A meta-analysis of c-Met expression in head and neck cancer

The prognostic role of c-Met expression in head and neck cancer patients were controversial among different studies. Thus, we performed a meta-analysis to evaluate the relationships between c-Met expression and survival and clinical parameters of head and neck cancer patients. The databases PubMed and Embase were searched. Summary hazard ratio (HR) and 95% confidence intervals (CIs) were calculated to analyze the correlations between c-Met expression and overall survival (OS), and disease-free survival (DFS). Furthermore odds ratios (ORs) and 95% CIs were used to describe the relationships between c-Met expression and different clinicophathological parameters. A total of 2,417 patients from 19 studies were enrolled in the final analysis. The results showed that c-Met expression was significantly associated with OS (HR: 1.65; 95% CI: 1.20–2.27, P = 0.002) while not DFS (HR: 1.48, 95% CI: 0.99–2.20, P = 0.055). And c-Met expression was associated with the N stage of head and neck cancer patients (OR: 2.11; 95% CI: 1.12–3.96, P = 0.02). These results suggested that c-Met expression was a risk factor for head and neck cancer; increased c-Met expression could be a predictor of a poorer prognosis for the patients.


INTRODUCTION
Head and neck cancer (HNC) are cancers that arise in the tissues of the head and neck, which includes cancers occurred in oral cavity, pharynx and larynx. Worldwide, approximately 550,000 individuals develop head and neck cancer and about 380,000 die from this disease every year [1]. In the United States, there would be more than 63,000 new cases of HNC and 13,000 deaths in 2017 [2]. Despite the incidence and mortality rate of head and neck cancer declined in the past two decades, there were no standard or routine screening tests for head and neck cancers. It is necessary to find new biomarkers that could help to screen high-risk individuals and predict the prognosis of the patients. c-Met is a member of receptor tyrosine kinase family, it also known as hepatocyte growth factor receptor (MET). It is overexpressed in multiple human cancers, activation of c-Met could mediate the migration, invasion, and angiogenesis of tumor cells [3]. By immunohistochemistry, Knowles et al found that c-Met and /or its ligand (hepatocyte growth factor, HGF) overexpressed in nearly 80% of primary HNC tumors [4]. The prognostic role of overexpression of c-Met had been investigated in head and neck cancer [5][6][7], but the results were inconsistent among different studies. Thus, we carried out a system review to evaluate the prognostic role of overexpression of c-Met in head and neck cancer patients.

Literature searching
A comprehensive literature retrieve was carried out in PubMed and Embase for the eligible studies published up to June 2017. The following keywords were used in the literature searching: "c-Met" and "head and neck cancer" or "oral cancer" or "pharyngeal cancer" or "laryngeal cancer" or "nasopharyngeal cancer" and "prognosis" or "survival". Furthermore, we manually searched the reference lists of relevant articles to identify the possible missing publications. Studies were included unless they met the following inclusion criteria: provided survival data in head and neck cancer patients stratified by c-Met expression and sufficient data to calculate an estimate of hazard ratio (HR) and a 95% confidence interval (CI); the expression of c-Met was measured in the head and neck cancer specimens; and all selected patients were pathologically confirmed. All publications were reviewed to avoid the duplicate data. If there were overlapping between studies, only the most recent or most complete data was included.

Data extraction and quality assessment
Five investigators were divided into two groups and they conducted the literature review and extracted data from each included publication separately. For each article, the following information was extracted: author's name, publication year, sample size, research country, tumor types, clinicopathological data (number of patients with different gender, T stage, N stage and clinical stage), c-Met expression detection methods, cut-off level of high expression, and survival data (HR and its 95% CIs for OS and DFS). If the survival data were not reported in the publications, we extracted it from the Kaplan-Meier curves by the methods of Jayne et al. [8]. The extracted data were crosschecked and any disagreements were resolved by discussion.

Statistical analysis
Overall survival (OS) and disease free survival (DFS) were the main outcomes to evaluate the prognostic value of c-Met expression, hazard ratios (HRs) and 95% confidence intervals (CIs) were used to combine as the effective value. For clinicopathological data, odds ratios (ORs) and 95% CIs were used to describe the relationships between c-Met expression and different clinicophathological parameters. Pooled HRs or ORs with 95% CIs were calculated using random-effects model due to the potential substantial heterogeneity among studies [9]. I 2 statistic was used to detect the heterogeneity, it indicated the presence of significant heterogeneity if I 2 ≥ 50% [10]. For survival datasets, we further investigated potential heterogeneity by subgroup analyses stratified by tumor types, region of the study, and calculation methods of HR. We used sensitivity analysis to assess the stability of the results. Potential publication bias was evaluated by the shape of funnel plots, an asymmetric plot indicates a possible publication bias [11]. The publication bias was further assessed by Egger's test, P < 0.05 was considered indicative of significant publication bias [12]. The "trim and fill" method was used to examine the potential impact of the publication bias on the interpretation of the data if publication bias was found [11]. All statistical analyses were performed by Stata statistical software (version 11.0; Stata Corporation, College Station, TX, USA).

c-Met expression and the survival of head and neck cancer
We used OS and DFS as the main outcomes to evaluate the impact of c-Met expression on the survival of head and neck cancer patients. In the collected dataset, there were 16 studies reported overall survival (OS) and 8 studies reported disease-free survival (DFS). Firstly, we analyzed the relationships between c-Met expression and OS in 16 studies. The results showed that the patients with higher c-Met expression had a poor prognosis than those Oncotarget 3 www.impactjournals.com/oncotarget with lower expression (HR: 1.65; 95% CI: 1.20-2.27, P = 0.002). Subgroup analysis was carried out based on tumor types, regions of the study and calculation methods of HR. Based on the region of study, we found significant association between c-Met expression and survival rate in Asian studies. In the subgroup analysis by statistical methodology, there was significant difference in the group of multivariable analysis. For different tumor types, we found c-Met expression was significant associated with the survival rate of HNC and NPC ( Figure 2).
Then HR and 95% CI were calculated to test the relationship between c-Met expression and disease-free survival (DFS). The results revealed that there was no significant association between c-Met expression and DFS rate (HR: 1.48; 95% CI: 0.99-2.20, P = 0.055). In the subgroup analysis, the significant association was just found in "Asia" and "Multivariate" (Figure 3). All the details were listed in Table 2.
To evaluate the reliability of the results, sensitivity analysis was carried out by deleted single study each time.
There was no significant change on the results of OS by removing any of the studies. As for DFS, the association changed into significant when we delete the study of "Lo Muzio", Figure 4. The results shown single study basically did not have any influence on the results. Begg's funnel plots were constructed to evaluate the publication bias; we did find the sign of publication bias by the shapes of the plots in OS while not in DFS ( Figure 5). The results of Egger's test were consistent with the plot (P = 0.006 for OS; P = 0.335 for DFS). The "trim and fill" method was used to adjust for the publication bias and to find the possible influence of the bias on the results. By this method, we estimated there would be one missing study, the filled dataset is much more symmetric than the original data ( Figure 6). Compared with the original results (HR: 1.65; 95% CI: 1.20-2.27, P = 0.002), the association between c-Met expression and the prognosis did not change when the missing study were added (HR: 1.56; 95% CI: 1.16-2.89, P = 0.004).

c-Met expression and clinical parameters
ORs and 95% CI were calculated to evaluate the associations between c-Met expression and different clinical parameters such as gender (male vs. female), T stage (T1/ T2 vs. T3/T4), N stage (N0 vs. N+) and clinical stage (I/II vs. III/IV). Pooled ORs showed that c-Met expression was association with gender, T stage, N stage and clinical stage. But the association was not significant except between c-Met expression and N stage (OR: 2.11; 95% CI: 1.12-3.96, P = 0.02). Sensitivity analysis and publication bias were carried out too. The results did not change after we deleted any single study (data not show). No publication bias was found in all of these analyses. The details were showed in Table 3.   [46]. It is overexpressed in a variety of tumors and plays an important role in malignant transformation, activated c-Met could mediate tumor growth, metastasis and angiogenesis [47]. It had been reported that c-Met could activate a lots of downstream signal factors, such as MAPK, PI3K/AKT and STAT3 [48,49]. Given its critical role in tumor malignant transformation, c-Met could be considered as a prognosis marker for cancer. But the prognosis role of c-Met in human head and neck cancer is unclear, the results is inconsistent among different studies, so we carry out this meta-analysis to combine these studies and get a more reliable result.
To our knowledge, the present meta-analysis is the first one to consider the relationships between c-Met expression and clinical pathological parameters and prognosis of human head and neck cancer. In this study, 2,417 patients from 19 studies were collected. For all kinds of head and neck cancer, the results showed that patients with higher c-Met expression had a bad overall survival rate (P = 0.002), but for the disease-free survival rate, the association was not significant (P = 0.055). Even though oral cavity and pharynx were the most common sites of the head and neck cancer, we did not find significant associations between c-Met expression and survival in oral and oropharyngeal cancer (OOC) and tongue cancer (TC), which could be the result of small sample size. So more studies with large number of patients were needed to Furthermore, we found the results were different between subgroup of univariable analysis and multivariable analysis. For overall survival (OS), we found there was significant association in the subgroup of "multivariate analysis", but for disease-free survival (DFS), the results showed there was significant association in the subgroup of "univariate analysis". In survival analysis, the "univariate" was used to evaluate the relations of the single factor to the survival rate, but the single factor was seldom informative due to relations among different factors and the potential interaction among the other factors to the survival rate. These issues could be handled by multivariate analysis, which could evaluate the effect of single factor without      the interaction of others. So the results of multivariate analysis were much more reliable than the results of univariate analysis.
As for the relationships between c-Met expression and clinical parameters of head and neck cancer, we found the expression of c-Met was associated with the N stage (N0 vs. N+). Patients with higher c-Met expression had a significant higher probability to occur lymph node metastasis. Based on these results, c-Met expression could act as a valuable prognosis marker for head and neck cancer patients.
Heterogeneity is an important indication to evaluate a meta-analysis. The heterogeneity test showed that there was obvious heterogeneity among different studies. In this study, we used random-effect model to calculate the pooled OR and HR to reduce the effect of the heterogeneity to the results. As for the source of heterogeneity, it may mainly come from different tumor types, studies regions and analysis types. So, we carry out the subgroup analysis to reduce the heterogeneity. The results showed the heterogeneity reduce or disappear in most subgroups. The results showed that Asian studies were quite homogeneous, while there was obvious heterogeneity in the Europe and North American studies. The possible reason of the high heterogeneity in these subgroups was that the number of the studies in these two groups is relatively small and these studies focused on different tumor types.
Sensitivity analysis and publication bias are the other two important ways to test the liability of the results of meta-analysis. These results showed our study were accurate and trustiness. But there are still some limitations in the present study. First, the number of the enrolled studies is relatively small, especially for each kind of cancer. Second, the sample size was small, most of the included studies just had limited patients (< 100).
In conclusion, c-Met expression was a risk factor for head and neck cancer patients, the patients wither higher c-Met expression had a significant poorer survival rate. It would be helpful to define the high-risk patients by determination of the expression of c-Met. But more studies with larger sample size were needed to expound the relationships between c-Met expression and prognosis of head and neck cancer patients.

Author contributions
H.S., L.M. and G.L. designed the research. L.L., Z.S., X. H., X. L., and L.S. collected articles and extracted data. L.L., Z.S., L.Z., X.Z. L.Y., and J.Y. performed the statistical analysis. L.L. and H.S. wrote the manuscript. All of the authors read and approved the final manuscript.

CONFLICTS OF INTEREST
The authors declare no competing financial interests.