Development of a Predictive Model for Cytokine Release Syndrome to Inform Risk Stratification and CRS Management Following Immunotherapy

Background:

Cytokine release syndrome (CRS) is a potentially life-threatening toxicity caused by immune activation. CRS can be triggered non-specifically by T-cell engaging therapies. Risk factors for CRS are expected to be disease-specific and prediction of an individual patient's CRS risk is not currently possible.

Glofitamab is a T-cell engaging bispecific antibody targeting CD20 and CD3 with a novel 2:1 format that has shown promising efficacy with a manageable safety profile in NP30179 (NCT03075696), an ongoing Phase I/II dose finding study evaluating glofitamab in patients with relapsed or refractory non-Hodgkin lymphoma (NHL). While CRS is observed with glofitamab, cases typically occur with grade 1/2 severity (per ASTCT, Lee et al 2019) with most occurrences confined to the first cycle of therapy. Data from NP30179 were used to develop a model to predict the occurrence of Grade ≥ 2 CRS after the first glofitamab dose to enable stratification of patients according to risk of CRS with possible future implications for intensity of monitoring for those at low risk.

Methods:

Glofitamab was administered intravenously over 4 to 8 hours as a fixed or step-up dosing regimen, as previously described (Hutchings, et al JCO 2021). Non-overlapping training and validation data sets were defined; the training data set included patients with aggressive (n=165) or indolent NHL (n=31) who received a first dose of 0.6-25 mg, and the model validation data set (n=51; 35 aggressive NHL) included patients who received a 2.5mg first dose.

The primary outcome was defined as Grade ≥2 CRS in the week after the first glofitamab dose, and included 65 events (n=58 training, n=7 validation). In the training data set we evaluated the association between the dose, putative risk factors (including demographics, medical history, disease characteristic variables, baseline laboratory values) and the occurrence of CRS. Univariate and multivariate models were applied in a stratified cross-validation setting to assess the most predictive and stable combination of risk factors.

Baseline and on-treatment cytokine levels were analyzed via ELISA in a subset of patients (n=89).

Results:

The temporal pattern of CRS occurrences revealed that the vast majority of first CRS events occur after the first dose of glofitamab. To predict the occurrence of Grade ≥ 2 CRS after the first glofitamab dose, a multivariate model was developed to include glofitamab first dose and a combined risk score, termed the "CRS risk score" (CRSRS), which is the weighted sum of binarized risk factor values at baseline (Figure). The predictive ability was tested in the validation data set in which the incidence of Grade ≥ 2 CRS was 14% (7/51). A low risk group (CRSRS <5.0) was identified to be 60% of the test cohort with patients within this group having only 5% chance (Negative Predictive Value=0.95, SE=0.03) of experiencing a Grade ≥2 CRS.

Induction of cytokines, including IL-6 and TNFα, was observed upon treatment with glofitamab and peak magnitude of cytokine induction was associated with CRS incidence and severity. Cytokine induction was evident by end of glofitamab infusion and peak levels of TNFα were observed by mid-infusion.

CRSRS and TNFα induction were evaluated together for the subset of patients with cytokine data available to determine whether risk classification incorporating both metrics could refine the risk stratification of patients. Early TNFα changes were layered on top of CRSRS (at the cutoff of 5.0) such that patients with less than 1.5-fold induction were classified as low risk, and patients with more than 8-fold induction were classified as high risk. In the training data set, this decision tree approach improved the performance of prediction compared to CRSRS alone.

Conclusions:

A model based on 8 baseline factors allowed an accurate classification of risk for Grade ≥2 CRS upon treatment with glofitamab. Addition of TNFα induction may improve the predictive value. The predictive performance was confirmed in a separate validation data set with additional analyses in independent cohorts ongoing. The CRSRS, alone or in combination with cytokine induction, represents a tool to predict the occurrence of Grade ≥2 CRS after the first glofitamab dose to enable stratification of patients according to risk of CRS with possible future implications for the intensity of monitoring for those at low risk.

K.V.K. and A.B. have contributed equally.