Figure 3.
Evaluation of biomarkers of toxicity. (A) Correlation between the CAR-HEMATOTOX score before infusion and the duration of severe neutropenia; (B-E) Measurement of the levels of NF-L, IL-1b, and ANG-1 and ANG-2 in the CSF of patients overtime, after infusion of CD19-CAR_Lenti cells. (F) The scatterplot illustrates how the rule identified by the JRip model classifies CRS categories (yes/no CRS) using samples from patients before the infusion of CAR T cells (day 0) in our data set. Samples for cytokine evaluation were unavailable for 2 patients. The red dashed line indicates the IL-6 cutoff at 6 pg/mL, and the blue dashed line indicates the IL-10 cutoff at 3.5 pg/mL. The background colors represent the predicted classification areas for patients with CRS negativity (blue) and CRS positivity (red), respectively; with a high Kappa value of 0.866, the model indicates substantial agreement beyond chance. The P value for exceeding the no-information rate (71%) is .02, suggesting that the model performs significantly better than a random classifier. The balanced accuracy is 95%, further highlighting its good performance. ANG-1/2, agiopoietin-1/2.

Evaluation of biomarkers of toxicity. (A) Correlation between the CAR-HEMATOTOX score before infusion and the duration of severe neutropenia; (B-E) Measurement of the levels of NF-L, IL-1b, and ANG-1 and ANG-2 in the CSF of patients overtime, after infusion of CD19-CAR_Lenti cells. (F) The scatterplot illustrates how the rule identified by the JRip model classifies CRS categories (yes/no CRS) using samples from patients before the infusion of CAR T cells (day 0) in our data set. Samples for cytokine evaluation were unavailable for 2 patients. The red dashed line indicates the IL-6 cutoff at 6 pg/mL, and the blue dashed line indicates the IL-10 cutoff at 3.5 pg/mL. The background colors represent the predicted classification areas for patients with CRS negativity (blue) and CRS positivity (red), respectively; with a high Kappa value of 0.866, the model indicates substantial agreement beyond chance. The P value for exceeding the no-information rate (71%) is .02, suggesting that the model performs significantly better than a random classifier. The balanced accuracy is 95%, further highlighting its good performance. ANG-1/2, agiopoietin-1/2.

or Create an Account

Close Modal
Close Modal