Helicobacter pylori infection reduces the risk of Barrett's esophagus: A meta‐analysis and systematic review

Abstract Introduction The prevalence of Helicobacter pylori infection (HPI) has been decreasing in developed countries, with an increasing prevalence of Barrett's esophagus (BE) and esophageal adenocarcinoma (EAC) at the same time. The aim of our meta‐analysis was to quantify the risk of BE in the context of HPI. Methods A systematic search was conducted in 3 databases for studies on BE with data on prevalence of HPI from inception until December 2016. Odds ratios for BE in HPI were calculated by the random effects model with subgroup analyses for geographical location, presence of dysplasia in BE, and length of the BE segment. Results Seventy‐two studies were included in the meta‐analysis, including 84 717 BE cases and 390 749 controls. The overall analysis showed that HPI reduces the risk of BE; OR = 0.68 (95% CI: 0.58‐0.79, P < .001). Subgroup analyses revealed risk reduction in Asia OR = 0.53 (95% CI: 0.33‐0.84, P = .007), Australia OR = 0.56 (95% CI: 0.39‐0.80, P = .002), Europe OR = 0.77 (95% CI: 0.60‐0.98, P = .035), and North‐America OR = 0.59 (95% CI: 0.47‐0.74, P < .001). The risk was significantly reduced for dysplastic BE, OR = 0.37 (95% CI: 0.26‐0.51, P < .001) for non‐dysplastic BE, OR = 0.51 (95% CI: 0.35‐0.75, P = .001), and for long segment BE, OR = 0.25 (95% CI: 0.11‐0.59, P = .001) in case of HPI. Conclusions This extensive meta‐analysis provides additional evidence that HPI is associated with reduced risk of BE. Subgroup analyses confirmed that this risk reduction is independent of geographical location. HPI is associated with significantly lower risk of dysplastic, non‐dysplastic, and long segment BE.


| BACKG ROU N D
Barrett's esophagus (BE) is the only known precursor for esophageal adenocarcinoma (EAC). 1 The prevalence of BE and incidence of EAC have been increasing in recent decades 2 and EAC often is evolved in BE. 1,3 At the same time, the prevalence of Helicobacter pylori infection (HPI) is decreasing in developed countries. 4 There are multiple individual studies, both with evidence for and against the risk reduction in case of HPI. In 3 of the 4 previous metaanalyses, HPI proved to reduce the risk of BE. [5][6][7] On the contrary, Wang et al 8 did not find a clear relationship between HPI and BE in their analysis. The 3 earlier meta-analyses used small subsets of studies; they included 5, 9, and 12 trials. [6][7][8] The most recent and extensive meta-analysis of Fischbach identified 49 trials with data on the association between HPI and BE. Besides proving the risk reduction, their other main findings were the significant heterogeneity among the studies included and a marked risk reduction in the case of CagApositive strains of H. pylori. The source of heterogeneity was one of the foci of their discussion and they concluded that both selection and information bias potentially contributed to their results. 5 The above meta-analyses have not published analytical results of subgroup analysis for geographical location of the study populations, for the segment length of the BE, and for the presence of dysplasia in BE. Our aim was to update the most recent meta-analysis which included studies until 2010 5 and to investigate and quantify the risk of BE in these subgroups.

| Protocol
An epidemiological meta-analysis and systematic review was performed using the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P). 9 The analysis was registered in advance on PROSPERO with registration number CRD42017077509. Additional articles were identified from the reference lists of eligible primary studies.

| Inclusion and exclusion criteria
All studies with relevant information on HPI prevalence in BE patients and controls within the same population were included in our analysis. All studies with abstracts in English were included; fulltext articles in languages other than English were read, appraised, and data were extracted by researchers who spoke and understood the respective language. Full-text articles and abstracts were both included. Different articles reporting data on the prevalence of HPI (proven by serological and/or histological studies and/or stool antigen testing and/or bacterial culture and/or rapid urease or urea breath test) and BE from the same population were thoroughly scrutinized and only one record with the highest number of BE cases was included in the meta-analysis. Articles from identical populations where the prevalence of HPI was more detailed for different lengths of BE were excluded from the overall analysis, but they were included in the subgroup analysis for BE segment length.
All types of observational studies, such as case control and crosssectional studies were included, regardless whether they were prospective or retrospective. Non-human studies and review articles were excluded.

| Data extraction
Numeric data were extracted by 3 investigators and manually populated onto a purpose designed Excel 2016 sheet (Office 365, Microsoft, Redmond, WA, USA). Data were collected on year of publication, study type, geographical location, number of cases and controls, and basic demographics (age, sex ratio) in both groups and method(s) of HPI testing. Most importantly, data were collected on the prevalence of HPI in BE cases and controls, also in dysplastic and non-dysplastic BE and in different segment lengths of BE, for further subgroup analysis. Data on prevalence of HPI by CagA-positive strains were also collected. Other relevant findings were mentioned in an additional column as free text. The data extraction was reviewed and conflicts were resolved by the first author.

| Statistical analysis
Helicobacter pylori infection prevalence data from individual studies were extracted and raw data (number of BE patients with HPI, number of patients without HPI, number of controls with HPI, number of controls without HPI) were calculated, followed by the calculation of odds ratios (ORs) and 95% confidence intervals (CIs) for risk of BE in case of HPI. Adjusted ORs from the original articles were not extracted. Pooled estimates were calculated with random effects model using the DerSimonian-Laird method. 10 Results of the metaanalysis were displayed graphically on forest plots. Heterogeneity was assessed using Cochrane's Q and the I 2 statistics, where Q exceeds the upper tail critical value of chi-square on k−1 degrees, and I 2 represents the percentage of effect size heterogeneity that cannot be explained by random chance. As suggested by the Cochrane Handbook, I 2 values were interpreted as moderate (30%-60%), substantial (50%-90%), and considerable (75%-100%) heterogeneity. 11 Publication biases of the included studies were checked by Egger's test 12 and by visual assessment of funnel plots.
All calculations were performed by Stata 11 data analysis and statistical software (Stata Corp LLC, College Station, TX, USA).

| Analysis of risk of bias and quality assessment
The assessments of both the risk of bias and the risk of quality were done at the outcome level.
A modified Newcastle-Ottawa Scale for case control studies was used for the quality assessment of the individual studies included in our meta-analysis, with the following items, and the result of the assessment was graphically demonstrated in a table with color codes, green: low risk of bias; yellow: moderate or unknown risk of bias; red: high risk of bias.
The questions for the risk assessment were as follows:

| Study selection
Our search strategy initially identified 1705 potential studies.
Removal of duplicates was followed by screening first the titles, and then the abstracts, leaving 96 studies for full-text review, including 8 additional studies identified in the reference lists of the primary eligible studies. [13][14][15][16][17][18][19][20] Thirteen studies were excluded, as they did not provide sufficient data (reasons for exclusion detailed in Appendix S1). Data were extracted from 83 studies 13-95 ; however, 11 of these studies had to be excluded from the statistical analysis as they contained data from same populations already described in other articles. 22,35,38,39,52,56,65,68,[80][81][82] Our final statistical analysis included 72 studies. Of the 72 articles, 2 studies contained data from populations already reported in the 70 studies; however, these had detailed data on the different prevalence of HPI in different segment lengths of BE, therefore these were only included in the subgroup analysis. 45,84 The study selection process is shown in Figure 1. The summary of the characteristics of the studies included in our analysis is shown in Table 1.

Helicobacter pylori infection
Our results confirmed an overall risk reduction OR = 0.68 (95% CI: 0.58-0.79, P < .001) by the calculation from the data of the 70 studies, including a total of more than 90 000 BE cases and nearly 400 000 controls. Heterogeneity was substantial, I 2 = 84.0%.   Figure S1.

| Risk of different segment lengths of Barrett's esophagus in case of Helicobacter pylori infection
Prevalence of HPI for different segment lengths of BE was detailed in 7 studies and data were suitable for meta-analy sis. 44,45,47,66,70,71,83,84,94 Two articles had detailed data on ultrashort segment BE (USSBE, <1 cm) 66,94 and they were not included in the short segment BE (SSBE) subgroup. We note that the new guideline of the British Society of Gastroenterology defines BE by at least 1 cm of metaplastic columnar lining, which questions the justification of the diagnosis of USSBE. 96 However, the metaanalytical calculation was performed for this subgroup as well.
In SSBE, the pooled OR was not statistically significant, but there is likely a risk reduction, OR = 0.63 (95% CI: 0.32-1.26, P = .191). The results on USSBE or intestinal metaplasia at the cardia are not suitable for any conclusion, but there does not seem to be a reduced risk. The results are detailed in Figure 3.

| Risk of dysplasia in Barrett's esophagus in case of Helicobacter pylori infection
Prevalence of HPI in association with the presence of dysplasia in BE was detailed in 7 studies and data were suitable for meta-analy

| Additional subgroup analyses to identify the source of heterogeneity
In order to understand the association between the risk of BE and the prevalence of HPI, further subgroup analyses were performed.
Stratification by the different control groups was possible for 4 subgroups of studies with population, gastro-esophageal reflux disease (GERD), endoscopy, and other controls as indicated in Table 1.
In subgroups of studies with population and GERD controls, the ORs were not significant. Only the studies with endoscopy controls showed a significant risk reduction OR = 0.48 (95% CI: 0.31-0.74, P = .001).
There was substantial and considerable heterogeneity among studies in all subgroups. The detailed results are shown in Figure S2.
Stratification by the H. pylori testing method was possible for 4 subgroups of studies with histology from the stomach, histology from the esophagus, serology, and rapid urease test as indicated in  Figure S4. A visual inspection of the funnel plot of the overall assessment from the 70 studies revealed asymmetry, Figure S5. There was no asymmetry on the funnel plots of the subgroup analyses.

| Risk of selection and information bias
The results of our quality and risk assessment by the modified Newcastle-Ottawa scale for case control studies are shown in Table 2.
It is important to note that our meta-analysis includes 78 studies of the meta-analysis by Fischer and our quality and risk assessment revealed both selection and information bias, which had been re-

5.
Comparability was poor in most of the studies, as only 23 of the studies had age-and sex-matched cases and controls and an additional 7 of them had either sex-or age-matched cases and controls. Some of the studies described significantly different proportion of races in the cases and controls and there is ample evidence that ethnicity influences the prevalence of both HPI and BE. 49,54,65 6. Only 3 studies stated clearly that the investigators were blinded to BE when testing HPI or vice versa. In some of the articles, the study design suggested that the single pathologist involved was obviously aware of the BE and the HPI status when assessing the histology slides for BE and HPI, while in other studies the endoscopist was aware of the BE diagnosis at the time when the rapid urease test was performed. However, the vast majority of the studies did not describe the process of HPI ascertainment; this is also a potential source of bias.

| Strengths of the analysis
To date, this is the largest and most comprehensive meta-analysis in this topic and includes data from 5 continents and 72 individual studies. To our best knowledge, this is the first meta-analysis on the effect of HPI on the length of BE and the presence of dysplasia in BE.

| Limitations of the analysis
Due to the limitations of the studies, the inconsistency of results, the indirect nature of the evidence, and the imprecision and reporting bias, the grade of evidence is low at best, based on the Grading of Recommendations Assessment, Development and Evaluation (GRADE) tool. Therefore, further research on this topic would very likely have an impact. 97

| Heterogeneity among the studies
Our subgroup analyses for geography, CagA status, segment length of BE, dysplastic BE, control groups, H. pylori testing method, and new diagnoses of BE revealed substantial and in cases considerable heterogeneity among the studies, apart from 3 subgroups in all analytical calculations.
There was no heterogeneity among studies from Australia, South America, and from studies where serology was used to detect HPI.
In the subgroup of Australia and South America, the small subsets of studies, 3 and 2 respectively, caution us to conclude that geography accounts for heterogeneity. However, subgroup analyses with smaller or more accurately specified geographic areas could show different results. blot, and 1 did not specify the exact method. None of the testing enzyme assays were the same. However, the homogeneity among these studies suggests that risk stratification of BE by HPI status could be best assessed by a serological test.

| Potential explanations of our findings
The role of H. pylori in the pathogenesis of BE is often described as controversial. 98 As mentioned before, our meta-analysis showed an inverse association between HPI and BE; however there are several previous studies with altogether different conclusions: reporting that HPI has no correlation with BE 33,69 or even a positive association 42,57 (describing HPI as a risk factor). Most papers (especially the ones with higher patient numbers) are in parallel with our findings. 32,36,77 If we accept that HPI leads to risk reduction, the following question arises: What could be the cause or mechanism behind this inverse association? This question is not only important from a theoretical, but also from a clinical standpoint: understanding the mechanism is crucial for evaluating the risks and benefits of H. pylori eradication therapy, in addition to bringing us closer to explaining the increasing incidence of BE.
Helicobacter pylori infection is a proven risk factor for gastric non-cardia adenocarcinoma and other cancers including lymphoma; however not much is known about its relationship with gastric cardia and EAC. 99 Epidemiological data show a simultaneous decline of HPI and increase in the aforementioned 2 tumor types. Along with the decrease of H. pylori positivity, the incidence of non-cardia adenocarcinomas is also falling. 100 F I G U R E 3 Forest plot of the random effect analysis of the 7 studies included in the subgroup analysis for different segment lengths of Barrett's esophagus. LS, long segment; SS, short segment; USS, ultrashort-segment Barrett's esophagus; OR, odds ratio, CI, 95% confidence interval; HP, Helicobacter pylori, weights of studies and heterogeneities are indicated too As to why and how exactly could HPI reduce the risk of BE development, several theories exist, but none of them are considered proven. Multiple articles attribute this fact to the effect of H. pylori on the gastric mucosa: the microorganism causes a corpus-predominant gastritis, which leads to decreased gastric output. In this case, the esophagus is less exposed to the harmful effect of gastric acid, thus it has a reduced risk for developing BE and EAC. 5,7,98,101 Several studies that did not find a negative correlation between HPI and BE only did so when looking at patients that were infected with a CagA-positive subgroup of H. pylori. 99 Other articles that found an inverse association between H. pylori and BE reported an Contrary to this theory, based on a population-based Swedish case control study, Ye et al speculates that it is unlikely that H. pylori lowers the risk of BE through the reduction of gastric acidity. They drew this conclusion because no correlation was found between gastric atrophy and EAC in their study; however, they did find a significant association between gastric atrophy and cardia adenocarcinoma. 102 In a meta-analysis on the subject, Fischbach et al describes another theory that aims to explain the inverse relationship between H. pylori and BE. They speculate that HPI is associated with reduced risk for obesity, thus not only reducing the likeliness for acidic reflux, but also the insulin level in the blood. This leads to the decreased production of Insulin-like Growth Factor (IGF), which normally acts as an agent that potentiates the proliferation of Barrett's epithelium. 5 With the reduced amount of circulating IGF due to H. pylori, BE is less likely to develop. 103 In contrast to these theories, Kountouras  An editorial in Gastroenterology elaborates on the possibility that the decline in H. pylori incidence might have other consequences, not necessarily limited to the field of gastroenterology.
For example, H. pylori might have an effect in regulating ghrelin and leptin, 2 hormones produced (partly in case of leptin) by the stomach and related to metabolism regulation. The article suggests that with the continuous fall of H. pylori incidence, we might see an increase in diabetes and obesity due to the dysregulation of these hormones. 100 Our results confirm the conclusion of previous meta-analyses [5][6][7] and we calculated a similar magnitude of risk reduction. Gisbert In summary, HPI is associated with a reduced risk of BE. Our new findings prove that the risk reduction in case of HPI seems to be independent of the geographical location and it is directly associated with the length of the BE segment and the presence of dysplasia in BE.