Why This Study Matters
Biliary tract cancers include intrahepatic cholangiocarcinoma, extrahepatic cholangiocarcinoma and gallbladder carcinoma. According to the authors, the prognosis remains poor, with most patients succumbing to advanced disease within around a year from diagnosis. Immune checkpoint inhibitors have produced low response rates as monotherapy; the source cites the multicenter phase II KEYNOTE-158 trial of pembrolizumab, in which objective responses occurred in 5.8% of 104 patients (a cross-study benchmark, not a comparison within this trial).
GM-CSF is a cytokine that influences immune cell function. The authors point to prior work combining GM-CSF with ipilimumab in melanoma and to a phase 1 prostate cancer study in which GM-CSF was associated with more CD8+ T cells in the tumor microenvironment. This trial tested whether adding it to pembrolizumab could improve outcomes in biliary cancer.
Study Design
This was an open-label phase II study at the University of California, San Francisco Helen Diller Family Comprehensive Cancer Center. It used a Simon's 2-stage design, with an interim analysis after Stage 1 (9 patients), followed by Stage 2 (18 patients). A third part, the "Biopsy Cohort" (15 patients), was added mid-study to collect paired on-treatment biopsies. There was no comparator arm.
All patients received pembrolizumab 200 mg intravenously every 21 days for up to 35 cycles or 24 months, plus GM-CSF 250 μg self-injected subcutaneously on days 1–14 for two cycles. In Stage 1 and the Biopsy Cohort, GM-CSF was given in cycles 2 and 3; in Stage 2, in cycles 1 and 2.
42 patients enrolled; 1 center (UCSF); single arm with no randomization; GM-CSF 250 μg on days 1–14 for two cycles.
Patient Population
Between May 2016 and December 2019, 42 patients enrolled. Eligible patients had advanced or locally advanced biliary cancer not eligible for resection after progression or intolerable toxicity on at least 1 prior line of systemic therapy, and had not received prior immune checkpoint inhibitors. Intrahepatic cholangiocarcinoma accounted for 28 patients (67%), extrahepatic tumors 26% and gallbladder carcinoma 7%. Stage IV disease was present in 38 patients (91%), with a median of 2 prior systemic therapies (range 1–6). Median age was 60.5 years (range 36–80). Ten patients (24%) had active or prior hepatitis B or C infection. Median follow-up was 13.5 months (range 1.7–66.5); data lock was 1 January 2021.
Primary Endpoint Results
The primary endpoint was objective response rate by local RECIST 1.1. The study was powered to detect an improvement from 5% (H0) to 20% (H1). Confirmed responses occurred in 5 patients (12%, 95% CI: 0.04, 0.26): 3 partial and 2 complete responses. Fifteen patients (36%) had a best response of stable disease and 22 (52%) had primary progressive disease. Eleven patients (26%) achieved clinical benefit (confirmed response or stable disease of at least 6 months). Median time to response was 3.7 months (range 1.9–5.6) and median duration of response was 9.8 months (range 5.1 to not reached).
| Endpoint | Result |
|---|---|
| Objective response rate (confirmed) | 12% (5/42); 95% CI: 0.04, 0.26 |
| Median PFS | 2.1 months (95% CI: 1.8, 4.2) |
| PFS at 6 months | 26% (95% CI: 0.14, 0.43) |
| Median OS | 13.1 months (95% CI: 8.1, 19.1) |
| Median duration of response | 9.8 months (range 5.1 to not reached) |
“The addition of GM-CSF to pembrolizumab was well tolerated but did not meet the prespecified response rate for efficacy.”
— Abstract, Kelley et al., Clinical Cancer Research 2026
Subgroup Analyses
The authors state that subgroup analyses were limited by small sample sizes. These are exploratory, and several findings are post-treatment associations rather than predictors.
| Subgroup | Result (as reported) |
|---|---|
| Best response: confirmed response / stable disease / primary progression (5 / 15 / 22 patients) | Median PFS 15.4, 5.4 and 1.8 months; median OS 44.2, 13.9 and 9.3 months (chi-square p<0.0001 for each) |
| Viral hepatitis infection (n=10) vs nonviral | Median PFS 9.2 vs 2.1 months (log-rank p=0.03); median OS 34.4 vs 10.8 months (log-rank p=0.052); ORR 0.30 vs 0.06 (Fisher's exact p=0.08) |
| Any immune-related adverse event vs none | Median PFS 4 vs 1.9 months (log-rank p=0.01); median OS 19.4 vs 7.4 months (log-rank p<0.0001) |
| Immune-related endocrinopathy (n=11) vs none | ORR 36% vs 3% (Fisher's exact p=0.01); PFS 10.6 vs 1.9 months (p=0.012); OS 38 vs 9.8 months (p=0.001) |
| Tumor PD-L1 positive (11 of 24 patients tested; 46%, 95% CI: 0.26, 0.66) | Greater likelihood of 6-month PFS (Fisher's exact p=0.02); trend toward higher response rate (p=0.08) |
| Anatomic subsite, prior lines of therapy, cohort | No relationship with ORR; no difference in PFS (p=0.86) or OS (p=0.21) by anatomic subsite |
Safety Profile
The most common treatment-related event was transient leukocytosis attributed to GM-CSF, in 31 patients (74%), all grade 1 or 2. Grade 3 or 4 treatment-related events occurred in 3 patients (7%). Serious adverse events of any cause occurred in 13 patients (31%), with 4 (10%) at least possibly treatment-related. Systemic corticosteroids were needed for immune-related adverse events in 5 patients (12%), and treatment was discontinued for related adverse events in 2 patients (5%). The authors report no grade 5 treatment-related adverse events.
| Treatment-related event (any grade) | n (%) | Grade 3-4 |
|---|---|---|
| Leukocytosis | 31 (74%) | 0 |
| Injection site reaction | 16 (38%) | 0 |
| Fatigue | 14 (33%) | 0 |
| Arthralgia | 14 (33%) | 0 |
| Back/bone pain | 14 (33%) | 0 |
| Rash | 13 (31%) | 0 |
| Diarrhea | 12 (29%) | 0 |
| Fever | 11 (26%) | 0 |
| Pruritus | 11 (26%) | 0 |
| Hypothyroidism | 7 (17%) | 0 |
| Ascites | 5 (12%) | 1 (2%) |
| Diabetes mellitus with DKA | 1 (2%) | 1 (2%) |
Selected rows from Table 2 of the source article.
Interpretation and Broader Context
The authors conclude that the combination was well tolerated and produced robust, durable responses in a subset of patients but did not reach the prespecified efficacy threshold. Two patients had complete responses; one had a history of hepatitis C infection and the other had a known MSI-high tumor, and both tumors were PD-L1 positive with intermediate-to-high tumor mutation burden (13 and 38 mut/Mb).
In six patients with evaluable paired biopsies, CD8+ naive/effector T cells increased in the tumor microenvironment after GM-CSF was added (paired t-test p<0.05), with enrichment of antigen-processing pathways. Because all biopsied patients received pembrolizumab first, the authors note they could not isolate GM-CSF's effect from pembrolizumab at the same time point.
Limitations
Limitations named by the authors include the pragmatic two-cycle GM-CSF schedule (the study was not a dose-finding study), the non-randomized design without a comparator arm, heterogeneity in prior therapy, and the fact that ORR is not a validated surrogate for survival in biliary tract cancers. The trial was conducted at a single center and enrolled 42 patients, and subgroup analyses were limited by small sample sizes. Findings apply only to patients without prior checkpoint inhibitor exposure. The authors suggest further study of GM-CSF, potentially with strategies to increase tumor-associated antigen release and presentation or in primary checkpoint-inhibitor resistance.