Detecting deviations from the efficacy and safety results of single-arm trials using real-world data: The case of a CAR-T cell therapy in B-cell lymphoma

Purpose: Personalized therapies are leading to an increasing number of marketing authorizations based on single-arm trials, which increases the demand for better post-authorization monitoring strategies. The aim of the present study was to estimate the power over time as data accrue in population-based registries for detecting deviations from the expected efficacy/safety of chimeric antigen receptor T cell (CAR-T) therapy approved for relapsed/refractory large B-cell lymphoma (RR-LBCL). Methods: The number of real-world RR-LBCL patients was projected over time in a general population of 5, 15, and 25 million citizens using lymphoma registry data. For each scenario, we computed the power over time for detecting significant deviations in efficacy (1-year overall survival [1yOS]) when comparing to historical controls (SCHOLAR-1 study; 1yOS, 28%) and RR-LBCL patients treated with CAR-T cell therapy in a single-arm trial (ZUMA-1; 1yOS, 59%) as well as deviations in selected adverse events (grade >= 3 aphasia) from the ZUMA-1 trial. We assumed a 10% absolute deviation in 1yOS (efficacy) and a relative increase of 50% in grade >= 3 aphasia (safety). Results: Assuming a general population of 5, 15, and 25 million, the accrual time needed to achieve 80% power for detecting a significant increase over the 1yOS reported in SCHOLAR-1 was 9, 4, and 3 years, respectively, while 80% power for detecting a significant decrease in 1yOS compared to ZUMA-1 required 10.5, 4.5, and 3 years of data accrual, respectively. However, corresponding estimates for aphasia were >20, 8, and 5 years, respectively. Conclusions: Projections of the statistical power for detecting important deviations in efficacy/safety from that reported in pivotal clinical trials(s) provide critical information about the expected performance of post-authorization monitoring programs.