An IL-10-incorporated predictive model for late immune effector cell-associated hematotoxicity following CAR-T therapy in B-cell malignancies

Abstract Introduction: Immune effector cell–associated hematotoxicity (ICAHT) is a common complication of chimeric antigen receptor T-cell (CAR-T) therapy. While pre-infusion lymphodepleting chemotherapy and profound inflammation post-infusion contributes to cytopenia, persistent late ICAHT (>day +30) may endure for months to years. Critically, this late-phase toxicity elevates infection risk, substantially compromising patient survival and quality of life. Predictive models for late ICAHT are therefore essential for proactive clinical management. Methods: We conducted a retrospective study of 151 patients with B-cell malignancies, including 67 who developed severe late ICAHT. To prevent inter-cohort overlap, patients were stratified into: a derivation cohort (n=50; treated 2018 - 2019) and a validation cohort (n=101; treated 2019 - 2022). All received CD19/22 CAR-T therapy as we described previously (Wang N et al. Blood. 2020); 8 had prior hematopoietic stem cell transplantation (HSCT). Serial serum samples collected post-infusion, especially during peak cytokine release syndrome (CRS), and were measured using a 70-plex cytokine panel (Meso Scale Discovery, Cat# K1508K). Neutrophil dynamics were tracked from day +30 to 3 years post-infusion. Neutropenia was graded per EHA/EBMT consensus (Rejeski et al., Blood 2023). Non-parametric testing and stepwise logistic regression identified cytokines associated with severe late ICAHT (ANC grades 3-4, >day +30). Model performance was validated independently. Results: In the derivation cohort (n=50), 18 patients (36%) developed severe late neutropenia, while 6 (12%) experienced grade ≥3 CRS. Comparative analysis of 70 cytokines revealed six significantly elevated cytokines in grade 3–4 vs. grade 0–2 ICAHT patients: IL-10, IL-1α, IL-2RA, IL-1β, IL-6, and IL-27 (all P<0.05). Stepwise regression identified IL-10 as the optimal predictor of severe late cytopenia (AUC = 0.75, P < 0.0001). Day +3 sampling yielded optimal predictive value; Youden index analysis established an IL-10 cutoff of 17.7 pg/mL (sensitivity 80.0%, specificity 62.5%). This cutoff demonstrated high accuracy in the validation cohort (AUC = 0.87, P < 0.0001; sensitivity 67.5%, specificity 88.24%). The established CAR-HT model (Rejeski et al., Blood 2021) demonstrated the following performance for late ICAHT prediction: derivation cohort (AUC = 0.79, sensitivity 50.0%, specificity 86.36%); validation cohort (AUC = 0.72, sensitivity 86.27%, specificity 48.98%). We thus developed the IL-10-HT model, assigning points based on IL-10 levels as follows: <20 pg/mL = 0 points; 20–50 pg/mL = 1 point; >50 pg/mL = 2 points. The scoring for the other 5 indicators followed the CAR-HT model definition, with a total score ≥2 points defining high-risk status. Integration of the CAR-HT score with IL-10 levels significantly improved predictive accuracy: derivation cohort (AUC = 0.90, sensitivity 81.82%, specificity 77.18%); validation cohort (AUC = 0.88, sensitivity 76.47%, specificity 86.33%). Conclusions: The IL-10-incorporated model effectively predicts severe late ICAHT following CAR-T therapy for B-cell malignancies. This approach enables early risk-stratified interventions, potentially improving post-discharge outcomes.