Gemcitabine–Cisplatin With or Without Durvalumab in Advanced Gallbladder Cancer: A Real-World Ambispective Cohort Study from a High-Incidence Region

Purpose: Gallbladder cancer is an aggressive malignancy with poor outcomes in advanced disease. Evidence specific to gallbladder cancer from routine clinical practice remains limited. We compared outcomes among patients treated with gemcitabine–cisplatin (GemCis) with or without durvalumab. Methods: This single-centre ambispective observational cohort included 98 patients with unresectable, recurrent, or metastatic gallbladder cancer treated from January 2023 through June 2025. Forty-eight received GemCis alone and 50 received GemCis plus durvalumab; treatment selection was based on drug availability and patient preference after counselling. Follow-up continued through December 2025. Progression-free survival (PFS) and overall survival (OS) were the primary endpoints. Results: Median PFS was 6.1 months with GemCis plus durvalumab and 4.1 months with GemCis alone (unadjusted HR 0.55, 95% CI 0.36–0.85; p = 0.007). Median OS was 14.2 and 9.7 months, respectively (unadjusted HR 0.58, 95% CI 0.35–0.99; p = 0.041). In 6-month landmark analyses, the PFS difference persisted (HR 0.39, 95% CI 0.19–0.81) but the OS difference did not (HR 1.15, 95% CI 0.61–2.17). In multivariable Cox models, the adjusted HR was 0.55 (95% CI 0.35–0.85) for PFS and 0.58 (95% CI 0.34–0.98) for OS. Objective response rates were 30.0% (15/50) and 18.8% (9/48), and disease control rates were 56.0% (28/50) and 39.6% (19/48), respectively. The response-rate differences were not statistically significant (Fisher’s exact p = 0.243 and p = 0.112, respectively). Immune-related adverse events were reported in 20% of patients receiving durvalumab, including grade ≥3 events in 2%. Conclusion: In this nonrandomized cohort, durvalumab-containing treatment was associated with numerically longer PFS and OS, but the OS difference was not maintained in landmark analysis. Residual confounding, adjustment limited to measured baseline factors, possible bias related to maintenance exposure, and unavailable systematic biomarker and subsequent-treatment data limit causal interpretation. The findings are hypothesis-generating.