Fig. 2

Construction and validation of cytokine-based clinical models for differentiating CRS-related fever from infectious fever. A Decision tree classification for differentiating between CRS and infection. B Receiver operating characteristic (ROC) curve of the decision tree classification in the training cohort and validation cohort. C An equation of clinical model for discriminating between CRS and infection. D ROC curve of stepwise logistic regression. E Decision curve analyses demonstrating the net benefit in training (left) and validation (right) cohorts. The black curve (“all”) represents that this model identifies that all the patients suffered from infectious fever, and the grey curve (“none”) represents that this model identifies that all the patients suffered from CRS-related fever. The upper and lower lines of each group represent a 95% confidence interval. F Flow chart of the optimized clinical management of fever after CAR-T cell infusion. AUC, area under the curve; CXCL, CXC motif chemokine ligand; IFN, interferon; IL, interleukin; VEGF, vascular endothelial growth factor