Cytokine Biomarkers for the Prediction of Cytokine Release Syndrome in Chimeric Antigen Receptor T-Cell Therapy: A Comprehensive Systematic Review and Meta-Analysis of Randomised Controlled Trials and Primary Studies

Background. Cytokine release syndrome (CRS) remains the defining acute toxicity of chimeric antigen receptor (CAR) T-cell therapy, and the capacity to identify, before clinical deterioration, the minority of recipients who will progress to high-grade disease would allow pre-emptive rather than reactive management. Individual cohorts have nominated a long list of candidate cytokine biomarkers, but their effect sizes, measurement windows and discriminative performance have never been quantitatively consolidated across the full breadth of CAR-T indications. Objective. To quantify the association between circulating cytokine biomarkers and severe CRS after CAR-T cell therapy, to compare cytokines with routinely available acute-phase and endothelial indices, and to estimate the discriminative accuracy of composite multi-analyte models. Methods. The study strictly adhered to the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) 2020 guidelines. Randomised controlled trials, comparative studies, cohort studies and case series of at least ten patients that reported a quantified cytokine biomarker and graded CRS were eligible; systematic reviews and meta-analyses were excluded from pooling and reserved for contextual discussion. Risk of bias was appraised with ROBINS-I and Cochrane RoB 2. DerSimonian–Laird random-effects models pooled odds ratios per ten-fold increment of each analyte, with prediction intervals, Egger and Begg tests, trim-and-fill, leave-one-out influence analysis, meta-regression and GRADE certainty rating. Diagnostic accuracy of composite models was synthesised bivariately with a summary receiver-operating characteristic curve. Results. Thirty-two primary studies comprising 4,381 participants were included, of which 27 contributed to quantitative synthesis. Interferon-gamma showed the strongest pooled association with severe CRS (odds ratio 2.65 (95% CI 2.13–3.30); I² 34%), followed by interleukin-6 (2.49 (95% CI 2.06–3.01); I² 35%), the monocyte–macrophage chemokine axis of interleukin-8 and MCP-1 (2.24 (95% CI 1.77–2.83)), interleukin-10 (2.15 (95% CI 1.78–2.60)), interleukin-2 and its soluble receptor (1.84 (95% CI 1.45–2.35)) and interleukin-15 (1.70 (95% CI 1.36–2.11)). Tumour necrosis factor-alpha was not significantly associated with severe CRS (1.18 (95% CI 0.96–1.46)). Acute-phase reactants performed materially less well than effector cytokines (1.50 (95% CI 1.30–1.74)), whereas the endothelial activation and stress index separated severe from non-severe disease with a standardised mean difference of 0.49 (95% CI 0.34–0.63). Pre-infusion composite models carried a pooled odds ratio of 2.01 (95% CI 1.70–2.38), significantly weaker than post-infusion sampling (1.60 versus 2.19; P for interaction = 0.002). Composite multi-analyte models achieved a pooled sensitivity of 77.2% (95% CI 70.5–82.7), specificity of 84.4% (79.0–88.6) and a diagnostic odds ratio of 21.0, with a mean reported area under the curve of 0.85. Certainty of evidence was low for most individual analytes and moderate for EASIX. Conclusions. Effector cytokines generated by the CAR T cell itself and by the myeloid compartment it recruits are reproducibly associated with severe CRS, and composite models combining two or three analytes with a routine haematological variable approach the discriminative performance required for clinical deployment. The evidence base is nonetheless constrained by platform heterogeneity, in-sample threshold derivation and the near-absence of randomised evidence for biomarker-triggered intervention. A biomarker-stratified randomised trial of pre-emptive cytokine blockade is now the logical and overdue next step.