Prognostic Value of Programmed Cell Death Ligand-1 Expression in Nasopharyngeal Carcinoma: A Meta-Analysis of 1,315 Patients

Background: The prognostic value of programmed cell death ligand-1 (PD-L1) in patients with nasopharyngeal carcinoma (NPC) remains controversial. Therefore, we conducted this meta-analysis to understand the role of PD-L1 in NPC. Method: We searched PubMed, Embase, Web of Science, and Cochrane Library up to April 2019. We determined the pooled hazard ratio (HR) and 95% confidence intervals (CIs) to assess the relationship between PD-L1 and various survival outcomes. Begg's funnel plot was used to assess any publication bias. Results: Eleven studies involving 1,315 patients were included in this meta-analysis. For overall survival (OS), the HR was 1.48 and 95% CI was 1.00–2.18 (p = 0.049). For disease-free survival (DFS), the HR was 1.51 and 95% CI was 0.85–2.69 (p = 0.162). For distant metastasis-free survival (DMFS), the HR was 1.75 and 95% CI was 0.64–4.79 (p = 0.277). For local recurrence-free survival (LRFS), the HR was 0.67 and 95% CI was 0.06–8.16 (p = 0.756). The results of prognosis of PD-L1 and OS were more significant after sensitivity analysis. The pooled odds ratio indicated that PD-L1 expression was not associated with T stage, N stage, M stage, overall stage, sex, age, smoking, or alcohol intake. No publication bias was found. Conclusion: Our meta-analysis showed that PD-L1 overexpression in NPC was associated with a poor OS and may be useful as a novel prognostic factor for NPC.


INTRODUCTION
Nasopharyngeal carcinoma (NPC) originates from the nasopharynx epithelium. NPC is rare compared with other types of cancer; moreover, its geographical distribution presents a unique pattern (1): it is a highly common type of head and neck cancer in the eastern and southeastern parts of Asia. NPC tends to metastasize to distant sites in the head and neck, and about 70% of the patients with NPC present with stage III or IV disease at the time of initial diagnosis (2). Radiotherapy (RT) is primarily used to treat NPC; however, concurrent chemoradiotherapy (CCRT) is recommended as the standard treatment for locoregionally advanced NPC (3). Although the 5-year survival rate of patients with NPC is ∼60-70%, NPC management remains challenging because of locoregional failure, recurrence, and distant metastasis after primary treatment (4). Currently, several prognostic parameters are used for NPC management, including age and sex (5). However, these parameters lack sensitivity or specificity in some patients with NPC, and are, therefore, insufficient for predicting survival outcomes. Thus, the identification of novel and efficient prognostic markers is highly important in NPC treatment.

Literature Search
Eligible studies were identified by searching PubMed, Embase, Web of Science, and Cochrane Library up to April 2019. The following search terms were used: "Programmed Cell Death Ligand 1" or "Programmed Death Ligand 1" or "PDL1" or "B7-H1" or "CD274" or "Programmed Cell Death 1" or "Programmed Death 1" or "PD-1" or "CD279" and "nasopharyngeal carcinoma" or "nasopharyngeal cancer" or "NPC." References in the retrieved articles were also manually searched for additional studies.

Inclusion and Exclusion Criteria
Eligible studies were identified according to the following inclusion criteria: (1) the NPC cases were pathologically confirmed; (2) PD-L1 expression was detected in NPC by using IHC; (3) studies provided the association between PD-L1 and survival outcomes and/or clinical characteristics; (4) studies provided the hazard ratio (HR) and 95% confidence interval (CI) for survival outcomes or sufficient information to calculate the HR and 95% CI in accordance with the methods by Parmar (24) and Tierney (25); (5) for studies (22,26,27) detected the expression of PD-L1 in both tumor cells (TCs) and immune cells (ICs), we only extracted the data of PD-L1 expression on TCs for analysis; and (6) original articles were published in English. The following were the exclusion criteria: (1) meeting abstracts, case reports, reviews, or letters; (2) duplicate studies; and (3) non-human studies.

Data Extraction and Quality Assessment
Two authors extracted relevant data independently, and disagreements were settled by discussion. The following data were extracted: the name of the first author; the year of publication; country/region; sample size; age; tumor stage; treatment modality; study design; study duration; and HR and 95% CI related to PD-L1 expression. The quality of eligible studies was evaluated according to the Newcastle-Ottawa Scale (NOS) (28). The scale includes three categories: selection, comparability, and outcome assessment. The top score is 9 points, and studies with a score >6 are regarded as high-quality studies.

Statistical Analysis
This meta-analysis was performed according to the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines (29). Pooled HR and 95% CI were determined to assess the relationship between the PD-L1 and various survival outcomes. The pooled odds ratio (OR) and 95% CI were determined to study the correlation between PD-L1 and clinicopathological features. Statistical heterogeneity among all studies was assessed using the I 2 and Chi-squared test. Heterogeneity was considered statistically significant when the I 2 > 50% or P < 0.10. A random-effects model was used when significant heterogeneity was present; alternatively, a fixed-effects model was used. Sensitivity analysis and subgroup analysis were  Frontiers in Oncology | www.frontiersin.org adopted to assess the heterogeneity and stability of the results. Begg's funnel plot was used to assess any publication bias (30). All statistical analyses were conducted with Stata version 12.0 statistical software (Stata Corporation, College Station, TX, US). A P < 0.05 was considered statistically significant.

Literature Search
A total of 166 related studies were initially retrieved. As shown in Figure 1, following the exclusion of 64 duplicate studies, 102 studies were screened. Eighty-six studies were excluded after title and/or abstract screening for the following reasons: not based on NPC (n = 35), reviews (n = 21), non-human studies (n = 17), and not related to PD-L1 (n = 13). Thereafter, the full text of 16 studies was assessed; five studies were excluded because of insufficient data (n = 2), duplicate studies from the same research group (n = 2), and no use of IHC (n = 1). Finally, 11 studies were included in this meta-analysis (18-23, 26, 27, 31-33).

Study Characteristics
Baseline characteristics of the eligible studies are summarized in Table 1. The studies were published from 2016 to 2019 and were all conducted in Asia. Six studies were conducted in China (20-23, 32, 33), and one each were in Hong Kong (18), Philippines (19), Thailand (31), Taiwan (26), and Japan (27), respectively.

The Association of PD-L1 Expression With Clinicopathological Features
To explore the correlation between PD-L1 expression and clinicopathological characteristics of NPC, the pooled OR and 95% CI were determined. As shown in

Subgroup Analysis and Sensitivity Analysis
Because several baseline characteristics varied among included studies, which may cause heterogeneity in meta-analysis. Therefore, subgroup analysis was conducted for OS and DFS. As shown in Table 2, PD-L1 expression was associated with poor OS in patients with stage II-III (n = 1, p = 0.016), stage III-IV (n = 1, p = 0.049), and recurrent/ previously-treated patients (n = 2, p = 0.044). However, PD-L1 was not significantly correlated to OS in both retrospective and prospective studies. Subgroup analysis also showed that PD-L1 remained a nonsignificant prognostic marker for DFS irrespective of tumor stage and disease type. Subgroup analysis was not performed for DMFS and LRFS because of the limited number of studies and uniformity among these studies. Sensitivity analysis by omitting one study in each turn was conducted to test the credibility of the prognostic value of PD-L1. As shown in Figure 3A, the results for OS were significantly altered when Li' s study (20)   was excluded. Therefore, we re-analyzed the prognostic value of PD-L1 for OS after exclusion of Li's study (20), the results were HR = 1.33, 95% CI = 1.04-1.70, p = 0.022; I 2 = 0, P = 0.513 ( Figure S1). The heterogeneity in the analysis of PD-L1 and OS was significantly reduced after deletion of Li's study (I 2 = 0, P = 0.513). No other individual study influenced the results for DFS, DMFS, and LRFS (Figure 3). The results of prognosis of PD-L1 and OS were more significant after sensitivity analysis, which suggested that PD-L1 could be a prognostic factor for OS in NPC patients.

DISCUSSION
Many studies demonstrated that PD-L1 plays an important role in tumor immune evasion (34,35). The upregulation of PD-L1 inhibits T cell function and triggers immune evasion in cancer (36). In the tumor microenvironment, tumor-associated PD-L1 increases the apoptosis of T cells and leads to the growth of immunogenic tumors (36). PD-L1 is mainly expressed on the surface of tumor cells and tumor-associated APCs in various types of cancer including pancreatic cancer (37), ovarian cancer (38), thymoma (39), and colorectal cancer (40). PD-L1 and its receptor PD-1 constitute the PD-1/PD-L1 immune checkpoint signaling pathway (41). Checkpoint inhibition targets regulatory pathways in T cells to promote anti-tumor immune responses (12,42). A recent phase II clinical trial including 44 patients revealed a promising activity of the PD-L1 antibody in NPC and a favorable 1-year OS rate (43). This study also suggested an association of PD-L1 expression with a higher response rate in patients with NPC (43).
To the best of our knowledge, this is the first meta-analysis to investigate the prognostic and clinicopathologic significance of PD-L1 expression in NPC. Survival data of 1,315 patients from 11 eligible studies were systematically analyzed. We found that PD-L1 overexpression was a significant prognostic factor for poor OS, whereas PD-L1 expression did not predict worse DFS, DMFS, or LRFS. Furthermore, PD-L1 expression was independent of T stage, N stage, M stage, overall stage, sex, smoking, age, or alcohol intake. Overall, this meta-analysis highlighted the potential of PD-L1 as a prognostic biomarker for poor OS in patients with NPC.
Previous meta-analyses exploring the prognostic value of PD-L1 in solid malignant tumors also indicated the unfavorable impact of PD-L1 on survival outcomes (10,44,45). A metaanalysis involving 13 studies showed that a high PD-L1 expression could predict a shorter OS (HR = 1.57, 95% CI = 1.09-2.27, P < 0.00001) and poorer DFS (HR = 2.07, 95% CI = 1.20-3.58, P = 0.009) in hepatocellular carcinoma (15). Another study also demonstrated a significant association of PD-L1 expression with a poor biochemical recurrence-free survival (BCR-FS) (HR = 1.78; 95% CI = 1.39 to 2.27; p < 0.00001) in prostate cancer (46). Wang et al. showed an association of PD-L1 expression with a poor OS in RCC (47). In addition, PD-L1 expression was found to be significantly associated with tumor stage, regional lymph node involvement, distant metastases, nuclear grade, and histologic tumor necrosis in RCC (47). The current meta-analysis showed the positive correlation of PD-L1 with OS, but not with DFS, DMFS, or LRFS. This may due to the limited sample size and the relatively short follow-up duration. Furthermore, we did not detect any association of PD-L1 with clinical factors in NPC. This requires further verification in future prospective studies.
There are several limitations in this study. First, all the eligible studies were conducted in Asia, which may reflect the high incidence of NPC in Asia. However, the results of this meta-analysis might be applied to patients in Asia. Second, although we selected eligible studies using uniform criteria, interstudy heterogeneity still exists in this meta-analysis. Therefore, subgroup analysis was performed to detect the source of heterogeneity. Third, the majority of the eligible studies were retrospective in design, which may compromise the validity of this study.

CONCLUSIONS
In summary, this meta-analysis showed that PD-L1 overexpression in NPC was associated with poor OS and may be useful as a novel prognostic factor. Nevertheless, because of the aforementioned limitations, well-designed, large-scale, prospective clinical trials are required to verify the findings of this meta-analysis.

DATA AVAILABILITY STATEMENT
All datasets generated for this study are included in the article/Supplementary Material.

AUTHOR CONTRIBUTIONS
YS designed the study, performed the literature search and screening, performed the data analyses, and wrote the manuscript. XL and YS designed the study, retrieved the literature and data, analyzed the retrieved data, and participated in the writing of manuscript. CS and JD assisted in the designing of the study, performed the literature search and screening, assisted in the data analyses, and participated in the writing of manuscript. JD designed the study and supervised the study. Every author approved the final version of this study.

SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.01111/full#supplementary-material Figure S1 | Meta-analysis of PD-L1 overexpression was associated with worse OS after deletion of Li's study.