Introduction
Advanced biliary tract cancer (BTC)—encompassing intrahepatic cholangiocarcinoma, extrahepatic cholangiocarcinoma, and gallbladder cancer—carries a dismal prognosis, with historically limited systemic options beyond gemcitabine plus cisplatin (GemCis). The emergence of immune checkpoint inhibitors targeting programmed death-1 (PD-1) and programmed death-ligand 1 (PD-L1) has transformed the first-line landscape. Two pivotal phase 3 trials—TOPAZ-1 evaluating durvalumab (Imfinzi, anti–PD-L1) and KEYNOTE-966 evaluating pembrolizumab (Keytruda, anti–PD-1), both combined with GemCis—have established chemoimmunotherapy as a global standard of care. Both regimens carry FDA approval, NCCN Category 1 Preferred designation, and dual ESMO Level I, Grade A recommendations 312. This narrative review synthesizes trial design, efficacy, safety, and biomarker evidence to inform practical first-line treatment selection, with relevance to US, EU, and China clinical practice.
Trial Design
TOPAZ-1 enrolled 685 patients with histologically confirmed unresectable, locally advanced, or metastatic BTC—including intrahepatic and extrahepatic cholangiocarcinoma and gallbladder cancer—across 105 centers in 17 countries. Ampullary carcinoma was explicitly excluded. Eligible patients were previously untreated at the unresectable/metastatic stage, or had disease recurrence at least 6 months after curative surgery; Eastern Cooperative Oncology Group (ECOG) performance status 0–1 was required. The treatment regimen consisted of durvalumab 1500 mg or placebo every 3 weeks alongside GemCis for up to 8 cycles, followed by durvalumab or placebo maintenance administered every 4 weeks until progression or unacceptable toxicity 87.
KEYNOTE-966 enrolled 1,069 previously untreated patients with locally advanced unresectable or metastatic BTC at 175 global centers, stratified by region (Asia vs. non-Asia), disease stage, and primary tumor site. Pembrolizumab 200 mg or placebo was administered every 3 weeks; cisplatin was capped at 8 cycles, while gemcitabine could be continued beyond 8 cycles at investigator discretion, and pembrolizumab was permitted for up to 35 cycles (approximately 24 months) 19. This design difference—flexible gemcitabine continuation and time-limited immunotherapy in KEYNOTE-966 versus fixed 8-cycle GemCis and ongoing durvalumab maintenance in TOPAZ-1—represents a meaningful operational distinction that limits direct efficacy cross-comparison.
Both trials designated overall survival (OS) as the primary endpoint, with progression-free survival (PFS), objective response rate (ORR), and duration of response (DoR) as key secondary outcomes 79.
Efficacy Outcomes
Table 1. Head-to-Head Summary of Key Efficacy and Safety Endpoints
| Domain | TOPAZ-1: Durvalumab + GemCis | KEYNOTE-966: Pembrolizumab + GemCis | Practical Interpretation |
|---|---|---|---|
| Trial population | 685 patients; unresectable/metastatic BTC; 17 countries; ampullary excluded; ECOG 0–1 8 | 1,069 patients; unresectable locally advanced/metastatic BTC; 175 global sites; stratified by region, stage, site 9 | Both enrolled broad first-line BTC; KEYNOTE-966 larger and more globally diverse |
| Treatment design | Durvalumab 1500 mg Q3W + GemCis x8 cycles; durvalumab maintenance Q4W thereafter 8 | Pembrolizumab 200 mg Q3W; cisplatin capped at 8 cycles; gemcitabine allowed >8 cycles; pembrolizumab up to 35 cycles 1 | Flexible chemotherapy in KEYNOTE-966 vs. fixed backbone in TOPAZ-1 may affect cumulative cytopenia and cross-trial PFS comparison |
| Overall survival | Updated median OS 12.9 vs 11.3 months; HR 0.76; 24-month OS 23.6% vs 11.5% 19 | Median OS 12.7 vs 10.9 months; HR 0.83 (95% CI 0.72–0.95; P=0.0034) 920 | Both statistically significant; numeric HR difference is cross-trial only and must not be interpreted as evidence of superiority |
| Progression-free survival | PFS HR 0.75 (95% CI 0.63–0.89) 7 | Median PFS 6.5 vs 5.6 months; HR 0.86 (95% CI 0.75–1.00; NS) 21 | TOPAZ-1 achieved a significant PFS HR; KEYNOTE-966 showed a modest, non-significant improvement; different chemotherapy duration may partly explain this |
| Response outcomes | ORR 26.7% vs 18.7%; median DoR 6.4 vs 6.2 months 8 | ORR 29% vs 29%; median DoR 8.3 vs 6.8 months 21 | Durvalumab improved ORR vs. control; pembrolizumab did not raise ORR but prolonged DoR among responders |
| Grade ≥3 toxicity | Grade 3/4 AEs 74–75% vs 75–78%; discontinuation due to TRAEs 8.9% vs 11.4% 198 | Grade 3–5 TRAEs 71.5% vs 69.3%; deaths due to AEs 6% vs 9%; treatment-related deaths 2% vs 1% 920 | Overall high-grade burden similar and largely chemotherapy-driven; both manageable with standard protocols |
| Immune-mediated toxicity | imAEs in 13.9% vs 4.7%; mostly low-grade; hypothyroidism most common; grade 3/4 imAEs 2.4%; OS HR preserved at 0.59 in patients with imAEs 10 | imAEs in ~24% vs ~8.5%; mostly grade 1–2; class-effect profile including pneumonitis, colitis, hepatitis, endocrinopathies 7 | Class-effect risks for both; early recognition and corticosteroid-based intervention preserve benefit; rare severe events |
| Biomarker/subgroup signals | Benefit consistent across PD-L1 expression, primary site, region, ECOG, age, race, disease status; exploratory IDH1/BRAF/BRCA signals are hypothesis-generating only 87 | Benefit observed across PD-L1 levels; HBV-positive (~31%) and Asian populations (~46%) included; ESMO: no validated biomarker favors one ICI over the other 31 | PD-L1 is not a validated treatment-selection marker between these regimens; subgroup CIs are wide; MSI-H/dMMR is rare (~1–2%) in BTC 22 |
Abbreviations: BTC = biliary tract cancer; GemCis = gemcitabine-cisplatin; OS = overall survival; HR = hazard ratio; NS = not statistically significant; PFS = progression-free survival; ORR = objective response rate; DoR = duration of response; TRAE = treatment-related adverse event; imAE = immune-mediated adverse event; ICI = immune checkpoint inhibitor; ECOG = Eastern Cooperative Oncology Group; HBV = hepatitis B virus. Cross-trial numeric comparisons are descriptive only; TOPAZ-1 and KEYNOTE-966 were not conducted head-to-head.
Overall Survival
Both regimens provide a clinically meaningful OS benefit over GemCis alone. The updated TOPAZ-1 analysis, after median follow-up of approximately 23 months, reported a median OS of 12.9 versus 11.3 months (HR 0.76), with 24-month OS rates of 23.6% versus 11.5%—a landmark signal suggesting durable long-tail benefit 19. The 2025 TOPAZ-1 update further indicated that more than twice as many patients in the durvalumab arm were alive at 3 years compared with the chemotherapy-only arm 16. KEYNOTE-966 demonstrated a median OS of 12.7 versus 10.9 months (HR 0.83; 95% CI 0.72–0.95; one-sided P=0.0034), meeting the prespecified statistical threshold 920. A cross-trial meta-analysis using restricted mean survival time estimated that both regimens provide approximately 1.2 additional months of mean OS at 24 months versus GemCis alone, with virtually identical effect magnitude across the two studies 12. This convergence supports treating the two regimens as parallel standards rather than ranking them on efficacy.
Progression-Free Survival and Response
TOPAZ-1 achieved a statistically significant PFS hazard ratio of 0.75, while KEYNOTE-966 reported a PFS hazard ratio of 0.86 (95% CI 0.75–1.00), which was not statistically significant 721. This divergence likely reflects, at least in part, KEYNOTE-966's allowance of gemcitabine continuation beyond 8 cycles—an important design distinction. For ORR, durvalumab improved the response rate compared with its control arm (26.7% vs. 18.7%), whereas in KEYNOTE-966, both arms showed an identical ORR of 29%, though pembrolizumab-treated responders showed a longer median DoR (8.3 vs. 6.8 months) 821. These patterns illustrate complementary but distinct response profiles that should not be interpreted as superiority for either agent.
Safety and Tolerability
The toxicity profiles of both combinations are dominated by GemCis-related hematologic toxicity—neutropenia, anemia, and thrombocytopenia—with immunotherapy-related adverse events superimposed as a class effect. In TOPAZ-1, grade 3/4 adverse events occurred in approximately 74–75% of durvalumab-treated patients versus 75–78% in the placebo arm, with treatment discontinuation due to treatment-related adverse events (TRAEs) of 8.9% versus 11.4% 198. Immune-mediated adverse events (imAEs) occurred in 13.9% of durvalumab recipients versus 4.7% in controls; the most common was hypothyroidism; grade 3/4 imAEs were rare (2.4%) 10. Crucially, a post hoc TOPAZ-1 analysis confirmed that the OS benefit was preserved in patients who experienced imAEs (OS HR 0.59 for imAE-positive patients) 10, supporting continuation of therapy with appropriate management.
In KEYNOTE-966, grade 3–5 TRAEs occurred in 71.5% versus 69.3% of patients, treatment-related deaths in 2% versus 1%, and imAEs in approximately 24% of pembrolizumab-treated patients versus 8.5% of controls, predominantly grade 1–2 79. No new safety signals emerged in either trial. Both agents carry the standard class-effect warning profile for PD-1/PD-L1 inhibitors: pneumonitis, colitis, hepatitis, endocrinopathies, nephritis, and dermatologic events. Practical management centers on early recognition, prompt initiation of corticosteroids at grade ≥2 severity, and endocrine replacement when appropriate—consistent with established institutional protocols applicable to either agent.
Biomarker-Defined Subgroup Evidence
No validated biomarker currently distinguishes between durvalumab and pembrolizumab for first-line BTC selection. ESMO explicitly states that there are no discernible clinical, biochemical, or molecular markers that favor one ICI over the other 3. TOPAZ-1 reported consistent OS benefit across PD-L1 tumor area positivity scores, all primary tumor sites, all geographic regions, ECOG strata, and disease stages; exploratory analyses suggested nominal OS benefit in patients with IDH1, BRAF, or BRCA1/2 alterations (HR <1), but these are hypothesis-generating signals only and should not influence regimen selection in routine practice 8. KEYNOTE-966 similarly enrolled patients irrespective of PD-L1 and demonstrated OS benefit across prespecified stratification strata including region and primary site 1.
A meta-analysis of 30 BTC immunotherapy studies (N=1,310) found that PD-L1 positivity was associated with superior PFS (HR 0.54) and OS (HR 0.58) but did not significantly increase ORR (OR 1.56), suggesting PD-L1 functions as a prognostic or weakly predictive marker rather than a decisive companion diagnostic 22. Critically, this finding applies across PD-1 and PD-L1 agents, does not support differential prescribing between durvalumab and pembrolizumab, and is limited by high risk of bias in the majority of included studies 22. Mismatch repair deficiency/microsatellite instability-high (dMMR/MSI-H) is rare in BTC (estimated at approximately 1–2%), and while clinically significant when present, it is insufficient to serve as a population-level selection marker between these first-line backbones 223.
Regulatory Status and Treatment Selection
Both agents are FDA-approved for first-line unresectable or metastatic BTC—durvalumab approved September 2, 2022, and pembrolizumab on October 31, 2023—based on statistically significant and clinically meaningful OS improvement in their respective randomized trials 11. In China, the National Medical Products Administration (NMPA) approved durvalumab + GemCis on November 7, 2023, making it the first immunotherapy regimen approved for BTC in China 17; Both durvalumab and pembrolizumab have received approvals in China for specific hepatobiliary malignancy indications; however, the exact approved indications and reimbursement status should be verified against the most recent NMPA labeling and national reimbursement updates 15. NRDL reimbursement status was not retrievable in the available sources. ESMO recommends both combinations as first-line therapy [I, A] without preference 3. An official 2024 CSCO guideline on biliary malignancy targeted and immunotherapy was identified, though its specific recommendation text was not accessible in the retrieved materials 23.
In the absence of a differentiating biomarker or a head-to-head trial, treatment selection is guided by practical factors. TOPAZ-1's fixed 8-cycle GemCis backbone followed by durvalumab maintenance may offer operational simplicity for programs preferring a clearly defined chemotherapy endpoint. KEYNOTE-966's design—with flexible gemcitabine continuation and time-limited pembrolizumab (up to 35 cycles)—may better accommodate patients tolerating gemcitabine well, though this flexibility must be balanced against cumulative cytopenic burden. Patient-specific considerations—including history of autoimmune disease, hepatic reserve, HBV infection, current immunosuppression, infusion logistics, and payer formulary access—may reasonably influence regimen selection in individual cases. An economic analysis from the China perspective found durvalumab + GemCis cost-effective at an ICER of approximately $26,401/QALY under modeled assumptions, while the same model did not favor durvalumab on cost-effectiveness grounds in the US context 14, reinforcing the relevance of geography and health system factors in access decisions.
Conclusion
TOPAZ-1 and KEYNOTE-966 established chemoimmunotherapy-based first-line treatment as a standard approach for patients with unresectable or metastatic BTC who are candidates for GemCis-based therapy, establishing that adding a PD-1 or PD-L1 inhibitor to GemCis yields consistent, statistically significant OS improvements with manageable toxicity. The trials differ in design nuances—population size, chemotherapy duration rules, maintenance structure, and data maturity—that preclude reliable cross-trial superiority claims. Both regimens are guideline-endorsed across the US, EU, and China, and the efficacy difference in absolute survival gain is comparable by pooled RMST estimation 12. PD-L1 expression, dMMR/MSI-H status, and tumor location do not currently guide selection between these two agents. In routine practice, regimen choice should be driven by local regulatory approval, reimbursement, infusion logistics, patient comorbidities, and institutional familiarity, with close attention to immune toxicity monitoring regardless of which agent is selected.