Researchers develop tool to predict CAR T-cell therapy-associated neurotoxicity
A new prognostic tool has been developed to help clinicians predict which patients may experience neurotoxic effects of chimeric antigen receptor T-cell therapy, according to a study published in JAMA Neurology.
The multivariable tool is based on a list of clinical predictors shown to be associated with a significant increase in the likelihood of neurotoxic events, researchers noted.
“We envisioned this tool being used to help clinicians and researchers risk-stratify their patients receiving CAR T-cell therapy,” Daniel B. Rubin, MD, PhD, a critical care neurologist for the Center for Neurotechnology and Neurorecovery at Massachusetts General Hospital and instructor of neurology at Harvard Medical School, told Healio.
The tool his team has developed has implications for both clinical care and clinical trial research, Rubin said.
“From a clinical standpoint, it allows individuals who may require additional resources to be recognized earlier, and perhaps even treated before developing more severe symptoms,” he said. “At the same time, it may help identify patients at low risk for neurologic complications, who may either be spared certain treatments or potentially even discharged from the hospital early.”
In terms of clinical trial research, Rubin said the predictive score can “serve as an objective tool” to identify patients at low or high risk for neurotoxicity, which has implications for clinical trial design.
“Consider an investigator studying a new intervention designed to prevent or mitigate neurologic complications from CAR T-cell therapy,” he said. “By being able to identify patients at high risk, the investigator could use this tool to enroll a patient population more likely to benefit from their proposed intervention.”
The prospective portion of the study included patients at Brigham and Women’s Hospital/Dana-Farber Cancer Institute who underwent CAR T-cell therapy with axicabtagene ciloleucel (Yescarta, Kite Pharma/Gilead) from April 2015 to February 2020 for relapsed or refractory lymphoma.
After applying inclusion and exclusion criteria, researchers assigned 204 patients (mean age, 60 years; 62% men) to one of two cohorts to test the prognostic score. The derivation cohort included 126 patients admitted from April 2015 to April 2019 and the internal validation cohort included 78 patients admitted from May 2019 to February 2020.
The cohorts were evenly matched in terms of baseline clinical characteristics, including age, sex, histologic subtype, incidence of cytokine release syndrome (CRS), incidence and grade of neurotoxicity, and median onset of neurotoxicity.
All patients underwent evaluation daily for CRS and neurotoxicity during their mandatory post-infusion inpatient monitoring. CRS was graded using the criteria established by Lee and colleagues, whereas neurotoxicity was graded using the Common Terminology Criteria for Adverse Events criteria, version 4.03.
Performance measures of the diagnostic tool to predict CAR-associated neurotoxicity — including accuracy, sensitivity, specificity and area under the curve — served as the primary outcomes.
A total of 73 patients (57.9%) in the derivation cohort and 45 patients (57.7%) in the validation cohort experienced symptoms of neurotoxicity.
Univariate data analysis showed the following variables to be significantly associated with developing neurotoxicity: age, maximum temperature, maximum serum C-reactive protein, maximum serum ferritin, maximum serum interleukin-6, maximum serum procalcitonin, minimum serum white blood cell count, histologic subtype of lymphoma (indolent or aggressive), number of doses of tocilizumab (Actemra, Genentech) received by treatment day 5, CRS severity by treatment day 5, and days to onset of CRS.
The multivariable analysis yielded a predictive score ranging from 0 to 14, with a score of zero indicating the lowest probability of developing neurotoxicity.
Testing of the predictive score among the internal validation cohort yielded an AUC of 74%, an accuracy rate of 77%, a sensitivity rate of 82%, and specificity rate of 70% (positive-to-negative likelihood ratio, 2.71:0.26).
“We can predict, with fairly good accuracy, which patients are likely to develop CAR T-cell-associated neurotoxicity using only simple clinical measures that are routinely available to clinical and research teams,” Rubin told Healio. “This is despite the fact that pathophysiologic mechanisms of CAR T-cell-associated neurotoxicity remain incompletely elucidated; as ongoing research more fully characterizes the molecular mechanisms underlying this novel neurologic syndrome, our ability to predict — and ideally prevent — these symptoms from occurring should only improve.”
Rubin and colleagues identified several limitations to the study and the resulting predictive score. They noted that the evaluation and risk score are limited to only those patients with relapsed or refractory lymphoma. In addition, clinical data of some patients was missing, including routine measurements of IL-6 levels, which were not available early in the study period. Finally, the study was from a single center, and the investigators acknowledged that further research would benefit from expansion to include more centers.
“We believe this score can be used in clinical practice; however, as with any innovation, it is important that it be applied in an appropriately judicious manner,” Rubin told Healio.
He said decisions must be “made in the context of a patient's full clinical picture, considering other medical conditions, the severity of the patient's underlying malignancy and other clinical factors.”
Rubin said his team is developing an online calculator that will make it easier for clinical care teams to use this prognostic scoring tool.
“Our hope is that as more clinicians become familiar with this tool and find it helpful, it will become more widely adopted,” he said.
For more information:
Daniel B. Rubin, MD, PhD, can be reached at firstname.lastname@example.org.