Censoring Patterns and Inconsistent Results in Checkpoint Inhibitor and Adjuvant BCG Trials for High-Risk Bladder Cancer.

Abstract Intravesical Bacillus Calmette-Guérin (BCG) is the standard-of-care treatment for high-risk non-muscle invasive bladder cancer (HR-NMIBC). However, patients with BCG-unresponsive disease are unlikely to benefit from further BCG. To overcome BCG unresponsiveness, the addition of immune checkpoint inhibition (ICI) to BCG has been investigated in three recently randomized controlled trials: CREST, POTOMAC, and ALBAN, studying sasanlimab, durvalumab, and atezolizumab, respectively. While CREST and POTOMAC demonstrated improvements in event-free survival (EFS), ALBAN was negative. Several potential explanations have been proposed for these discrepant findings. We propose an additional explanation: censoring. In the experimental arm, toxicity-related discontinuation may preferentially remove frailer patients, artificially enriching the remaining population for healthier individuals potentially biasing EFS. We explored censoring patterns across trials. In POTOMAC and CREST, early censoring at 12 months was greater in the immunotherapy arms. By contrast, ALBAN showed no excess early censoring in the experimental arm. We conducted sensitivity analyses based on reconstructed Kaplan-Meier curves, and modifying only a few patients in POTOMAC and CREST eroded the EFS benefit as reported. Differential informative censoring may represent an important, underrecognized explanation for the inconsistent efficacy results reported across ICI plus BCG trials. More broadly, given the expanding use of immunotherapy in earlier disease settings and the reliance on time-to-event endpoints in open-label trials with differential toxicity, systematic attention to censoring mechanisms is essential for accurate interpretation.