/μL from D0 to D30) was 13 days (95% CI, 10-16 days; Figure 1C). Severe prolonged neutropenia (ANC <500/μL for ≥14 days D0-D30) was seen in 77 (49%) patients (Figure 1D). Of 146 patients with evaluable D30 ANC and platelet counts, 91 (62%) had grade 3 or 4 neutropenia (ANC <1000/μL) and 58 (40%) patients had grade 3 or 4 thrombocytopenia (platelet count <50 × 10³/μL) at D30 (Figure 1E). Collectively, 73 (47%) did not have neutrophil recovery and 74 (47%) did not have platelet count recovery (platelet count >50 × 10³/μL for 3 consecutive days) by D30 (Figure 1F-G, respectively). Ultimately, only 61 (39%) patients had both ANC and platelet recovery by D30 (Figure 1H). Per early ICAHT grading,¹ 83 (53%) experienced grade 3 or higher ICAHT (Figure 1I), which slightly varied across trial (≥grade 3 ICAHT range, 41%-75%) (supplemental Figure 1A-D).

Serial ANC analysis from LD to D30 post-CAR T-cell infusion revealed no clear day of recovery for ANC or platelets (Figure 2A). Stratified by ICAHT grade, those with high-grade ICAHT (grades 3-4) started at a lower baseline ANC without count recovery (Figure 2B). In contrast, patients with low-grade ICAHT (grade 0-2) had a higher median ANC and experienced only a transient decrease and gradual recovery (Figure 2B). Last, we evaluated the impact of disease burden. Expectedly, patients with high baseline BM disease had a lower baseline ANC and continuously low counts postinfusion without subsequent count recovery (Figure 2C), differing markedly from the pattern in low BM disease (Figure 2C).

Serial platelet count analysis revealed a gradual decrease at CAR T-cell infusion through day 25, followed by gradual recovery (Figure 2D). However, patients with high-grade ICAHT had lower baseline platelet counts that did not recover by day 30 (Figure 2E). Patients with low BM disease had higher baseline platelet counts, and despite a gradual decrease postinfusion, levels consistently exceeded the platelet levels of those with high BM disease (Figure 2F).

We subsequently sought to evaluate for associations between clinical characteristics and ICAHT grading. Although prior HSCT and prior CAR T-cell use did not impact ICAHT grade, patients with grade 3 or higher CRS more frequently experienced grade 3 or higher ICAHT (69% vs 50%, respectively; P = .07; supplemental Table 2). As patients with higher CRS grade more often had higher disease burden, the corresponding ICAHT grade may largely reflect underlying disease (Figure 2G-I).

Additionally, as postinfusion nonresponse or residual disease can impact the potential for count recovery, a subanalysis comparing nonresponders (n = 42) with those achieving a CR (n = 114) revealed no significant difference in severe prolonged neutropenia (P = .15; supplemental Figure 2A), despite nonresponder patients exhibiting more cumulative days of neutropenia (20.5 vs 12.0 days; P = .04; supplemental Figure 2B).

Following the cytopenia and ICAHT characterization, we applied the original CAR-HT scoring (supplemental Table 3) to our B-ALL cohort and assessed its association with the outcome of severe prolonged neutropenia. Because of missing baseline ferritin in 91 (58%) patients and missing CRP values in 14 (9%) patients, 108 patients had an assignable CAR-HT risk group and were further analyzed (supplemental Methods). An ROC curve generated to analyze CAR-HT score performance in classifying risk for severe prolonged neutropenia showed discrimination with the outcome (area under the curve [AUC] = 0.74, P = .001; Figure 3A). Accordingly, when applying the original CAR-HT risk-stratification cutoff to our B-ALL cohort, we found an association between CAR-HT risk group and severe prolonged neutropenia (P = .008; Figure 3B). However, as 87% (n = 94) of the patients with B-ALL fell into the HR group, which is substantially higher than the 49% of patients with B-ALL that experienced severe prolonged neutropenia, it raised concerns that CAR-HT was limited in discriminating for hematotoxicity in B-ALL. The imbalance in risk group assignment remained even when increasing cutoff thresholds (eg, to CAR-HT score of ≥3) (supplemental Figure 3A-B). This is in stark contrast to its use in LBCL, MCL, and MM, where the HR population comprised 46%, 46%, and 44% of the cohorts, respectively, and was more equally balanced (Figure 3C).

Separated by risk score, the median duration of severe neutropenia in the CAR-HT LR group (n = 14) was 1.5 days (95% CI, 0-15 days) compared with 22.5 days (95% CI, 16-26 days; P < .0001) for the HR group (n = 94) (supplemental Figure 4A). Despite double the length of follow-up in the LBCL cohort (through D+60), the median duration of neutropenia in B-ALL cohort HR patients was substantially longer (median duration was 16.5 days [95% CI, 13-43 days] in LBCL). Furthermore, there was no clear distinction between CAR-HT LR and HR in terms of neutrophil and platelet count recovery (supplemental Figure 4B-C). Therefore, despite the association between risk groups and severe prolonged neutropenia, as most patients with B-ALL were considered HR (87%), applicability of CAR-HT was limited.

Given the skewed proportion of B-ALL categorized as HR in CAR-HT, we sought to examine key differences between the LBCL cohort on which CAR-HT was established and our B-ALL cohort (Table 1). In addition to having a younger age in the B-ALL cohort compared with the LBCL cohort, the B-ALL cohort had a substantially lower median baseline ANC (1370/μL vs 2540/μL; P < .0001) and lower median baseline platelet count (124 vs 164 × 10³/μL; P = .0002). Notably, baseline ferritin in our B-ALL cohort was almost triple that of the LBCL cohort (median, 1421 vs 501 ng/mL; P < .0001). The more substantial baseline cytopenias and BM involvement in B-ALL compared with LBCL, therefore, provided rationale for modifying the original CAR-HT model to better serve a B-ALL population.

To preserve the established utility of the CAR-HT model, we assessed each individual variable for a univariate correlation with the continuous variable of cumulative days of neutropenia (ANC <500/μL) from D0 through D30. Lower platelets, ANC, and hemoglobin at time of LD all associated with a higher number of cumulative days of neutropenia (r = −0.66, P < .0001; r = −0.52, P < .0001; and r = −0.43, P < .0001, respectively) (Figure 3D-F). Baseline CRP and ferritin were also analyzed against duration of neutropenia post-CAR T-cell infusion, showing a positive correlation with CRP (r = 0.34, P < .0001; Figure 3G) and a weaker correlation for ferritin (r = 0.25, P = .05; Figure 3H). Importantly, although other variables were available for nearly all patients, only 65 (42%) had ferritin measurements within the designated baseline time frame, limiting its utility for modelling. ROC curves of each baseline feature and the binary outcome of severe prolonged neutropenia were also studied (supplemental Figure 5A-E) and showed that all variables were associated with the primary outcome of severe prolonged neutropenia (P < .05) except for ferritin (P = .2; supplemental Figure 5E), potentially because of substantially higher baseline values compared with LBCL. With the simple removal of ferritin from the CAR-HT model, the association with severe prolonged neutropenia was retained, with improved risk group classification (supplemental Figure 6A-C).

Given the association and biologic relevance of BM disease with cytopenias and ICAHT severity, we also assessed continuous baseline BM disease with duration of neutropenia and found that higher disease burden associated with increased duration of neutropenia (r = 0.64, P < .0001; Figure 3I). Low vs high BM disease also associated with duration of neutropenia, OS, and EFS (supplemental Figure 7A-C). Considering these findings, we merged the relevant factors from CAR-HT with baseline BM disease to propose an optimal B-ALL–specific CAR-HT model, called ALL-Hematotox (ALL-HT) (Figure 4A). For incorporation of BM disease, an iterative process led to selection of standard categorization of disease (M1 marrow [<5% blasts] = 0 points; M2 marrow [5%-25% blasts] = 1 point; and M3 marrow [>25% blasts] = 2 points) as the optimal approach for ALL-HT (supplemental Methods; supplemental Table 4).

The ROC curve comparing ALL-HT score with severe prolonged neutropenia showed a significant association with an AUC of 0.84 (P < .0001; Figure 4B). The optimal cutoff assignment for HR vs LR was determined to be 4, which was identified by optimizing sensitivity and specificity using the Youden (J) statistic in the ALL-HT ROC curve (Figure 4B; supplemental Table 5). The threshold for HR assignment was lowered to ≥3 for patients without a CRP. Applying these thresholds to determine risk by ALL-HT score, we observed a strong association between risk assignment and severe prolonged neutropenia. Specifically, most patients with severe prolonged neutropenia were categorized to HR (P < .0001; Figure 4C). Importantly, ALL-HT was able to identify patients who were at HR for severe prolonged neutropenia despite the similarity in baseline demographic features between LR and HR groups (supplemental Table 6) with more appropriate risk stratification, with 47% (n = 73) falling into the HR group, of whom only 21% did not have prolonged severe neutropenia. When looking at cumulative days of neutropenia post-CAR T-cell infusion (D0-D30), the ALL-HT LR group (n = 83) had a median of 4 days (95% CI, 2-8 days) of neutropenia (ANC <500/μL), whereas the HR group (n = 73) had a median of 26 days (95% CI, 22-28 days) of neutropenia (P < .0001; Figure 4D). Lastly, when stratifying patients by ALL-HT risk group across daily ANC and platelet counts, a distinct, nonoverlapping pattern of recovery of both ANC and platelet counts in the post-CAR T-cell period highlights the ability of ALL-HT to distinguish risk groups for hematotoxicity (Figure 4E-F).

We next analyzed ALL-HT risk stratification and association with secondary outcomes, including CRS, neurotoxicity, BM CR, OS, and EFS. ALL-HT scoring was not associated with incidence of CRS (P = .12), severe CRS (≥grade 3) (P = .67), or neurotoxicity (P = .57) (supplemental Figure 8A-C). However, ALL-HT risk group was associated with BM CR (P = .03; supplemental Figure 8D) and BM MRD-negative CR (P = .03; supplemental Figure 8E), with HR patients having lower MRD-negative CR rates (74% in LR vs 58% in HR). Using Kaplan-Meier estimates, we observed substantially shorter OS in the ALL-HT HR group compared with the LR group (2-year OS rate: 23% vs 42%; median OS: 9.8 vs 24.0 months; log-rank P = .0002; Figure 4G). Furthermore, we noted shorter EFS in the ALL-HT HR group compared with the LR group (2-year EFS rate: 10% vs 27%; median EFS: 3.2 vs 7.1 months; log-rank P = .001; Figure 4H).

On the basis of our ALL-HT model, we sought to evaluate its performance in external data sets, reflecting a broad range of patient populations (eg, pediatric, AYA, and adult) and CAR T-cell products (eg, CD19 targeting with CD28z and 4-1BBz costimulation). In collaboration with investigators at MSK and SCH, we applied ALL-HT to their patients with r/r B-ALL receiving CAR T cells. Across 106 patients at MSK (median age, 27.5 years [range, 0.7-73.8 years]) and 84 patients at SCH (median age, 12.4 years [range, 1.5-26.0 years]) (Table 2), the duration of neutropenia was similar between the training (NCI), MSK, and SCH cohorts (median, 13 days [95% CI, 10-16 days]; median, 14 days [95% CI, 12-17 days]; and median, 12 days [95% CI, 10-20 days], respectively) (Table 2). There was also a similar proportion of patients with severe prolonged neutropenia in the training, MSK, and SCH cohorts as well (49%, 54%, and 48%, respectively; Table 2; supplemental Table 7). ROC curves for the ALL-HT score and the outcome of severe prolonged neutropenia were significant for both the MSK and SCH validation cohorts (AUC = 0.89, P < .0001; and AUC = 0.90, P < .0001, respectively; Figure 5A-B; supplemental Table 8). Additionally, the proportion of patients with severe prolonged neutropenia was significantly higher in the ALL-HT HR group compared with the LR group (MSK: 90% vs 20%, P < .0001; SCH: 74% vs 14%, P < .0001; Figure 5C-D). In both the MSK and SCH cohorts, there was a greater number of days of neutropenia in the ALL-HT HR group compared with the LR group (MSK median, 28 vs 10 days, P < .0001; SCH median, 25 vs 8 days, P < .0001; Figure 5E-F). Lastly, ALL-HT HR patients in both validation cohorts had a shorter OS compared with ALL-HT LR patients (MSK log-rank P = .006; SCH log-rank P = .0025; Figure 5G-H), comparable to findings in our NCI cohort. ALL-HT HR was associated with a shorter EFS in the MSK cohort only (supplemental Figure 9A-B).

Through this effort, a series of models and associations were explored—including CAR-HT, CAR-HT without ferritin, and BM disease alone. Ultimately, the combination of all variables in ALL-HT provided superior discrimination over BM disease alone, CAR-HT, and CAR-HT without ferritin for severe prolonged neutropenia in the training cohort (highest AUC), which was confirmed in both external validation cohorts (supplemental Figure 10A-C).

Prolonged cytopenias are among the most common adverse effects of B-cell targeted CAR T cells and are associated with a significant increase in risk of infection and morbidity.^1,17,18 Despite extensive analysis of hematotoxicity in adults with LBCL, MCL, and MM,^5-7,19 data on post-CAR T-cell hematotoxicity in r/r B-ALL are limited,²⁰ and to the best of our knowledge, early ICAHT grading has not been described in B-ALL. Additionally, the applicability of CAR-HT in B-ALL is unknown.

Through this effort, we found that preinfusion baseline cytopenias were pronounced in r/r B-ALL, particularly in those with high-burden disease (≥5% marrow involvement). Additionally, the incidence of grade 3 or higher early ICAHT in B-ALL was much higher than in B-cell lymphomas.⁴ Despite limiting the duration of follow-up to half that of comparator lymphoma cohorts,⁵ the median duration of severe neutropenia post-CAR T-cell infusion was more than a week longer in patients with B-ALL. Additionally, almost half (49%) of patients with B-ALL experienced severe prolonged neutropenia in the first month postinfusion, with 47% failing to achieve neutrophil recovery by D30. Given impaired hematopoiesis from BM infiltration, the high prevalence of baseline cytopenias is unsurprising, but the extent of postinfusion cytopenias at D30 is notable.

Factors contributing to postinfusion cytopenias include a high rate of inflammation or immune dysregulation²¹ (as reflected by a higher incidence of severe CRS,²² particularly in high-disease burden) and decreased BM reserve. The latter is often a result of intensive prior treatment lines, including radiation, which further contributes to delayed BM recovery (independent of any impact from CAR T cells). As illustrated by the substantially higher ferritin in patients with B-ALL, potentially because of a combination of inflammation, disease, and/or extensive prior transfusions, even baseline biomarkers can be markedly different in B-ALL compared with LBCL and may not accurately reflect underlying hyperinflammation. Collectively, these data provide a foundation for future study of ICAHT and cytopenias in patients with B-ALL and inform clinical trial design and expectations for anticipated depth and duration of postinfusion cytopenias, particularly in those with high-disease burden and baseline cytopenias.

On the basis of the importance of predicting post-CAR T-cell hematotoxicity, although CAR-HT showed statistical association with severe prolonged neutropenia in B-ALL, its inability to discriminate patients by risk (nearly 90% of patients with B-ALL were HR) highlighted a clear need for a new model. Importantly, however, as most individual variables (ANC, platelets, hemoglobin, and CRP) of CAR-HT were meaningfully associated with severe prolonged neutropenia in B-ALL, maintaining the existing model as feasible, considering its broad use (eg, CAR-HT without ferritin), was critical. Given the limited association of ferritin with severe prolonged neutropenia in this cohort, potentially because of substantial baseline elevations and missing values, BM disease involvement emerged as a key predictor of outcomes, facilitating the replacement of ferritin with BM disease to establish ALL-HT. The ALL-HT model was found to more appropriately distinguish LR from HR than CAR-HT and had a significant association with cumulative days of neutropenia and severe prolonged neutropenia. Importantly, the utility of the ALL-HT score was confirmed in 2 external data sets across a broad range of patients with B-ALL and a multitude of CAR T-cell products, highlighting the external validity of the score. Moreover, the practical implication of the ALL-HT model is that by retaining the key features of CAR-HT, only minimal modification is required by CAR T-cell providers who are already using CAR-HT across other B-cell subtypes, facilitating more rapid implementation.

In addition to having meaningful associations with hematotoxicity, ALL-HT HR patients had a shorter OS in the training cohort and in both validation cohorts compared with ALL-HT LR patients. However, ALL-HT was not associated with incidence of CRS, severe CRS, or neurotoxicity, highlighting a need for future models to predict these other key outcomes.

On the basis of the retrospective nature of this study, limitations related to missing variables (eg, ferritin, CRP) are inherent. Nonetheless, our analysis accounts for missing values when building and validating this model. Moreover, we would advise that future patients receiving CAR T cells have ferritin and CRP checked at baseline—as these may become more clinically relevant, particularly as patients start receiving CAR T cells earlier in therapy with lower disease burden.²³ As consistent follow-up beyond D30 was limited, especially as patients transitioned to their home institution and/or rapidly proceeded to HSCT, future efforts will include evaluation of late ICAHT and use of the ALL-HT beyond D30 to explore patterns of delayed hematologic recovery. Additionally, although our collective data are derived from patients receiving a host of CAR T-cell constructs, including those which are commercially available, this heterogeneity is reflective of the B-ALL CAR T-cell landscape, potentially making these findings even more broadly applicable. Last, although ICAHT reflects cytopenias collectively due to LD and CAR T cells, given that all patients received comparable LD regimens containing only slight differences in dosing, we did not correlate the specific dosing regimen to ICAHT. To address this question, future trials need to implement pharmacokinetic measurements as differences even with the same dosing regimen have been reported.

In conclusion, we highlight the high incidence of early ICAHT in B-ALL. Indeed, the most salient finding is in establishing the undisputable impact of BM disease involvement on post-CAR T-cell cytopenias. This adds to emerging data demonstrating the critical importance of preinfusion BM disease involvement in B-ALL on post-CAR T-cell outcomes, including survival,^14,15 toxicity,^24-26 relapse,²⁷ and now cytopenias. Ultimately, the eloquent combination of peripheral cytopenias with BM disease burden to develop ALL-HT, which predicts for hematotoxicity alongside survival, is biologically sound. Further application and validation of ALL-HT in prospective studies and consistent use of ICAHT are critical next steps. As the CAR-HT model has also been shown to predict severe infection, duration of hospitalization, progression-free survival, and OS,¹⁹ future efforts will assess for associations with other clinical outcomes (eg, hospitalizations and infections), particularly as this may provide insights in mitigation of predicted sequelae (eg, preemptive interventions to limit infection)^19,28,29 in patients deemed to be high risk.

The authors gratefully acknowledge the study participants and their families, referring medical care teams, the faculty and staff of the National Institutes of Health (NIH), Clinical Center (CC), who provided their expertise in the management of the study participants, and the data managers, research nurses, and patient care coordinators involved with this work. The authors acknowledge Crystal L. Mackall, Alan S. Wayne, Daniel W. Lee, and Terry J. Fry, along with Cindy L. Delbrook, for their leadership in implementing these studies at the NIH/National Cancer Institute (NCI). The authors acknowledge Naveed Ahmed for support of data collection at MSKCC. The authors additionally acknowledge Craig Sauter and Sujata Patil (Cleveland Clinic) for their generous input throughout this analysis.

This work was supported in part by the Intramural Research Program, Center for Cancer Research, NIH/NCI and NIH/CC (ZIA BC 011823). Research support was provided by the NIH Medical Research Scholars Program, a public-private partnership supported jointly by the NIH and contributions to the Foundation for the NIH from private donors. Roni Shouval reports grant support from the NIH/NCI Memorial Sloan Kettering Cancer Center support grant (P30 CA008748); and NIH/NCI K08-CA282987 award.

The content of this publication does not necessarily reflect the views of policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US government.

Contribution: M.S.N. and N.N.S wrote the first version of the manuscript; S.K.S., K.R, A.J.L, and M.S. contributed to manuscript writing; M.S.N., S.K.S, and N.N.S. performed primary data analysis; M.S.N., S.K.S., K.R., A.J.L., C.A., B.Y., K.A.W., A.M.A., R.S., K.C., and H.S. gathered primary data necessary for this analysis; M.S.N. conducted primary statistical analysis; M.S.N., S.K.S., K.R., A.J.L., Y.V., J.H.P., C.A., R.A.G., M.S., and N.N.S. provided critical input for data analysis; S.K.S., H.S., B.Y., and N.N.S. provided patient care and oversight of National Institutes of Health/National Cancer Institute clinical trials from which participants in the training cohort were enrolled on; A.J.L, K.C., R.A.G., C.A., and J.H.P. provided patient care and oversight for the external validation cohorts; no nonauthor wrote the first draft or any part of the manuscript; R.S provided additional support for data; and all authors contributed to the final manuscript and have agreed to be coauthors.

Conflict-of-interest disclosure: This work was supported in part by the Intramural Research Program, Center of Cancer Research, National Institutes of Health (NIH), National Cancer Institute and NIH Clinical Center, (ZIA BC 011823; N.N.S). N.N.S. receives research funding from Lentigen, VOR Bio, and CARGO Therapeutics. N.N.S. has attended advisory board meetings (no honoraria) for VOR, ImmunoACT, and Sobi. K.R. receives research funding, consultancy, honoraria, and travel support from Kite/Gilead; honoraria from Novartis; consultancy from Bristol Myers Squibb/Celgene; and honoraria and travel support from Pierre-Fabre. Y.V. received a 1-time consultancy fee from EastRx. M.S. has received industry research support from Amgen, Bristol Myers Squibb/Celgene, Gilead/Kite, Miltenyi Biotec, Molecular Partners, Morphosys, Novartis, Roche, Seattle Genetics, and Takeda; has served as a consultant/scientific advisory board member at Autolus, AvenCell, CanCell Therapeutics, CDR-Life, Genmab US, Ichnos Sciences, Incyte Biosciences, Interius BioTherapeutics, Janssen, Millennium Pharmaceuticals, Miltenyi Biomedicine, Molecular Partners, Nektar Therapeutics, Novartis, Pfizer, Ridgeline Discovery, Sanofi, Scare, and Takeda; and serves on the speakers' bureau at Amgen, AstraZeneca, Bristol Myers Squibb/Celgene, Gilead/Kite, GSK, Janssen, Novartis, Octapharma, Pfizer, Roche, Springer Healthcare, and Takeda. Educational grants were received to develop the app "MyTcell" from Bristol Myers Squibb, Gilead, Janssen, Novartis, Roche, and Takeda. The remaining authors declare no competing financial interests.

Correspondence: Nirali N. Shah, MHSc, Pediatric Oncology Branch, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, MD 20892; email: nirali.shah@nih.gov.