The prognostic significance of pretreatment serum γ-glutamyltranspeptidase in primary liver cancer: a meta-analysis and systematic review

Aim: To assess the prognostic value of the pretreatment serum γ-glutamyltranspeptidase (GGT) level in patients with primary liver cancer (PLC). Methods: Relevant studies were systematically searched online on Web of Science, PubMed, and Embase databases published until 9 October 2018. The end points were overall survival (OS), recurrence-free survival (RFS), and disease-free survival (DFS). Meta-analysis was conducted using hazard ratio (HR), and its 95% confidence interval (CI) as effect measure. Results: A total of 33 eligible studies with 9238 patients with PLC were included in this meta-analysis. The synthesized analysis showed that that higher serum GGT level was significantly related to poorer OS (HR: 1.79, 95% CI: 1.66–1.93, P<0.01), RFS (HR: 1.60, 95% CI: 1.46–1.77, P<0.01), and DFS (HR: 1.52, 95% CI: 1.33–1.73, P<0.01) of patients with PLC. Subgroup analyses demonstrated that the negative prognostic impact of higher serum GGT level on OS and RFS was still of significance regardless of ethnicity, pathological type, sample size, cut-off value, first-line treatment, and analysis type. Conclusion: The pretreatment serum GGT might be a predictive factor of poor prognosis for PLC patients.


Introduction
Primary liver cancer (PLC), including hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC), is one of the most common human malignant neoplasms [1]. Although a comprehensive therapy integrating surgical resection, thermal ablative techniques, chemotherapy, and molecular-targetted therapy has been applied to deal with patients with PLC in recent years, the long-term survival of patients remains rather unfavorable [2,3]. Hence, it is of great importance to identify biomarkers for accurately predicting the prognosis of patients with PLC, which may contribute to optimizing individual treatment and then improve the long-term outcomes.
γ-Glutamyltranspetidase (GGT) is a cell-membrane bound enzyme that modulates the metabolism of glutathione (GSH), catalyzes the degradation of extracellular GSH, and subsequently facilitates amino-acid recovery for intracellular GSH synthesis [4]. GGT has been recognized to enhance cellular antioxidant defenses [4]. In addition, several researchers reported that GGT might be involved in contributing to the tumor initiation, progression, invasion, and drug resistance [5][6][7]. More importantly, increased serum level of GGT was found to be linked with worse prognosis in several human malignancies, including PLC.
Up to date, numerous studies have reported that serum GGT level was related to the prognosis of patients with PLC , while the limitation of sample size in individual studies might affect the reliability

Data extraction
Two reviewers extracted the required data from all eligible studies independently and inconsistencies were worked out by discussion. The required information included first author's family name, publication time, country, recruitment time, median age, sample size, disease stage, primary treatment type, GGT cut-off level, follow-up time, and HR estimations for prognostic indicators, including overall survival (OS), DFS, and RFS.

Quality assessment
The methodological quality of all eligible studies was evaluated by two reviewers using Newcastle-Ottawa scale (NOS) independently [39]. The maximum of 9 stars was applied to assess the selection, comparability as well as exposure, and outcome of each included study. In this meta-analysis, we defined studies with no less than 7 stars as high quality and 6 stars as moderate quality.

Statistical analysis
We measured the effects of serum GGT level in OS, DFS, and RFS using the pooled HRs and 95% CIs. Heterogeneity was evaluated with I 2 test. The random effect model was chosen for pooling analysis if significant heterogeneity existed (I 2 >50% or P-value of heterogeneity test <0.05). If not (I 2 ≤ 50% or P-value of heterogeneity test), the fixed effect model was applied to perform pooling analysis. When the pooled HRs and 95% CIs were >1, it indicated that PLC patients with higher level of serum GGT had poorer prognosis as compared with those with lower level. Publication bias was evaluated using Begg's funnel plot and Egger's test [40,41]. When there is significant publication bias, trim-and-fill method was utilized to evaluate the effect of publication bias on the robustness of the pooled HR [42]. In addition, subgroup analysis and sensitivity analysis were also performed to assess the influence of each study on the pooled HR. P<0.05 was considered as statistically significant. The subgroup analysis and sensitivity analysis in this meta-analysis were fulfilled by means of STATA version 12.0 (Stata Corporation, College Station, TX), and the other statistical analysis were carried out with Review Manager (RevMan) version 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen).

Search results
The processes of searching and filtering of publications were presented in Figure 1. Initially, a total of 503 articles were identified. Then we used Endnote X7 software to exclude 217 duplicated publications with 297 articles left for further identification. Subsequently, after scanning the titles, abstracts, and publication types of these 297 articles, 230 articles were excluded for meeting abstracts, reviews, case reports, or comments (n=39), irrelevant topics (n=181), and cell and animal experiments (n=10), and 60 articles remained for full-text review. In the process of full-text review, 34 studies were further excluded for no available data (n=29) and enrolling the same or overlapping population (n=5). Ultimately, a total of 33 eligible studies with 9238 patients were included in this meta-analysis 43,44].

Study characteristics and quality assessment
The detailed characteristics of the 33 eligible studies were shown in Table 1. These studies were originally published between 2011 and 2018. Of these studies, 30 studies with 8056 patients were from China, and 3 studies with 1182 patients from Italy, France, and U.S.A. A total of 28 studies reported about HCC and 5 studies focussed on ICC. In addition, OS was mentioned in 29 studies, RFS was found in 13 articles, and DFS was reported in 5 studies. We assessed the quality of the 33 included studies in our meta-analysis following the guideline of the NOS. The included studies were given 6-8 scores, indicating that the eligible studies were of moderate to high quality.

Relationship between serum GGT and DFS in PLC patients
The association between serum GGT level and DFS was investigated in five of the included studies [10,13,14,35,37].

Subgroup analysis
In order to explore the potential sources of heterogeneity of the combined HR for OS, we conducted subgroup analyses through stratifying eligible studies by ethnicity (Asian and non-Asian), pathological type (HCC and ICC), sample size (≤300 and >300), cut-off value (≤50, 50-75, 70-100, and >100), disease stage (only unresectable), primary treatment (Liver resection, transcatheter arterial chemoembolization (TACE), and ablation), and analysis type (multivariate analysis and univariate). It should be noted that the subgroup of disease stage (unresectable) only covered four included studies that referred to OS in patients with unresectable HCC. Each of the remaining 25 eligible studies referring to OS enrolled patients with different disease stages, or even did not provide clear information about disease stages, so these studies showed no similarity in terms of disease stage and we could not classify these studies into subgroup of disease stage for pooling analysis. The same situation is also for subgroup of primary treatment.  We established this subgroup according to TACE, liver resection, and ablation, which only covered 26 eligible studies that referred to OS. However, the remaining three eligible studies that referred to OS showed no similarity in primary treatment, so we could not classify these studies into subgroup of primary treatment for pooling analysis either. From the results, we found that there was no subgroups in which the heterogeneity of the combined HR for OS (Table 2) disappeared completely, indicating that those factors might not be the sources of heterogeneity of the pooled HR for OS. Although we failed to identify the possible sources of the pooled HR for OS, we demonstrated that our pooled result was robust, since the pooled HR for OS was >1 continuously, and its correspondent CI did not overlap 1 in all subgroups ( Table 2). Although no significant heterogeneity was observed for the pooled HR for RFS, we still performed subgroup analysis to test whether our pooled HR for RFS was stable and dependable. The results showed that significant heterogeneity was still not detected and the pooled HR for RFS was >1 continuously with its corresponding CI not overlapping 1 in any subgroup ( Table 3), suggesting that our pooled result was reliable.

Figure 4. Forest plot of the HR for the relationship between pretreatment serum GGT level and DFS in PLC patients
For limited number of the eligible studies about DFS, subgroup analysis was not conducted to investigate the sources of the pooled HR for DFS.

Sensitivity analysis
Sensitivity analysis was used to evaluate the robustness of the pooled results. The results showed that the pooled HRs for OS ( Figure 5), RFS ( Figure 6A), and DFS ( Figure 6B) did not alter substantially when the included studies were sequentially omitted in each step, revealing the robust stability of our pooled results.  Abbreviation: df, degree of freedom.

Publication bias
The assessment for publication bias was fulfilled by Begg's funnel plot and the Egger's tests. As our results show, there were obvious asymmetries for Begg's funnel plots of OS ( Figure 7A) and RFS ( Figure 7B). Additionally, the Egger's test

Figure 7. The publication bias assessment of the pooled HRs for OS (A) and RFS (B)and the trim-and-fill analysis of the effect of publication bias on the pooled HRs for OS (C) and RFS (D)
also suggested that there was significant publication bias for OS (P<0.01) and RFS (P<0.01) in this meta-analysis. Therefore, we applied trim-and-fill method to assess the impacts of the publication bias on the reliability of the pooled HR for OS and RFS. From the results, we observed that the adjusted funnel plots for OS and RFS turned symmetric ( Figure 7C,D). Furthermore, the result of trim-and-fill method showed that the adjusted pooled HRs for OS and RFS were still >1, and meanwhile their corresponding CIs did not include 1. From the results stated above, we concluded that the publication bias did not substantially affect the robustness of the pooled HR for OS and RFS. Considering the limited number of the included studies about DFS, we did not perform the Begg's test and Egger's test to assess the publication bias for DFS.

Discussion
To our best knowledge, the present study is the first meta-analysis to synthetically analyze the prognostic significance of serum GGT in patients with PLC. The results of this meta-analysis demonstrated that a higher serum GGT is a useful biomarker for poorer OS, RFS, and DFS in PLC patients and this result would not change substantially when any eligible study was sequentially omitted in sensitivity analysis. Moreover, our subgroup analyses validated that higher serum GGT remained an effective prognostic indicator in spite of ethnic background, sample size, pathological type, cut-off value, primary treatment, and analysis type. Several potential mechanisms by which GGT affects tumor biology have been investigated. As a membrane-bound enzyme, GGT plays an essential role in maintaining the production of intracellular glutathione (GSH), which protects cells from reactive oxygen compounds and free radicals as a key antioxidant element [45]. Therefore, GGT may contribute to the formation of tumor microenvironment that protects tumor cells form oxidative stress and drug cytotoxic effects [46,47]. Furthermore, reactive oxygen species (ROS), as a kind of carcinogenic factor, could up-regulate GGT expression through the redox regulation of many genes [5]. Therefore, it could be postulated that oxidative stress, as a part of the tumor microenvironment, might promote tumor tissues to produce GGT, and an elevated GGT expression might protect cells from the impacts of oxidative stress. For instance, up-regulation of GGT could help prostate epithelial cell overcome hydrogen peroxide-induced apoptosis [48], but this should be further studied in tumor cells. Additionally, GGT could also induce the production of additional source of endogenous ROS, leaving cells exposed to persistent oxidative stress, subsequently inducing aberrant CpG island methylation, DNA damage and genome instability, and ultimately promoting several carcinogenic processes, such as cellular growth, proliferation, and survival [5,7,49,50]. Evidences showed that several inflammatory cytokines, including tumor necrosis factor α, interferon-α, and interferon-β, could induce GGT expression [45,46]. Additionally, it was also reported that serum GGT level associates with the active status, fibrosis and cirrhosis stage of chronic hepatitis [51,52] and functions as a biomarker of the inflamed liver microenvironment in hepatitis [53]. Therefore, it may be hypothesized that GGT reflects or participates in the tumor-associated inflammatory responses to predict the prognosis of tumor patients. Abnormal Ras signaling transduction plays a key role in promoting cancer progression and closely correlates with the prognosis of cancer patients [54], and oxidative stress-induced activation of Ras-mitogen-activated protein kinase pathways could up-regulate GGT expression in colon cancer cells [55,56]. Thus, it is possible that elevated GGT reflects the abnormal activation of Ras signaling pathway to associate with the prognosis of tumor patients. However, the exact direct mechanisms of elevated GGT in cancer initiation and progression was rarely declared, so more studies should be performed in this regard.
A larger sample size was one of the strengths of our meta-analysis, which made our study have more statistical power than any of the included studies. In addition, our meta-analysis was performed by analyzing a massive dataset from multicenters and this made our conclusions more generalizable. However, there were also several limitations in our meta-analysis. First, most important limitation was that there was significant heterogeneity for data synthesis of OS. Although subgroup and sensitivity analyses were performed, the main source of heterogeneity was not identified. It may be possible that the significant heterogeneity derived from the inconsistencies in patient characteristics and study designs, and thus more homogeneous studies are required to validate our findings. Second, our study was a literature-based analysis, and thereby had a risk of publication bias, in which predominantly positive results had a tendency to be published, ultimately exaggerating our estimation for the relationship between serum GGT and survival. Third, the cut-off values applied to define the elevated GGT level across the included studies were inconsistent. For one thing, it might introduce the heterogeneity into our meta-analysis and weaken the reliability of our conclusions. For another thing, the different cut-off values made it difficult for doctors to make clinical decisions based on GGT level of patients with PLC. Finally, sometimes, serum GGT could have been influenced by other non-neoplastic conditions, such as the diseases of hepatobiliary tract, pancreatic, heart disease, and alcohol abuse. Moreover, some of the included studies in this meta-analysis did not clearly state that PLC patients with those conditions were excluded or explore the effects of those factors on the prognosis. Last but not the least, many cancer prognosis-associated variables, such as tumor grade, tumor differentiation, and additional treatment, were not available for sensitivity and subgroup analyses, since most of included studies did not provide enough information about these variables. Therefore, the pooled results in this meta-analysis may face unavoidable bias risks.
In conclusion, this meta-analysis demonstrated that higher serum GGT level was associated with poorer prognosis for PLC patients. The serum GGT was an economical and effective prognostic biomarker, which could be applied for risk stratification and formulating individualized treatments for PLC patients. Considering the limitation of our meta-analysis analysis, more prospectively well-designed studies are demanded to confirm our findings and meanwhile investigate the correspondent mechanisms deeply.