Immune effector cell-associated neurotoxicity syndrome following CAR T-cell therapy: a review of recent advances

Abstract

Background

Chimeric antigen receptor (CAR) T-cell therapy has transformed the treatment landscape for hematologic malignancies. However, immune effector cell-associated neurotoxicity syndrome (ICANS) remains a significant and potentially lethal complication, affecting approximately 27% to 65% of patients and challenging the therapeutic risk-benefit profile.

Main body

This review synthesizes recent advances in the epidemiology, pathophysiology, diagnosis, and management of ICANS. Incidence varies significantly by product design, with anti-CD19 therapies and CD28-containing constructs demonstrating markedly higher toxicity rates compared to other targets and 4-1BB-based designs. The pathophysiological mechanism centers on blood-brain barrier disruption driven by systemic cytokine release and direct cellular injury. Monocytes and macrophages act as principal effectors, releasing interleukin-1 and granulocyte-macrophage colony-stimulating factor, which trigger endothelial activation and neuroinflammation. Clinical manifestations typically appear within the first week post-infusion, ranging from mild language disturbances to life-threatening cerebral edema. Current management has evolved from reactive symptom control to proactive strategies. Severity-based algorithms guide the use of corticosteroids and intensive care support, while emerging prophylactic approaches, particularly interleukin-1 receptor blockade with anakinra, show promise in reducing severe neurotoxicity without compromising anti-tumor efficacy. Furthermore, diagnostic precision is improving through the use of novel biomarkers, such as chimeric antigen receptor-positive extracellular vesicles, and machine-learning models that predict toxicity days before symptom onset.

Conclusions

The management of ICANS is shifting towards a precision medicine paradigm. By integrating predictive biomarkers, artificial intelligence, and novel prophylactic interventions, clinicians can better stratify risk and implement early treatments. Future research focusing on next-generation constructs with engineered safety features will be essential to decouple therapeutic efficacy from neurotoxicity, ultimately optimizing outcomes for patients with advanced hematologic cancers.

Introduction

Although Chimeric antigen receptor (CAR) T-cell therapy has transformed the management of hematological malignancies, important toxicities complicate the success story. One such toxicity that stands out as potentially lethal and challenges the risk-benefit equation of CAR T-cell therapy in its entirety is immune effector cell-associated neurotoxicity syndrome (ICANS). It affects 27–65% of patients who are treated with CAR T-cell therapy; these wide ranges exist because of inter-study variations in patient cohorts and variable criteria definitions [1–3].

The symptoms occur 4–6 days after the infusion process and display signs of encephalopathy, such as disorders of language, frontal lobes, apraxia, and disorientation, which last up to 8 days [3]. Although most patients recover fully from neurological deficits after proper management, the condition poses a significant economic burden due to its costs of management, which range from US$1,930 in grade 1 cases to US$177,343 in cases accompanied by cytokine release syndrome [4].

This review synthesizes current understanding of ICANS pathophysiology, clinical management, and emerging therapeutic strategies, highlighting areas where evidence remains incomplete and future research directions are most urgently needed.

Methods

This narrative review was conducted to synthesize recent advances in the understanding and management of immune effector cell-associated neurotoxicity syndrome (ICANS) following CAR T-cell therapy. A comprehensive literature search was conducted across the PubMed, Web of Science, and Google Scholar databases from January 2020 through March 2025 to capture the most recent developments in this rapidly evolving field.

The primary search strategy employed combinations of the following key terms: “CAR T-cell,” “chimeric antigen receptor,” “ICANS,” “immune effector cell-associated neurotoxicity syndrome,” “neurotoxicity,” “cytokine release syndrome,” “blood-brain barrier,” “CD19,” “BCMA,” and “hematologic malignancies.” Additional targeted searches were conducted for specific subtopics, including “CAR T biomarkers,” “anakinra prophylaxis,” “tocilizumab neurotoxicity,” “EEG CAR T,” and “machine learning CAR T prediction.”

Studies were included if they reported on ICANS epidemiology, pathophysiology, risk factors, clinical presentation, diagnostic approaches, or management strategies in patients receiving CAR T-cell therapy for hematologic malignancies. We included data from prospective clinical trials, retrospective cohort studies, real-world analyses, and mechanistic investigations. Both adult and pediatric populations were considered to provide comprehensive coverage of ICANS across age groups. Preclinical studies were included when they offered mechanistic insights relevant to clinical observations or informed emerging therapeutic strategies.

Studies focusing exclusively on CAR T-cell therapy for solid tumors were excluded, as ICANS patterns and incidence in this context remain poorly characterized and may differ substantially from hematologic malignancy settings. Case reports were generally excluded unless they described novel clinical phenotypes or provided unique mechanistic insights. Non-English language publications were excluded. Reference lists of identified articles and relevant review papers were manually screened to identify additional pertinent studies.

Epidemiology and risk stratification

Incidence patterns and product-specific differences

A comprehensive meta-analysis of 75 trials revealed overall ICANS incidence rates of 26.9% for all-grade and 10.5% for high-grade cases across hematologic malignancies [2]. However, this aggregate statistic masks significant variation across CAR construct designs and target antigens. Anti-CD19 therapies demonstrate markedly higher neurotoxicity rates compared to other targets, with a 4.6-fold increase in high-grade ICANS risk versus anti-BCMA products.

The most striking differences are between specific CD19-targeted products. Recent meta-analyses show axicabtagene ciloleucel demonstrates substantially higher ICANS rates (approximately 42% all-grade, 20–28% high-grade) compared to tisagenlecleucel (approximately 16–20% all-grade, 4–10% high-grade), with axicabtagene ciloleucel associated with significantly higher odds of severe ICANS grade ≥ 3 (OR, 4.03; P < 0.001) [5]. These differences persist even after controlling for patient characteristics, suggesting that CAR design features, particularly co-stimulatory domains, fundamentally influence neurotoxicity risk [6]. Notably, these comparisons should be interpreted with caution. The cited meta-analyses primarily included patients with large B-cell lymphoma, and the odds ratio for severe ICANS (OR 4.03; P < 0.001) was derived from pooled analyses that may not have consistently controlled for all confounders [5]. Furthermore, reported incidence rates vary substantially depending on whether studies employed Common Terminology Criteria for Adverse Events (CTCAE), American Society for Transplantation and Cellular Therapy (ASTCT) consensus, or institution-specific grading criteria; comparative analyses demonstrate that when evaluating toxicity grade-by-grade, only 55% of patients received concordant ICANS grades across different grading systems, and CTCAE criteria identified 45% of patients as having neurotoxicity compared to only 17% with ASTCT-based assessments in one retrospective analysis [7, 8]. Beyond co-stimulatory domain design, observed product differences likely reflect confounding variables, including approved indications, baseline tumor burden, prior treatment exposure, lymphodepletion regimens, and variability in institutional experience and supportive care practices [9, 10].

The critical role of co-stimulatory domain selection

The choice between CD28 and 4-1BB co-stimulatory domains represents one of the most consequential design decisions in CAR engineering. CD28-based constructs promote preferentially glycolytic metabolism, supporting rapid effector responses, while 4-1BB constructs preserve mitochondrial fitness and promote memory-like differentiation [11]. These metabolic differences translate into clinically meaningful outcomes: the mitochondrial fitness and memory-like differentiation promoted by 4-1BB co-stimulation support sustained anti-tumor activity and reduced T-cell exhaustion [11], resulting in superior survival benefits (median OS: 15.0 vs. 8.5 months; median PFS: 7.0 vs. 3.0 months) and significantly lower neurotoxicity rates (31.11% vs. 60.28%) compared to CD28-containing constructs [12]. Mechanistically, CD28 activates the PI3K-Akt and MAPK pathways to drive rapid glycolytic metabolism and effector memory differentiation, whereas 4-1BB signals through TRAF adaptor recruitment and NF-κB activation to promote oxidative metabolism, fatty acid oxidation, and central memory phenotype maintenance [13]. The accelerated expansion kinetics and stronger cytokine production characteristic of CD28 co-stimulation likely contribute to its higher neurotoxicity profile, as CAR T-cell proliferation itself generates cytokine surges that can trigger blood-brain barrier disruption and neuroinflammation [14, 15]. Furthermore, CD28-based constructs induce Th2 polarizing genes and secrete higher levels of Th2 cytokines (IL-4, IL-5), whereas 4-1BB-based constructs demonstrate Th1 polarization with minimal Th2 cytokine elevation, a difference that may contribute to their differential toxicity profiles [16].

Emerging next-generation designs offer hope for further reductions in toxicity. A phase 1 trial of third-generation CD19 CAR T-cells combining CD28 and TLR2 domains demonstrated no ICANS of any grade while maintaining efficacy, suggesting that novel co-stimulatory combinations may decouple therapeutic benefit from neurotoxic risk [12] (Table 1).

Table 1.

ICANS incidence by CAR T-cell product

Product	Target	Co-stimulatory Domain	All-Grade ICANS	High-Grade ICANS	Notes	Ref.
Overall (meta-analysis)	Various	Mixed	26.9%	10.5%	Across 75 trials	[2]
Axicabtagene ciloleucel	CD19	CD28	42%	20–28%	Significantly higher risk	[5]
Tisagenlecleucel	CD19	4-1BB	16–20%	4–10%	Lower toxicity profile	[5]
Anti-CD19 products	CD19	Mixed	Higher	4.6x increase vs. BCMA	Target-dependent risk	[2]
Anti-BCMA products	BCMA	Mixed	Lower	Baseline reference	Lower neurotoxicity	[2]
Third-generation (CD28 + TLR2)	CD19	CD28 + TLR2	0%	0%	Phase 1 trial results	[17]

Key Finding: Axicabtagene ciloleucel shows significantly higher odds of severe ICANS ≥ 3 (OR 4.03, P < 0.001) compared to tisagenlecleucel

Risk factors and predictive models

Beyond product-specific factors, several patient and treatment variables influence ICANS risk. High tumor burden emerges as a consistent risk factor, with successful tumor burden reduction through enhanced bridging therapy reducing neurotoxicity incidence from 5% to < 1% in multiple myeloma patients [18]. However, the relationship between baseline characteristics and ICANS development remains complex, with some studies finding no significant differences in baseline parameters between patients who develop ICANS versus those who do not [1].

Sophisticated predictive models incorporating multiple biomarkers show promise for risk stratification. Amidi et al. [10] developed a forecasting model combining clinical predictors, including maximum daily temperature, C-reactive protein, IL-6, and procalcitonin, achieving 96.7% accuracy for ICANS prediction five days in advance. Similarly, recent multivariable models combining cytokines with clinical variables achieved AUCs of 0.83 (any-grade) and 0.80 (grade 2–4) ICANS prediction [19]. However, it should be noted that current evidence linking clonal hematopoiesis to ICANS risk derives from small retrospective analyses and is insufficient to inform clinical treatment decisions or patient selection (Table 2).

Table 2.

Risk factors and predictive models

Category	Factor	Impact	Ref.
Product Design	CD28 vs. 4-1BB domain	CD28 → higher risk	[11, 12]
	Anti-CD19 vs. other targets	CD19 → 4.6x higher	[2]
Patient Characteristics	Age > 65 years	OR 1.77 (95% CI 1.39–2.26)	[20]
	High tumor burden	Increased risk	[18]
Genetic Factors	Clonal hematopoiesis (DNMT3A, TET2, ASXL1)	58.9% vs. 25% for grade ≥ 3	[21, 22]
	Specific mutations (PPM1D, DNMT3A, TET2, ASXL1)	Associated with onset	[21, 22]
Predictive Model	Multi-biomarker model (temp, CRP, IL-6, PCT)	96.7% accuracy 5 days advance	[10]
	ML cytokine profiles	AUC 0.83 (any-grade ICANS); AUC 0.80 (grade 2–4 ICANS)	[19]

Key Finding: Clonal hematopoiesis (DNMT3A, TET2, ASXL1 mutations) is associated with a dramatically increased risk (58.9% vs. 25% for grade ≥ 3 ICANS). Multi-biomarker predictive models achieve 96.7% accuracy in forecasting ICANS 5 days in advance, enabling proactive intervention

Pathophysiology: unraveling the neuroinflammatory cascade

Blood-brain barrier disruption as the central event

ICANS’ pathophysiology centers on disruption of the blood-brain barrier (BBB) through multiple converging mechanisms. Monocytes and macrophages serve as the principal effectors in ICANS pathogenesis, producing the majority of IL-1 and IL-6 following CAR T-cell activation [23, 24]. These myeloid-derived cytokines drive endothelial activation and decreased tight junction molecule expression, leading to blood-brain barrier disruption [25]. While CD19-expressing pericytes within the BBB represent a plausible off-tumor target based on single-cell sequencing studies, the relative contribution of direct pericyte targeting versus cytokine-mediated injury remains debated and requires further investigation [9]. Critically, preclinical studies demonstrate that IL-1 blockade with anakinra abolishes both CRS and neurotoxicity, whereas tocilizumab (IL-6 receptor antagonist) prevents CRS but fails to protect against delayed lethal neurotoxicity, underscoring the primacy of monocyte-derived IL-1 in ICANS pathogenesis [23, 26].

The complement system plays an increasingly recognized role in this process. Elevated post-CAR T-cell terminal complement complex (sC5b-9) levels associate significantly with clinically significant ICANS, with multivariable analysis showing increased risk per 100-unit increase in sC5b-9 levels [27, 28]. This complement-coagulation interaction creates a thromboinflammation state that exacerbates BBB disruption and systemic inflammatory responses [27].

Cytokine-mediated neuroinflammation

The cytokine landscape in ICANS differs meaningfully from that seen in cytokine release syndrome, despite superficial similarities. While IL-6 is prominently elevated in both conditions, ICANS exhibits distinct patterns of GM-CSF and IFN-γ elevation, leading to hyperactivation of innate myeloid cells and their infiltration into the CNS [29]. GM-CSF appears particularly critical, with CAR T cells significantly upregulating GM-CSF receptors upon antigen-specific stimulation, from < 10% to > 30% [30].

Interestingly, the relationship between cytokine levels and clinical symptoms proves complex. Children developing severe motor deficits like quadriparesis paradoxically showed lower levels of pro-inflammatory cytokines, including IFN-γ, compared to those without these symptoms [31]. These findings challenge simplistic models of cytokine-mediated toxicity and reveal that ICANS pathogenesis involves multiple parallel and interconnected pathways beyond inflammatory cytokine elevation alone.

Multiple pathophysiological pathways in ICANS

Recent mechanistic studies have identified at least six distinct but interconnected pathways contributing to ICANS pathogenesis:

1. Endothelial Activation and Blood-Brain Barrier Disruption. Systemically elevated cytokines activate brain endothelial cells to release Weibel-Palade bodies containing angiopoietin-2 and von Willebrand factor, disrupting the angiopoietin-Tie2 axis that typically maintains endothelial quiescence and tight junction integrity [32]. Biomarker studies demonstrate that elevated neurofilament light chain and glial fibrillary acidic protein correlate with endothelial activation markers, including von Willebrand factor and soluble ST2, confirming the link between endothelial dysfunction and blood-brain barrier impairment [33].

2. Pericyte Dysfunction and Loss. Cytokine exposure causes pericyte stress with consequent VEGF and IL-6 release that further activates endothelial cells in a positive feedback loop [32], while CD19-expressing pericytes may be directly targeted and depleted by CD19 CAR T-cells, compromising neurovascular unit integrity through on-target, off-tumor toxicity [32, 34].

3. Astrocyte Injury and Dysfunction. Cytokine-stimulated astrocytes undergo cellular swelling, causing abnormal osmotic forces and cerebral edema, while simultaneously producing VEGF-A that exacerbates blood-brain barrier disruption and endfoot process detachment [24, 32].

4. Direct Neuronal Damage. Elevated serum levels of neurofilament light chain, a marker of axonal injury, correlate with ICANS severity, indicating direct neuronal damage distinct from inflammatory mechanisms [33], while infiltrating macrophages and activated microglia produce quinolinic acid and glutamate that activate NMDA receptors, inducing excitotoxicity and seizures [32]. Cerebrospinal fluid and serum analyses in patients with ICANS have revealed elevations in markers of astroglial injury and excitotoxicity. Glial fibrillary acidic protein and neurofilament light chain, markers of astrocyte activation and axonal damage, respectively, are elevated in both pediatric and adult patients during ICANS and correlate with neurotoxicity severity [35]. Significantly, these biomarkers also correlate with markers of endothelial activation, including von Willebrand factor and soluble ST2, confirming the mechanistic link between endothelial dysfunction and neural injury [33]. Infiltrating macrophages and activated microglia produce excitotoxic mediators, including quinolinic acid and glutamate, which activate N-methyl-D-aspartate (NMDA) receptors and may contribute to seizure activity and neuronal injury [36].

5. Thrombotic Microangiopathy. Tumor lysis products, including HMGB1, histone H3, and ATP, activate and injure endothelial cells, exposing tissue factor and collagen that trigger coagulation cascades, with inadequate von Willebrand factor cleavage potentially causing microvascular thrombosis and consumptive coagulopathy [24].

6. Microglial Activation Pathways. Microglial activation via TGF-β-activated kinase-1 (TAK1) and p38 MAPK signaling drives TNF-α and GM-CSF production, and pharmacological TAK1 inhibition reduces microglial activation and improves cognitive function in preclinical ICANS models [37]. Recent clinical data also implicate complement activation, with elevated post-CAR T terminal complement complex (C5b-9) levels associated with clinically significant ICANS, suggesting complement-mediated endothelial injury as an additional contributory pathway [38].

These pathways do not operate in isolation but rather form an interconnected network where dysfunction in one component amplifies damage to others, ultimately manifesting as the heterogeneous clinical presentations observed in ICANS (Fig. 1).

Fig. 1.

Pathophysiologic cascade of ICANS following CAR T-Cell therapy. (1) Antigen recognition triggers T-cell proliferation, amplified by high tumor burden. (2) Blood-brain barrier (BBB) disruption occurs through dual mechanisms: direct CD19 + pericyte targeting and systemic cytokine release (IL-6, GM-CSF, IFN-γ), causing endothelial activation. (3) BBB breakdown allows peripheral inflammatory mediators to enter the central nervous system. (4) The resulting neuroinflammation manifests as clinical ICANS with characteristic features including language disturbances, encephalopathy, apraxia, disorientation, and frontal lobe dysfunction

Genetic susceptibility and individual variation

Emerging evidence suggests genetic factors influence ICANS susceptibility, though research remains limited. Specific mutations in PPM1D, DNMT3A, TET2, and ASXL1 genes associate with ICANS onset, while clonal hematopoiesis, particularly mutations in DNMT3A, TET2, and ASXL1, significantly increases grade ≥ 3 ICANS risk (58.9% vs. 25%) [21, 22]. These findings suggest that pre-existing hematopoietic abnormalities may predispose to severe neurotoxicity, potentially through altered cytokine signaling regulation.

Clinical presentation and diagnostic approaches

Contemporary grading frameworks

The American Society for Transplantation and Cellular Therapy (ASTCT) developed consensus grading criteria in 2019 to harmonize ICANS assessment across institutions and clinical trials [39]. The ASTCT framework uses the Immune effector Cell-associated Encephalopathy (ICE) score, a 10-point bedside assessment tool that quantifies cognitive impairment across five domains: orientation to year/month/city/hospital (4 points), naming three objects (3 points), following simple commands (1 point), writing a standard sentence (1 point), and counting backwards from 100 by 10 (1 point) [39].

ICANS severity is determined by the most severe finding across multiple domains, including ICE score, level of consciousness, motor symptoms, seizures, and signs of raised intracranial pressure. Grade 1 corresponds to ICE scores of 7–9 with spontaneous awakening; Grade 2 to ICE 3–6 with awakening to voice; Grade 3 to ICE 0–2 (or awake with global aphasia) with awakening only to tactile stimulus; and Grade 4 to patients who are unrousable or require vigorous or repetitive tactile stimuli to arouse [39]. Critically, the ICE score emphasizes quantifying language and writing deficits, distinguishing it from Common Terminology Criteria for Adverse Events (CTCAE) grading, which focuses more broadly on activities of daily living impairment and may identify substantially different patient populations [7, 8, 39].

In pediatric populations (typically < 12 years), the Cornell Assessment of Pediatric Delirium (CAPD) serves as the recommended screening tool [40, 41]. CAPD is an 8-item observational instrument with scores ranging from 0 to 32, where scores ≥ 9 indicate possible delirium in typically developing children [41, 42]. The tool demonstrates high sensitivity (81.8%) but modest specificity (44.8%) in mechanically ventilated children, with performance varying by developmental status [42]. Importantly, CAPD functions as a screening tool rather than a diagnostic instrument; its high sensitivity enables early detection of at-risk patients who require closer monitoring and formal evaluation, though positive screens must be confirmed through comprehensive clinical assessment [41, 42]. For children with developmental delay, higher cutoff scores (≥ 17) improve diagnostic accuracy [42].

Evolving clinical spectrum

ICANS presents with heterogeneous manifestations ranging from mild cognitive symptoms to life-threatening cerebral edema. The classical presentation includes apraxia, expressive aphasia, disorientation, and hallucinations, typically developing 4–6 days post-infusion [1]. However, 10% of patients experience delayed onset after three weeks, highlighting the need for extended monitoring protocols [43].

Age appears to be a significant risk factor for ICANS development, with patients over 65 years old showing an increased incidence compared to younger adults (OR 1.77; 95% CI 1.39–2.26 for any grade ICANS). However, while chronological age influences initial ICANS risk, real-world outcomes suggest that advanced age alone does not preclude successful CAR-T treatment or significantly impact overall treatment outcomes beyond the increased initial susceptibility to neurotoxicity [20]. In pediatric populations, changes in the Cornell Assessment for Pediatric Delirium score may serve as early indicators, occurring 24–72 h before formal ICANS diagnosis [40].

Advanced diagnostic approaches

Neuroimaging findings correlate strongly with ICANS severity, with abnormalities present in only 7% of patients without ICANS but 100% of those with grade 4 disease [44]. Common patterns include signal changes in the thalami, external capsule, brainstem, and subcortical white matter, often categorized as ischemic changes, hemorrhages, or cerebral edema [45, 46].

Electroencephalography emerges as a particularly valuable diagnostic tool. The Visual EEG-ICANS (VE-ICANS) grading scale demonstrates strong correlation with ICANS severity (R = 0.58) and excellent discrimination (AUC = 0.91 for ICANS ≥ 2) [47]. Frontal intermittent rhythmic delta activity shows 88% sensitivity for ICANS detection and resolves with neurotoxicity improvement, offering an objective biomarker for monitoring [48].

Biomarker development

Several promising biomarkers enable earlier detection and severity assessment. Pre-infusion plasma neurofilament light chain levels predict ICANS development, while matrix metalloproteinase-9 and glial fibrillary acidic protein serve as markers of BBB disruption and brain injury [20, 49]. Most notably, CAR + extracellular vesicles represent immediate predictors of ICANS, with specific concentration thresholds predicting ICANS development four days in advance with superior sensitivity and specificity compared to traditional markers [50]. However, these biomarkers remain largely exploratory, derived predominantly from single-center studies with cohorts ranging from 11 to 199 patients [20, 49, 50]. Current evidence lacks prospective multicenter validation, and no biomarker-guided intervention trials have demonstrated clinical utility sufficient for routine implementation [20]. The transition from discovery to clinical translation requires standardized assay platforms, validation across diverse patient populations and CAR constructs, and demonstration that biomarker-driven interventions improve patient outcomes [36] (Table 3).

Table 3.

Diagnostic biomarkers for ICANS

Biomarker	Timing	Sensitivity/Specificity	Clinical Utility	Ref.
CAR + Extracellular Vesicles	4 days pre-ICANS	Superior to traditional markers	Immediate predictor	[50]
Neurofilament Light Chain	Pre-infusion	Predictive	Early risk stratification	[20]
MMP-9	During ICANS	Marker of BBB disruption	Severity assessment	[20, 49]
GFAP	During ICANS	Marker of brain injury	Severity assessment	[20, 49]
Terminal Complement Complex (sC5b-9)	Post-CAR T	Significant association	Risk per 100-unit increase	[27, 28]
EEG (VE-ICANS scale)	Real-time	AUC 0.91 for ICANS ≥ 2	Objective monitoring	[47]
Frontal Intermittent Rhythmic Delta	During ICANS	88% sensitivity	Diagnostic and monitoring	[48]
Cornell Assessment (Pediatric)	24–72 h pre-diagnosis	Early indicator in children	Pediatric screening	[40]

Key Finding: CAR + extracellular vesicles represent the most promising early predictor, detecting ICANS 4 days pre-onset with superior sensitivity/specificity compared to traditional biomarkers. EEG monitoring using the VE-ICANS grading scale demonstrates excellent discriminatory ability (AUC 0.91 for ICANS ≥ 2), providing an objective real-time assessment for clinical decision-making

Current management paradigms

Severity-based treatment algorithms

Current ICANS management follows established severity-based algorithms developed through consensus guidelines. Corticosteroids represent the standard initial therapy for ICANS, with evidence-based dosing according to severity grade [26, 51]. Tocilizumab, while effective for concurrent cytokine release syndrome, is used primarily for CRS management rather than ICANS itself, as IL-6 receptor blockade has limited efficacy for neurotoxicity due to insufficient penetration of the blood-brain barrier and failure to address IL-1-mediated CNS inflammation [23, 26, 52]. Grade 1 disease requires only supportive care and close monitoring, while grade 2 disease warrants corticosteroid therapy (dexamethasone 10–20 mg intravenously every 6 h). Grade 3 ICANS uses the same dexamethasone regimen, while grade 4 necessitates intensive care unit management with high-dose methylprednisolone (1 g IV daily for ≥ 3 days) and consideration of additional immunosuppressive agents [26].

The timing of corticosteroid intervention has evolved toward earlier, more proactive approaches. Early corticosteroid administration with tocilizumab for low-grade cytokine release syndrome significantly reduced high-grade complications without worsening ICANS incidence or compromising therapeutic efficacy [53]. This paradigm shift from reactive to proactive management reflects growing confidence that early intervention preserves CAR T-cell function while preventing severe toxicity.

Supportive care considerations

Management of severe ICANS requires sophisticated intensive care approaches addressing multiple organ systems. Seizure activity occurs in 7.1% of CAR T-cell patients, with 75% being nonconvulsive, supporting routine continuous EEG monitoring in moderate-to-severe cases [54]. Fluid management and electrolyte balance require careful attention, as abnormalities can prolong mechanical ventilation and increase mortality in critically ill patients [55].

The economic implications of intensive ICANS management are substantial, with ICU stays representing key cost drivers. Approximately one-third of CAR T-cell patients require ICU admission (27–34%) [56], where ICANS represents a significant indication alongside cytokine release syndrome [57]. After excluding CAR T-cell acquisition costs, inpatient care, including ICU hospitalizations, accounts for 57.5% of remaining costs, with median total hospitalization costs exceeding US$300,000-US$400,000 [58]. This reality necessitates specialized multidisciplinary teams incorporating oncology, neurology, and intensive care expertise [59] (Fig. 2).

Fig. 2.

Evidence-based ICANS treatment algorithm. Management strategy based on ASTCT grading criteria. Left panel: cytokine-guided medication selection distinguishing IL-6-driven pathology (tocilizumab for CRS) from IL-1-driven cases (anakinra). Right panel: risk-stratified approach from prophylaxis through ASTCT grade-specific interventions (Grade 1: supportive care; Grade 2: dexamethasone with EEG monitoring; Grade 3: ICU admission with methylprednisolone; Grade 4: intensive immunosuppression with consideration of anakinra, GM-CSF neutralization, or safety switch activation). Response assessment every 12–24 h guides escalation to second-line (anakinra) or third-line (intrathecal) therapy for refractory cases

Emerging therapeutic strategies

Prophylactic interventions

The paradigm shift toward prophylaxis represents one of the most promising developments in ICANS management. Anakinra, an IL-1 receptor antagonist, has shown particular promise in clinical trials. Key evidence derives predominantly from single-arm phase 1–2 studies in relapsed/refractory B-cell lymphomas, with cohort sizes ranging from 31 to 60 patients [60, 61]. Prophylactic subcutaneous anakinra administered from day 2 to at least day 10 post-CAR infusion reduced severe ICANS rates to 9.7% while maintaining 77% overall response rates in these initial studies [60]. Similar findings have been reported across predominantly lymphoma populations receiving CD19-directed CAR T-cell products, with grade ≥ 3 ICANS rates of approximately 10% [2, 61, 62].

However, prophylactic strategies require careful validation. Current evidence base consists primarily of non-randomized, single-arm cohort studies without direct comparative arms, and data remain limited for other disease indications, including acute lymphoblastic leukemia, multiple myeloma, and non-CD19 targeted CAR constructs [61, 63]. Multicenter randomized controlled trials are needed to validate efficacy and safety across diverse CAR products, disease types, and patient populations before anakinra prophylaxis can be considered standard of care [64]. The relationship between inflammation and therapeutic efficacy complicates prophylactic approaches, as CAR T-cell activation drives both therapeutic response and adverse events [65]. Successful prophylaxis must thread the needle between toxicity prevention and efficacy preservation.

Novel therapeutic targets

Beyond traditional immunosuppression, targeted approaches addressing specific pathophysiologic mechanisms show promise. GM-CSF neutralization remains largely in preclinical and early clinical investigation and is not currently recommended for routine clinical practice outside of research protocols [30, 66]. Nonetheless, preclinical data demonstrate that GM-CSF neutralization prevents CAR-induced toxicities while improving efficacy by reducing myeloid activation, and additionally enhances CAR T-cell proliferation and antitumor activity through direct effects on the CAR T-cells themselves [30]. This dichotomy highlights the complexity of cytokine networks in CAR T-cell biology.

Complement inhibition represents another rational target given evidence of complement activation in ICANS pathogenesis. However, complement-targeted interventions remain in early exploratory stages and are not recommended for routine clinical use [27, 67]. Early studies suggest complement pathway modulation might prevent BBB disruption and subsequent neuroinflammation, though clinical validation remains preliminary and requires prospective interventional trials [27].

Engineering solutions

Next-generation CAR designs incorporate intrinsic safety features to prevent toxicity, though these engineered solutions remain primarily developmental and are not yet available for routine clinical use outside of specialized clinical trials [68, 69]. Inducible safety switches, such as the inducible Caspase 9 system, provide definitive ‘circuit breakers’ for life-threatening, refractory toxicity in CAR T-cell therapy [68]. Similarly, “tunable” CAR systems using synthetic biology approaches enable real-time activity modulation through external small molecules [70].

Base editing and prime editing technologies enhance safety by enabling precise genetic modifications without double-strand breaks, facilitating the development of safer allogeneic products with reduced toxicity profiles [71]. The Prime-Assisted Site-Specific Integrase Gene Editing (PASSIGE) platform represents a novel non-viral approach to multiplex CAR T-cell engineering. PASSIGE leverages prime editing (a CRISPR-based technology that enables precise genomic modifications without inducing double-strand breaks) to simultaneously knock out multiple genes and integrate large transgenes at targeted genomic loci. In preclinical studies, this platform achieved > 90% knockout efficiency of the endogenous T cell receptor alpha constant (TRAC) and beta-2-microglobulin (B2M) loci in human T cells through multiplex prime editing. In comparison, co-delivery of a non-viral DNA donor template with PASSIGE components enabled targeted integration of a 3.5 kb CD19-CAR expression cassette at the TRAC locus in > 60% of cells [72]. Critically, this one-step, modular platform achieved these editing efficiencies without impacting T cell viability, phenotype, or functionality, and avoided the safety concerns associated with DSB-inducing nucleases, including p53 activation, chromothripsis, and gross chromosomal rearrangements that can occur with multiplex editing using conventional approaches. By eliminating the need for viral vectors and double-strand breaks, PASSIGE potentially reduces manufacturing complexity, cost, and toxicity risks while enabling flexible generation of allogeneic off-the-shelf CAR T-cell products [72].

Special considerations and future applications

CNS-directed therapy

As CAR T-cell therapy expands into CNS malignancies, understanding neurotoxicity in this unique context becomes critical. Patients with pre-existing CNS involvement were historically excluded from trials, but emerging evidence suggests that while ICANS incidence may be elevated, severity does not appear significantly impacted by pre-existing CNS disease [73]. This finding challenges the notion of absolute contraindications and supports carefully monitored CAR T-cell therapy in CNS lymphomas.

Intrathecal and intraventricular delivery routes show promise for CNS malignancies, potentially offering superior CNS penetration with reduced systemic toxicity. Studies demonstrate that intrathecal CAR T-cell infusion achieves comparable safety profiles to intravenous delivery while potentially reducing CNS relapse rates [74].

Long-term sequelae and monitoring

While most patients achieve complete neurological recovery, emerging evidence suggests potential long-term complications that may require extended surveillance. Rare cases of late Parkinsonism-like movement disorders have been reported following severe ICANS, and 27% of patients develop prolonged neurologic symptoms lasting over four weeks [75]. These findings underscore the need for comprehensive long-term neurological monitoring protocols.

CAR T-cell persistence may contribute to prolonged symptoms, with persistent central memory CD4 + CAR T-cells detectable one year after infusion, coinciding with recurrent late-onset inflammatory toxicities [76]. This observation suggests that CAR T-cell kinetics and persistence patterns may predict long-term neurological outcomes.

Critical analysis and future directions

Gaps in current understanding

Despite significant advances, several critical gaps limit optimal ICANS management. The relationship between inflammation and therapeutic efficacy remains incompletely understood, complicating the development of prophylactic strategies. While early intervention appears beneficial, the optimal timing, intensity, and duration of immunosuppressive therapy remain unclear.

Biomarker validation represents another critical need. While numerous promising candidates have emerged, few have undergone rigorous validation in independent cohorts or proven clinically actionable. The field requires biomarker-guided treatment algorithms that can reliably identify high-risk patients and guide personalized intervention strategies.

Machine learning and precision medicine

Artificial intelligence and machine learning approaches show transformative potential for ICANS risk assessment and early intervention. A pioneering combined hidden Markov model and lasso-penalized logistic regression approach developed using data from 199 CAR T-cell patients achieved area under the curve values of 96.7% and 93.2% for predicting ICANS and severe ICANS, respectively, when forecasting 5 days, with the model providing well-calibrated probabilities of ICANS onset and accurately predicting the time course 1–7 days in advance [10]. Recent applications of gradient-boosted classification trees have enabled early prediction of severe ICANS after standard-of-care CD19 CAR T-cell therapy [77], while imaging-based prognostic models integrating pre-infusion PET/CT radiomics features with clinical variables demonstrate improved accuracy for predicting neurotoxicity occurrence [78].

AI applications in CAR T-cell therapy extend beyond toxicity prediction to encompass CAR design optimization, dose and schedule optimization through reinforcement learning, morphological and phenotypic analysis of CAR cells using computer vision, and extraction of clinical insights from literature using natural language processing [79]. Synthetic data generated by artificial intelligence trained on clinical trial datasets enables privacy-preserving analysis for predicting severe cytokine release syndrome and analyzing co-occurrence patterns of CRS and ICANS, facilitating insights that would otherwise be inaccessible due to data privacy constraints [64].

The integration of multi-omics data, including genomics, transcriptomics, proteomics, and metabolomics, combined with machine learning algorithms, may enable truly personalized risk assessment. Current biomarker research emphasizes the need for integrative models that account for heterogeneous immune traits, endothelial activation markers, blood-brain barrier disruption indicators, and neuronal injury biomarkers [36]. However, such approaches require substantial computational resources, standardized data collection protocols across institutions, and rigorous external validation in diverse patient populations and real-world clinical settings before implementation.

Regulatory and implementation challenges

The economic burden of ICANS management poses significant healthcare system challenges, with total adverse event management costs contributing US$40,368-US$47,270 annually per patient [80]. These costs, combined with manufacturing complexities and regulatory requirements, create access disparities that may limit equitable distribution of CAR T-cell therapy.

Future research must address these implementation challenges alongside scientific advances. Value-based care models, manufacturing optimization, and regulatory pathway streamlining will prove as critical as scientific breakthroughs for realizing CAR T-cell therapy’s full potential.

Conclusions

ICANS represents a paradigmatic challenge in modern cancer therapy: a potentially life-threatening toxicity that paradoxically accompanies therapeutic breakthrough. Current evidence supports a complex, multifactorial pathophysiology involving BBB disruption, cytokine-mediated neuroinflammation, and complement activation, with significant variation based on CAR design features and patient characteristics.

Management approaches have evolved from reactive to increasingly proactive strategies, with early intervention protocols showing promise for preventing toxicity without compromising efficacy. Prophylactic approaches, particularly IL-1 receptor antagonism, represent transformative advances, though optimal implementation strategies require further validation.

The future lies in precision medicine approaches that integrate predictive biomarkers, machine learning algorithms, and engineered safety features to enable individualized risk assessment and intervention. Success will require continued collaboration between basic scientists, clinicians, engineers, and regulatory authorities to navigate the complex balance between therapeutic efficacy and patient safety.

As CAR T-cell therapy expands into new diseases and patient populations, the lessons learned from ICANS management will prove invaluable for optimizing this revolutionary therapeutic modality. The challenge ahead is not merely to prevent ICANS, but to do so in ways that preserve and enhance the remarkable therapeutic potential of CAR T-cell therapy for patients with otherwise incurable malignancies.

Author contributions

Mohammad Amin Ansarian: Literature search and selection, data extraction and analysis, writing, original draft preparation, visualization, project administration. Mahsa Fatahichegeni: Literature search and selection, data extraction and analysis, writing—original draft preparation, visualization. Yuqi Wang: Literature search, data extraction, writing, review, and editing. Tongxin Zhang: Literature search, data extraction, writing, review, and editing. Juan Ren: Data extraction, writing, review, and editing. Xiaoning Wang: Conceptualization, supervision, writing, review and editing, funding acquisition. All authors have read and agreed to the published version of the manuscript.

Funding

National Key R&D Program, Grant/Award Number: 2022YFC2502700.

Data availability

The data generated in this study are available within the article and its supplementary data files.

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing financial interests.