Prognostic value of ZEB-1 in solid tumors: a meta-analysis

An electronic literature retrieval of studies investigating the expression of ZEB-1 and clinical prognosis in Web of Science, PubMed and EBSCO was conducted until September 2018 to identify latent eligible articles. The search keywords for the subject heading terms were: “Zinc-finger E-box binding homeobox 1” or “ZEB-1” or “dEF1” or “ZFHX1A” or “Nil-2-a” or “TCF8” or “AREB6” or “BZP” and “cancer” or “tumor” or “neoplasm” or “carcinoma,” and “prognosis” or “survival”. Only studies involving patients with solid tumors were included. Collectively, 840 entries were identified.

The inclusion criteria were as follows: (1) diagnosis of human solid tumor; (2) the role of ZEB-1 in the tumor progression must be investigated; (3) description of the relevance of ZEB-1 expression to OS; (4) availability of data to obtain a hazard ratio (HR) and 95% confidence interval (CI); (5) publication in English. The exclusion criteria were as follows: (1) duplication; (2) conference abstract, review, and book; (3) basic research and animal experiments; (4) lack of usable data. There were two independent investigators (B.R Chen and Z.P Zhu) carefully scrutinized each candidate article. Disagreements were resolved through discussion between the two investigators until a consensus was reached.

The following data were extracted by two independent investigators (B.S Chen and W.P Ye): name of first author, year of publication, country, type of cancer, case number, sex of patients, median age of patients, detection method, cut-off value of the overexpression of ZEB-1, follow-up period, survival analysis, obtained of HR, and Newcastle-Ottawa Scale (NOS) scores. Each HR and corresponding 95% CI were extracted based on the tables or Kaplan-Meier curves describing groups of patients with high or low expression of ZEB-1. Besides, all of the included researches were cohort studies. The NOS [23] was used to estimate the quality of the included studies in our meta-analysis. Studies with a NOS score ≥ 6 were defined as high quality research. Finally, any conflicting results were resolved through by discussion between the two authors.

Survival outcomes were the most essential endpoints. Therefore, the HR and 95% CI were directly extracted from the original publications or calculated by means of the Kaplan– Meier method according to Tierney’s method. [24] Subsequently, the RevMan 5.3 analysis software (Cochrane Collaboration, Copenhagen, Denmark) was used to evaluate the prognostic value of ZEB-1 in solid tumors. The heterogeneity of pooled results was assessed through Higgin’s I² statistics and Cochran’s Q test. [25, 26] I² < 50% and P > 0.05 indicated the absence of severe heterogeneity and hence, a fixed-effects model (Mantel-Haenszel) was adopted. Otherwise, random-effects model (DerSimonian and Laird) was be performed (I² ≥ 50% and P ≤ 0.05). Subgroup and sensitivity analyses were performed to evaluate the sources of heterogeneity. Begg’s funnel plot was conducted to estimate the risk of publication bias through STATA 12.0 software. A P < 0.05 denoted statistical significance.

The detailed steps of literature selection are shown in Fig. 1. Initially, 840 relevant studies were identified. Eventually, 13 eligible articles published between 2011 and 2017 were included in this meta-analysis, and their main features are summarized in Table 1. A total of 1616 patients with solid tumors from China, Japan, United States of America, Germany, and Norway were analyzed. Besides, there were involved in seven types of solid tumors, including three colorectal cancer, [10‐12] two gastric cancer, [13, 14] two hepatocellular carcinoma, [15, 16] two pancreatic cancer, [17, 18] two esophageal squamous cell carcinoma, [19, 20] one oral cavity carcinoma, [21] and one intrahepatic cholangiocarcinoma. [22] The included studies were performed to analyze OS. HR and 95% CI data were directly obtained from six studies. Data from remaining seven studies were extracted using Kaplan–Meier survival curves. All 13 studies reported a NOS score > 6 (Table 2).

The combined analysis of published data from nine univariate analyses suggested that a high expression level of ZEB-1 was correlated with poor OS in patients with solid tumors (pooled HR: 1.66; 95% CI: 1.45–1.90). Notably, there was no significant heterogeneity reported (P = 0.39, I² = 5%), thus, a fixed-effects model was performed (Fig. 2).The combined analysis of published data from seven multivariate analyses demonstrated that overexpression of ZEB-1 was associated with worse survival outcome in patients with solid tumors (pooled HR: 2.28; 95% CI: 1.58–3.30), and there was significant heterogeneity found (P < 0.01, I² = 82%), hence, a random-effects model was utilized (Fig. 3). Subsequently, a sensitivity analysis was performed to explore the sources of heterogeneity. Exclusion of any individual study from the analysis did not significantly change the results (data not shown). Furthermore, a subgroup analysis was conducted based on various types of cancer to detect heterogeneity. The results indicated a negative effect of high ZEB-1 expression on OS (univariable analysis) in patients with colorectal cancer (pooled HR: 1.80; 95% CI: 1.28–2.54) (Fig. 4a), esophageal squamous cell carcinoma (pooled HR: 1.71; 95% CI: 1.38–2.13) (Fig. 4b) and pancreatic cancer (pooled HR: 1.63; 95% CI: 1.19–2.23) (Fig. 4c). In addition, a similar result was found on OS (multivariate analysis) in patients with colorectal cancer (pooled HR: 2.75; 95% CI: 1.65–4.56) (Fig. 5a), gastric cancer (pooled HR: 2.17; 95% CI: 1.33–3.55) (Fig. 5b) and hepatocellular carcinoma (pooled HR: 2.07; 95% CI: 0.73–5.92) (Fig. 5c). Additionally, subgroup analysis of OS according to different detection methods of ZEB-1 expression showed that high levels of ZEB-1 protein and mRNA expression were both correlated with poor OS in patients with solid tumors (Additional file 1).

In this study, Begg’s funnel plot and Begg’s test were performed to evaluate potential publication bias. The data indicated that there was no significant publication bias for OS in both the univariate (Fig. 6a) and multivariate analyses (Fig. 6b).