Three-Factor Prediction Model for Grade 2+ Cytokine Release Syndrome in Large B-Cell Lymphoma Patients Receiving Epcoritamab Monotherapy

Background: Epcoritamab (epcor) is a subcutaneously administered bispecific antibody (bsAb) targeting CD3xCD20 approved for the treatment of both patients (pts) with 3L+ large B-cell lymphoma (LBCL) and 3L+ follicular lymphoma (FL) as monotherapy and is being further developed in other B-cell malignancies. Cytokine release syndrome (CRS) is the most common toxicity of bsAbs, including epcor. The diagnosis of CRS requires fever, with higher-grade events including hypotension, hypoxia, and/or organ dysfunction, which can be life-threatening. A predictive CRS model was developed for another CD3xCD20 bsAb, glofitamab (Gritti G et al. Blood Adv. 2024); however, it had suboptimal performance when applied to epcor LBCL data. This highlights the need for improved models to accurately identify pts at low risk for developing severe CRS to facilitate outpatient dosing. Methods: EPCORE™ NHL-1 (NCT03625037) is a multicohort, single-arm, phase 2 trial; the LBCL expansion cohort enrolled 157 adult pts with relapsed or refractory CD20+ LBCL, including diffuse LBCL (DLBCL) (n=139), high-grade B-cell lymphoma (n=9), Primary mediastinal B-cell lymphoma (n=4), and FL grade (G) 3B (n=5). Pts received SC epcor with 2 step-up doses: priming (0.16 mg) and intermediate (0.8 mg), followed by 48 mg full doses in 28-d Cycles (C): QW, C1-3; Q2W, C4-9; Q4W, C≥10 until progressive disease or unacceptable toxicity. Pts received CRS prophylaxis with steroids (predominantly prednisone), acetaminophen, and diphenhydramine in C1. ORR/CR rates by IRC were 63/40% (Thieblemont et al, EHA 2024). CRS was the most reported AE (51% all G [32% G1, 16% G2, 3% G3]) and most frequently occurred after the first full dose of epcor (C1 day [D] 15). Multiple logistic regression with 10-fold cross-validation and clinical interpretation were used to develop an epcor-specific model with risk scoring to identify factors predictive of developing G2+ CRS after the first full dose of epcor (C1D15). Potential factors included baseline characteristics (age, performance status, medical history, prior treatment, and disease burden) and laboratory data. Performance of the glofitamab 5-factor CRS model was assessed by applying the model to the epcor data in C1 and evaluating the negative predictive value (NPV) and the statistical significance of the model variables. Results: The epcor model development included a total of 147 pts with relapsed or refractory LBCL who received the first full dose (C1D15), of which 24 (16%) pts had G2+ CRS following that dose. A 3-factor model was identified using the following characteristics: (1) prior CAR T use (Yes vs No; odds ratio [OR] 0.11; P=0.0042); (2) extranodal disease (Yes vs No; OR 3.89; P=0.0420); and (3) baseline total metabolic tumor volume (TMTV) (≥80 mL vs <80 mL; OR 2.66; P=0.0659). A risk score (range, −2.2 to 2.4) assigns pts with a score of <1 as “low risk.” Of the 70 pts assigned as low risk by the model, 68 did not have G2+ CRS after the first full dose of epcor (NPV of 97%). The recall of the model was 92%. Validation of the model with other external datasets is planned. Previous studies have reported the glofitamab 5-factor CRS model to have an NPV of 97%; however, it did not perform well for predicting G2+ CRS for the epcor LBCL cohort. The resulting NPV was 80% and all factor P-values were >0.5. Conclusions: The epcor predictive 3-factor model uses baseline characteristics that can be calculated from local pretreatment assessments: prior CAR T, extranodal disease, and TMTV. This model was developed prior to the implementation of EPCORE NHL-1 C1 optimization measures (dexamethasone and fluids), which have further reduced the rates and severity of CRS in pts with LBCL. The model yielded an NPV of 97%, which could still be useful in identifying those already at lower risk of G2+ CRS, irrespective of the choice of steroids. The availability of TMTV may vary based on the clinical setting; however, as TMTV emerges as an important prognostic factor in LBCL, it is expected to become more readily available (Vercellino et al, Blood 2020). Once validated, adoption of this epcor-specific prediction model may support clinicians to tailor CRS monitoring, improving patient outcomes and resource utilization.