Trimodal therapy versus radical cystectomy in muscle-invasive bladder cancer: The overlooked influence of informative censoring—A systematic review and meta-analysis.

Abstract

729 Background: Informative censoring (IC) is a statistical bias that occurs when dropout rates differ between study arms and the mechanism of dropout is not random but related to unmeasured or unaccounted-for factors. Comparative analyses of trimodal therapy (TMT) versus radical cystectomy (RC) in muscle-invasive bladder cancer (MIBC) have produced conflicting results, potentially reflecting such bias. We conducted a systematic review and meta-analysis to quantify the potential impact of informative censoring on reported oncological outcomes in studies comparing these two treatment modalities. Methods: A comprehensive literature search was performed to identify studies including patients with cT2–T4, any N, M0 MIBC treated with TMT or RC. Eligible studies were evaluated for the risk of informative censoring affecting oncological endpoints. Published Kaplan–Meier curves were digitally reconstructed to extract time-to-event data. The inverse Kaplan–Meier method and log-rank test were applied to detect evidence of IC. A simulation analysis was performed to estimate the minimum proportion of events among censored patients required to offset the potential bias introduced by IC. Results: Six studies encompassing 8,594 patients were included in the analysis; only four utilized propensity score matching. Censoring imbalances were detected in 10 of 12 reported outcomes, favoring RC in five cases and TMT in five. Overall, censoring patterns favored RC for overall survival (OS), metastasis-free survival, and disease-free survival (log-rank, all p < 0.01). These results remained consistent despite stratification by propensity score matching and neoadjuvant chemotherapy use. For OS, the minimum event proportion required to offset the influence of IC ranged from 16% to 33%. After adjustment for the potential presence of IC, RC retained a significant survival advantage over TMT (5-year OS: 42% vs. 30%; p < 0.001). Conclusions: Most retrospective studies comparing TMT and RC are at high risk of IC, which tends to bias results in favor of RC. RC was associated with superior oncological outcomes even after accounting for this bias. These findings underscore the methodological limitations of comparing TMT and RC in non-randomized settings and call into question the acceptance of TMT as an equivalent alternative to RC based on retrospective data alone.