KEYNOTE-158 (biliary tract cohort)
In previously treated advanced biliary tract cancer, what activity does pembrolizumab show in the KEYNOTE-158 cohort?
Trial design and population
Key results
ORR was 5.8% (6/104; 95% CI 2.1–12.1); median PFS 2.0 months and median OS 7.4 months. Median duration of response was not reached in the initial report, with responses lasting at least 6.2 months.
Source-reported outcome visualization
Objective response rate · Best overall response
Bars display source-reported values for the named endpoint; where the endpoint is a median, bars compare medians and do not represent probabilities. No unreported rates are inferred.
Absolute Effect
Single-arm response rate; no comparator-based absolute benefit or NNT can be calculated.
NNT: Not applicable — single-arm cohort. No NNT is derived from median survival/PFS differences.
NNH / harm: Grade 3–5 treatment-related adverse events occurred in 13.5%; one grade 5 renal failure was reported. Consider immune-mediated toxicities and biomarker-specific indications.
NNT/NNH are only displayed where the endpoint and time horizon support a defensible absolute-risk calculation. Otherwise, the source-reported outcome is visualized.
Safety and treatment burden
Grade 3–5 treatment-related adverse events occurred in 13.5%; one grade 5 renal failure was reported. Consider immune-mediated toxicities and biomarker-specific indications.
Follow-up, maturity and limitations
Unselected BTC cohort response was low. Do not conflate this result with pembrolizumab activity in MSI-H/dMMR or TMB-high tumours, nor with KEYNOTE-028.
Clinical interpretation
Current-practice context · reviewed 20 October 2026
This page separates original trial findings from contemporary treatment context. Confirm current disease-specific guidelines, local regulatory status, biomarker requirements, prior treatment exposure and patient-specific contraindications before applying the evidence. Historical comparators may no longer represent current standard practice; subgroup findings and immature endpoints should not be generalized.