Prognostic Implication of EBV Infection in Gastric Carcinomas: A Systematic Review and Meta-Analysis

Background and objectives: This study aims to elucidate the prognostic implications of Epstein–Barr virus (EBV) infection in gastric carcinomas (GCs) through a systematic review and meta-analysis. Materials and Methods: In total, 57 eligible studies and 22,943 patients were included in this meta-analysis. We compared the prognoses of EBV-infected and non-infected GC patients. The subgroup analysis was performed based on the study location, molecular classification, and Lauren’s classification. This study was checked according to the PRISMA 2020. The meta-analysis was performed using the Comprehensive Meta-Analysis software package. Results: EBV infection was found in 10.4% (95% confidence interval (CI) 0.082–0.131) of GC patients. The EBV-infected GC patients had a better overall survival compared with the EBV-non-infected GC patients (hazard ratio (HR) 0.890, 95% CI 0.816–0.970). In the subgroup analysis based on molecular classification, no significant differences were found between EBV+ and microsatellite instability and microsatellite stable (MSS)/EBV− subgroups (HR 1.099, 95% CI 0.885–1.364 and HR 0.954, 95% CI 0.872–1.044, respectively). In the diffuse type of Lauren’s classification, EBV-infected GCs have a better prognosis compared with the EBV-non-infected GCs (HR 0.400, 95% CI 0.300–0.534). The prognostic impact of EBV infection was found in the Asian and American subgroups but not in the European subgroup (HR 0.880, 95% CI 0.782–0.991, HR 0.840, 95% CI 0.750–0.941, and HR 0.915, 95% CI 0.814–1.028). Conclusions: EBV infection is a favorable survival factor for GCs. However, the prognostic implications of EBV infection in the new molecular classification are not clear.


Introduction
Gastric carcinoma (GC) is a common malignancy of the gastrointestinal tract and is a heterogeneous disease with different phenotypes, genotypes, and tumor behaviors [1,2]. GC carcinogenesis is associated with infectious agents, including Helicobacter pylori and Epstein-Barr virus (EBV). Burke et al. first reported a correlation between EBV infection and GC in 1990 [3], but the molecular classification of GC has only recently been introduced, with follow-up studies also being only recently published. Based on this molecular classification, GCs can be categorized as EBV-positive, microsatellite instability-high, genomically stable, and chromosomally instable. In the study by Burke et al., EBV-positive GC comprised 9% of overall GC cases. All EBV-positive tumors clustered together and exhibited an extreme CpG island methylator phenotype distinct from that in the microsatellite Published articles were searched in the PubMed database on 15 August 2022. The keywords used in the search were "gastric carcinoma or gastric cancer or stomach cancer" and "Epstein-Barr virus or EBV". The titles and abstracts of all searched articles were screened for the inclusion and exclusion of each article. Articles that studied the correlation between EBV infection and the overall survival of GCs were included. Case reports, nonoriginal articles, or articles not written in English were excluded. This study was performed by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). The PRISMA checklist is shown in the Supplementary Table S1. In addition, this review was not registered.

Data Extraction
All data were obtained from eligible studies [2, by two independent authors (J.S.P. and N.Y.K.). The extracted data included the authors' information, study location, number of analyzed patients, EBV positivity rates, and information about survival. EBV positivity rates were investigated according to study location and tumor location. For the subgroup analysis, information about survival was additionally gathered based on tumor subtypes, such as the presence of lymphoid stroma, Lauren's classification, and molecular classification. For a quantitative aggregation of the survival results, we obtained the hazard ratio (HR) using one of the following methods. In studies not quoting the HR or its confidence interval (CI), these variables were calculated from the presented data using the HR point estimate, log-rank statistic or its p-value, and the O-E statistic (difference between the number of observed and expected events) or its variance. If those data were unavailable, HR was estimated using the total number of events, the number of patients at risk in each group, and the log-rank statistic or its p-value. Finally, if the only useful data were in the form of graphical representations of survival distributions, the survival rates were extracted at specified times to reconstruct the HR estimate and its variance under the assumption that patients were followed up at constant intervals [60]. The published survival curves were read independently by two researchers in order to reduce variability. The HRs were then combined into an overall HR using Peto's method [61].

Statistical Analyses
The meta-analysis was performed using the Comprehensive Meta-Analysis software package (Biostat, Englewood, NJ, USA). The rates of EBV-infected GCs were evaluated in eligible studies, and the differences in overall survival between EBV-infected and noninfected GCs were investigated. Subgroup analyses for EBV positivity rates were performed based on study and tumor locations. In survival analysis, subgroup analyses were performed based on Lauren's classification and molecular classification. In addition, the overall survival between patients with EBV-infected GCLS and those with non-GCLS were compared. We checked the heterogeneity between the studies based on the Q and I 2 statistics, expressed as p-values. Additionally, we conducted a sensitivity analysis to assess the heterogeneity of the eligible studies and the impact of each study on the combined effects. In the meta-analysis, as the eligible studies used a variety of different populations, a random effects model (rather than a fixed effects model) was determined to be more suitable. In the case of 1 included literature, the p-value of heterogeneity was 1. In addition, the effect size of the fixed effect and the random effect model is the same. The statistical difference between subgroups was evaluated with a meta-regression test. We used Begg's funnel plot and Egger's test to assess the publication bias; if significant publication bias was found, the fail-safe N and trim-fill tests were additionally used to confirm the degree of publication bias. The results were considered statistically significant at p < 0.05.

Selection and Characteristics of the Studies
Through a database search, 1500 articles were obtained. During their review, 1443 articles were excluded. Five hundred fifty-nine articles lacked sufficient information. Of the remaining articles, 385 articles were excluded as they were non-original articles. Another 499 articles were excluded; the reasons for this include non-human studies (n = 256), articles on other diseases (n = 195), written in a language other than English (n = 46), and duplication (n = 2). Finally, 57 eligible articles were selected and included in the meta-analysis ( Figure 1 and Table 1). These studies included 22,943 patients with GC. were compared. We checked the heterogeneity between the studies based on the Q and statistics, expressed as p-values. Additionally, we conducted a sensitivity analysis to as sess the heterogeneity of the eligible studies and the impact of each study on the combine effects. In the meta-analysis, as the eligible studies used a variety of different population a random effects model (rather than a fixed effects model) was determined to be mor suitable. In the case of 1 included literature, the p-value of heterogeneity was 1. In add tion, the effect size of the fixed effect and the random effect model is the same. The statis tical difference between subgroups was evaluated with a meta-regression test. We use Begg's funnel plot and Egger's test to assess the publication bias; if significant publicatio bias was found, the fail-safe N and trim-fill tests were additionally used to confirm th degree of publication bias. The results were considered statistically significant at p < 0.05

Selection and Characteristics of the Studies
Through a database search, 1500 articles were obtained. During their review, 144 articles were excluded. Five hundred fifty-nine articles lacked sufficient information. O the remaining articles, 385 articles were excluded as they were non-original articles. An other 499 articles were excluded; the reasons for this include non-human studies (n = 256 articles on other diseases (n = 195), wri en in a language other than English (n = 46), an duplication (n = 2). Finally, 57 eligible articles were selected and included in the meta analysis ( Figure 1 and Table 1). These studies included 22,943 patients with GC.

Meta-Analysis
First, the EBV positive rates of GCs were investigated and analyzed through a metaanalysis. The estimated EBV positivity rate of GCs was 0.104 (95% CI 0.082-0.131; Table 2). The p-value of the heterogeneity test was less than 0.001. In addition, there was no significant publication bias in an Egger's test (p = 0.860). In the subgroup analysis based on study location, the EBV positivity rates in the Asian, American, and European subgroups were 0.114 (95% CI 0.080-0.158), 0.107 (95% CI 0.079-0.142), and 0.084 (95% CI 0.070-0.100), respectively. In the subgroup analysis, there was no significant publication bias. The EBV positive rates of antrum, body, and cardia/fundus were 0.083 (95% CI 0.055-0.123), 0.130 (95% CI 0.095-0.176), and 0.182 (95% CI 0.126-0.257), respectively. Comparisons of the overall survival between the patients with EBVaGC and those with non-EBVaGC were performed. Those with EBVaGC showed better overall survival than those with non-EBVaGC (HR 0.882, 95% CI 0.813-0.957; Table 3). The p-value of the heterogeneity test was less than 0.001. In addition, there was no significant publication bias in an Egger's test (p = 0.297). Next, a subgroup analysis was performed based on the histologic subtype, Lauren's classification, and molecular classification. There was a significant difference in the overall survival between the patients with EBVaGC and those with non-EBVaGC in GCLS, but not in non-GCLS (HR 0.178, 95% CI 0.059-0.537 and HR 0.870, 95% CI 0.620-1.222, respectively). The Egger's test was not performed due to the number of included articles in GCLS and non-GCLS subgroups. In the diffuse type of Lauren's classification, the patients with EBVaGC had better prognosis than those with non-EBVaGC (HR 0.400, 95% CO 0.300-0.534). However, there was no significant difference in the overall survival between the patients with EBVaGC and those with non-EBVaGC in the intestinal type of Lauren's classification (HR 1.364, 95% CI 0.425-0.4379). In addition, Egger's test was not performed due to the number of included articles in intestinal and diffuse type subgroups. In the molecular classification, there were no significant differences in the overall survival between the patients with GCs who were EBV positive and MSIhigh (MSI-H) and those with GCs who were microsatellite stable (MSS) and EBV negative (HR 1.099, 95% CI 0.885-1.364 and 0.954, 95% CI 0.872-1.044, respectively). There was no significant publication bias in this subgroup analysis. In the Asian and American subgroups, the patients with EBVaGC showed better prognosis than those with non-EBVaGC (HR 0.880, 95% CI 0.782-0.991 and HR 0.840, 95% CI 0.750-0.941, respectively; Figure 2A,B). However, there were no significant correlations between EBV infection and overall survival in the European subgroup (0.915, 95% CI 0.814-1.028; Figure 2C).

Discussion
The estimated EBV positivity rate in the present study was 10.4%. In the eligible studies, the EBV positivity rates ranged from 2.0% to 40.1% [2,. A meta-analysis is useful for elucidating the prognostic role of EBV in GCs as it has a diverse distribution of EBV positivity. To the best of our knowledge, this meta-analysis was the first to evaluate the prognostic role of EBV positivity based on the molecular classification of GCs. Our results show that the prognostic impact of EBV differed between the overall cases and the molecular classification.
Although the estimated EBV positivity rate was 10.4%, the EBV positivity rate in the eligible studies reached 40.1%. In addition, the EBV positivity rates differed among the various study locations. Therefore, it is important to evaluate the prognostic impact of EBV positivity in various subgroups. In our previous study, EBV infection significantly correlated with sex, GCLS, programmed death ligand-1 expression, and tumor-infiltrating lymphocytes [62]. In another meta-analysis, a correlation between EBV infection and GC prognosis was reported [63]. Liu et al. reported that EBV was significantly correlated with a favorable prognosis of GCs, which was similar to our results [63]. However, this meta-analysis was conducted in 2015 and included eligible studies published from 1996 to 2014 [63]. Another previous meta-analysis included 24 eligible studies. Furthermore, the present meta-analysis included 57 eligible studies. In addition, a subgroup analysis based on molecular classification was not performed [63]. Recently, the molecular classification of GCs has been introduced and investigated [1]. As EBV infection is important in this molecular classification, a detailed evaluation based on molecular classification is needed.
Previous studies have reported that the patients with EBVaGC had a better prognosis compared with those with non-EBVaGC. In the present meta-analysis, EBVaGC was significantly correlated with better prognosis compared with non-EBVaGC (HR 0.890, 95% CI 0.816-0.970). We additionally performed a subgroup analysis based on various factors such as study location, Lauren's classification, and molecular classification. In the subgroup analysis based on the study location, statistical significance was found in the Asian and American subgroups, but not in the European subgroup. This result was similar to that of a previous meta-analysis [63]. According to Lauren's classification, intestinal-type GCs had more favorable survival than diffuse-type GCs [64,65]. The prognostic impacts were also identified in the diffuse type of Lauren's classification and GCLS (HR 0.400, 95% CI 0.300-0.534 and HR 0.178, 95% CI 0.059-0.537, respectively). Koriyama et al. reported the prognostic significance of EBV involvement according to Lauren's classification [33]. They suggested that the opposite prognostic impact of EBV was identified in the diffuse and intestinal types of Lauren's classification. Interestingly, a favorable prognosis was identified in the diffuse type, but not in the intestinal type, according to Lauren's classification. However, our results differ from those of a previous study.
The different tumor behaviors of GCs can be affected by their histologic and molecular characteristics [1,66], and recently, a molecular classification of GCs has been introduced [1]. The patients with EBV-positive or MSI-high GC had better prognosis compared with those with EBV-negative or MSS GC [1]. Kohlruss et al. analyzed prognosis according to the new molecular classification, but no statistical differences were found between the molecular subgroups [67]. The prognostic impact of EBV according to this molecular classification is unclear in individual studies. The present study also performed a subgroup analysis based on this molecular classification. No significant difference was observed in the prognosis between the EBV positive and MSI-H subgroup and the MSS and EBV negative subgroup (HR 1.099, 95% CI 0.885-1.364 and HR 0.954, 95% CI 0.872-1.044, respectively). Since the nature of the tumor may differ depending on the molecular classification, the therapeutic effect of chemotherapy may also vary. Hence, further studies are needed to determine the prognostic role of EBV according to the molecular subtypes or GC therapy.
The patients with EBVaGC have better prognosis compared with those with non-EBVaGC. Similar results were obtained in our meta-analysis. However, individual studies reported different results, while some showed contrasting results. The intestinal type of EBVaGC had a worse prognosis than the intestinal type of non-EBVaGC [33]. In addition, Lin et al. reported a correlation between EBV and worse prognosis [38]. Interestingly, several studies have shown no significant difference in the survival rates between the patients with EBVaGC and those with non-EBVaGC. A correlation between EBV positivity and better prognosis was reported in a few eligible studies [17,18,33,40,42,45,46,51,58]. Furthermore, other studies have shown different results regarding the prognostic significance of EBV positivity. Therefore, this meta-analysis may be useful for interpreting the prognosis of EBVaGC. In a previous meta-analysis, EBV positivity was significantly correlated with sex [62]. The factors associated with tumor progression and behavior were not correlated with EBV positivity [62]. However, in the present study, EBVaGC was significantly correlated with favorable prognosis. A significant correlation was also found between EBV and GCLS levels. We also compared the prognostic impact of EBV infection between the patients with GCLS and those without GCLS. EBV was associated with a favorable prognosis in the patients with GCLS but not in those with non-GCLS (HR 0.178, 95% CI 0.059-0.537 and HR 0.870, 95% CI 0.620-1.222, respectively). To understand this discrepancy, an analysis of the results of the subgroup analysis can be helpful.
This study has some limitations. First, some studies show that EBVaGC is associated with worse prognosis. This discrepancy might be due to the different populations evaluated and methods used for detecting EBV. The significance of the prognostic implication of EBV was reported in 7 out of 32 Asian studies, 1 out of 11 American studies, and 1 out of 12 European studies. The individual studies had sample sizes ranging from 43 to 1328 participants. The EBV positivity ranged from 2.0% to 40.1%. As the individual studies included a variety of different populations, the results may vary between eligible studies. However, our meta-analysis did not confirm this cause. Second, differences were found in the number of studies conducted before and after the introduction of the molecular classification. The amount of analyzed results according to the molecular classification were too few compared with the total number of analyzed results. Third, the patients with GCLS and those with diffuse-type GC and EBV infection had better prognosis. However, since the number of studies presenting this information is too small, further studies are warranted. Fourth, the EBV positive rates were differed by the tumor location. However, we could not perform the detailed analysis for the prognostic impacts of EBV based on the tumor location due to insufficient information. Further study on the impact of the tumor location will be useful for predicting the prognostic role of EBV.

Conclusions
Although EBV infection was significantly correlated with favorable survival in the patients with GCs, the details of this patient group differed according to the new molecular classification and by subtype. However, this meta-analysis may still be considered significant as the prognostic role of EBV may vary depending on the GC subtype. Hence, further detailed studies on the subtypes with fewer cases are needed.