An Inflammatory Biomarker Signature Reproducibly Predicts CAR-T Treatment Failure in Patients with Aggressive Lymphoma across the Zuma Trials Cohorts

Background: Disease progression and relapse remain significant challenges to chimeric antigen receptor (CAR)-T cell therapy in relapsed/refractory (R/R) large B-cell lymphoma (LBCL). Using an unsupervised machine learning approach applied to pre-infusion labs and cytokines assessing organ function and inflammation, we previously identified unique signatures (inflammatory vs. non-inflammatory clusters). The inflammatory cluster was enriched for patients with elevated inflammatory markers, increased cell turnover, or end-organ sequelae of inflammation (e.g., hepatitis, bone marrow suppression) and was strongly associated with CAR-T treatment failure in the clinical setting (Raj S et al., ASH 2022).

Methods: Here, we validated the model, named InflaMix, and evaluated its utility in a large cohort of patients with LBCL (n=352) treated with axicabtagene ciloleucel (axi-cel) as part of the ZUMA-1 (NCT02348216; n=182, 52%) and ZUMA-7 (NCT03391466; n=170, 48%) registrational studies. InflaMix was applied using 14 pre-infusion, day 0 laboratory and cytokine (hemoglobin, platelets, white blood cells, CRP, ferritin, IL-6, IL-10, TNF∝, d-dimer, aspartate transaminase, alkaline phosphatase, total bilirubin, LDH, albumin) measurements.

Results: InflaMix assigned 46% and 30% of patients to the inflammatory cluster in the ZUMA-1 and ZUMA-7 cohorts, respectively. Patients in the inflammatory clusters were more likely to have had more prior lines of therapy and higher baseline LDH than those in the non-inflammatory clusters (p < 0.001). Assignment to the inflammatory cluster vs. non-inflammatory cluster was significantly associated with an increased risk of not achieving complete response (CR) by day 100 across both ZUMA-1 (odds ratio [OR] 1.68 [95% confidence interval (CI) 1.31 to 2.64) and ZUMA-7 (OR 1.57 [95% CI 1.25 to 2.53]) cohorts (Figure A). Importantly, the inflammatory cluster was also significantly associated with inferior overall survival compared with the non-inflammatory cluster (Figure B) in ZUMA-1 and ZUMA-7 (hazard ratio [HR] 1.65 [95% CI 1.38 to 2.14] and HR 1.75 [95% CI 1.39 to 2.59], respectively), as well as decreased progression-free survival in ZUMA-1. All analyses were adjusted for age, pre-lymphodepletion LDH as a surrogate for baseline disease burden, and history of primary refractory disease, and adjustment for prophylactic steroid treatment in the ZUMA-1 cohort; prophylactic steroids were administered immediately before and after infusion in sub-cohort 6 of ZUMA-1.

Conclusion: InflaMix identifies a unique inflammatory biomarker signature that can be derived using a simple blood draw prior to CAR-T infusion. We show that the signature reproducibly stratifies patients with LBCL by risk for CAR-T treatment failure and inferior survival is valid in patients treated with axi-cel as second-line treatment or later. Potential clinical applications of InflaMix include prognostication and informed design of clinical trials to target high-risk populations.