Data from this meta-analysis were based on previous published papers; therefore, this study did not require ethical approval or patient consent.

The primary outcome of this meta-analysis was to investigate the prognostic value of PD-L1 for OS (defined as the time from diagnosis or initiation of treatment to patient death, irrespective of cause), PFS (defined as the time from diagnosis or initiation of treatment until tumor recurrence/progression or any-cause death), and LRC (defined as the time from diagnosis or initiation of treatment to the first loco-regional event) in patients treated by upfront surgery or (chemo)radiotherapy for OPSCC.

The secondary endpoint was the prognostic significance of PD-L1 on the above clinical outcome variables stratifying patients by type of treatment, upfront surgery or upfront (chemo) radiotherapy, and HPV status, as detected by p16 immunohistochemistry (IHC), HPV-DNA in situ hybridization or PCR.

This systematic review and meta-analysis were conducted following the preferred reporting items for systematic reviews and meta-analyses (PRISMA) checklist. Medline/PubMed (via Ovid), Cochrane, Embase (via Ovid), Web of Science (Core Collection) and Scopus were searched from inception through the end of July 2021. The research was conducted according to PRISMA criteria (36). The following keyword search was conducted: “head and neck” OR “facial” OR “mouth” OR “oral cavity” OR “oropharyngeal” OR “pharyngeal” OR “oropharynx” OR “pharynx” AND “squamous cell carcinoma*” OR “carcinoma*” OR “tumor*” OR “cancer*” OR “neoplasm*” AND “PD-L1” OR “B7-H1” OR “CD274” AND “prognosis” OR “risk” OR “recurrence*” OR “mortality” OR “survival” OR “outcome*”. The reference lists of articles included in this review as well as narrative reviews published in the last 10 years were also manually searched to minimize the risk of missing data. Two authors (PBR and EF) independently screened all titles and abstracts generated by the search and then evaluated the full texts of all the relevant articles identified against the inclusion criteria ( Figure 1 ); a third author (JP) settled discordances when present. Any disagreement between the assessors on the suitability of articles for inclusion were tackled thorough discussion between assessors, or failing this, by referral to the other authors.

Studies were included in the analysis if they met the following criteria: 1) the study reported the prognostic role of PD-L1 in primary non-metastatic OPSCC; 2) the patients were treated according to the current standard of care; 3) studies in which PD-L1 expression was detected by IHC; 4) the study reported the association of PD-L1 and patient outcome with sufficient survival data to extract hazard ratios (HRs) and 95% confidence intervals (CI). Non-English studies were excluded. Studies containing aggregated data or duplicated data from previously published work were excluded, as were review articles, case reports, editorials, and letters. Two authors (JP, PBR) independently assessed the quality of the included studies with the Newcastle-Ottawa Scale (37). Low-quality articles (Newcastle–Ottawa Scale (NOS) score <4) were also excluded.

The standard error of the log hazard ratio was derived from the log CIs. The summary HR (sHR) and corresponding 95% CI were calculated according to random-effects models of DerSimonian and Laird (38), which incorporates both within-and between-study variability, as a weighted average of the estimated HRs, by giving each study a weight proportional to its precision. Statistical heterogeneity among studies was evaluated using the I² and τ² statistics (38). Influence analysis was performed when sHRs were estimated from five or more studies: sHR was calculated by omitting one study at a time. Publication bias was assessed through a funnel plot (39). The results of the meta-analysis were presented graphically using forest plots, plotting the individual paper and sHR with corresponding 95% CI. Statistical significance was defined as p <0.05 (two sided).

A total of 2078 potentially relevant articles were identified with our initial search strategy. After screening the titles and abstracts of these articles, 2042 studies were excluded because they were deemed repetitive or unqualified. After reading 36 potentially eligible articles in detail, 13 studies met our inclusion criteria. However, since the study by Salomon et al. (27) excluded patients with HPV-negative OPSCC, thus introducing a potential selection bias, it was excluded from the analysis, thus leaving 12 articles in the final analysis (17, 20–25, 28–30, 32, 34). No additional studies were obtained by checking the reference lists of these articles. Figure 1 presents a detailed diagram of the above screening process.

The characteristics of the included studies are reported in Table 1 . The sample sizes of these studies ranged from 65 to 303 patients for a total of 1522 patients. Different cut-off values for the percentage of PD-L1-positive TCs were utilized in 12 studies (17, 20–25, 27–30, 34), ranging from 1% to 50%, while one study used the Combined Positive Score (CPS) ≥1 to define a PD-L1 positive OPSCC (32). The rates of PD-L1 positive OPSCC in the above studies ranged from 7.7% to 88.7% with overall 910 patients (53.4%) being PD-L1 positive. With the exception of one study that only selected HPV positive OPSCC (27), all other studies included consecutive patient series selected regardless of HPV status.

All included studies reported a satisfactory quality (Newcastle­Ottawa Scale score ≥6), with a median of 8 ( Table 2 ). The most frequent sources of potential bias were specific selection criteria that could have impaired external validity (e.g., restriction to tonsillar cancer or to patients undergoing radiation therapy) and the lack of reporting information of the completeness of follow-up. Only two studies did not report multivariable estimates.

All articles provided HRs and 95% CIs of OS in OPSCC. As shown in Figure 2A , no evident heterogeneity was observed in these studies (I ^{2 =} 11%; p = 0.34). The results revealed that positive PD-L1 expression was related to a significantly improved OS in patients with OPSCC (sHR = 0.63, 95% CI: 0.50-0.79). Differently, only seven articles provided HRs and 95% CIs for PFS in OPSCC (17, 20–23, 25, 29). As depicted in Figure 2B , there was no significant heterogeneity among these 7 articles (I ^{2 =} 0%; p = 0.45). The meta-analysis indicated that positive PD-L1 expression was related to greater PFS in patients with OPSCC (sHR = 0.62, 95% CI: 0.49–0.79). Finally, the association between PD-L1 expression and LRC was investigated in four studies (22, 23, 29, 32) and a significant heterogeneity was not detected among them (I ^{2 =} 48%; p = 0.12) ( Figure 2C ). Although a significant relationship did not emerged, results showed that LRC was improved in OPSCC patients that were PD-L1 positive (sHR = 0.92, 95% CI: 0.55–1.53).

The subgroup analysis based on HPV status showed that PD-L1-positive expression was related to better OS both in HPV-negative and HPV-positive OPSCC. However, few studies (22, 23, 28, 30, 32) reported HRs and 95% CI for OS according to HPV-status and the difference reached the statistical significance only in the subgroup of HPV-positive OPSCC (sHR=0.37, 95% CI 0.19-0.73) ( Figures 3A, B ).

The subgroup analysis based on treatment modality [upfront surgery or upfront (chemo)radiotherapy] revealed a better OS in PD-L1 positive tumors, regardless of treatment type ( Figures 4A, B ). Nevertheless, the effect was greater and significant only in the group treated with up-front surgery (sHR = 0.34, 95% CI: 0.18–0.65). Unfortunately, data available for the present analysis was limited since only six studies were included (21, 24, 25, 28, 29, 32).

We inspected publication bias through a funnel plot. The result ( Figure 5 ) indicated a lack of publication bias (test for asymmetry in funnel plot: p=0.89). Furthermore, influence analysis was implemented by eliminating one by one studies to determine whether a single study could significantly influence the summary result. The results showed no relevant impact on the overall result by any study ( Figure 6 ), with sHRs ranging from 0.59 by excluding Steuer et al. to 0.66 by excluding either Sato et al. (26) or Kwon et al. (25).

The strength of our study relies on the inclusion of very recent data, adding further precision to the evidence that PD-L1 expression in TCs might be a favorable prognostic marker in OPSCC. However, several limitations exist in our meta-analysis. First, since the number of included studies was small for some outcomes and for sub-group analyses, bias may have been introduced. This occurred especially when the studies were greatly unbalanced in sample size, as summary estimates were more sensitive to single studies. Further, heterogeneity estimation was challenging in this context; in addition to imprecision, the I ² statistic may also suffer of estimation bias (58), claiming for caution in its interpretation. Second, some information was not available in all included studies; for instance, data collected from published papers often lacked individual results for HPV status and/or treatment modality; hence, we failed to conduct a stratified analysis of the influence of PD-L1 expression on the response to surgery and/or radio-therapy in subgroups of HPV-positive and -negative OPSCC patients. In addition, since it has recently been shown that the level of PD-L1 expression was discrepant between primary HNSCC and corresponding distant metastases (59), only primary OPSCC have been considered. Third, this meta-analysis might not include studies that were not published due to negative results, which are often rejected or not even submit. Nonetheless, no publication bias emerged according to funnel plot inspection. Fourth, to compare OPSCC patient survival based on PD-L1 expression status might depend on the variability of IHC assay. In fact, accurate measurement of PD-L1 protein levels in FFPE tumor samples could be problematic due to antibodies discrepancy, differences in types of platforms used for the IHC staining and scoring criteria. A significant degree of intratumoral heterogeneity of PD-L1 expression might also represent an important limitation (60). Finally, the studies included in this meta-analysis have utilized a variety of techniques for determining the HPV status, including p16 IHC, HPV-DNA in situ hybridization or PCR. Hence, a standardized method for quantifying PD-L1 expression and determining HPV status is urgently needed.

The present meta-analysis showed that PD-L1 expression, as measured by IHC, was significantly associated with a better outcome in primary OPSCC treated according to the current standard of care. Particularly, HPV-positive OPSCC most benefited from PD-L1 expression with those expressing PD-L1 showing a reduction of 63% in the risk of death compared to the PD-L1 negative counterpart. Thus, PD-L1 might represent a useful biomarker, outside the context of ICI immunotherapy, to stratify prognosis in OPSCC in addition to HPV status.