Car-T therapies show potential benefit as second-line treatment in relapsed/refractory large B-cell lymphoma: A bayesian meta-analysis of randomized controlled trials

Abstract

Abstract Introduction The advent of chimeric antigen receptor T-cell (CAR-T) therapies has transformed the management of relapsed/refractory large B-cell lymphoma (R/R LBCL), offering a promising alternative to standard-of-care (SOC) second-line treatments. Despite encouraging results from individual randomized phase 3 trials, such as TRANSFORM, ZUMA-7, and BELINDA, variability in outcomes across studies highlights the need for a comprehensive synthesis. This Bayesian meta-analysis aims to integrate evidence from these trials to estimate the overall effect of CAR-T therapies on overall response rate (ORR), event-free survival (EFS), and overall survival (OS), while accounting for heterogeneity and uncertainty using a shrinkage approach. Methods We conducted a Bayesian meta-analysis incorporating data from three randomized phase 3 trials: TRANSFORM, ZUMA-7, and BELINDA, comparing CAR-T therapies with SOC in R/R LBCL patients. Outcomes assessed included ORR (log relative risk), EFS, and OS (log hazard ratios). A random-effects model with a hierarchical Bayesian framework was employed, utilizing shrinkage estimation to adjust individual study effects toward a pooled mean. Non-informative priors were used for the pooled effect and between-study variance (tau), with 95% credible intervals (CrI) and prediction intervals derived from posterior distributions. Heterogeneity was quantified using tau, and sensitivity analyses explored the impact of trial-specific factors such as bridging therapy and manufacturing delays. Data were analyzed using R version 4.5.1. Results The analysis included 1,123 patients across the three trials. For ORR, the pooled log relative risk was 0.380 [95% CrI: -0.342, 1.018], with shrinkage-adjusted estimates of 0.531 [95% CrI: 0.268, 0.793] for TRANSFORM, 0.510 [95% CrI: 0.348, 0.671] for ZUMA-7, and 0.000 [95% CrI: -0.351, 0.351] for BELINDA, indicating a potential benefit with CAR-T, though with wide uncertainty. The prediction interval was 0.387 [-1.042, 1.717], and heterogeneity was moderate (tau = 0.36 [0.00, 1.14]). For EFS, the pooled log hazard ratio was -0.60 [95% CrI: -1.62, 0.45], with estimates of -1.08 [-1.51, -0.65] for TRANSFORM, -0.92 [-1.18, -0.66] for ZUMA-7, and 0.07 [-0.13, 0.27] for BELINDA, suggesting a reduced event risk with CAR-T, but the CrI crossed zero. The prediction interval was -0.60 [-2.73, 1.57], with higher heterogeneity (tau = 0.73 [0.24, 1.62]). For OS, the pooled log hazard ratio was -0.22 [95% CrI: -0.72, 0.28], with estimates of -0.28 [-0.73, 0.17] for TRANSFORM, -0.31 [-0.64, 0.01] for ZUMA-7, and -0.09 [-0.43, 0.24] for BELINDA, showing no clear survival benefit, with a prediction interval of -0.22 [-1.19, 0.75] and low heterogeneity (tau = 0.18 [0.00, 0.87]). Conclusion This Bayesian meta-analysis suggests a potential benefit of CAR-T therapies over SOC for ORR and EFS in R/R LBCL, though confidence intervals indicate uncertainty and variability across trials. The lack of a clear OS advantage underscores the need for longer follow-up and further investigation into factors driving heterogeneity, such as trial design and patient characteristics. These findings support the continued evaluation of CAR-T in clinical practice and research, with a focus on optimizing delivery and patient selection to maximize efficacy.