Survival prediction in relapsed/refractory B-cell lymphoma treated with CAR-T therapy: An AI-optimized cox regression model based on CCL5, scgf-β, and TRAIL

Abstract

Abstract Objective: To identify prognostic biomarkers associated with long-term survival in patients receiving chimeric antigen receptor-T cell (CAR-T) therapy, using integrative omics analysis to inform clinical risk stratification and guide precision treatment. Methods: We conducted a single-center retrospective study involving 161 patients with relapsed or refractory (R/R) aggressive B-cell lymphoma who received CAR-T therapy between November 2017 and July 2025. A total of 220 plasma samples were collected either at baseline or during the peak phase of cytokine release syndrome (CRS) following CAR-T cell infusion. Protein dynamics were assessed using high-throughput proximity extension assays (PEA) and multiplex bead-based cytokine profiling. Kaplan–Meier survival analysis and multivariate Cox proportional hazards modeling (with P < 0.05 considered statistically significant) were performed to identify prognostic biomarkers and construct a survival prediction model. The training cohort comprised 87 patients, while the most recent 74 patients constituted an independent validation cohort. Patients who received commercial CAR-T products met the approved clinical indications, and participants enrolled in clinical trials provided written informed consent in accordance with the Declaration of Helsinki. Results: Proteomic analysis revealed key pathway alterations, including chemokine–integrin signaling. Patients with ≥2-year survival exhibited distinct protein expression profiles compared to those with <2-year survival, involving pathways associated with myeloid cell migration and T-cell functionality. Core regulatory proteins included CXCL8, CXCL10, and EGF. Absolute quantification identified 15 significantly differentially expressed proteins (e.g., CCL3–5, CCL27), primarily enriched in TNF signaling, NK cell chemotaxis, and myeloid activation pathways. Multivariate analysis confirmed CCL5 as an independent prognostic biomarker (HR = 0.13; 95% CI: 0.03–0.57; P = 0.007). A survival prediction model incorporating CCL5, SCGF-β, and TRAIL demonstrated excellent discriminative performance (C-index = 0.980). Risk stratification based on this model identified three survival groups with significant differences in both the training cohort (P = 0.014) and the validation cohort (P < 0.001). Conclusion: CCL5 is an independent prognostic biomarker for CD19 CAR-T therapy in B-cell lymphoma. The integrative prediction model based on CCL5, SCGF-β, and TRAIL offers a robust tool for survival prediction and supports precision risk stratification and individualized therapeutic decision-making in the context of CAR-T therapy.