Classification of patients with relapsed/refractory large B-cell lymphoma who do not develop early CRS/NE toxicity using ZUMA clinical trial data

Background We aimed to develop an actionable and feasible prospective clinical model to estimate toxicity risk to assist chimeric antigen receptor (CAR) T-cell therapy providers with the management of patients with relapsed and/or refractory large B-cell lymphoma. Methods We conducted an observational, retrospective cohort study using secondary data from 390 patients treated with the CD19 CAR T-cell therapy axicabtagene ciloleucel under two prospective clinical trials, ZUMA-1 and ZUMA-7; these clinical trials enrolled patients with relapsed/refractory large B-cell lymphoma between 2015 and 2019. Using machine learning and statistical methods, we developed a classification model for identifying patients unlikely to experience early cytokine release syndrome (CRS) and neurological events (NE) of any grade. Results We found the use of prophylactic corticosteroids to be an important factor in remaining CRS-free and NE-free within the first 3 days post-treatment (p<0.001). We identified a top model for no early CRS/NE using a set of six pre-lymphodepletion clinicopathologic features: number of lines of prior systemic therapy, age, baseline tumor burden (as measured by sum of the product of the diameters), C-reactive protein, aspartate transaminase, and hemoglobin, which achieves a positive predictive value of 0.71 in the holdout validation cohort. Additionally, we find that predicted probabilities generated from the model are strongly associated with incidence of Grade 2 or higher NE. Conclusions We illustrated that routine clinicopathologic variables can be used to identify patients at low risk of developing early post-treatment CRS and/or NE. Such knowledge can be used to help treating centers prospectively manage patient care, including consideration of outpatient treatment.