The Association between Barrett’s Esophagus and Helicobacter pylori Infection: A Meta-Analysis

Abstract

Objective

The effect of Helicobacter pylori on Barrett’s esophagus is poorly understood. We conducted a meta-analysis to summarize the existing literature examining the effect that H pylori has on Barrett’s esophagus.

Design

We performed a comprehensive search to identify studies pertaining to the association between H pylori and Barrett’s esophagus. We conducted meta-regression analyses to identify sources of variation in the effect of H pylori on Barrett’s esophagus.

Results

Our analysis included a total of 49 studies that examined the effect of H pylori on Barrett’s esophagus and 7 studies that examined the effect of cag A positivity on Barrett’s esophagus. Overall, H pylori, and even more so cag A, tended to be protective for Barrett’s esophagus in most studies; however there was obvious heterogeneity across studies. The effect of H pylori on Barrett’s esophagus varied by geographic location and in the presence of selection and information biases. Only 4 studies were found without obvious selection and information bias, and these showed a protective effect of H pylori on Barrett’s esophagus (Relative Risk = 0.46 [95% CI: 0.35, 0.60]).

Conclusions

Estimates for the effect of H pylori on Barrett’s esophagus were heterogeneous across studies. We identified selection and information bias as potential sources of this heterogeneity. Few studies without obvious selection and information bias have been conducted to examine the effect of H pylori on Barrett’s esophagus, but in these, H pylori infection is associated with a reduced risk of Barrett’s esophagus.

Background

The incidence of esophageal adenocarcinoma has steadily increased over the past three decades in developed countries, while the five-year survival rate continues to be low (only 16% in the US and 10% in Europe).^1,2 Barrett’s esophagus, which is estimated to affect less than 2% of the general population,^3–5 is considered to be the precancerous lesion for esophageal adenocarcinoma.^6–8 Yet little is known about the etiological process leading to Barrett’s esophagus.

Helicobacter pylori has been implicated as a risk factor for precancerous lesions in the stomach which affect the acid producing parietal cells.^9–11 However, inconsistent evidence exists regarding the effect of H pylori on gastroesophageal reflux disease, the primary putative risk factor for Barrett’s esophagus and esophageal adenocarcinoma, and the current evidence regarding the effect of H pylori on Barrett’s esophagus remains poorly understood. Some previous studies have reported that H pylori is a risk factor for Barrett’s esophagus,^12–14 while other studies have reported that H pylori has no effect on Barrett’s esophagus^15–16 and others still report a protective effect.^17–19 Previous meta-analyses, using small subsets of studies on this topic, failed to investigate sources of the heterogeneity of effects across studies.^20–22 To better understand the conflicting results, we conducted a meta-analysis to evaluate potential sources of heterogeneity for estimates of the effect of H pylori on Barrett’s esophagus, and to summarize the effect that H pylori has on Barrett’s esophagus within homogeneous patient groups.

Methods

We conducted a meta-analysis to summarize the effect that various factors have on Barrett’s Esophagus.^{4, 12–19, 23–62} However, the study presented here focuses on the analyses examining the effect of H pylori on Barrett’s Esophagus.

A. Data sources

Sources of studies included the literature databases Medline (PubMed and Ovid) and Science Citation Index.⁶³ Studies were searched from the inception of each database through December 31, 2010 using the keywords ‘Barrett’s esophagus,’ ‘Barrett’s metaplasia’ or ‘Barrett’s oesophagus,’ and ‘Helicobacter pylori’ or ‘Campylobacter pylori. Two collaborators used information from the references’ titles and abstracts to identify potentially eligible studies from the literature database searches. We supplemented these searches with backward citation tracking of eligible primary studies, reviews of Barrett’s esophagus, and hand-searches of conference proceedings published in Gut and Gastroenterology.

B. Study selection

All eligible studies satisfied the following inclusion criteria:

1. Barrett’s esophagus could be used as an outcome in the analysis.

2. Relevant information was provided to allow the estimation of a relative risk (risk ratio or odds ratio) and a variance measure for the association between H pylori and Barrett’s esophagus.

3. The individual was used as the unit of analysis for estimating the relative risk.

4. Information must have been available in English or Spanish (at least as an abstract).

5. A sample size of more than 4 subjects for each comparison group was utilized.⁶⁴ Therefore, case reports were not included.

We excluded studies based on the following criteria:

1. The study was not conducted on humans.

2. Barrett’s esophagus was not mentioned in the abstract.

3. The results came from a review article.

When data from multiple reports were identified, we included only the report with the most complete relevant data.

C. Data extraction

All potentially eligible studies were randomly assigned to two of the three primary data extractors for independent preliminary screening. Each extractor first reviewed the report to confirm eligibility, and if ineligible, provided the primary reason for ineligibility. The two assigned primary extractors then independently extracted relevant data from studies judged to be eligible.

We created an extraction database to collect relevant information regarding each eligible study such as citation information; how H pylori and Barrett’s esophagus were measured; the measure of effect and confidence intervals; sample sizes; study location (geographic location, country, etc); design (cross-sectional, basic control or population-based case-control); characteristics of the study population (e.g. prevalence of the modifiable risk factor); comparison group (endoscopic patients, asymptomatic patients; gastroesophageal reflux disease patients; etc); data analyses conducted (adjustment for confounding, etc); likelihood for selection and information bias; and other potential sources of heterogeneity across studies.

Selection bias was considered likely if the comparison group without Barrett’s esophagus did not represent the base population (in terms of the distribution of the exposure) from which the cases of Barrett’s esophagus came. The base population was defined as those who, if they developed Barrett’s esophagus, would be a case with Barrett’s esophagus in the study. For example, if Barrett’s esophagus cases were chosen from patients undergoing an upper endoscopic examination at clinic A in Los Angeles, CA, then the base population would be individuals who, if they had Barrett’s esophagus, would undergo an upper endoscopy examination at clinic A in Los Angeles, CA and would be identified as a case in the study. If the comparison group (without Barrett’s esophagus) was chosen from those undergoing upper endoscopy examinations at clinic A in Los Angeles, CA, then it is likely that selection bias occurred since the distribution of H pylori in endoscopy patients is likely to differ from H pylori in the base population.

Information bias was suggested when the measurement of H pylori or Barrett’s esophagus was likely to be inaccurate. Variables indicating how H pylori and Barrett’s esophagus were measured included an indicator of incident (versus prevalent) measurement, methods used to measure or define H pylori (urea breath test, rapid urease test, culture, histology, serology, fecal test, etc), location of biopsy samples for H pylori measure (gastric biopsies, esophageal biopsies only, or no biopsies), whether the methods for H pylori measurement was consistent for Barrett’s esophagus cases and the non-Barrett’s comparison group, and whether Barrett’s esophagus was diagnosed by first seeing Barrett’s mucosa at endoscopy, and then histologically confirming intestinal metaplasia in biopsy specimens taken at the site where the Barrett’s mucosa was observed.

D. Data cleaning and screening

Studies judged to be eligible by the two data extractors were then assigned to an additional screener for data cleaning and screening. When a discrepancy was observed between the two extractions, the third collaborator reviewed the original report to resolve the discrepancy and correct any errors.

E. Data analysis

The effect measures of interest were relative risks that compared the risk of Barrett’s esophagus among individuals who tested positive for H pylori to the risk of Barrett’s esophagus among individuals who tested negative for H pylori. We assumed that the Barrett’s esophagus was a rare outcome and therefore used proportions, risk ratios, or odds ratios from eligible studies to estimate this relative risk.

We examined the distribution of the measures of effect using visual and tabular displays and tests of homogeneity to reveal systematic variation in the relative risks of H pylori on Barrett’s esophagus across studies.⁶⁴ Furthermore, we investigated potential publication bias by using funnel plots.⁶⁵

We conducted meta-regression analyses to identify factors that influence variation in the estimated relative risks across studies within the pool of available data (previously described under data extraction) and to define subgroups for which the effect of H pylori on Barrett’s esophagus did not show observable residual heterogeneity in the measure of effect across studies. The relative risk estimates were then modeled by the fixed effects of factors that had the greatest potential to explain variation in effects and for which there was sufficient data and variation across studies. Quality scoring, which has well-described methodological shortcomings,^66–68 was not included in this meta-analysis. Instead we examined whether specific characteristics of data quality could predict variation in the measure of effect across studies. We also examined the association between Barrett’s esophagus and cag A positive strains of H pylori.

Results

A total of 2,661 abstracts were screened for preliminary eligibility. Of these, 2,487 were judged to be ineligible. Data extraction was performed on the remaining 174 studies, 40 of which were subsequently judged to be ineligible and 134 eligible for the effect of any factor on Barrett’s esophagus; 49 of these pertained to the effect of H pylori on Barrett’s esophagus and were included in this anlaysis. ^{4, 12–19, 23–62} The most common reasons for ineligibility were lack of reported data to estimate a relevant measure of effect (46%) and sources of data exclusively coming from review articles (41%).

The included studies utilized populations from around the world, including Europe (51%), the United States (20%), Canada, Japan, Malaysia, China, Peru, Chile, Pakistan and Australia. The pooled overall random effects estimate of the relative risk was 0.73 (95% CI 0.60, 0.88). However, heterogeneity was observed both in the visual graph of the measures of effect across studies and the test for heterogeneity (p<0.0001). Figure 1 illustrates the distribution of the estimated effect of H pylori on Barrett’s esophagus across the 49 studies that specifically examined this effect. The funnel plot did not suggest that publication bias was a major concern.

Figure 1.

Estimates of the effect of H pylori on Barrett’s esophagus using estimates from all available studies.

Most of the 49 studies (92%) which examined the effect of H pylori on Barrett’s esophagus used convenience comparison groups or other populations which would not be likely to estimate the true distribution of H pylori in the base population from which the cases of Barrett’s esophagus arose (Table 1), and therefore were considered to have biased estimates of effect due to selection bias. Misclassification of H pylori and Barrett’s esophagus (information bias) was also common; 33% of studies did not specifically state that they diagnosed Barrett’s esophagus by histologically confirming intestinal metaplasia in biopsy specimens taken at the site where the Barrett’s mucosa was observed at endoscopy, and 14% (n=7) diagnosed H pylori histologically but did not use a biopsy sample from either the antrum or the corpus of the stomach to measure H pylori (6 of these 7 studies only measured H pylori in the esophagus or gastro-esophageal junction). The pooled estimate for the studies that used gastric biopsies was 0.66 (95% CI: 0.53, 0.83); test for heterogeneity p<0.0001) while the estimate from studies that measured H pylori only in the esophagus or gastro-esophageal junction was 1.39 (95% CI: 0.52, 3.52; test for heterogeneity p>0.0001).

Table 1.

Summary of studies with estimates for the effect of Helicobacter pylori on Barrett’s esophagus

1st Author’s Last Name, Year	Country	Observed Study Design*	Gold Standard Measure of BE Reported?	Measure of H pylori	Comparison Group without BE†	Selection Bias?	Factors Adjusted†	N	Relative Risk (95% CI)
Abbas Z, 1995	Pakistan	Case-control	No (prevalent)	Rapid urease, histology	Endoscopy patients with reflux esophagitis	Likely	None	58	0.57 (0.20, 1.62)
Abouda G, 2003	UK	Case-control	No (prevalent)	Rapid urease, serology	Endoscopy patients without GERD	Likely	None	85	1.71 (0.56, 5.28)
Anderson L, 2008	Ireland	Population-based Case-control	Yes (newly diagnosed)	Serology	Random sample from General Practice Master Index	Unlikely	None	468	0.47 (0.32, 0.68)
Blaser M, 1991	Canada	Case-control	Not specified	Serology/histology	Healthy hospital or military employees, blood donors, and elderly nursing home residents.	Likely	None	116	1.07 (0.52,2.23)
Chacaltan, 2009	Peru	Case-control	Yes (prevalent)	Histology	Asymptomatic military personnel without erosive esophagitis, and peptic ulcer disease	Likely	None	922	0.79 (0.25, 2.59)
Corley D, 2008	USA	Population-based Case-control	Yes (prevalent)	Serology	Random sample from Kaiser Permanente HMO	Unlikely	Age, gender, location, BMI, ethnicity, smoking, education, multivitamins	591	0.42 (0.26, 0.70)
Csendes A, 1997	Chile	Case-control	No (prevalent)	Histology	Endoscopy patients with dyspepsia, but no GERD	Likely	None	290	1.00 (0.55, 1.83)
El-Serag H, 1999	USA	Case-control	Yes (prevalent)	Histology	Endoscopy patient with erosive esophagitis	Likely	None	108	0.87 (0.36, 2.11)
Fassan M, 2009	Italy	Case-control	Not specified	Histology	Patients with dyspepsia	Likely	None	420	0.25 (0.16, 0.40)
Ferrández A, 2006	Spain	Case-control	Yes (prevalent)	Serology, rapid urease, histology, PCR	Healthy blood donors	Likely	None	317	2.38 (1.23, 4.59)
Goldblum J, 2002	USA	Case-control	Yes (newly diagnosed)	Rapid urease, histology, culture	Endoscopy VA patients without GERD	Likely	None	130	0.52 (0.25, 1.09)
Hackelsberger A, 1998	Germany	Case-control	Yes (prevalent)	Rapid urease, histology	Endoscopy patients	Likely	None	363	0.79 (0.43, 1.46)
Henihan R, 1998	Ireland	Case-control	Yes (prevalent)	Histology (esophagus)	Reflux esophagitis	Likely	None	122	18.3 (1.01,329.99)
Hirota W, 1999	USA	Cross-sectional	Yes (prevalent)	Histology (esophagus)	Endoscopy patients	Likely	None	842	0.42 (0.15, 1.18)
Inomata Y, 2008	Japan	Case-control	Yes (prevalent)	Serology, rapid urease, histology	Endoscopy patients with esophagitis	Likely	None	227	0.40 (0.19, 0.84)
Johansson J, 2007	Sweden	Case-control	Yes (prevalent)	Histology (gastrooesphageal junction)	Endoscopy patients (H pylori was not measured for the population controls)	Likely	Age, gender, BMI, heartburn, smoking and alcohol	519	1.7 (0.7, 4.6)
Jonaitis L, 2008	Lithuania	Case-control	Yes (prevalent)	Rapid urease	Endoscopy patients with esophagitis	Likely	None	227	0.40 (0.19, 0.84)
Kala Z, 2007	Czech Republic	Case-control	No (prevalent)	Rapid urease	Endoscopy patients with esophagitis	Likely	None	86	2.72 (0.98, 7.54)
Kiltz U, 1999	Germany	Case-control	Yes (prevalent)	Rapid urease, serology	Endoscopy patients	Likely	None	355	0.75 (0.33, 1.70)
Laheij R, 2002	Netherlands	Cross-sectional	No (prevalent)	Rapid urease, histology, culture	Endoscopy patients without reflux esophagitis	Likely	Gender, corpus gastritis	551	0.92 (0.30, 2.50)
Lam K, 2008	USA	Case-control (nested within a Cross-sectional study)	Yes (prevalent)	Serology	Asymptomatic endoscopy patients	Likely	None	269	0.51 (0.32, 0.68)
Loffeld R, 1992	Netherlands	Case-control	Yes (prevalent)	Histology (esophagus)	Healthy blood donors	Likely	None	469	3.88 (2.21, 6.79)
Loffeld R, 2000	Netherlands	Case-control	No (newly diagnosed)	Serology	Endoscopy patients	Likely	None	490	0.53 (0.26, 1.06)
Loffeld R, 2004	Netherlands	Case-control	No (prevalent)	Histology	Endoscopy patients	Likely	None	4154	0.69 (0.50, 0.96)
Lord R, 2000	Australia	Case-control	Yes (prevalent)	Histology	Endoscopy patients	Likely	None	305	0.43 (0.23, 0.81)
Martinek J, 2003	Czech Republic	Case-control	Not Specified	Histology, rapid urease	Endoscopy patients	Likely	None	290	0.74 (0.25. 2.24)
Meng X, 2008	Not stated	Case-control	Not Specified	PCR	Endoscopy patients with dyspepsia	Likely	None	132	5.14 (1.67, 15.87)
Monkemuller K, 2008	Germany	Case-control	Yes (prevalent)	Histology	Non-erosive reflux disease patients with heartburn	Likely	None	194	0.94 (0.23, 1.00)
Nandurkar S, 1997	Australia	Cross-sectional	Yes (prevalent)	Histologyof biopsy tissue from the gastrooesphageal junction	Endoscopy patients	Likely	None	158	0.73 (0.27, 1.98)
Newton M, 1997	England	Case-control	Yes (prevalent)	Histology, rapid urease	Asymptomatic endoscopy patients	Likely	None	41	0.59 (0.15, 2.39)
Paull G, 1988	USA	Case-control	Yes (prevalent)	Histology	Endoscopy patients	Likely	None	51	0.85 (0.28, 2.58)
Peng S, 2009	China	Case-control	Yes (prevalent)	Rapid urease	Endoscopy patients with esophagitis	Likely	None	137	0.98 (0.37, 2.55)
Peng, 2010	China	Case-control	Yes (prevalent)	Rapid urease	Endoscopy patients with uninvestigated reflux symptoms	Likely	None	469	0.90 (0.70, 1.16)
Rajendra S, 2004	Malaysia	Case-control	Yes (prevalent)	Rapid urease, histology	Symptomatic endoscopy patients	Likely	None	1864	1.12 (0.77, 1.64)
Rajendra S, 2007	Malaysia	Case-control	Yes (prevalent)	Serology	Endoscopy patients without GERD	Likely	None	108	0.59 (0.27, 1.26)
Rex D, 2003	USA	Case-control	Yes (newly diagnosed)	Rapid urease	Asymptomatic Colonoscopy patients	Unlikely	None	812	0.27 (0.10, 0.77)
Ronkainen J, 2005	Sweden	Cross-sectional	Yes (prevalent)	Serology, histology, culture	Random sample of the adult population	Unlikely	None	1000	0.91 (0.31, 2.63)
Rugge M, 2001	Italy	Case-control	Yes (prevalent)	Histology	Endoscopy patients with non-ulcer dyspepsia	Likely	None	106	0.43 (0.20, 0.94)
Schenk B, 1999	Netherlands	Case-control	No (prevalent)	Histology	Endoscopy patients with GERD	Likely	None	137	2.17 (1.19, 3.96)
Sirigu F, 1994	Italy	Case-control	Yes (prevalent)	Histology	Age and gender matched endoscopy patients	Likely	None	82	1.45 (0.27, 7.67)
Sonnenberg A, 2010	USA	Cross-sectional	Yes (prevalent)	Histology	Endoscopy patients	Likely	Age, sex, state of residence and insurance type	78985	0.42 (0.35, 0.49)
Toruner M, 2004	Turkey	Case-control	Yes (prevalent)	Histology	Endoscopy patients	Likely	None	335	1.18 (0.54, 2.58)
Vicari J, 1998	USA	Case-control	No (prevalent)	Serology, histology	Endoscopy patients without GERD	Likely	None	105	0.54 (0.24, 1.21)
Vieth M, 2000	Germany	Case-control	No (prevalent)	Histology	Endoscopy patients with non-ulcer dyspepsia	Likely	None	1766	0.60 (0.49, 0.73)
Watari J, 2009	Japan	Case-control	Yes (prevalent)	Histology, culture	Patients with gastric metaplasia	Likely	None	140	0.01 (0.001, .20)
Werdmuller B, 1997	Netherlands	Case-control	Not specified	Histology, culture, serology, rapid urease	Endoscopy patients with GERD	Likely	None	412	0.29 (0.08, 1.06)
Weston A, 2000	USA	Case-control	Yes (prevalent)	Histology	Endoscopy patients with GERD	Likely	None	506	0.62 (0.43, 0.89)
White N, 2008	Canada	Case-control	Yes (prevalent)	Histology (esophagus)	Endoscopy patients	Likely	None	68	0.47 (0.07, 3.00)
Zaninotto G, 2002	Italy	Case-control	Yes (prevalent)	Histology	Endoscopy patients with GERD	Likely	None	66	0.77 (0.23, 2.58)

^*If not specified, studies are not population-based. Study design observed by data extraction may differ from the study design reported in the manuscript.

^† BMI= body mass index; GERD = Gastroesophageal reflux disease

In weighted meta-regression analyses using all 49 studies, selection bias, H pylori infection measured only using a biopsy sample from the stomach, and study location were identified as sources of heterogeneity. As illustrated in Figure 2, the effect within the group of 4 studies with appropriate measurement of H pylori and without a likely source of selection bias was consistently protective for BE; the random effects summary estimate for the relative risk was 0.46 (95% CI: 0.35, 0.60).^4,17,25,51 Likewise, the effect of H pylori on BE was consistently protective in studies conducted in the United States; 0.46 (95% CI: 0.40, 0.53), (heterogeneity test p=0.50) (Figure 3). Strongly protective measures of effect (RR = 0.01 and 0.04) were reported for the two studies from Japan. Residual heterogeneity was observed for all other strata of these variables; effect estimates ranged from 0.25 to 5.14 for studies conducted outside of the United States or Japan with a likely source of selection bias which used gastric biopsy samples from the antrum or corpus to measure H pylori.

Figure 2.

Estimates of the effect of H pylori on Barrett’s esophagus in studies with appropriate measurement of H pylori and without a likely source of selection bias..

Figure 3.

Estimates of the effect of Helicobacter pylori on Barrett’s esophagus using estimates from studies in the United States.

We also examined the effect of cag A positivity on Barrett’s esophagus. Data from the subgroup of 7 studies which examined the effect of cag A positivity on BE, suggested a protective effect of cag A on BE in all but 1 study (Table 2, Figure 4).^{17,25,29,49,52,56,69} The pooled random estimate for the relative risk in these 7 studies was 0.38 (95% CI: 0.19, 0.78), but the relative risk for the study by Ferrández et al. (estimated to be 1.5 [95% CI: 0.93, 2.46]) differed greatly from the other studies.²⁹

Table 2.

Summary of studies with estimates for the effect of CagA + Helicobacter pylori on Barrett’s esophagus

1st Author’s Last Name, Year	Country	Observed Study Design	Gold Standard Measure of BE Reported?	Comparison Group without BE†	Selection Bias?	Factors Adjusted†	N	Measure of Effect (95% CI)
Anderson L, 2008	Ireland	Population-based Case-control	Yes	General Practice Master Index from 4 general practices in the region matched by age and gender	Likely, since age and gender were not adjusted for in the analysis	None	468	0.62 (0.43, 0.89)
Corley D, 2008	USA	Population-based Case-control	Yes	Kaiser Permanente members in Northern California	Unlikely	Age, gender, location, BMI, ethnicity, smoking, education, multivitamins	516	0.08 (0.02, 0.35)
Ferrández A, 2006	Spain	Case-control	Yes	Healthy blood donors	Likely	None	317	1.51 (0.93, 2.46)
Rajendra S, 2007	Malaysia	Case-control	Yes	Endoscopy patients without GERD and without acid suppression therapy	Likely	None	108	0.69 (0.32, 1.49)
Rugge M, 2001	Italy	Case-control	Yes	Endoscopy patients	Likely	None	54	0.18 (0.05, 0.62)
Vicari J, 1998	USA	Case-control	No	Endoscopy patients with Non-ulcer dyspepsia	Likely	None	41	0.21 (0.04, 1.13)
Vaezi M, 2000	USA	Case-control	No	Endoscopy patient without symptomatic GERD	Likely	None	143	0.11 (0.02, 0.52)

^*If not specified, studies are not population-based.

^†BMI = body mass index; GERD = Gastroesophageal reflux disease

Figure 4.

Estimated effect of Cag A status on Barrett’s esophagus.

Discussion

In the current analysis we examined the effect of H pylori on Barrett’s esophagus across 49 studies. Our analyses suggest that H pylori tended to be protective for Barrett’s esophagus. However, heterogeneity was easily observable across studies, and the effect varied in the presence of selection bias, with the use of esophageal tissue instead of gastric tissue to diagnose H pylori, and in different geographic locations.

The effect of information bias through inaccurate measurement of H pylori using tissue from the esophagus instead of the stomach was observed to be a source of heterogeneity. Studies that defined H pylori exclusively from esophageal biopsies were more likely to find a positive association between H pylori and Barrett’s esophagus, whereas the rest of the studies that measured H pylori using gastric biopsies tended to show a protective association between H pylori and Barrett’s esophagus. Since the occurrence of H pylori in the esophagus would not likely reflect the occurrence of H pylori in the stomach, a person’s true H pylori status is likely to be misclassified when tissue from the esophagus is used.

A strong protective effect of gastric H pylori on Barrett’s esophagus was observed without observable heterogeneity within the subgroup of 4 studies without a likely source of selection or information bias 0.46 (95% CI: 0.35, 0.60), where the definitions of Barrett’s esophagus as well as H pylori adhere to internationally accepted standards.⁷⁰ However, 45 of the 49 studies had obvious sources of selection bias (some with information bias as well), so only 4 studies were used to estimate this measure of effect.^4,17,25,51 Additional studies are needed with appropriately chosen comparison groups. For example, for case-control studies, the controls should be chosen to represent the exposure distribution of the base population from which the cases arose. Endoscopy patients, healthy blood donors, and patients with conditions positively or negatively associated with H pylori would not be appropriate comparison groups. Controls from identified catchment populations from which the cases arose would be the most appropriate choice.

The effect of cag A positive H. pylori on Barrett’s esophagus also tended to be protective. Only one of the seven studies examining the association between cag A and Barrett’s esophagus did not show a protective effect. In the study by Ferrández et al which used healthy blood donors as controls, the relative risk was estimated to be 1.5 (95% CI: 0.93, 2.46).²⁹ Blood donors typically are healthier in many ways and younger than most populations and therefore may have led to selection bias since they may have a lower prevalence of H pylori than the base population from which the cases arose. On the other hand, the protective effect observed in 5 of the 6 other studies was also likely affected by selection bias.

Potential candidates for sources of heterogeneity for the effect of H pylori on Barrett’s esophagus include corpus inflammation and atrophy. Some evidence suggests these factors may modify the effect of H pylori on related disease outcomes such as symptomatic GERD and erosive esophagitis.⁷¹ However, existing studies examining the effect of H pylori on Barrett’s esophagus did not evaluate effect modification by type or distribution of gastritis on the risk of Barrett’s esophagus. Figure 5 illustrates possible mechanisms whereby H pylori may decrease the risk for Barrett’s esophagus. H pylori infection’s hypothesized protective effect on Barrett’s esophagus may occur due to its association with multifocal atrophic gastritis and the resulting damaging effect it has on the acid producing parietal cells.^9–11 With a loss of parietal cells, the esophagus is less likely to be exposed to the harmful effects of gastric acid and acid reflux, erosive esophagitis and then Barrett’s esophagus is less likely to occur. Also in this hypothesized protective pathway for Barrett’s esophagus, both a low intake of fruits and vegetables and older age is associated with H pylori infection,^72–746 while younger age is associated with lower calorie consumption and increased physical activity, both leading to a lower risk for obesity.^75,76 With a lower risk for obesity, lower visceral fat and lower insulin levels both may lead to a reduction in Barrett’s esophagus (Figure 5).^77–80 Using this figure, after eliminating intermediates and controlling for age and dietary factors such as fruit and vegetable consumption, there may be no remaining potential confounders for the effect of H pylori on Barrett’s esophagus. Therefore, age and dietary factors such as fruit and vegetable intake should be controlled for, yet only 3 (7%) of the studies adjusted for age,^17,19.34 and only 1 adjusted for a relevant dietary factor (Corley, et al).¹⁷

Figure 5.

The putative pathways involving Helicobacter pylori and a decreased risk for Barrett’s Esophagus.

Two previous meta-analyses have reported a protective effect of H pylori on Barrett’s esophagus across only 5 and 9 studies.^20,21 These summary measures of effect were reported in the presence of heterogeneous effects without searching for sources of the variation of effects across studies. As noted by Wang et al in a more recent meta-analysis, the reported protective summary measure of effect may have been affected by the types of subjects used as the comparison group.²² Wang et al limited their meta-analysis of the effect of H pylori on Barrett’s esophagus to studies which used “normal healthy subjects as controls,” which included blood donors in 3 studies and endoscopy controls in 9 studies.²² As we discussed previously, the use of blood donors as controls is likely to lead to a biased (higher) estimate of effect due to selection bias. Likewise, the prevalence of H pylori infection among endoscopy patients would not likely represent the prevalence of H pylori in the base population from which the cases arose. The current meta-analysis is the first to examine the effect of selection and other biases on the observed effect of H pylori on Barrett’s esophagus across studies.

Conclusion

Our meta-analysis of 49 studies suggests that H pylori infection is associated with a reduced risk of Barrett’s esophagus. However, the effect is heterogeneous across studies. We identified methodological issues such as selection and information bias as potential sources of heterogeneity. In total, we found four studies without obvious selection and information bias and these showed a protective effect of H pylori on Barrett’s esophagus.