Can one scoring system fit all? Comparative validation of CAR-HEMATOTOX, ALL-HEMATOTOX, and eIPM for predicting immune effector cell-associated hematotoxicity following CAR-T therapy in hematologic malignancies.

Abstract

Objective: Immune effector cell-associated hematotoxicity (ICAHT), characterized by prolonged cytopenia and delayed hematopoietic recovery, is a common complication following chimeric antigen receptor T (CAR-T) cell therapy. However, the applicability of existing predictive models, CAR-HEMATOTOX (CAR-HT) for lymphoma, acute lymphoblastic leukemia-HEMATOTOX (ALL-HT) for B-ALL, and the early ICAHT prediction model (eIPM), remains uncertain across different hematologic malignancies.

Methods: We prospectively analyzed 119 patients who received CAR-T therapy between January 2022 and June 2025, including B-ALL (n=62), T-ALL/non-Hodgkin's lymphoma (NHL) (n=25), and multiple myeloma (MM, n=32). The CAR-HT, ALL-HT, and eIPM models were evaluated for their ability to predict ICAHT severity and survival outcomes.

Results: Grade 3 ICAHT occurred in 32.3% of B-ALL, 40.0% of T-ALL/NHL, and 25.0% of MM patients, while grade 4 rates were 33.9%, 20.0%, and 6.3%, respectively. CAR-HT classified 67.2% of patients as high-risk, and ALL-HT identified 56.3% of ALL/NHL patients as high-risk. In both models, high-risk groups experienced significantly more prolonged neutropenia than low-risk groups (CAR-HT: 17.7 vs. 5.3 d, P<0.001; ALL-HT: 21.3 vs. 7.7 d, P<0.001). Both eIPMpre and eIPMpost strongly correlated with grade 3-4 ICAHT (P<0.001). Importantly, survival analysis showed that eIPMpre stratification distinguished outcomes: 1-year overall survival (OS) was 65% in medium+high-risk vs. 84% in low-risk patients (P=0.006), and 1-year disease-free survival (DFS) was 44% vs. 73% (P<0.001). Similar predictive accuracy was observed with eIPMpost.

Conclusions: The CAR-HT, ALL-HT, and eIPM models consistently identify patients at high risk for severe ICAHT across B-ALL, T-ALL/NHL, and MM. Among these, the eIPM stands out as a promising universal tool for survival prediction. These models provide valuable prognostic insights that can guide supportive care and inform treatment planning in CAR-T therapy.