The study protocol was registered on PROSPERO (CRD42023452422). This meta-analysis was carried out in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria. A systematic search of the PubMed, Cochrane Library, Web of Science, Google Scholar, and Embase databases was performed to identify and retrieve literature in English that had been published from the time the databases were created until July 17, 2023. The following terms were used to search the databases to identify relevant literature: “biliary tract cancers”, “bile duct neoplasm”, “bile duct carcinoma”, “gallbladder carcinoma”, “cholangiocarcinomas”, “PD1”, “PDL1”, and “immunotherapy”. The American Society of Clinical Oncology (ASCO) and the European Society of Medicine Oncology (ESMO) reference lists for reviews and conference materials, such as abstracts and posters, up to June 31, 2023, were also manually searched.

The following were the study inclusion requirements: (1) clinical studies with human patients included; (2) patients pathologically diagnosed with BTC were treated with anti-PD1/PDL1 together with immunotherapy or other anti-PD1/PDL1, either combined with chemotherapy/radiotherapy or not; (3) progression-free survival (PFS), overall survival (OS), ORR, and adverse events (AEs) were reported; (4) quality studies based on the Newcastle–Ottawa Scale (NOS), with studies having a medium or high quality rating.

The following were the exclusion requirements: (1) editorials, letters, reviews, meta-analyses, protocols, and case reports; (2) either no comprehensive findings were supplied or the results were not clear; and (3) duplicate research. After identifying the initial papers, a full-text check was conducted to examine whether the papers met the inclusion and exclusion criteria. Two independent researchers performed the above processes, and their search results were consistent.

We used the NOS as a quality assessment indicator since most of the relevant research involved retrospective or single-arm studies. Studies with an NOS rating of 7–9 were considered high quality, while those with an NOS rating of 4–6 were considered medium quality.

The target data were taken from the retrieved publications that were included in the study by two independent researchers. If disagreements existed, the other authors of this study collaborated to find solutions and reach agreement. The following information was taken out of the studies: names of the authors, year of publication, country, patient ages, total case numbers, the number of patients with ICC, ECC, and GC, the treatment regimens, the treatment line, and the median PFS, OS, and incidence of grade 3–5 AEs. If the PFS or OS was not described in detail (for example, the range of the OS was missing), the data were taken from Kaplan–Meier (K–M) curves using Engauge Digitizer V11.4.

The PFS, OS, and treatment-related toxicity (AEs) were the primary endpoints. The above endpoint proportions were pooled and analyzed. The I² value was used to assess the heterogeneity between studies. A random effect result was used with an otherwise fixed-effect outcome. I² > 50% was deemed significantly heterogeneous. Sensitivity analysis was performed to find potential studies that could cause significant heterogeneity. A visual examination by funnel plot and quantitative analysis utilizing Egger’s test of the intercept were used to evaluate publication bias. All the statistical analyses were conducted with Stata (Version 14). A p-value of 0.05 was considered statistically significant.

After searching the aforementioned databases, 233 studies were initially identified. Among these, 28 studies were duplication studies, 10 studies were not clinical studies, 123 studies were deemed irrelevant, 52 studies had a sample size of <10, and 5 studies had no clinical outcomes, and these were excluded from our meta-analysis. Finally, 15 studies remained with 665 patients and these were included in our meta-analysis (16–30). Supplementary Figure 1 shows the inclusion process flowchart for the studies.

Among the 15 included studies, there were 5 prospective studies (no randomized controlled trial) (16–18, 24, 25) and 10 retrospective studies (19–23, 26–30). All the studies underwent quality appraisal according to the NOS ( Supplementary Table 1 ), with 13 ranked as medium quality, and 2 as high quality. Finally, the 15 studies all passed the quality inclusion criterion and were thus all included in the meta-analysis.

The 15 included studies involved 665 patients in total, among whom 660 were finally included in the efficacy analysis in this meta-analysis, while the other 4 patients were excluded from the PFS and OS analysis but included in the adverse events analysis. The overall patient characteristics are presented in Table 1 . Only 12 of the studies reported the patients’ median age, for which the pooled median age was 59.7 years old (range 25–82). Among the 661 patients, 434 (65.7%) patients had ICC, 43 (6.5%) ECC, and 184 (27.8%) GC. Besides ICI combination therapies, 177 (26.8%) patients received concurrent chemotherapy, 51 (7.7%) concurrent radiotherapy, and 23 (3.4%) locoregional therapy, which were not described in detail. In total, 251 patients (38.0%) received other treatments besides ICI. Moreover, 367 (55.5%) patients received ICI combination therapies as first-line treatment, and 298 (44.5%) received the therapies as non-first-line treatments.

The main outcomes are presented in Table 2 . All the studies, except that of Lei et al. (19), reported tumor responses (n = 535). Overall, only 3 patients (0.6%) achieved a complete response (CR), while 182 (34.0%) achieved a partial response (PR), and 230 patients (43.0%) had a stable disease (SD) state. The overall ORR and disease control rate (DCR) were 34.6% and 77.6%, respectively.

Twelve studies reported the median follow-up periods, which ranged from 9.8 to 31.9 months. The overall median PFS was 6.06 months [95% confidence interval (CI): 4.91–7.21] ( Figure 1A ). Only 7 studies reported the 6-month PFS, which was 49.6%. Also, only 5 studies reported the 12-month PFS, which was 20.1%. The median OS was 12.11 months (95% CI: 10.66–13.55) ( Figure 1B ). The 6-month OS was reported in 4 studies, and it was 87.7%. The 12-month OS was reported in 9 studies, and it was 51.5%.

To analyze the effect of other treatments besides ICI combination therapy, 12 studies were included in the analysis. The study by Ding et al. (26) did not separately report the clinical outcomes of the patients who underwent other treatments or did not. Among the patients who were treated with ICI combination therapy alone, the estimated median PFS was 4.53 months (95% CI: 3.68–5.39), and the estimated median OS was 11.29 months (95% CI: 9.72–12.87). Among the patients who were treated with ICI combination therapy combined with other treatments, the estimated median PFS was 8.19 months (95% CI: 7.16–9.21) and the estimated median OS was 14.56 months (95% CI: 11.10–18.01). The median PFS and OS were significantly longer in patients treated with ICI combination therapy combined with other treatments (z = 9.69, p < 0.001, and z = 16.17, p < 0.001) ( Figures 2A, B ). The ORR and DCR for patients who were treated with ICI combination therapy combined with other treatments were 42.4% and 90.6%, respectively, while the ORR and DCR for patients who were treated with ICI alone were 28.7% and 71.3%, respectively. The ORR and DCR were thus significantly higher for patients who received other treatments as well (z = 18.3, p < 0.001, and z = 20.42, p < 0.001) ( Table 3 ).

To analyze the effect of the treatment lines, 12 studies were included in the statistical analysis that reported data on the treatment lines. The study by Cousin et al. (16) did not report the first-line and non-first-line clinical outcomes separately, and so was not included in the analysis. Among the patients who were treated as first line in the included studies, the estimated median PFS was 7.55 months (95% CI: 5.82–9.28) and the estimated median OS was 14.7 months (95% CI: 11.7–17.7). Among the patients who were treated as non-first line, the estimated median PFS was 5.47 months (95% CI: 3.85–7.09) and the estimated median OS was 11.54 months (95% CI: 9.26–13.82). The median PFS and OS were significantly longer in patients treated as first-line patients compared with non-first-line patients (z = 11.19, p < 0.001, and z = 14.17, p < 0.001) ( Figures 2C, D ). The ORR and DCR for patients who were treated as first-line patients were 47.6% and 81.1%, respectively. The ORR and DCR for patients who were treated as non-first-line patients were 26.6% and 78.1%, respectively. The ORR was significantly higher for patients who were treated as first line (z = 19.45, p < 0.001). However, the DCR showed no significant difference between the first-line and non-first-line patients (z = 1.42, p = 0.407) ( Table 3 ).

The study by Wang et al. (22) only reported the number of adverse events (AEs) that occurred, and it was not possible to ascertain the specific number of patients who experienced AEs. Therefore, the remaining 14 studies were included in the safety analysis.

The overall pooled grade ≥3 AEs rate was 38.2% (95% CI: 0.268–0.497) ( Figure 3A ). In patients who underwent concurrent chemotherapy or radiotherapy, the grade ≥3 AEs rate was 43.0% (95% CI: 0.22–0.63) and the grade ≥3 AEs rate in those who did not receive additional treatment was 36% (95% CI: 0.26–0.50). The grade ≥3 AEs rates showed no significant difference between the 2 groups (p = 0.10) ( Figure 3B ). Likewise, the treatment line did not influence the grade ≥3 AE rate (first line 36.7% vs. non-first line 39.0%, p = 0.24) ( Figure 3C ).

Sensitivity analysis was performed leave one-out approach after the exclusion of studies with a high potential risk of publication bias from the meta-analyses. The stabilities of the PFS, OS, and AEs data were evaluated. According to the results of the sensitivity analysis conducted by STATE, 3 studies (16, 17, 28) were removed at a time to assess the PFS, and the median PFS was found to be 6.60 (95% CI: 5.44–7.75), and the results were comparable to the overall median PFS (6.60 vs. 6.06 months, p = 0.36).

Two studies (17, 18) were removed at a time to estimate the median OS, and the result was 11.84 months (95% CI: 10.61–13.08). The results were comparable to the overall median PFS (11.84 months vs. 12.11 months, p = 0.43).

Four studies (22, 28–30) were removed at a time to estimate the grade ≥3 AEs rates, and the result was 33.7% (95% CI: 0.22–0.43). The results were comparable to the overall grade ≥3 AEs rate (38.2% vs. 38.2%, p = 0.09).

The sensitivity analysis confirmed the stability of our findings.

Egger’s tests were conducted for the median PFS, median OS, and grade ≥3 AEs rates ( Supplementary Figure 2 ). The median PFS (p = 0.211) and median OS (p = 0.108) showed no publication bias. However, a possibility of publication bias (p = 0.016) was identified for the grade ≥3 AEs rates.