接受 CAR-T 治疗的复发/难治性 B 细胞 ALL 患者治疗反应预测模型的识别

Univariate and multivariate logistic regression analyses were conducted to identify the independent factors of CR and MRD-negative CR. The predictive models for the probability of remission were constructed based on the identified independent factors. Discrimination and calibration of the established models were assessed by receiver operating characteristic (ROC) curves and calibration plots, respectively. The predictive models were further integrated and validated in the internal series. Moreover, the prognostic value of the integration risk model was also confirmed.

The predictive model for CR was formulated by the number of white blood cells (WBC), central neural system (CNS) leukemia, TP53 mutation, bone marrow blasts, and CAR-T cell generation while the model for MRD-negative CR was formulated by disease status, bone marrow blasts, and infusion strategy. The ROC curves and calibration plots of the two models displayed great discrimination and calibration ability. Patients and infusions were divided into different risk groups according to the integration model. High-risk groups showed significant lower CR and MRD-negative CR rates in both the training and validation sets ( p < 0.01). Furthermore, low-risk patients exhibited improved overall survival (OS) (log-rank p < 0.01), higher 6-month event-free survival (EFS) rate ( p < 0.01), and lower relapse rate after the allogeneic hematopoietic stem cell transplantation (allo-HSCT) following CAR-T cell infusion ( p = 0.06).

We have established predictive models for treatment response estimation of CAR-T therapy. Our models also provided new clinical insights for the accurate diagnosis and targeted treatment of r/r B-ALL.