Development and validation of a logistic regression model to predict carHLH. (A) Fraction of carHLH according to patient demographics, disease characteristics, and treatment course. Squares represent the observed percentage with carHLH for a given factor, with the lines representing 95% confidence intervals for the difference of proportions from the reference category. (B) Logistic regression model for prediction of carHLH. (C) ROC curves for predictive model (incorporating IFNγ on day of CRS⁺2, and baseline BM T:NK ratio). The threefold cross validation mean AUC was 0.86. AUC values for each fold are indicated in the legend (n = 39). *Score rounded to 2 decimal places. **Limited cytokine data were available for validation cohort. disc, discovery cohort; N/A, not applicable (for patients to whom a model did not apply); ROC, receiver operating characteristic; TCS, T-cell selection during CAR product manufacturing; val, validation cohort.