Blood Cell Crashes After CAR-T Therapy Follow a Predictable Risk Score, Study Finds

Chimeric antigen receptor T-cell therapy has rewritten the outlook for patients with relapsed or refractory B-cell malignancies, offering durable remissions where chemotherapy had failed. Yet the same engineered immune cells that hunt down malignant B cells can also drag down the blood counts that keep patients alive. A new retrospective cohort study from Beijing Tongren Hospital, published in BMC Cancer, maps exactly how deeply and how long those blood counts fall after CAR-T infusion, and identifies a simple pre-treatment score that flags who is most likely to suffer the worst drops. The findings arrive at a moment when the field is shifting its attention from the dramatic early toxicities of the therapy, such as cytokine release syndrome, toward the quieter but often more dangerous problem of prolonged bone marrow suppression.

The research team, led by Henan Wang and Xindi Liu with corresponding authors Liang Wang and Jia Cong, followed 96 consecutive patients treated between November 2022 and March 2025. All had relapsed or refractory B-cell malignancies, the population in which CAR-T products have shown their greatest success. The median age was 56.5 years, with an interquartile range of 46 to 66.5, reflecting the typical demographic of lymphoma patients referred for cellular therapy. Rather than relying on the older Common Terminology Criteria for Adverse Events alone, the investigators graded blood toxicity using the recently formalized immune effector cell-associated hematotoxicity framework, known as ICAHT, which distinguishes neutropenia, or N-ICAHT, from thrombocytopenia, or T-ICAHT, and tracks each on its own timeline.

That distinction matters clinically. Neutropenia strips patients of their main defense against bacterial and fungal infections, while thrombocytopenia raises the risk of spontaneous bleeding, including intracranial hemorrhage in the most severe cases. The two complications have different kinetics, different risk factors, and different management strategies, yet many earlier studies lumped them together under a single hematologic toxicity label. By separating them, the Beijing team could ask whether the predictors of a crashing neutrophil count differ from the predictors of a crashing platelet count, and the answer turned out to be yes, at least in degree.

The efficacy backdrop of the cohort was consistent with published experience. The overall response rate reached 78.3 percent at one month after infusion and settled to 65.5 percent at three months. Classical acute toxicities were common but rarely severe: cytokine release syndrome occurred in 74.0 percent of patients, yet only 2.1 percent experienced grade 3 or higher episodes, and immune effector cell-associated neurotoxicity syndrome appeared in 10.4 percent of patients, of which 5.2 percent were grade 3 or higher. These figures confirm that in experienced centers the inflammatory toxicities that dominated early CAR-T literature are now largely manageable, leaving hematologic recovery as the dominant safety concern.

The neutrophil data tell a stark story. Grade 3 to 4 neutropenia peaked at day 4 after infusion, when 71.9 percent of patients had severely depleted counts, and although most recovered, 34.7 percent still had severe neutropenia at day 28. Persistent suppression a month after infusion is the point at which infectious risk compounds, since patients remain on prophylactic antivirals and often prophylactic antibiotics, and any breakthrough infection can escalate rapidly. To predict who would fare worst, the team stratified patients using the CAR-HEMATOTOX score, a pre-infusion composite that incorporates baseline blood counts, inflammatory markers such as ferritin and C-reactive protein, and disease burden. The score was calculable in 86 patients, and 31 of them, or 36.0 percent, fell into the high-risk category of 2 or above.

The stratification worked. Severe early N-ICAHT, defined as grade 3 to 4 neutropenia in the early window after infusion, occurred in 16.1 percent of high-risk patients compared with 3.6 percent of low-risk patients, a statistically significant difference with a p value of 0.013. In multivariable logistic regression, where candidate predictors were first filtered through LASSO regression to avoid overfitting, a CAR-HEMATOTOX score of 2 or higher was the only significant predictor of severe early neutropenia, with an adjusted odds ratio of 4.04 and a 95 percent confidence interval of 1.01 to 22.17. For late neutropenia, a different factor emerged: patients who had received bridging therapy, meaning anti-cancer treatment given between lymphodepletion planning and CAR-T infusion, had an adjusted odds ratio of 4.25, with a confidence interval of 1.64 to 10.98, for developing late-phase neutropenia.

Thrombocytopenia told an even sharper story. High-risk patients by CAR-HEMATOTOX developed severe early T-ICAHT at a rate of 74.2 percent versus 25.5 percent in low-risk patients, and late T-ICAHT at 80.6 percent versus 47.3 percent, both differences highly significant with p values below 0.001. The CAR-HEMATOTOX score was the strongest adjusted predictor of both early and late thrombocytopenia, with adjusted odds ratios of 8.22 and 4.20 respectively. The authors suggest this stronger association for platelets may reflect the score’s inclusion of baseline platelet count and inflammatory burden, both of which directly bear on platelet production and consumption. Perhaps the most provocative observation was that 9 of the 10 patients who developed neurotoxicity, 90 percent, also developed thrombocytopenia. The team frames this as hypothesis-generating, noting it may point to shared endothelial injury pathways linking brain vascular leakage and platelet decline, while cautioning that residual confounding cannot be excluded in an observational design.

The study also addressed a long-running practical debate: whether the supportive care used to rescue blood counts might itself fan the inflammatory toxicities of CAR-T. Growth factor support with granulocyte colony-stimulating factor, G-CSF, was given to 78.1 percent of patients, and thrombopoietin receptor agonists, TPO-RA, to 48.9 percent. Neither intervention was associated with an increased rate of grade 3 or higher cytokine release syndrome or neurotoxicity. This is reassuring for clinicians who worry that stimulating recovering myeloid cells could amplify the cytokine storm, and it supports the increasingly common practice of early, aggressive count support in the weeks after infusion. The authors are careful to note that within the limits of the observational design, this finding cannot establish causality in either direction, but it removes a major practical objection to using these drugs.

What the study cannot do, by its own design, is settle questions of causation or generalize beyond a single center. As a retrospective cohort at one Chinese hospital, it captures real-world practice, including the bridging therapies and supportive care patterns actually used, but it lacks the randomization needed to prove that the CAR-HEMATOTOX score modifies risk rather than merely marking it. The confidence interval around the early neutropenia odds ratio, stretching from barely above 1 to more than 22, reflects the modest sample size of 96 patients. Still, the consistency of the signal across both neutropenia and thrombocytopenia, and its alignment with prior validation work on the score in European and American cohorts, strengthens the case that pre-infusion risk stratification should become routine.

The practical implications are immediate. Patients flagged as high risk before infusion could be considered for closer platelet monitoring, earlier thrombopoietin receptor agonist initiation, lower thresholds for transfusion, and heightened vigilance for neurologic symptoms given the observed overlap with ICANS. The finding that bridging therapy predicts late neutropenia adds weight to the argument for minimizing or carefully selecting pre-infusion treatment in patients who can safely proceed directly to CAR-T. As CAR-T therapy expands beyond academic centers into community oncology, tools like the CAR-HEMATOTOX score offer a way to bring sophisticated risk assessment to settings without specialized cellular therapy infrastructure. The Beijing cohort suggests that the era of treating CAR-T hematotoxicity as an unpredictable hazard is ending, replaced by a framework in which the patients most likely to crash are identifiable before the first engineered cell is infused.

Subject of Research: Hematologic toxicity, including neutropenia and thrombocytopenia, following CAR-T cell therapy for B-cell malignancies

Article Title: Analysis of hematologic toxicity profiles following CAR-T cell therapy: a retrospective cohort study

Article References: Wang, H., Liu, X., Li, F., Liang, Y., Zeng, L., Zhu, J., Wang, L., & Cong, J. (2026). Analysis of hematologic toxicity profiles following CAR-T cell therapy: a retrospective cohort study. BMC Cancer. https://doi.org/10.1186/s12885-026-16932-0

Image Credits: AI Generated

DOI: 10.1186/s12885-026-16932-0

Keywords: CAR-T cell therapy, ICAHT, neutropenia, thrombocytopenia, CAR-HEMATOTOX, hematologic toxicity, B-cell malignancy, cytokine release syndrome, ICANS, G-CSF, thrombopoietin receptor agonist, retrospective cohort study