Fever After CAR T-Cell Therapy: New Push for Early Detection of Toxicity

T-cell–redirected immunotherapies, including chimeric antigen receptor (CAR) T-cell therapy and bispecific antibodies, have transformed the treatment of relapsed or refractory hematologic malignancies, achieving durable remissions in otherwise refractory B-cell acute lymphoblastic leukemia, B-cell lymphoma, and multiple myeloma. But the clinical benefit of these therapies is inseparable from a distinct spectrum of immune-mediated toxicities. Cytokine release syndrome (CRS) and immune effector cell–associated neurotoxicity syndrome (ICANS) are the two best-characterized complications after CAR T-cell therapy and are also observed after bispecific antibody treatment, although their incidence and severity patterns generally differ. A new review published in Blood Research synthesizes current concepts in the biology, recognition, risk prediction, early detection, and personalized management of these syndromes, arguing that management has evolved from reactive treatment of established toxicity toward early detection, risk-adapted monitoring, and individualized intervention.

CRS is not simply a direct consequence of T-cell expansion but a multicellular process involving tumor cells, effector T cells, monocytes and macrophages, and the vascular endothelium. After antigen recognition, CAR T cells undergo local activation and expansion, while bispecific antibodies promote repeated or continuous engagement of endogenous T cells and malignant cells. Early cytokine signals such as interleukin (IL)-2, IL-7, and IL-15 support the proliferation and survival of immune effector cells. Subsequently, inflammatory mediators including IL-1, GM-CSF, interferon-γ, and tumor necrosis factor-α activate monocytes and macrophages, which in turn produce high levels of IL-6, IL-1, nitric oxide, and additional inflammatory cytokines, amplifying the systemic response. Preclinical and translational studies show that myeloid cells are not merely bystanders but central amplifiers of CRS. Among the downstream mediators, IL-6 plays a central clinical role, as reflected by the efficacy of IL-6 receptor blockade with tocilizumab—yet IL-6 is only one component of a broader inflammatory network, which explains why IL-6 blockade may be insufficient in severe or refractory cases in which alternative inflammatory pathways remain active.

Endothelial activation links cytokine signaling to the clinical phenotype of CRS. In clinical cohorts of patients receiving CD19 CAR T-cell therapy, severe CRS has been associated with distinct cytokine kinetics and biomarkers of endothelial activation. Inflammatory cytokines increase vascular permeability and promote capillary leakage, hypotension, hypoxia, coagulopathy, and, in severe cases, multiorgan dysfunction. Clinical severity therefore reflects not only the magnitude of immune activation, influenced by tumor burden and baseline inflammation, but also host susceptibility, including endothelial vulnerability and limited organ reserve. ICANS, meanwhile, is biologically related to CRS but is not simply a neurologic extension of it. Systemic inflammation increases endothelial permeability, facilitating the entry of cytokines and immune cells into the central nervous system, where local cytokine accumulation, microvascular injury, and neuroinflammation may lead to encephalopathy, aphasia, tremors, seizures, cerebral edema, and, rarely, fatal neurologic deterioration. Tocilizumab has limited penetration into the central nervous system and does not directly suppress many inflammatory pathways implicated in ICANS, which is why corticosteroids remain the mainstay of treatment for clinically significant neurotoxicity.

Recognizing these syndromes early depends on structured assessment. The ASTCT consensus grading system provides a unified framework across clinical trials and routine practice. CRS grading rests on three core clinical elements: fever, hypotension, and hypoxia. Grade 1 is fever without hypotension or hypoxia; grade 2 adds hypotension not requiring vasopressors and/or hypoxia requiring low-flow oxygen; grade 3 requires a vasopressor and/or high-flow oxygen; and grade 4 involves life-threatening hypotension requiring multiple vasopressors and/or positive-pressure ventilation. ICANS grading incorporates the immune effector cell–associated encephalopathy (ICE) score, a bedside tool assessing orientation, naming, command following, writing, and attention, with a maximum score of 10. Grade 1 generally corresponds to an ICE score of 7–9, grade 2 to 3–6, grade 3 to 0–2 with severe impairment, and grade 4 to coma, life-threatening seizures, or cerebral edema. CRS and ICANS should be graded independently because their timing, clinical course, and optimal management may differ. Clinically, CRS is typically the earliest toxicity, beginning with fever within the first several days after CAR T-cell infusion or during step-up dosing of bispecific antibodies, and early ICANS may manifest as inattention, impaired handwriting, difficulty finding words, or expressive aphasia before more obvious deterioration in arousal.

The stakes of severe toxicity extend beyond acute resource use. In a large CIBMTR registry analysis of patients with B-cell lymphoma receiving CD19 CAR T-cell therapy, grade ≥ 3 CRS and grade ≥ 3 ICANS were independently associated with inferior overall survival, with adjusted hazard ratios of 1.79 and 1.34, respectively. In a systematic review and meta-analysis of non-relapse mortality after CAR T-cell therapy, CRS and ICANS accounted for 4.7% and 5.2% of reported non-relapse deaths, respectively, although infections remained the predominant cause. The review emphasizes that the association between CRS or ICANS and survival should be interpreted in the context of baseline disease burden, inflammatory status, and other adverse patient- and disease-related factors.

Efforts to shift from reactive management to proactive risk prediction have produced a growing set of models. High tumor burden remains one of the most reproducible predictors of severe toxicity, and baseline systemic inflammation reflected by elevated C-reactive protein or ferritin has also been associated with increased risk. Treatment-related factors further modify toxicity profiles: CD28-based CAR T-cell products generally show earlier expansion kinetics and higher rates of severe CRS and ICANS than 4-1BB-based constructs, while bispecific antibodies are more commonly associated with lower-grade CRS during initial step-up dosing and substantially lower rates of severe neurotoxicity. The Endothelial Activation and Stress Index (EASIX), combining lactate dehydrogenase, serum creatinine, and platelet count, and its modified forms have shown moderate ability to predict CAR T-cell–related toxicity, but performance has been inconsistent across cohorts. A recent systematic review and meta-analysis found that most prediction models relied on overlapping clinical and laboratory variables, yet overall discrimination was only modest, particularly for ICANS, and all included models were judged to have a high risk of bias.

More disease- and platform-specific tools are emerging. In the French DESCAR-T registry of patients with large B-cell lymphoma treated with axicabtagene ciloleucel or tisagenlecleucel, a CRS prognostic scoring system incorporated bulky disease, platelet count below 150 × 10⁹/L, C-reactive protein above 30 mg/L, and absent or ineffective bridging therapy, while the ICANS score incorporated female sex, low platelet count, absent or ineffective bridging therapy, and axi-cel use; these outperformed EASIX-based approaches in that setting, though broader validation is needed. A Japanese nationwide registry analysis proposed CRS-PRE, a pre-infusion model based on estimated glomerular filtration rate, ferritin, C-reactive protein, and lactate dehydrogenase, which stratified patients into low-, intermediate-, and high-risk groups for grade ≥ 2 CRS with observed incidences of 2.8%, 26.0%, and 50.0%, respectively. Imaging-based assessment may also contribute: patients with higher CAR-IMPI values, a metabolic prognostic index incorporating metabolic tumor volume, age, and stage, also had higher rates of grade ≥ 3 CRS, high-grade ICANS, and intensive care unit admission. Until such tools are prospectively validated, the review concludes, they should complement rather than replace clinical judgment.

Early detection itself is being augmented by technology. In a prospective study using twice-daily single-channel electroencephalography after CAR T-cell infusion, an automated polymorphic delta activity score identified ICANS with good discrimination and preceded clinical symptoms by a median of 68 hours. A separate pilot study of a rapid neurocognitive testing tool, CART-NS, suggested that brief serial cognitive testing detects subtle neurotoxicity before deterioration of the ICE score, with early Stroop test changes showing promising predictive performance. Neither approach is yet ready to replace ICE-based monitoring, but they illustrate how objective neurophysiologic tools may eventually support earlier detection in high-risk patients. Monitoring intensity should be tailored to platform and baseline risk, with more intensive early observation for patients receiving CD28-costimulated products, those with high tumor burden, elevated inflammatory markers, thrombocytopenia, or renal dysfunction, and with particular vigilance around initial and step-up doses of bispecific antibodies.

On the treatment side, the most informative prospective experience supporting earlier intervention came from the axi-cel safety management cohorts in ZUMA-1. Cohort 4 incorporated levetiracetam prophylaxis and earlier corticosteroid and/or tocilizumab intervention, including treatment of grade 1 CRS if no improvement occurred after three days, and reduced the incidence of severe CRS and neurologic events compared with pivotal cohorts without an apparent adverse effect on CAR T-cell expansion or ongoing response. In cohort 6, prophylactic dexamethasone 10 mg once daily from day 0 through day 2 was associated with no grade ≥ 3 CRS, low rates of severe neurologic events, and high response rates, with longer follow-up continuing to support durable activity. However, the review cautions against indiscriminate corticosteroid exposure: retrospective studies suggest higher cumulative doses and prolonged use are associated with inferior outcomes, though confounding by indication is a major limitation, and time-dependent analyses link greater cumulative exposure to higher infection risk and late hematologic toxicity. Prophylactic tocilizumab in ZUMA-1 cohort 3 reduced high-grade CRS but not high-grade neurologic events, and IL-1 blockade with prophylactic anakinra has yielded mixed results, with a recent real-world weighted analysis showing no reduction in severe ICANS and raising concerns about infectious complications.

For refractory CRS or ICANS, which are uncommon but clinically challenging, the review identifies anakinra and siltuximab as the most clinically developed salvage options. In a multicenter retrospective study of 43 patients treated for refractory CRS or ICANS after CAR T-cell therapy, anakinra was feasible at higher intravenous doses with limited drug-related toxicity, CAR T-cell efficacy appeared preserved with an overall response rate of 77%, and higher-dose anakinra was associated with lower day-28 treatment-related mortality—though the retrospective design and small cohort preclude definitive conclusions. In a multicenter retrospective analysis of 54 patients treated with siltuximab, which neutralizes circulating IL-6 directly rather than blocking its receptor, the CRS grade improved in 75% of patients with CRS and the ICANS grade improved in 60% of those with ICANS. The review stresses that escalation should be phenotype-directed—distinguishing uncontrolled systemic CRS, refractory ICANS, and immune effector cell–associated hemophagocytic lymphohistiocytosis-like syndrome—and that the goal is proportionate intervention: sufficient to control toxicity while preserving antitumor efficacy and minimizing infection, cytopenia, and unnecessary treatment disruption.

Subject of Research: Early detection and personalized management of cytokine release syndrome and ICANS after CAR T-cell therapy and bispecific antibodies

Article Title: CRS and ICANS: Early detection and personalized management

Article References: Kim, J. (2026). CRS and ICANS: Early detection and personalized management. Blood Research, 61(1), Article 51. https://doi.org/10.1007/s44313-026-00167-w

Image Credits: AI Generated

DOI: 10.1007/s44313-026-00167-w

Keywords: cytokine release syndrome, ICANS, CAR T-cell therapy, bispecific antibodies, tocilizumab, anakinra, siltuximab, ICE score, ASTCT grading, risk prediction models, corticosteroids, hematologic malignancies