This study was a single-arm, phase II trial assessing the safety and antitumor activity of camrelizumab plus GEMOX in patients with advanced BTC(online supplemental file 2). Eligible patients were aged 18–75 years with histologically confirmed stage IV advanced BTC including cholangiocarcinoma and gallbladder cancer. Additional inclusion criteria included an Eastern Cooperative Oncology Group performance status of 0 or 1 and presence of at least one measurable lesion assessed using the Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST version 1.1). Patients were required to have an estimated life expectancy of 12 weeks or more and adequate organ function (neutrophil count of ≥1.5×10⁹ cells/L, hemoglobin concentrations of ≥90 g/L, platelet cell count of ≥100×10⁹ cells/L, bilirubin <1.5×ULN, blood glutamate transaminase <2.5×ULN (could be extended to 5×ULN in case of liver metastases), creatinine clearance rate >60 mL/min (calculated using Cockcroft-Gault formula) and left ventricular ejection fraction ≥50%). Patients who received previous treatment with drugs specifically targeting T-cell costimulation or checkpoint pathways and those with previous or concurrent malignancies (except curatively treated skin basal cell carcinoma or cervical carcinoma in situ) were excluded from the study.

Enrolled participants received camrelizumab (3 mg/kg, total dose ≤200 mg, Intravenous drips (ivd), D1/2W) plus gemcitabine (800 mg/m², ivd, D1/2W) and oxaliplatin (85 mg/m², ivd, D2/2W). Chemotherapy lasted for no more than 12 cycles. Once chemotherapy intolerance occurred or at end of 12 cycles, patients with objective response or stable disease would continue to receive camrelizumab (3 mg/kg, total dose ≤200 mg) as single agent until confirmed disease progression, death, unacceptable toxicity, withdrawal of consent, investigator’s decision or completion of 24 months of study. We assessed response to treatment every 8 weeks using CT or MRI based on RECIST (version 1.1) criteria.

Adverse events were monitored and graded using National Cancer Institute Common Terminology Criteria for Adverse Events (version 4.0). All adverse events, from the time of treatment allocation till 90 days after cessation of treatment, were reported.

The primary endpoints were 6-month progression-free survival (PFS) rate, determined by RECIST version 1.1 and safety. Secondary end points were objective response rate (ORR), PFS, OS and duration of response. PFS was defined as the time from first drug administration to the first documented disease progression according to RECIST version 1.1 or to death from any cause, whichever occurred first. OS was defined as time from first drug administration to death from any cause. Duration of response was defined as time from first RECIST response to progression in patients who achieved a partial response or better. Exploratory endpoints included the association between survival and PD-L1 expression, tumor mutational burden (TMB), blood tumor mutational burden (bTMB), change of ctDNA and immune microenvironment.

Optional baseline biopsy specimens and blood samples were obtained from patients for exploratory biomarker assessment. Specifically, genomic DNA from tumor tissues were extracted and matched with white blood cell samples, followed by next generation sequencing and analysis of the TMB. Tissue PD-L1 expression, CD8⁺ cell, macrophages and natural killer cells level were measured by multiple immunofluorescence. The ctDNA sequencing from blood samples before and 8–10 weeks after treatment was also performed.

A total of 35 patients would provide 80% power to detect a 6-month progression-free rate of 60% at a one-sided 2.5% alpha level under the null hypothesis of the 6-month progression-free rate equals to 40%. Considering a 10% discontinuation rate, 38 assessable patients were enrolled in the study. Activity was assessed in all patients who received at least one dose of camrelizumab, had measurable disease at baseline and had one or more post-baseline scans or discontinued because of progressive disease or a treatment-related adverse event (full analysis set). Safety was assessed in all patients who received at least one dose of camrelizumab (all-patients-as-treated population).

For objective response, 95% CIs were generated using the exact binomial distribution. Patients without response data were considered as non-responders. We used the Kaplan-Meier method for estimating PFS, OS and duration of response. Exploratory, post-hoc analysis of the association between clinical response and PD-L1 expression, TMB, bTMB or tumor immune microenvironment was done in the full analysis set. For all analyzes, p value <0.05 was considered to be statistically significant, and a CI of 95% was used (95% CI). The SPSS V.22.0 software was used for statistical analysis.

Between February 2018 and April 2019, 54 patients were screened for eligibility. Of these patients, 16 (30%) were deemed ineligible because they did not meet the inclusion criteria or met the exclusion criteria (figure 1). One patient withdrew his consent before first treatment. Thirty-seven enrolled patients were treated with at least one dose of camrelizumab and were included in the primary analysis. The baseline characteristics of the enrolled participants are shown in online supplemental table S1.

As of data cut-off on August 1, 2020, the median duration of follow-up was 11.8 months (IQR 7.4–19.1) and on this date, 1 (2.7%) out of the 37 participants who had received at least one dose of camrelizumab were still receiving the treatment. The most common reasons for treatment discontinuation were progressive disease in 30 (81%) participants and adverse events in 2 (5%) participants (figure 1).

As data cut-off, 30 (81%) out of 37 participants had died or had disease progression. At 6 months, 18 out of the 37 patients were still alive and progression-free, giving a 6-month PFS of 50% (95% CI 33 to 65). The median progression-free survival (mPFS) was 6.1 months (95% CI 5.1 to 6.8; figure 2A). As data cut-off, 29 (78%) out of the 37 participants in the study had died and the mOS was 11.8 months (95% CI 8.3 to 15.4; figure 2B). The mPFS was 6.9 months in patients with gallbladder cancer versus 5.4 months in patients with cholangiocarcinoma (figure 2C). The mOS was 13.0 months in gallbladder cancer versus 11.2 months in cholangiocarcinoma (figure 2D).

An objective response was recorded in 20 (54%) out of 37 participants (95% CI 38 to 69; table 1) who had received at least one dose of camrelizumab plus GEMOX. 20 (54%) participants had partial response, 13 (35%) had stable disease and 3 (8%) had progressive disease. One patient discontinued treatment because of treatment-related adverse event before first radiographic assessment. Disease control was reported in 33 (89%; 95% CI 75 to 96) out of the 37 treated participants (table 1). The median time to response was 2.7 months (range 1.8–13.7; figure 3A). As data cut-off, 1 of the 20 responses were ongoing and the median duration of response was 4.8 months (95% CI 1.4 to 19.5 months). 30 (83%) out of 36 assessable patients had a decrease in tumor size as compared with baseline and the median change from baseline was −35% (figure 3B).

At least one treatment-related adverse event occurred in 36 (97%) of 37 participants (grade 3–4 in 26 (70%) patients; table 2). The most common treatment-related adverse events were fatigue (27 (73%)) and fever (27 (73%)). The most frequent grade 3 or worse adverse events were hypokalemia (7 (19%)), fatigue (6 (16%)), nausea, neutropenia, gamma-glutamyl transferase (GGT) increased and biliary tract infection (5 (14%) each). Two (5%) participants discontinued treatment after an adverse event, including one (3%) patient with immune-related pneumonitis and one (3%) with skin reaction.

In the prespecified exploratory analysis, we evaluated the association between tumor immune microenvironment and clinical response to camrelizumab plus GEMOX in subsets of participants with tissue for immunohistochemistry. The proportion of patients who achieved an objective response was similar in the overall cohort (20 (54%) of 37 participants) and the subgroup with available PD-L1 expression data (18 (58%) of 31 participants), as were event rates for PFS (35 (95%) of 37 participants in the overall cohort; 29 (94%) of 31 participants in the PD-L1 subgroup). Association between biomarkers and clinical response along with survival are shown in online supplemental table S2. Objective responses were recorded in 80% (4 of 5 patients) with tumor proportion scores of at least 1% versus 55.8% (14 of 26 patients) with tumor proportion scores less than 1%, and 63% (5 of 8 patients) with immune proportion scores of at least 1% versus 57% (13 of 23 patients) with immune proportion scores less than 1%. The mPFS was 9.0 months in patients with tumor proportion scores of at least 1% versus 6.0 months with tumor proportion scores less than 1% (online supplemental table S2, figure S1A) and the mOS was 17.8 months versus 11.9 months (online supplemental table S2, figure S1B). The mPFS was 8.7 months in patients with immune proportion scores of at least 1% versus 6.0 months with immune proportion scores less than 1% (online supplemental table S2, figure S1C) and the mOS was 19.9 months versus 11.1 months (online supplemental table S2, figure S1D). No difference was observed in the number of cells and in the fraction of immune cells in center or invasive margin between responders and non-responders (p>0.05; online supplemental figure S2A,B).

In another prespecified exploratory analysis, we also evaluated the association between genomic alteration and clinical response to camrelizumab plus GEMOX in subsets of participants with baseline tissue for gene sequencing. The proportion of patients who achieved an objective response was similar for the overall cohort (20 (54%) of 37) and the subgroup with available gene sequencing data (18 (53%) of 34). Similar observations were obtained for PFS (35 (95%) of 37 participants in the overall cohort; 33 (97%) of 34 participants in the gene sequencing subgroup). Objective responses for participants were recorded in 60% (6 of 10) patients with top 25% TMB versus 50% (12 of 24) patients with bottom 75% TMB (online supplemental table S2). No association was observed between TMB and PFS or OS (PFS, HR, 0.89; 95% CI 0.41 to 1.92; p=0.76; online supplemental figure S3A); OS, HR, 0.90; 95% CI 0.39 to 3.05; p=0.79; online supplemental figure S3B). The median TMB was 5.3 muts/mb in responders versus 5.7 muts/mb in non-responders (p=0.92; online supplemental figure S3C). Distribution of genetic variations associated with the response to camrelizumab plus GEMOX is depicted in figure 4. The p values for the association between each gene alteration and PFS and OS are depicted in online supplemental figure S3D). Only ARID1A showed similar trend in the association with PFS and OS. The mPFS was 4.3 months in patients with ARID1A mutation versus 6.3 months in patients without ARID1A mutation (HR, 2.39; 95% CI 1.04 to 5.48; p=0.04; online supplemental figure S3E). The mOS was 7.2 months in patients with ARID1A mutation versus 13.2 months in patients without ARID1A mutation (HR, 2.60; 95% CI 1.11 to 6.10; p=0.03; online supplemental figure S3F).

We also evaluated the association between the dynamic change of ctDNA and clinical response to camrelizumab plus GEMOX in subsets of participants with post-treatment blood samples for ctDNA gene sequencing. The proportion of patients who achieved an objective response was also similar for the overall cohort (20 (54%) of 37) and the subgroup with available ctDNA data (17 (57%) of 30). Objective responses were observed in 10 out of 13 (77%) patients with negative post-treatment ctDNA versus 7 out of 17 (41%) patients with positive post-treatment ctDNA (online supplemental table S2). bTMB was correlated with tissue TMB (online supplemental figure S4A). There was no difference in bTMB between responders and non-responders (online supplemental figure S4B). The mPFS was 4.1 months in patients with top 25% bTMB versus 6.4 months in patients with bottom 75% bTMB (HR, 2.57; 95% CI 1.08 to 6.12; p=0.03; online supplemental figure S4C). No association between TMB and OS was observed (online supplemental figure S4D). However, the mPFS was 4.3 months in patients with positive post-treatment ctDNA versus 7.3 months in patients with negative post-treatment ctDNA (HR, 2.83; 95% CI 1.27 to 6.28; p=0.007; online supplemental figure S4E). The mOS was 9.1 months in patients with positive post-treatment ctDNA versus 13.0 months in patients with negative post-treatment ctDNA (HR, 1.77; 95% CI 0.78 to 3.99; p=0.16; online supplemental figure S4F). To further demonstrate the role of post-treatment ctDNA, we further studied the association between post-treatment ctDNA and PFS in responders and non-responders. In responders, patients with positive post-treatment ctDNA also had shorter PFS than patients with negative post-treatment ctDNA (5.0 versus 8.4 months, HR, 3.22; 95% CI 1.10 to 9.40; p=0.02; online supplemental figure S5A), while in non-responders, no difference was observed due to the small sample size (online supplemental figure S5B).