Association between PD-L1 expression and head and neck cancer prognosis: a meta-analysis

Background: Programmed cell death ligand 1PD-L1) plays an important role in tumor cell immune escape, and it has been extensively studied in head and neck cancer. However, its prognostic impact on patients with head and neck cancer remains controversial, so we sought to investigate this issue through a comprehensive meta-analysis. Methods: To assess the significance of PD-L1 on the survival of patients with head and neck cancer, we collected articles reported in PubMed, EMBASE, and Cochrane Library, until January 31, 2019. We also used the Newcastle Ottawa Scale (NOS) for literature quality evaluation. Results: The study included a total of 4551 patients affected by 6 different types of head and neck cancer reported in 26 articles. Our study found that the association between the expression of PD-L1 and the prognosis of head and neck tumors was highly heterogeneous (P < 0.00001, I2 = 80.0%); therefore, the random effects model was applied to combine the effect sizes. Based on the combined hazard ratios (HR)of 1.15 (95% CI: 0.88 to 1.50, P = 0.32), the expression of PD-L1 in head and neck tumors may not be a factor associated with poor prognosis. Conclusions: Our results suggest that PD-L1 expression cannot predict the overall survival of patients with oral, or esophageal cancer. Through subgroup analysis, we found that the expression of PD-L1 may be a poor prognostic factor for some head and neck cancers. carcinoma; carcinoma PTC, papillary thyroid cancer.

oral cavity, larynx, hypopharyngeal and esophageal cavity, nasal and paranasal cavity, temporal bone, salivary glands and so on, and most of these subtypes are squamous cell carcinoma. The global incidence of HNC has been increasing for decades, but the reasons are not fully understood [3].Due to the absence of early symptoms and lack of effective screening techniques, most patients are diagnosed with advanced cancer [1].
Programmed death factor 1(PD-1) and its ligands are currently one of the research focuses of tumor targeted therapy. When PD-1 binds to the ligand PD-L1, the function of T cells is reduced by various mechanisms, including antagonizing T cell receptors to inhibit their signal transduction, changing the normal metabolism of T cells to inhibit their differentiation, and reducing the antioxidant function of T cells to inhibit their killing [4,5].
At present, several PD-1 inhibitors and PD-L1 inhibitors have entered clinical trials, and some are used in cancer treatment. Nivolumab was the first PD-L1 inhibitor to be marketed and approved by the FDA for the treatment of metastatic melanoma, non-small cell lung cancer and advanced renal cell carcinoma [6].
To better prolong the survival of patients with HNC, the selection of effective immunological molecular biomarkers and immunotherapy programs will open up new strategies for the treatment of advanced HNC. Many studies have found that high expression of PD-L1 in HNC patients is associated with poor prognosis and does not depend on the tumor origin. PD-L1 is considered to be a prognostic biomarker [7]. Therefore, we conducted this meta-analysis to determine the prognostic impact of PD-L1 on patients with HNC.

Search method
All articles were searched in the PubMed, EMBASE, and Cochrane Library databases, up to January 31, 2019. We used the search terms ("head and neck tumor" OR "oral" OR "tonsil" OR "parathyroid" OR "thyroid" OR "laryngeal" OR "ear" OR "nose" OR "tongue" OR "salivary gland" OR "esophageal" OR "facial" OR "nasopharyngeal" OR "hypopharyngeal") and ("pd-l1" OR "b7-h1"OR "programmed death ligand 1" OR "cd274" OR "b7homolog1" OR "pdl1") and ("cancer" OR "carcinoma" OR "adenocarcinoma"). Selected publications were all written in the English language. In addition, the references included in the articles were manually reviewed to obtain other potentially relevant articles.

Inclusion criteria
The inclusion criteria included that the clinical and pathological data of all cases were complete, and the pathological examination confirmed the diagnosis of HNC. Additionally, immunohistochemistry (IHC) staining for the detection of PD-L1 expression in tumor tissues, and raw data about PD-L1 expression and HR and 95% CI for the prognostic correlation in cancer patients were provided in the study.

Exclusion criteria
The exclusion criteria consisted of non-original research articles, such as reviews, abstracts, letters, and case reports; copying reports or similar data; Newcastle-Ottawa Scale (NOS) scores <5 points and other unusable documents; the subjects were not human patients; the sample size was less than 30; and the full text could not be obtained.

Data extraction
Data extraction and quality assessment were performed by two of the authors. The data included in each study were extracted independently. The authors extracted the author's name, publication year, country in which the study was conducted, sample size, cancer type, and critical criteria for over expression (the definition of positive PD-L1), hazard ratios (HRs) and 95% confidence intervals (CI).

Quality assessment
According to the NOS, the quality of each of the included studies was evaluated [8].The total score was 9 points, and articles with scores below 5 points were deemed to be of low-quality. The evaluation process was independently conducted by 2 authors, and differences in evaluation were resolved by negotiation. When the opinions were inconsistent, a 3rd author made the assessment. Finally, the appropriate literature was selected for meta-analysis.

Statistical analysis
HR and CI were used to assess the association between PD-L1 expression levels and head and neck tumor prognosis. Heterogeneity testing was performed using Revman 5.3 software. The Chi-square test was to estimate the between-study heterogeneity. When P ≥ 0.1 and I 2 ≤ 50%, the studies were considered homogeneous, and the fixed effect model was selected and analyzed. When P < 0.1 and I 2 > 50%, the studies were considered to be heterogeneous, and the source of heterogeneity was further analyzed by subgroup analysis or sensitivity analysis. If the heterogeneity could not be eliminated, the random effects model was chosen. Additionally, a funnel plot was drawn to test the publication bias of the assessment study [9].

Description of trials included in the meta-analysis
A total of 320 articles were retrieved from the relevant databases, all the documents were saved to Endnote and evaluated, and 111 duplicates were excluded. After reading the titles and abstracts of the articles, 80 preliminary screenings were obtained. After reading the full text, according to the inclusion and exclusion criteria, 26 articles were selected  Table 1.

Prognosis
A total of 26 studies reported that the levels of PD-L1 expression were associated with prognosis. Twelve studies showed that PD-L1 over expression is associated with poor prognosis in HNC. Our analysis showed that the expression level of PD-L1 was not associated with prognosis in HNC (HR = 1.15, 95% CI = 0.88-1.50, P = 0.32, random effect) (Fig. 2).

Subgroup analysis
Further subgroup analysis was based on the patient's origin (China and not China), the IHC assessment of PD-L1 percentage cutoff, and the cancer type. The conclusions from the subgroup analysis were similar to those of the overall analysis (more details are listed in Table 3, 4 and 5.). However, we found that PD-L1 may be a poor prognosis factor for oral squamous cell carcinoma and other types of head and neck cancer.

Prognostic study publication bias and sensitive analysis
The funnel plot test evaluated the publication bias. The result shows incomplete symmetry ( Fig. 6), suggesting that there may be publication bias in the prognostic study. Sensitive analysis did not change the results.

Discussion
The study included a total of 4551 patients affected by 6 different types of head and neck cancer reported in 26 articles. Our study found that the association between PD-L1 expression and head and neck tumor prognosis is highly heterogeneous; therefore, a random effects model was applied. We found that the expression of PD-L1 in head and neck tumors may not be a factor in poor prognosis. Through subgroup analysis, we found that the expression of PD-L1 was not related with the prognosis of oral cancer, esophageal cancer, and nasopharyngeal carcinoma, but it may be a poor prognostic biomarker for other types of head and neck cancer.
Many meta-analyses report the prognostic value of PD-L1 in various types of cancer [36][37][38][39], and a large number of studies have reported the association between PD-L1 and the prognosis of head and neck tumors , but the results are not completely consistent. Therefore, this paper used a meta-analysis to explore the association between PD-L1 and HNC prognosis. Head and neck tumors consists mainly of head and neck squamous cell carcinoma. Recently, yang [40] reported that PD-L1 expression by IHC is not recommended as a predictor of HNSCC survival and that positive PD-L1 expression may predict better PFS in patients with advanced HNSCC. Our study collected more studies for meta-analysis to investigate the association between PD-L1 and HNC prognosis, and similar results were obtained. It should be emphasized that the heterogeneity of the results cannot be ignored. Heterogeneity may be due to different sample sizes of the included studies, the different relevant detection methods, including antibody selection, standardized definition of outcome indicators and differences in population characteristics and types and sources of cancer. The most important is the different standards for PD-L1 expression. During our research, we found that the cut-off value was not standardized.
There are two ways to define the cut-off value in the study. One method is the integral comprehensive measurement method; the value corresponding to each positive intensity × the percentage of the intensity cells, and then sum again. The second method is to use the value corresponding to the percentage of positive cells. Therefore, based on the cutoff value, we performed a subgroup analysis. We found that the heterogeneity of the cutoff group of 5% is small (Fig. 4). In addition, we performed a subgroup analysis based on the type of cancer and found that for other types of head and neck cancers, it seems that PD-L1 expression was associated with prognosis.

Conclusion
Based on the above meta-analysis, we preliminarily estimated that the expression of PD-L1 may not be significantly related to the prognosis of oral squamous cell carcinoma, nasopharyngeal carcinoma, and esophageal squamous cell carcinoma. However, for some of other HNC, we cannot deny the existence of the correlation.       Meta-analysis forest plots of the survival subgroups of HNC with different cutoffs.

Figure 5
Meta-analysis forest plots of the survival subgroups of different types of HNC.

Figure 6
The funnel plot test evaluating the publication bias.