Bayesian reanalysis of the NIAGARA trial using reconstructed individual patient data: perioperative durvalumab in cisplatin-eligible muscle-invasive bladder cancer

OBJECTIVES: To provide decision-focused estimates of the benefit of perioperative durvalumab in cisplatin-eligible muscle-invasive bladder cancer using Bayesian modeling of reconstructed individual patient data (rIPD) from the phase 3 NIAGARA trial. PATIENTS AND METHODS: Individual patient data were reconstructed from published Kaplan-Meier curves and event counts (durvalumab, n = 533; control, n = 530). Bayesian piecewise-exponential models with prespecified intervals (0-12, 12-24, 24-36, and 36-60 months) were used to estimate interval-specific hazards, posterior probabilities of benefit, absolute survival differences, number needed to treat (NNT), and restricted mean survival time (RMST). RESULTS: For event-free survival (EFS), the overall hazard ratio (HR) was 0.68 (95% credible interval [CrI], 0.56-0.82). The absolute improvement in EFS with durvalumab was 6.3%, 8.5%, and 10.8% at 12, 24, and 36 months, respectively, corresponding to median NNT values of 15.8, 11.7, and 9.2. The gain in RMST for EFS was 2.48 months over 36 months and 5.98 months over 60 months. For overall survival (OS), the overall HR was 0.74 (95% CrI, 0.59-0.93). The absolute improvement in OS was 2.7%, 7.3%, and 7.5% at 12, 24, and 36 months, respectively, with a median NNT of 13.8 at 24 months. The gain in RMST for OS was 1.66 months over 36 months and 3.31 months over 60 months. CONCLUSION: This Bayesian rIPD reanalysis complements the primary NIAGARA report by translating relative effects into clinically meaningful absolute benefits over prespecified perioperative time points, supporting perioperative counseling and institutional decision-making.