PubMed, Embase, the Science Citation Index, Cochrane databases and the Ovid Database were systematically searched for studies investigating the tumor clinicopathological and prognostic relationship between CTCs and H&N cancer, with no restrictions on language, place of publication or date of publication (up to November 2015). The main search terms used were “circulating tumor cells”, “disseminated tumor cells”, “head and neck cancer”, “nasopharyngeal”, “nasal”, “oral”, “oropharyngeal”, “hypopharyngeal”, “laryngeal”, and “larynx”.

To make our analysis reliable, we screened the titles and abstracts for all searched papers, and full text was perused for potential eligible studies according to the following inclusion criteria: 1) containing patient cases of H&N cancer; 2) measuring the presence of CTCs in peripheral blood (PB); 3) investigating the clinicopathological and prognostic significance of CTCs in H&N cancer patients with at least one of the outcome measures of interest. Studies were excluded in our study: 1) duplicated publications; 2) no outcomes of interest that were provided or can’t be calculated for prognostic evaluation.

Two reviewers (Chen RW and Zhou Y) independently evaluated each papers and extracted data, and any disagreements were resolved via discussion, with a third investigator if necessary. The following information was extracted: first author, publication year, population characteristics (i.e., country, number, sex and age), tumor clinicopathological characteristics (i.e. anatomical sites, pathologic differentiation and TNM stage), sampling times (preoperative or postoperative), detection methods (RT-PCR array, Non-RT-PCR array including Cell Search system or immunocytochemistry), CTCs positive rate, detection markers, endpoints and survival data. For studies with multiple markers or detect methods, each of the cohorts was considered an independent data set. However, for those studies with multiple sample times (i.e. pre-treatment and intra/post-treatment), we use data from pre-treatment samples because those data were usually dependent from various treatment regimens.

Statistical analysis was performed using Review Manager 5.2 (Copenhagen: The Nordic Cochrane Centre; The Cochrane Collaboration, 2012). To evaluate the association between CTCs and clinicopathological characteristics, the estimated odd ratios (ORs) were extracted from enrolled publications. To statistically assess the prognostic significance (DFS and OS), we extracted the estimated hazard ratios (HRs) and associated 95% CIs when available. If the HR and its variance were not reported directly in the original study, these values were calculated from available reported data using software designed by Tierney JF [8] . When HRs was presented by both univariate and multivariate analyses, the multivariate ones were employed due to adjustment for confounding factors. Heterogeneity among the studies was tested using the χ² test and I² statistic. A value of I² < 25%, within 25% - 50% or more than 50% was regarded as low, moderate, or significant heterogeneity, respectively. The random-effects mode was explored to perform the analyses, because this model obtained more conservative results than the fixed-effect model [9] . A two-sided P < 0.05 was considered statistically significant. Furthermore, subgroup analyses were made to explore the inherent heterogeneity. Lastly, the potential publication bias was evaluated using the funnel plot. All of the studies included in this research were assessed by referring to the Newcastle-Ottawa quality assessment scale (NOS) for cohort studies.

The primary literature research initially yielded 2320 articles. After screening the titles, abstracts, language and other information, 2161 studies were excluded and 159 potential studies were reviewed further. An additional 137 studies were then excluded which were reviews (n = 18), laboratory studies (n = 97) or studies of other tumors (n = 22). Upon detailed evaluation of the remaining 22 studies, 5 studies had to be excluded because the outcome of interest could not be calculated. Finally, a total of 17 articles were considered to be appropriate for the meta-analysis (Figure 1).

The eligible 17 studies comprising 816 patients diagnosed as head and neck squamous cell carcinoma were published between 1999 and 2014 [10] - [26] . RT-PCR arrays were used to evaluate CTCs status in 6 studies, Cell Search system in 6 studies and immunocytochemistry (ICC) and other methods in the remaining studies. One study from the United Kingdom applied both RT-PCR and ICC and compared the agreement between these two methods, which was considered as two independent cohorts in the analysis. The samples for CTCs array were collected pre-treatment in 14 studies, only 2 studies [11] [21] were collected post-treatment and 1 study not reported. The main characteristics of the included studies are summarized in Table 1. The quality of the in-