Clinical Predictors of Neurotoxicity After Chimeric Antigen Receptor T-Cell Therapy

Abstract

Importance: Chimeric antigen receptor (CAR) T-cell therapy for relapsed or refractory hematologic malignant neoplasm causes severe neurologic adverse events ranging from encephalopathy and aphasia to cerebral edema and death. The cause of neurotoxicity is incompletely understood, and its unpredictability is a reason for prolonged hospitalization after CAR T-cell infusion.

Objective: To identify clinical and laboratory parameters predictive of neurotoxicity and to develop a prognostic score associated with its risk.

Design, setting, and participants: This single-center diagnostic/prognostic accuracy study was conducted at Brigham and Women's Hospital/Dana Farber Cancer Institute from April 2015 to February 2020. A consecutive sample of all patients undergoing CAR T-cell therapy with axicabtagene ciloleucel for relapsed or refractory lymphoma were assessed for inclusion (n = 213). Patients who had previously received CAR T cells or who were treated for mantle cell lymphoma were excluded (n = 9). Patients were followed up for a minimum of 30 days from the date of CAR T-cell infusion.

Main outcomes and measures: The primary outcomes were measures of performance (accuracy, sensitivity, specificity, area under the curve) of a diagnostic tool to predict the occurrence of CAR-associated neurotoxicity, as graded by the Common Terminology Criteria for Adverse Events criteria.

Results: Two hundred four patients (127 men [62.2%]; mean [SD] age, 60.0 [12.1] years) were included in the analysis, of which 126 (61.8%) comprised a derivation cohort and 78 (38.2%), an internal validation cohort. Seventy-three patients (57.9%) in the derivation cohort and 45 patients (57.7%) in the validation cohort experienced neurotoxicity. Clinical and laboratory values obtained early in admission were used to develop a multivariable score that can predict the subsequent development of neurotoxicity; when tested on an internal validation cohort, this score had an area under the curve of 74%, an accuracy of 77%, a sensitivity of 82%, and a specificity of 70% (positive:negative likelihood ratio, 2.71:0.26).

Conclusions and relevance: The score developed in this study may help predict which patients are likely to experience CAR T-cell-associated neurotoxicity. The score can be used for triaging and resource allocation and may allow a large proportion of patients to be discharged from the hospital early.

Figure.. Logistic Plots of the Multivariable Predictive Score Demonstrating Model Performance

A, The multivariable predictive model fit to the derivation cohort (n = 126); logistic plot of predictive score vs outcome. The blue line is the probability of neurotoxicity from the multivariable model. Position of markers along the y-axis of each marker is randomly jittered to allow visualization of each data point. B, The results of the multivariable prediction model applied to the validation cohort (n = 78). The model has maximum discriminating power using a score of 6 as cutoff. False positives are blue and false negatives are orange; all other data points demonstrate accurate predictions. In panels A and B, open circles denote individual patients who did not experience neurotoxicity; solid circles denote patients who did experience neurotoxicity. The multivariable prediction model can be used to predict which patients in the validation cohort are unlikely to develop moderate or severe neurotoxicity (grade ≥2; our institutional cutoff for discharge or readmission) (C) and which are unlikely to develop severe neurotoxicity (grade ≥3) (D). The blue line indicates the probability of not developing grade of 2 or greater (C) or grade of 3 or greater (D) neurotoxicity based on the multivariable model. Open circles denote individual patients who did not experience grade of 2 or greater (C) or grade of 3 or greater (D) neurotoxicity; solid circles denote patients who did experience grade of 2 or greater (C) or grade of 3 or greater (D) neurotoxicity. A cutoff of 6 (shaded in blue) provides a high sensitivity and high negative predictive value for the development of grade of 2 or greater (C) or grade of 3 or greater (D) neurotoxicity.