A systematic search was conducted in the PubMed, EMBASE, Cochrane Library, and Web of Science databases. The search terms included “head and neck squamous cell carcinoma”, “Squamous Cell Carcinoma of the Larynx”, “randomized clinical trial”, “immune checkpoint inhibitors”, “PD-L1 inhibitor”,”PD-1 inhibitor”, “CTLA-4 inhibitor”, “pembrolizumab”, “camrelizumab”, “nivolumab”, “ipilimumab”, “durvalumab”, “tremelimumab”, “toripalimab” and “tislelizumab”. ( Supplementary Table 2 ). The search covered publications from database inception until September 1, 2024, utilizing a combination of free-text and MeSH terms, without language restrictions.

Inclusion Criteria:

OS, defined as the time from randomization to death from any cause; progression-free survival (PFS), defined as the time from randomization to disease progression or death from any cause; ORR, defined as the proportion of patients achieving an objective response; and grade 3 or higher AEs (AEs ≥3) as defined by the Common Terminology Criteria for Adverse Events (CTCAE) of the National Cancer Institute.

Exclusion Criteria:

RCTs were screened based on titles and abstracts before inclusion. All included RCTs were double-checked by two reviewers to ensure that the data were the most recently published.

Three researchers independently extracted data from the RCTs in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Any discrepancies were resolved through discussion with a fourth author. The data extracted from each article included the trial name, NCT number, publication journal, randomization ratio, year of publication, trial phase, tumor stage, histological type, sample size, patient age and gender distribution, racial composition, PD-L1 expression status, Eastern Cooperative Oncology Group (ECOG) performance status, and treatment regimens for both the experimental and control groups. The outcomes extracted included hazard ratios (HRs) with 95% confidence intervals (CIs) for OS and PFS, and odds ratios (ORs) with 95% CIs for ORR and AEs ≥ 3.

The quality of the included RCTs was assessed using the Cochrane Risk of Bias Tool (2.0). This tool evaluates five domains: risk of bias arising from the randomization process, risk of bias due to deviations from the intended interventions, risk of bias from missing outcome data, risk of bias in the measurement of the outcome, and risk of bias in the selection of the reported result. Each RCT was categorized into one of three risk levels: low risk, high risk, and having “some concerns.”

The primary outcomes are OS and PFS, while the secondary outcomes include ORR and AEs≥3. The effect size for OS and PFS is expressed as HRs with 95% CIs, and the effect size for ORR and grade 3 or higher AEs is expressed as ORs with 95% CIs.

A NMA was conducted using a Bayesian model in R software, employing the “rjags” and “gemtc” packages to evaluate the efficacy and safety of immunotherapy in first and second-line treatments for R/M HNSCC (17). A random-effect model was used, establishing three independent Markov chains, each with 20,000 burn-ins and 100,000 sample iterations, with a thinning interval for each chain. The overall ranking probability of different treatment regimens’ efficacy and safety was derived from the Markov chain iterations, and results were visualized through graphical representations. Funnel plots were created using STATA 18.0 software to assess publication bias. To verify the accuracy of indirect comparisons in the NMA, a pairwise meta-analysis based on frequentist methods was conducted comparing head-to-head studies and NMA indirect comparisons ( Supplementary Table 3 ).

Additionally, Revman 5.4 software was used to perform a pairwise meta-analysis based on frequentist methods, aiming to re-evaluate the efficacy and safety of first-line or second-line immunotherapy versus chemotherapy alone in R/M HNSCC patients with and without PD-L1 expression. Heterogeneity was assessed using the Q-test and I² statistic, with I² ≤ 50% or P ≥ 0.1 indicating low heterogeneity, and I² > 50% or P < 0.1 indicating high heterogeneity. For studies with high heterogeneity, a random-effects model was applied; otherwise, a fixed-effect model was used. Sensitivity analyses were performed on high heterogeneity studies by sequentially excluding studies from the model and comparing heterogeneity changes before and after exclusion to ensure result reliability. If significant heterogeneity changes were observed, the sources of heterogeneity were analyzed. Subgroup analyses were conducted based on OS and PFS outcomes for different PD-L1 positive patients receiving first-line/second-line treatment, comparing the efficacy of various immunotherapy regimens versus chemotherapy. The significance level was set at α = 0.05.

In the initial literature search, a total of 769 records were identified from the databases. After screening the abstracts to remove duplicates and irrelevant articles, 554 studies remained eligible for full-text review. Among these, 497 were excluded due to irrelevance (n = 226), review/meta-analyses (n = 259), or non-English publications (n = 12). Of the remaining 57 reports, 48 were excluded during eligibility assessment for reasons such as non-RCT designs (n = 21), study protocols (n = 12), duplicate clinical trials (n = 7), or inappropriate control groups (n = 8). Ultimately, 9 studies met our eligibility criteria ( Figure 1 ), enrolling a total of 5,946 patients who received any of the following 8 treatments: nivolumab plus ipilimumab (nivo-ipi), durvalumab plus tremelimumab (durva-treme), durvalumab (durva), pembrolizumab (pem), pembrolizumab plus chemotherapy (pem-chemo), nivolumab (nivo), tremelimumab (treme), and standard of care (soc). Detailed information on all included studies is provided in Tables 1 , 2 , and Supplementary Table 4 .

Quality assessment using the ROB 2.0 tool showed that 5 of the 9 included studies were evaluated as low risk, while 4 were classified as having some concerns. CheckMate 141 has certain variations in baseline patient data, leading to ‘some concerns’ regarding deviations from intended interventions. CheckMate 141, CheckMate 651, EAGLE, Keynote-048, and Keynote-040 are all open-label studies that were not double-blinded. Additionally, more than 10 patients in each study either withdrew or were lost to follow-up, thereby raising concerns regarding potential deviations from the intended interventions. Overall, all the RCTs were meticulously designed, demonstrating a high level of research quality. The specific assessment results are detailed in Supplementary Figures 1 , 2 .

Ranking analysis based on Bayesian ranking profiles was conducted ( Figures 10 – 12 ; Supplementary Tables 6 – 9 ). Among patients with R/M HNSCC without PD-L1 selection, nivolumab was most likely to rank first for OS with a cumulative probability of 64.17%.Pem-chemo ranked first for PFS (69.42%) and ORR (35.06%), while pembrolizumab monotherapy was most likely to rank first for grade ≥3 AEs (56.42%). Notably, pem-chemo demonstrated the best efficacy in first-line treatment, ranking first for both PFS and ORR, and second for OS.

For second-line treatment in R/M HNSCC patients, nivolumab was most likely to rank first for OS (89.91%), PFS (75.10%), and ORR (76.94%), while also ranking first for grade ≥3 AEs (37.14%). In second-line treatment, nivolumab exhibited excellent performance in both efficacy and safety, ranking first across all key outcomes: OS, PFS, ORR, and grade ≥3 AEs.

Among R/M HNSCC patients with PD-L1 expression ≥1%, nivolumab as a first-line treatment was most likely to provide the best OS benefit (63.37%), while pem-chemo was most likely to offer the greatest PFS benefit (91.76%). In second-line treatment, nivolumab demonstrated the most favorable outcomes for both OS (91.76%) and PFS (98.96%).For patients with PD-L1 expression ≥20%, pem-chemo was the most likely first-line therapy to provide the best OS (49.65%) and PFS benefit (91.42%).

The results of the pairwise meta-analysis based on the frequentist approach are consistent with the corresponding aggregated results within the Bayesian framework ( Supplementary Table 3 ). Heterogeneity was assessed using the Q test and I² statistics, indicating high heterogeneity with I² > 50% ( Figures 2 , 3 ). After sequential exclusion of individual studies, the heterogeneity did not significantly decrease, suggesting the reliability of the conclusions. A funnel plot was used to analyze publication bias with OS as the outcome indicator, showing a symmetrical distribution of study points without scattered distribution, indicating a low likelihood of publication bias in this study ( Supplementary Figures 15 , 16 ).

Although this study draws several important conclusions, it is important to acknowledge a few limitations. First, there were variations in the SOC regimens used across the different control groups. For instance, the CheckMate 651 and KESTREL trials utilized cisplatin or carboplatin plus 5-fluorouracil in combination with cetuximab, while CheckMate 141 employed methotrexate, docetaxel, or cetuximab. While both regimens are recognized as standard first-line treatments, these differences may have introduced some degree of bias into the results. Secondly, as mentioned earlier, less than 1% of the included patients were Black, and fewer than 3% were Hispanic. The applicability of the study’s conclusions to Black individuals or other minority populations requires further consideration. Third, despite our efforts to include all relevant RCTs investigating immunotherapy for R/M HNSCC, the limited number of available trials means that some interventions were represented by only a single RCT, which may restrict the robustness of our conclusions. Fourth, as only two studies reported the safety outcomes of combination therapy compared to SOC, the finding that dual ICIs significantly reduced the incidence of grade ≥3 AEs compared to SOC should be interpreted with caution. Fifth, the included RCTs employed different methods for assessing PD-L1 expression, with some using the tumor proportion score (TPS) and others using the combined positive score (CPS). Given that CPS provides a more comprehensive reflection of the tumor microenvironment and PD-L1 expression status, it is generally preferred. However, in trials such as KESTREL or CheckMate 714, which only reported TPS, we were constrained to using TPS for PD-L1 evaluation. Finally, head and neck cancers comprise a diverse array of subtypes, such as oropharyngeal cancer, hypopharyngeal cancer, and laryngeal cancer. While survival outcomes may differ among these subtypes, the limited number of studies precludes subgroup analyses, necessitating cautious interpretation of the findings.

Despite these limitations, this study offers a thorough and comprehensive summary of randomized controlled trials on first- and second-line immunotherapy for R/M HNSCC.