Golgi protein 73 versus alpha-fetoprotein as a biomarker for hepatocellular carcinoma: a diagnostic meta- analysis

Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related death worldwide. There were estimated 748,000 new cases of liver cancer worldwide in 2008, causing 696,000 deaths [1]. As most patients with HCC are diagnosed at an advanced stage with underlying liver dysfunction [2], the mortality rate of HCC is similar to the incidence rate. Early detection of HCC is therefore extremely important in improving the survival of patients. Alpha-fetoprotein (AFP), together with hepatic ultrasonography, is the most common marker used in clinical practice to detect HCC in cirrhotic patients, and has been considered the gold-standard serum marker for screening patients at high risk for HCC, as well as for the diagnosis and monitoring of responses to HCC treatment for over 40 years. But the clinical value of AFP is challenged in recent years due to low sensitivity and specificity [3‐5]. In addition, AFP levels greater than 500 ng/ml are correlated with the tumor size: 80% of small HCC show no increase of AFP concentration [6]. Some patients with cirrhosis and/or hepatic inflammation could have an elevated level of AFP without the presence of tumors. Therefore, serum markers with better diagnostic accuracy are needed for HCC.

Golgi protein 73 (GP73, also known as Golph2), is a resident Golgi-specific membrane protein expressed by biliary epithelial cells in normal liver, and its expression is increased markedly in chronic liver diseases, especially in HCC cells [7]. There have been studies reporting the use of serum GP73 as a serum marker for HCC, but the results are heterogeneous and even conflicting. The objective of the present review was to synthesize and analyze the results from systematic selection of research papers that evaluated the diagnostic accuracy of serum GP73 by directly comparing it with AFP for the diagnosis of HCC.

A total of 149 articles were found, of which 20 publications dealing with GP73 for the diagnosis of HCC were considered to be eligible for inclusion in the analysis. After full-text review, 12 studies were excluded: five articles were excluded because they did not allow the calculation of sensitivity or specificity [10‐14]; five articles were excluded because of lacking AFP test [7, 15‐18], and two were excluded because they had a suspected overlap study population with another study in which results were most completely reported (Table 1) [19, 20]. Finally, 8 studies were available for the meta-analysis, including 5,988 patients who received both serum GP73 and AFP tests (Figure 1) [21‐28]. The characteristics of each study are shown in Table 2.

We found and assessed eight studies that directly compared the diagnostic accuracy of serum GP73 and AFP in the same patient population. The results demonstrate that GP73 is a comparable marker for HCC to AFP. However, the studies showed numerous methodological limitations, a broad range of diagnostic accuracy values and heterogeneity. The methodological quality of the eight studies is poor, and one of the studies did not even report the age and sex of the patients included. Five of the studies stated that serum GP73 was advantageous over AFP, while the remaining three studies reported opposite results. Because of the limited number and the different cutoff values of the studies included in this meta-analysis, we were not able to explore causes for the existing big heterogeneity by meta-regression.

The increasing incidence of HCC worldwide has sparked a new interest in HCC serum markers. A number of novel candidate markers have been suggested. GP73 is a 400-amino acid 73-kDa transmembrane glycoprotein that normally resides in epithelial cells of many human tissues [32]. Higher levels of serum GP73 were first found in patients with hepatitis B virus-related HCC by Block in 2005 [7]. Although the precise mechanism of GP73 elevation in the circulation remains obscure, serum GP73 has gained great interest for its potent role in the diagnosis of HCC. Western blotting, immunoblotting and ELISA are three major methods used to assay GP73. The former two are semiquantitative and laborious, but the results of ELISA are disappointing. Six studies [18, 21, 23, 24, 27, 33] using ELISA method failed to find significant elevation of serum GP73 in HCC groups compared with that in liver cirrhosis groups. Recently described GP73-specific serum autoantibodies might interfere with ELISA analysis. Researchers have found several isoforms of GP73 that correspond with different patterns or levels of glycosylation [34]. Whether measuring HCC-specific GP73 isoform helps improve the diagnostic accuracy still needs further research.

The goal of cancer-screening is to detect tumors at an early stage when successful treatment is possible [35]. An ideal biomarker should be specific and noninvasive. The field of HCC serum markers is primed for new candidate markers that have the potential to replace their flawed leader AFP [36]. To avoid substantial confounding factors in comparisons, we only used studies that directly compared GP73 with AFP in the same patients. However, fully paired studies are usually scant. Despite the big heterogeneity, we still found that GP73 may be a comparable marker to AFP for HCC. However, cancer is a diverse class of diseases that differ widely in their causes and biology, and therefore it is unlikely for a single biomarker to detect all cancers of a particular organ with high specificity and sensitivity. The diagnostic value of GP73 combined with AFP for HCC was stated in Wang's and Mao's reports, and the results seem better than the single marker. However, further observation is needed in more studies.

Only three studies reported diagnostic utility of serum GP73 compared AFP by tumor stage and the conclusions were conflicting. The diagnostic ability of GP73 to detect HCC in an early stage needs further observation.

One of the major goals of meta-analysis is to explore reasons for heterogeneity rather than computation of a single summary measure [37]. Notably, the differential cutoff values within the studies could cause a threshold effect. Different study populations appeared to have substantially influenced the diagnostic sensitivity [38]. In our meta-analysis, the prevalence of HCC, cutoff value, consecutive patient selection and assay method were used in the meta-regression analysis to assess the effect of study methodology and threshold on DOR. Because of the small number of studies included, we used the permutation test recommended by Higgins and Thompson [39]. But we failed to find the reason responsible for the existing heterogeneity. The following reasons may hamper the statistical analysis of sources of heterogeneity: First, many reports of studies on diagnostic accuracy lacked information on key elements of design and conduct. Without complete and accurate reporting, we cannot correctly identify potential sources of bias and variability. Second, few studies using direct comparisons are available. The small numbers also made subgroup analysis unavailable. Third, the diagnostic meta-analysis is also threatened by publication bias. Investigating publication bias for diagnostic tests is problematic. Funnel plot-based tests used to detect publication bias in review of randomized controlled trials have proven to be seriously misleading for diagnostic studies, and alternatives have poor power [40].

In conclusion, our meta-analysis found that GP73 is a comparable marker to AFP as an independent diagnostic tool for the diagnosis of HCC. The combination value of GP73 and AFP still needs further research. The currently existing studies on GP73 have not yet met the most stringent criteria defined by the Early Detection Research Network. Future studies should be designed to prospectively test markers in appropriately selected risk populations.