Predictive Model for Infection Risk in CAR-T Therapy Patients

Background:

Chimeric antigen receptor T-cell (CAR-T) therapy is a powerful treatment for hematologic malignancies, but it often leads to serious infections, a major cause of non-relapse mortality. These infections typically occur within the first 30 days post-therapy, a period also characterized by cytokine release syndrome (CRS). Early fever after CAR-T therapy can indicate either an infection or CRS, complicating diagnosis, especially in neutropenic patients. A predictive model is needed to accurately identify infections for prompt and effective management.

Methods:

This multicenter retrospective study involved three phases: model development, internal validation, and external validation. The internal cohort included 443 patients treated with CAR-T therapy at Tongji Hospital, Huazhong University of Science and Technology, from June 2016 to December 2023. The study excluded patients with pre-existing infections, fever unrelated to CRS or infection, incomplete data, or no fever within 30 days post-CAR-T therapy. An external validation cohort of 30 patients was recruited from three hospitals. Febrile events (temperature >38°C) within 30 days post-infusion were recorded, with daily assessments of blood counts and inflammation markers such as C-reactive protein (CRP), interleukin-6 (IL-6), and ferritin. Febrile episodes were classified as infection or CRS. A multivariate logistic regression model identified independent predictors of infection and established a scoring system.

Results:

Among the 443 patients, 514 febrile episodes were recorded, with 86% occurring during the first episode. Infections accounted for 21% of initial episodes, often co-occurring with CRS. The predictive model for infection during the first febrile episode included factors like pre-fever neutropenia (N), ferritin change (FCV) on the second day of fever, lymphocyte count (L), platelet count (Plt), IL-6 level, and CRP level. The model achieved an area under the curve (AUC) of 0.9, with 87% sensitivity and 81% specificity in the training set. The external validation cohort showed an AUC of 0.93, with sensitivity and specificity of 75% and 84%, respectively. The Hosmer-Lemeshow test confirmed a good model fit (P = 0.950), and decision curve analysis indicated a significant net benefit.

Conclusion:

This study developed and validated a robust risk assessment model for predicting infections within 30 days post-CAR-T therapy. The model, based on routine blood tests and febrile episodes, is practical for clinical use and offers high accuracy across different patient cohorts.

No relevant conflicts of interest to declare.