This study followed the PRISMA guidelines (14) strictly and adopted established literature searching processes for systematic reviews (15), though it was not registered. We searched literature published between January 1, 2018, and November 8, 2024, as this period followed the significant increase in research following the FDA approval of Kymriah for relapsed/refractory B-ALL treatment in 2017 (16), which marked the formal entry of CAR-T cell therapy into clinical practice. Research published since 2018 has tracked the progression of CAR-T cell therapy from clinical trials to routine practice (17). Subsequent studies have systematically examined treatment optimization, safety protocols, and long-term outcomes, reporting key metrics like complete remission and minimal residual disease negativity rates to assess contemporary B-ALL management (18).

Our literature search employed three databases: PubMed, Web of Science, and Embase. The search strategy combined Medical Subject Headings (MeSH) terms and free-text keywords, focusing on four core concepts: CAR-T therapy (“CAR-T OR CAR T OR chimeric antigen receptor T”), biomarkers (“biomarker* OR predict* OR prognosis*”), treatment outcomes (“response OR efficacy”), and target disease (“B-ALL OR B-cell acute lymphoblastic leukemia OR B-lineage acute lymphoblastic leukemia”). Detailed search strategies for each database are provided in Supplementary Material S2 .

Inclusion criteria required studies: (1) on B-ALL patients undergoing CAR-T therapy investigating biomarker associations with efficacy or toxicity; (2) randomized trials, observational studies (prospective or retrospective), or case series (≥10 patients); (3) peer-reviewed English articles published since January 2018.

Exclusion criteria specified: (1) non-B-ALL populations or studies of therapies without CAR-T; (2) studies lacking biomarker analysis; (3) systematic reviews, meta-analyses, review articles, or conference abstracts; (4) non-English publications or inadequately translated works. Complete inclusion and exclusion criteria are detailed in Supplementary Material S3 .

Ke YH led the study selection process, which involved screening titles and abstracts, then reviewing full texts. Zhou F was regularly consulted to resolve eligibility uncertainties. All screening decisions were documented per PRISMA guidelines.

Data extraction utilized standardized forms capturing: (1) baseline characteristics (sample size, sex/age distribution, clinical status, treatment history); (2) therapeutic protocols (CAR-T product specifications, preconditioning); (3) biomarkers (types, detection methods, thresholds); and (4) clinical outcomes (efficacy/safety endpoints). All extractions adhered to predefined classification criteria.

Significant heterogeneity prevented meaningful statistical pooling across the included studies. Study designs varied considerably, including single-arm phase I/II trials (54.5%), retrospective cohort studies (36.4%), and prospective cohort analyses (9.1%). Patient ages varied widely(4 months to 76 years). Treatment protocols differed in CAR-T product targeting (CD19 single vs. CD19/CD22 dual) and costimulatory domains (4-1BB vs. CD28). Toxicity assessment methods lacked standardization, using ASTCT (15 studies), Lee (10 studies), and Penn grading systems (8 studies). Given this substantial heterogeneity, we conducted a narrative synthesis instead of a meta-analysis.

Study quality was evaluated using design-specific tools: cohort studies underwent Newcastle-Ottawa Scale assessment (evaluating cohort selection, comparability, and outcome ascertainment), while single-arm trials were appraised via the JBI checklist for single-arm studies (assessing design, implementation, and reporting). All quality assessments strictly followed established tool guidelines, with detailed assessment criteria and individual study ratings provided in Supplementary Material S4 .

The selection process for this systematic review is shown in the PRISMA flow diagram ( Figure 1 ). A total of 626 records were initially identified from three databases, and 33 studies were finally included after screening and eligibility assessment.

This systematic literature review included 33 studies (18 single-arm phase I/II trials, 12 retrospective cohort studies, and 3 prospective cohort analyses), encompassing 2,095 patients with relapsed/refractory B-ALL. Individual study populations ranged from 12 to 254 patients, with 9 studies enrolling over 100 patients, among which Zhang et al (19) reported the largest cohort (n=254). The age spectrum extended from 4 months (20) to 76 years (21). Male representation varied from 25% (22) to 74% (23). All enrolled patients had relapsed/refractory B-ALL with multiple prior lines of therapy, with some patients who had received allogeneic hematopoietic stem cell transplantation, blinatumomab, and inotuzumab.

Regarding CAR-T product characteristics, CD19 emerged as the predominant target, with products including Tisagenlecleucel, Sino19 (24–26), and AUTO1 (27). Target strategies encompassed single-targeting of CD19 or CD22, and dual-targeting of CD19/CD22 (28, 29), utilizing either 4-1BB or CD28 costimulatory domains. The standard therapeutic dose was 5×10^6 cells/kg, with total dose calculations for patients weighing over 50kg, preceded by fludarabine-cyclophosphamide conditioning regimen.

Treatment outcomes showed complete remission rates of 64.9%-100% (27 studies ≥80%) and minimal residual disease (MRD) negativity rates of 45%-100% (17 studies >80%). Safety profiles revealed CRS incidence of 55%-97.9% (grade ≥3: 10.2%-45%) and ICANS incidence of 19%-65% (grade ≥3: 2.6%-45%), evaluated using ASTCT criteria (15 studies), Lee criteria (10 studies), and Penn grading (8 studies). Long-term follow-up revealed 1-year overall survival (OS) rates of 54.8%-78.86% and progression-free survival (PFS) rates of 44.8%-69.89%. Two-year OS ranged from 59.6% to 71.4%, while 4-year OS and event-free survival (EFS) reached 69.3% and 59.0%, respectively. At 36 months, the duration of response was 56.26% (95%CI: 32.81%-74.31%) with OS of 54.72% (95%CI: 30.90%-73.38%). Notably, Next-generation sequencing (NGS) -MRD-negative patients demonstrated significantly superior 2-year EFS compared to MRD-positive patients (68%, 95%CI: 54%-86% vs 23%, 95%CI: 8.8%-62%).Detailed characteristics and outcomes for all included studies are presented in Table 1 .

The 24 biomarkers identified across studies were classified into three functional groups based on their predictive associations with treatment effects and toxicity outcomes, addressing the benefit-risk interplay in CAR-T therapy. Dual-predictive biomarkers (n=10) concurrently predicted both therapeutic effects and toxicity risks, with some markers indicating favorable efficacy outcomes alongside increased toxicity risk, while others predict poor efficacy outcomes coupled with lower toxicity risk. Favorable biomarkers (n=12) were exclusively associated with treatment efficacy outcomes, encompassing both favorable prognostic indicators (biomarkers predicting higher complete remission rates, sustained clinical benefits, and improved survival, such as CD8+ naive T cells, low Tregs ≤5.94%, PD-1+LAG-3+ expression >5.2%, B-cell aplasia, EPICART, low LDH ≤210 U/L, high platelets ≥100,000/mL, MIP3α elevation, and TH2 cytokines) and unfavorable prognostic indicators (biomarkers predicting reduced survival and poor therapeutic response, such as TP53 mutations, EP300 mutations, elevated LDH >210 U/L, low platelets, and severe bone marrow fibrosis grade ≥3), without significant toxicity linkages. Unfavorable biomarkers (n=2) were exclusively associated with toxicity manifestations and adverse event profiles (IL-6 and IKZF1 mutations), independent of therapeutic effects. The comprehensive classification framework is illustrated in Figure 2 .

Our systematic review identifies bone marrow tumor burden as a principal prognostic factor in B-ALL CAR-T outcomes. Patients with high tumor burden (≥40% blasts) exhibited reduced complete remission rates and long-term survival, consistent with established mechanisms of T-cell exhaustion and immunosuppressive microenvironments (42). These findings support the use of bridging therapies such as inotuzumab and blinatumomab for effective cytoreduction (56, 57).

Immune cell composition further refined prognostic stratification. A threshold of 5.94% for CD4+CD25+CD127low regulatory T cells (Tregs) within total CD4+ T cells reliably discriminated between favorable and unfavorable outcomes. Treg levels above this limit may suppress CAR-T function through IL-10 and TGF-β secretion, direct cell contact inhibition, and metabolic interference. In contrast, enriched early memory T-cell subsets (Tscm and Tcm) emerged as favorable biomarkers, demonstrating enhanced proliferative capacity, reduced exhaustion, and sustained persistence—properties critical for long-term efficacy (58). Tcm-enriched CAR-T products combined with fludarabine significantly improved persistence (p<0.01) and progression-free survival (p=0.001) (48).

Beyond biomarkers, treatment-related factors influenced outcomes. Newer-generation CAR-T products yielded superior remission rates, while dual-targeting or sequential infusion strategies achieved higher MRD negativity (32). These findings indicate sequential infusion of CD19/CD22 multi-targeted CAR-T cells may offer promising personalized approaches for refractory/relapsed B-ALL. While CD19 expression intensity lacked prognostic impact, prior blinatumomab exposure potentially compromised CAR-T efficacy (56, 59).

Cytokine monitoring post-infusion enhances predictive accuracy for both toxicity and response. Beyond IL-6, which remains indicative of severe cytokine release syndrome (CRS) at levels exceeding 1000 pg/mL, a multi-cytokine panel (IL-8, IL-10, IL-15, IFN-γ, TNF receptor p55) within 36 hours improved prediction of severe CRS. The combination of IL-15 ≥50 pg/mL and EGF <120 pg/mL specifically identified patients at 100% risk of severe neurotoxicity, supporting preemptive management in high-risk cases.

A TH2-oriented cytokine profile (IL-4, IL-5, IL-13) correlated with complete remission and prolonged survival, suggesting a protective immunologic milieu. Integrative models combining cellular and cytokine variables—such as CAR+TH2+ frequency and memory T-cell counts—achieved 70% sensitivity with less than 5% false-positive rates, indicating strong potential for clinical translation (26).

Somatic mutations create critical therapeutic windows in cancer treatment. TP53 mutations, among the most significant alterations, disrupt normal tumor suppression mechanisms (60). This vulnerability has led to the development of MDM2 inhibitors, which work by preventing MDM2-mediated p53 degradation. APG-115 demonstrates this strategy, effectively activating p53 and p21 while boosting antitumor immunity. The strategy proves most effective when TP53 remains functionally intact, allowing restored p53 signaling to eliminate malignant cells (61).

EP300 dysfunction presents another compelling therapeutic target in acute lymphoblastic leukemia. The EP300-ZNF384 fusion protein drives B-ALL progression through aberrant activation of genes like IL3RA (62). Unlike conventional approaches, targeting this epigenetic dysregulation offers a more precise intervention. Early studies suggest that disrupting EP300-mediated transcription can reactivate silenced tumor suppressors and improve treatment responses, particularly in fusion-positive cases where normal epigenetic control has been compromised.

Epigenetic regulation contributed to CAR-T therapy efficacy through the EPICART methylation signature, which affected gene expression patterns in responding versus non-responding patients (53). Translationally, pre-infusion CD3+ T-cell counts and CD4+/CD8+ ratios offered actionable thresholds for patient stratification (25), while achievement of minimal residual disease negativity by day 28 emerged as a critical early indicator of long-term leukemia-free survival (30).

Dynamic MRD monitoring is pivotal for outcome prediction in CAR-T therapy. Our findings confirm that day 28 MRD status serves as a critical prognostic marker, with MRD-negative patients achieving significantly superior 2-year event-free survival compared to those with MRD≥10^-6 (68% vs 23%) (43, 44). Higher MRD burdens (≥1%) consistently correlate with both poor clinical outcomes and increased toxicity risks (30, 39–41, 44, 45). Standardizing detection methods (NGS vs flow cytometry) and thresholds (10^-4 vs 10^-6) is critical for cross-study harmonization (11, 13). Digital PCR (dPCR) has emerged as a superior tool for CAR-T copy quantification due to high sensitivity and absolute quantification without standard curves. While dPCR, NGS, and flow cytometry all detect MRD, dPCR offers distinct advantages for low-frequency MRD detection (63).

CD19 dynamics and prior blinatumomab exposure may influence outcomes. Emerging surface markers (CD5, CD123, CD33, CD70, CD38, BCMA) and T-cell distributions show promise pending multicenter validation (59), suggesting substantial potential for improving CAR-T prediction systems (9, 64).

IL-6 and ferritin are cornerstone CRS severity predictors. IL-6 levels >1000 pg/mL indicate severe CRS risk, while daily levels >15.2 pg/mL predict grade ≥3 ICANS (9). Tocilizumab (IL-6R antagonist) suppresses CRS in ASTCT grade 2 with significant IL-6 elevation (≥4-fold increase) (65). However, administration with <4-fold IL-6 increase may elevate severe CRS risk (66, 67). Ferritin >10,000 ng/mL necessitates early intervention, predicting severe toxicity and reduced survival (45% vs 100% at 1 year) (22). The m-EASIX score (LDH×CRP/PLTs) >6.2 further refines risk stratification, enabling preemptive high-risk management (23).

Heterogeneity in toxicity grading systems (ASTCT, Lee, Penn criteria) complicates cross-study comparisons. In JULIET, CRS grading showed significant discordance: 38% downgraded (Lee vs Penn) and 36% downgraded (ASTCT vs Penn), particularly for grade 2 events (68).

This review highlights several limitations affecting the current evidence. Heterogeneity in sample sizes (12–254 patients) and follow-up duration (6–48 months) may affect the generalizability of results. Inconsistent use of toxicity grading systems (e.g., ASTCT, Lee, and Penn criteria) complicates cross-trial comparisons. Furthermore, variability in biomarker assays and thresholds (e.g., tumor burden cutoffs ranging from ≥5% to ≥50%) impedes unified clinical application.

Most studies were retrospective and did not adequately control for confounding variables via multivariate modeling. Research on mutations such as TP53 and EP300 has focused predominantly on prognostic association rather than mechanistic dissection or therapeutic targeting. Similarly, although IL-6 and ferritin are established toxicity markers, intervention thresholds remain poorly defined. Most biomarkers have been studied in isolation; few analyses incorporate multi-parameter models reflective of clinical complexity.

Future studies should prioritize prospective, multi-center designs using standardized endpoints and assay methods. There is a particular need for integrative biomarker models that incorporate cellular, soluble, genetic, and clinical variables to improve risk stratification and support personalized treatment strategies.