Helicobacter pylori Infection and Barrett ' s Esophagus: A ... - Nature

2 downloads 0 Views 187KB Size Report
Jan 6, 2009 - the majority of cases arise in Barrett ' s epithelium (1,2) . e. Helicobacter pylori Infection and Barrett ' s Esophagus: A Systematic Review and ...
REVIEW

492

CLINICAL REVIEWS

nature publishing group

CME

Helicobacter pylori Infection and Barrett’s Esophagus: A Systematic Review and Meta-Analysis Changcheng Wang, MD, MSc1, Yuhong Yuan, MD, PhD1 and Richard H. Hunt, MD, FRCP, FRCPC, FACG, AGAF1

OBJECTIVES:

The majority of distal esophageal adenocarcinomas are believed to arise in patients with Barrett’s esophagus (BE). Helicobacter pylori (H. pylori ) infection plays an etiological role in gastric carcinogenesis, but any possible role in BE is uncertain. We aimed to explore the possible relationship between H. pylori infection and BE by meta-analysis.

METHODS:

Observational studies comparing the prevalence of H. pylori infection in patients with BE and healthy controls conducted in adult populations and published in all languages were identified through MEDLINE, EMBASE, and Cochrane database searches up to week 5, 2008. H. pylori infection had to be confirmed by histology and/or serology and/or RUT and/or culture. Studies were excluded if no raw data for outcomes of interest were available or controls were patients with disease or duplicate publications. Summary effect size was calculated as odds ratio (OR) and 95% confidence intervals (CIs) by the random-effects model using Review Manager 4.2.8.

RESULTS:

Of 519 citations identified, a total of 12 case–control studies compared the prevalence of H. pylori infection in BE (n = 550) and controls (9 studies included controls with normal endoscopy and 3 studies used healthy blood donors as control, n = 2,979). There was no significant difference in the overall prevalence of H. pylori infection between BE and controls (42.9% vs. 43.9%, OR = 0.74, 95% CI 0.40–1.37, P = 0.34), but with significant heterogeneity. Subgroup analysis showed that the prevalence of H. pylori infection was significantly lower in BE than in endoscopically normal healthy controls (23.1% vs. 42.7%, OR = 0.50, 95% CI 0.27–0.93, P = 0.03) with significant heterogeneity observed between studies. The heterogeneity was eliminated by excluding a single Asian outlier study. In contrast, H. pylori infection was significantly increased in BE patients in the three studies using healthy blood donors as “normal controls” (71.2% vs. 48.1%, OR = 2.21, 95% CI 1.07–4.55). In BE patients, the prevalence of H. pylori infection was significantly lower in the esophagus than in the stomach (3.3% vs. 24.7%, OR = 0.14, 0.03–0.67) in three studies.

CONCLUSIONS: H. pylori infection and BE are inversely related when compared with endoscopically normal

controls but not blood donor controls. Limited evidence suggests that there is no clear association between H. pylori infection and BE. To determine more accurately the effect size of H. pylori infection in BE, high quality prospective case–control studies with age-matched, endoscopically normal healthy controls are needed. Am J Gastroenterol 2009; 104:492–500; doi:10.1038/ajg.2008.37; published online 6 January 2009

INTRODUCTION Barrett’s esophagus (BE) is considered the most important risk factor for the development of adenocarcinoma of the

distal esophagus and esophagogastric junction. The incidence of this cancer has been rising rapidly in recent decades, and the majority of cases arise in Barrett’s epithelium (1,2). The

1 Division of Gastroenterology, Department of Medicine, McMaster University Health Science Centre, Hamilton, Ontario, Canada. Correspondence: Richard H. Hunt, MD, FRCP, FRCPC, FACG, AGAF, Division of Gastroenterology (Rm 4W8A), Department of Medicine, McMaster University Health Science Centre, Hamilton, Ontario L8N 3Z5, Canada. E-mail: [email protected] Received 18 April 2008; accepted 29 August 2008

The American Journal of GASTROENTEROLOGY

VOLUME 104 | FEBRUARY 2009 www.amjgastro.com

A Systematic Review and Meta-Analysis

METHODS Literature search

A computer-aided search, supplemented by a manual literature search was performed in the MEDLINE, EMBASE, and Cochrane databases published in all languages for human studies up to week 05, 2008, with the following MeSH terms or keywords: (Barrett’s (o)esophagus or Barrett (o)esophagus or BE) and (Helicobacter pylori or H. pylori or Campylobacter pylori). Published abstracts from conference proceedings from the Digestive Disease Week (2005–2007), American College of Gastroenterology Annual Meeting (2005–2007), British Society of Gastroenterology Annual Meeting (2005–2007), and UEGW (2006–2007) were scanned for potential relevance and subsequent full publications were sought. Inclusion and exclusion criteria

The following inclusion criteria were used to select observational (case–control or cohort or cross-sectional) studies reporting raw data on H. pylori infection and BE: studies conducted in adult populations; controls were defined as normal (healthy) controls by the authors. H. pylori infection had to be confirmed by histology and/or serology and/or rapid urease test (RUT) and/or culture. Exclusion criteria were: studies without raw data for outcome of interest; studies without controls or controls who were patients with disease (e.g., GERD, peptic ulcer or functional dyspepsia diagnosed by the authors), and duplicate publications. No language restriction was set and abstracts were not excluded if inclusion criteria were met. © 2009 by the American College of Gastroenterology

Data extraction

Data were extracted by two independent reviewers (C.W. and Y.Y.) and agreement was reached by discussion with conflict resolution by the third reviewer (R.H.H.). Major extracted items included the study design, characteristics of case and control populations, definition of BE cases and controls, and testing methods for H. pylori infection. Quality assessment

Study quality was assessed by a series of validity criteria, including study design, patient selection, definition of cases and controls, the biopsy sites, and methods for testing for H. pylori infection. All studies included were observational studies, which were qualitatively assessed by referring to the transparent reporting of evaluations with nonrandomized designs (TREND) statement (16) and the meta-analysis of observational studies in epidemiology (MOOSE) guideline (17). No score of quality was given by these guidelines but the quality of the study was considered and discussed when interpreting the results especially when heterogeneity was seen. Statistical analysis

Our primary objective was to compare the prevalence of H. pylori infection in BE with that in normal controls. The secondary objective was to compare the prevalence of H. pylori infection in the esophagus and stomach in patients with BE. The proportion with infection was calculated for each group. Odds ratios (OR) and 95% confidence intervals (CIs) were calculated under a randomeffects model (18) using Cochrane software Review Manager 4.2.8. A level of significance less than 0.05 was considered statistically significant. The test for heterogeneity was analyzed by Cochrane’s 2-test; a P value of less than 0.10 was considered to be significant for heterogeneity between studies. Publication bias was assessed by StatsDirect statistical software (Version 2.4.5).

RESULTS Searches generated 519 citations and 13 studies which reported normal controls were identified. A large sample study (n = 307) (19) was further excluded because it mixed normal subjects with patients with endoscopic gastritis in the control group. The authors did not provide raw data to allow separation and did not provide this information when requested by us. In total, 12 studies (8,10,20–29) met inclusion criteria and compared the prevalence of H. pylori infection between BE cases and normal healthy controls. Of these, nine used endoscopically normal controls and three used healthy blood donors as controls without endoscopic examination (Figure 1; Table 1). Although most endoscopically normal controls presented with some form of gastrointestinal (GI) symptoms, we accepted the authors’ definition of “normal controls,” considering the difficulties in getting a true normal population as controls with the need for endoscopic examination. Of the three studies that required biopsies from several different sites for assessing H. pylori infection and provided more than one The American Journal of GASTROENTEROLOGY

REVIEW

estimated annual risk for adenocarcinoma of the distal esophagus ranges from 0.2% to 2.0%, a risk that is 30–125 times that in an age-matched population (3–5). Helicobacter pylori (H. pylori) infection plays an etiological role in gastric carcinogenesis, but any potential role in esophageal disease, including BE is still controversial. Several studies (6–10) have found that H. pylori infection may “protect” against the development of gastroesophageal reflux disease (GERD) and BE, whereas other studies (11,12) suggest that the presence of H. pylori infection does not alter the natural history of BE. Moreover, one study suggested (13) that neither gastric nor esophageal infection with H. pylori is a requisite for the development of adenocarcinoma in BE. These conflicting results question the relationship between H. pylori infection and BE and whether H. pylori infection protects against the development of BE. To date, two meta-analyses (14,15) have reviewed the prevalence of H. pylori infection in BE and found that the prevalence of H. pylori infection was significantly lower in BE than in controls. However, these studies mixed a variety of controls (including GERD patients and patients with a variety of diseases), which may overestimate or underestimate the true effect size of H. pylori infection in BE. Therefore, we sought and analyzed studies with normal healthy subjects as controls and aimed to explore the possible relationship between H. pylori infection and BE by meta-analysis.

493

REVIEW

494

Wang et al.

Citations searched from three databases and four conferences (n =519)

Retrieved observational studies (n =30)

Studies without controls, reviews, basic researches, duplicate papers, etc. (n =489) Studies using GERD or functional dyspepsia or other abnormal diseases as controls (n =17) or normal subjects mixed with gastritis (n =1) ∗

Studies with normal controls (n =12)

Endoscopically normal controls (n = 9)

“Healthy blood donors” as normal controls (n =3)

Figure 1. Literature search flow chart. *This study was excluded in the meta-analysis because we were not successful in obtaining information about the raw data even after contacting the author.

H. pylori infection rate, we took the infection rate in the antrum as the H. pylori infection rate (21,22,24) for the analysis. All BE cases were defined as the presence of histologically confirmed specialized columnar (intestinalized) epithelium above the esophagogastric junction, irrespective of the length of the columnar epithelium-lined esophagus. The exception was one study, which noted the presence of a segment of distal esophagus lined by gastric epithelium for at least 3 cm in length (21). This was also the only study that mentioned the length of the columnar lined epithelium. Two studies reported the prevalence of H. pylori infection in long-segment BE (≥3 cm) and short-segment BE ( < 3 cm): one was from endoscopically normal controls (26) and the other from healthy blood donor controls (29), but we were not able to perform any further analysis. The prevalence of H. pylori infection in BE and normal controls

Of the included 12 studies, 550 patients with BE and 2,979 controls were studied. There was no significant difference in the prevalence of H. pylori infection between BE and normal controls overall (normal endoscopy and blood donors) (42.9% (236/550) vs. 43.9% (1,308/2,979), OR = 0.74, 95% CI

Table 1. Characteristics of studies comparing the prevalence of H. pylori infection in BE and normal controls (as defined by the authors) Study (reference) (country)

Controls (n)

Biopsy sites for detecting H. pylori

H. pylori testing methods

48 age and sex-matched patients with normal esophagus

Esophagus and antrum

Histology

100

190 non-specific dyspeptic symptoms or healthy and asymptomatic controls with normal endoscopy

Esophagus, antrum, fundus and duodenum

Histology

Newton 1997 (22) (UK)

16

25 asymptomatic patients for investigating anemia with normal endoscopy

Esophagus, antrum, and fundus

Histology and RUT

Werdmuller 1997 (23) (The Netherlands)

13

399 patients with normal endoscopy

Antrum

Histology and RUT and culture and serology

Kiltz 1999 (10) (Germany)

35

320 patients with normal endoscopy

Antrum and corpus

RUT and serology

Loffeld 2000 (8) (The Netherlands)

36

454 patients with normal endoscopy

Antrum

Serology and/or histology or RUT or culture

Laheij 2002 (24) (The Netherlands)

23

528 symptomatic patients with normal endoscopy

Antrum and corpus

Histology and RUT and culture

Martinek 2003 (25) (Czechoslovakia)

31

259 patients with normal endoscopy

Antrum and corpus

RUT and/or histology

Inomata 2006 (26) (Japan)

36

80 asymptomatic subjects for screening with normal endoscopy

Antrum and corpus

Histology and RUT and serology

Blaser 1991 (27) (USA)

58

58 age and sex-matched healthy subjects or blood donors without endoscopy

Antrum

Histology and serology

Loffeld 1992 (28) (The Netherlands)

64

405 age-matched healthy blood donors without endoscopy

Esophagus and stomach

Histology (BE) or serology (blood donors)

104

213 age and sex-matched healthy blood donors without endoscopy

No

Serology

Sirigu 1994 (20) (Italy) Csendes 1997 (21) (Chile)

Ferrandez 2006 (29) (Spain)

BE cases (n)a 34

BE, Barrett’s esophagus. a All cases were histologically proved Barrett’s esophagus.

The American Journal of GASTROENTEROLOGY

VOLUME 104 | FEBRUARY 2009 www.amjgastro.com

the H. pylori infection rate between those diagnosed with BE and normal endoscopy. Three studies retrospectively collected cases and controls, and H. pylori infection information from the patients’ chart. Subgroup analyses were performed according to enrollment of patients prospectively (OR = 0.40, 95% CI 0.18– 0.91; test for heterogeneity: P = 0.0003) and/or retrospectively (OR = 0.75, 95% CI 0.26–2.19; test for heterogeneity: P = 0.03), and a significant difference in H. pylori infection was seen between BE and controls only in studies which enrolled patients prospectively. The significant difference was seen in those studies published in and after year 2000 (OR = 0.28, 95% CI 0.09– 0.84; test for heterogeneity: P = 0.001) but not in the studies published before year 2000 (OR = 0.85, 95% CI 0.51–1.42; test for heterogeneity: P = 0.21); a significant difference was seen in European populations (seven studies (n = 188), OR = 0.62, 95% CI 0.39–0.98; test for heterogeneity: P = 0.17), but not in nonEuropean populations (two studies (n = 136), OR = 0.20, 95% CI 0.01–5.51; test for heterogeneity: P < 0.00001). H. pylori detection methods were different in the included studies and the H. pylori infection rate was more likely to be higher in the studies using more test methods. However, no subgroup analysis for this was possible due to variations between studies (range from 1 to 4 methods used in studies). Only two studies provided raw data for the prevalence of CagA + strains of H. pylori; one with endoscopically normal controls (8), a second with blood donors without endoscopic examination as controls (29). Therefore, no subgroup analysis was conducted for CagA + strains.

0.40–1.37, P = 0.34), with significant heterogeneity observed between studies (I2 = 84.5%, P < 0.00001). Subgroup analysis showed that the prevalence of H. pylori infection was significantly lower in BE than in endoscopically normal controls (23.1% (75/324) vs. 42.7% (983/2,303), OR = 0.50, 95% CI 0.27–0.93), P = 0.03), with significant heterogeneity observed between studies (I2 = 74.0%, P = 0.0002). In contrast, H. pylori infection was significantly increased in BE cases compared with controls in the three studies using healthy blood donors as “normal controls” (71.2% vs. 48.1%, OR = 2.21, 95% CI 1.07– 4.55, P = 0.03) and significant heterogeneity was also observed (I2 = 73.3%, P = 0.02) (Figure 2). Although the heterogeneity decreased, it could not be eliminated by excluding the latter three studies and they were excluded in the further sensitivity or subgroup analyses in order to minimize the control selection bias (see “Discussion”). In the sensitivity analysis for endoscopically normal controls, the heterogeneity was eliminated by excluding one outlier study (OR = 0.04) (26) (eight studies, OR = 0.68, 95% CI 0.46–1.02; test for heterogeneity: I2 = 34.8%, P = 0.15). Most studies were small, with less than 50 BE cases, and the only one study from Chile had 100 BE cases and 190 endoscopically normal controls. We performed a sensitivity analysis to exclude this largest sample study and similar results were seen for with and without this study (21) (both P = 0.03; OR = 0.50 and 0.45, respectively). Several subgroup analyses were performed. Six studies prospectively enrolled patients by including all consecutive patients referred to the endoscopy room and then compared

Outcome: Prevalence of H. pylori infection in BE and healthy controls Sub-category

BE (n/N )

Control (n/N )

OR (random) 95% CI

Weight (%)

01 Prevalence of H. pylori infection in BE and endoscopically normal controls 13/34 12/48 Sirigu 1994 20/100 38/190 Csendes 1997 4/16 9/25 Newton 1997 3/13 204/399 Werdmuller 1997 8/35 91/320 Kiltz 1999 14/36 248/454 Loffeld 2000 6/23 281/528 Laheij 2002 4/31 43/259 Martinek 2003 3/36 57/80 Inomata 2006 324 2303 Subtotal (95% CI) Total events: 75 (BE), 983 (control) Test for heterogeneity: ² = 30.71, df = 8 (P =0.0002), I² = 74.0% Test for overall effect: Z = 2.18 (P =0.03) 02 Prevalence of H. pylori infection in BE and healthy blood donors 27/58 26/58 Blaser 1991 43/64 140/405 Loffeld 1992 91/104 159/213 Ferrandez 2006 226 676 Subtotal (95% CI) Total events: 161 (BE), 325 (control) Test for heterogeneity: ² = 7.48, df = 2 (P =0.02), I² = 73.3% Test for overall effect: Z = 2.15 (P =0.03)

8.26 9.37 6.72 7.03 8.69 9.10 8.28 7.74 7.12 72.31

1.86 1.00 0.59 0.29 0.75 0.53 0.31 0.74 0.04 0.50

[0.72, [0.55, [0.15, [0.08, [0.33, [0.26, [0.12, [0.25, [0.01, [0.27,

4.81] 1.83] 2.39] 1.06] 1.70] 1.06] 0.80] 2.24] 0.13] 0.93]

8.99 9.49 9.21 27.69

1.07 3.88 2.38 2.21

[0.52, [2.21, [1.23, [1.07,

2.23] 6.79] 4.59] 4.55]

100.00

550 2979 Total (95% CI) Total events: 236 (BE), 1308 (control) Test for heterogeneity: ² = 70.87, df = 11 (P < 0.00001), I ² = 84.5% Test for overall effect: Z = 0.95 (P =0.34) 0.1

0.2

0.5

1

Favors BE

2

5

OR (random) 95% CI

0.74 [0.40, 1.37]

10

Favors control

Figure 2. The prevalence of H. pylori infection in Barrett’s esophagus (BE) and normal controls.

© 2009 by the American College of Gastroenterology

The American Journal of GASTROENTEROLOGY

495

REVIEW

A Systematic Review and Meta-Analysis

496

Wang et al.

Outcome: Prevalence of H. pylori in esophagus and stomach in BE

REVIEW

Study

Stomach (n/N )

3/34 0/100 2/16

13/34 20/100 4/16

44.67 20.92 34.42

0.16 [0.04, 0.62] 0.02 [0.00, 0.33] 0.43 [0.07, 2.76]

150

150

100.00

0.14 [0.03, 0.67]

Sirigu 1994 Csendes 1997 Newton 1997 Total (95% CI)

OR (random) 95% CI

OR (random) 95% CI

Esophagus (n/N )

Weight (%)

Total events: 5 (Esophagus), 37 (Stomach) Test for heterogeneity: ² = 3.88, df = 2 (P = 0.14), I ² = 48.4% Test for overall effect: Z = 2.46 (P = 0.01) 0.1

0.2

0.5

Favors esophagus

1

2

5

10

Favors stomach

Figure 3. The prevalence of H. pylori infection in the esophagus and stomach in Barrett’s esophagus (BE) cases.

Bias assessment plot

Table 2. The prevalence of H. pylori infection at the different sites in the stomach in nine endoscopically normal studies in BE Antrum

Corpus

Fundus

Sirigu, 1994

13/34

NA

NA

Csendes, 1997

20/100

NA

0/31

Newton, 1997

4/16

NA

3/16

Werdmuller, 1997

3/13

NA

NA

a

Kiltz, 1999

8/35 14/36

NA

NA

Laheij, 2002

6/23

9/23

NA

a

Inomata, 2006

4/31

NA

3/36

NA

NA, not available. a

The biopsy sites were in the antrum and the corpus in these studies.

The prevalence of H. pylori infection in the esophagus and stomach in BE

Three studies (20–22) provided raw data for comparison of H. pylori prevalence in the esophagus and stomach of BE patients. The pooled prevalence of H. pylori infection at 3.3% (5/150) was significantly lower in the esophagus than in the stomach 24.7% (37/150) in BE cases (OR = 0.14, 95% CI 0.03–0.67, P = 0.01) with homogeneity seen between studies (P = 0.14) (Figure 3). In the post hoc analysis, we also pooled the H. pylori infection rate for the different biopsy sites in the stomach in BE cases. The pooled prevalence of H. pylori infection in the antrum (9 studies), and the corpus (4 studies) and the fundus (2 studies) was 23.1% (75/324), 19.8% (24/121), and 6.4% (3/47), respectively. No available data were identified for the cardia (Table 2) and no comparisons between the biopsy sites were made. Publication bias

No publication bias was detected in all 12 included studies (Horbold-Egger: bias = 5.52. (92.5% CI = − 0.44–11.48), The American Journal of GASTROENTEROLOGY

0.4

0.6

NA

Loffeld, 2000

Martinek, 2003a

Standard error

Study

0.2

0.8 –2

0

2

4

6

Log (Odds ratio) Figure 4. Funnel plot of 12 included studies.

P = 0.10) (Figure 4) or in 9 studies with endoscopically normal controls (Horbold-Egger: bias = 0.57 (92.5% CI = − 6.52– 7.66), P = 0.87) or in 3 studies using healthy blood donors as controls (Horbold-Egger: bias = 12.40 (92.5% CI = − 26.20– 51.00), P = 0.22).

DISCUSSION Because H. pylori infection is associated with several upper GI diseases, in our meta-analysis we excluded studies which used GERD (12,30–32) or functional dyspepsia patients (32,33) or patients with a disease diagnosis (6,9,34–37) as controls, in order to investigate the true effect size of H. pylori infection in BE. Our results suggest that overall the H. pylori infection rate is nonsignificantly lower in BE cases compared to normal controls, according to the authors’ definitions (OR = 0.74, 95% CI 0.40–1.37). However, we found that the selection of the control group has a major impact on the results. Two major different populations of controls were reported in the included studies and the direction of effect size was inverse. When only endoscopically normal controls were included, the prevalence of H. pylori infection was significantly lower VOLUME 104 | FEBRUARY 2009 www.amjgastro.com

in BE cases than in endoscopically normal controls (23.1% vs. 42.7%, OR = 0.50, 95% CI 0.27–0.93). In contrast, H. pylori infection was significantly increased in BE compared with controls in the three studies which used healthy blood donors without endoscopic examination as “normal controls” (71.2% vs. 48.1%, OR = 2.21, 95% CI 1.07–4.55). The reasons for the different results between these two subgroups are possibly explained by: (i) H. pylori infection in the blood donors being different from that in the general population. Sorberg et al. (38) found that the overall adjusted OR for H. pylori seropositivity among blood donors was decreased by 43% (95% CI 28–55%) compared with population-based controls. This may partly explain the prevalence of H. pylori infection in blood donors being unexpectedly lower than that in BE in the three studies. (ii) Controls in the three studies did not have an endoscopic examination and therefore possible cases of BE could not be excluded. BE was present in 1.6% of the general Swedish population (39). In European epidemiological studies, blood donors are often used to represent the general population as all European countries have unpaid donors (40,41) but this population may still be different from the population who were found to have a normal endoscopy. That is, the population of blood donors may harbor a small proportion of BE as in the general population. (iii) “Healthy” blood donors may not be entirely normal healthy subjects. Menegatti et al. (42) found that in 1,010 healthy blood donors, endoscopy revealed 113 with silent or asymptomatic H. pylori-associated diseases (e.g., peptic ulcer). These findings potentially introduce control selection bias even though the H. pylori infection rate in blood donors was similar to that in the endoscopic controls. Owing to the reasons discussed above and the significantly different ORs seen between endoscopically normal controls and healthy blood donors, the selection of blood donors as controls may be less or not at all suitable as proxies for the general population in studies of H. pylori infection. Furthermore, some other factors may contribute to the unexpectedly high H. pylori infection rate in BE patients in blood donor studies, compared with BE patients in studies in which endoscopically normal subjects were used as controls (71.2% vs. 23.1%). For example, all three blood donor controlled studies were age-matched case– control studies and the H. pylori infection effect size is likely to be different from unmatched studies in which age may be a confounder. We also noticed that the H. pylori detection method was different between BE (by histology) and blood donors (by serology) in one study (28) and therefore variability of acquisition in recruiting patients may influence the interpretation of the results. To date, two meta-analyses have reviewed the relationship between H. pylori infection and BE. Both studies reported that the prevalence of H. pylori infection was significantly lower in BE than that in controls, which suggested that H. pylori infection might have a protective role in BE (14,15). However, their studies included GERD, functional dyspepsia and a variety of non-BE diseases as controls, which could not accurately reflect the real effect size of H. pylori infection in BE vs. healthy © 2009 by the American College of Gastroenterology

controls. Moreover, we found that their searches were probably incomplete; for example, the Rokkas’s meta-analysis (14) only included seven studies for the analysis. Therefore, these results should be interpreted with caution because these controls may lead to an overestimate or underestimate of the real effect size of H. pylori infection in BE due to potential selection bias. In our study, although a significant difference in the rate of H. pylori infection is seen between BE and endoscopically normal controls, the current evidence is far from suggesting that H. pylori infection has a “protective” role in the development of BE because significant heterogeneity is seen, and the association shown in case–control studies cannot be interpreted as “causal.” In the sensitivity analysis for endoscopically normal controls, the heterogeneity was eliminated by excluding one outlier study (26), which was the source of heterogeneity. This Japanese study differed by inclusion of patients who had both BE and reflux esophagitis. H. pylori infection may give rise to different outcomes in Asian, European, or American populations and therefore contribute to the heterogeneity. As the Inomata study (26) was the only Asian study, we cannot conclude whether H. pylori infection is more likely to be lower in Asian BE patients compared to those in Western countries. However, the subgroup analysis suggested that the significant difference in H. pylori infection between BE and controls was significant in European populations but not in non-European populations. There is marked geographic heterogeneity in the association between H. pylori infection and GERD (43); when a strong negative association has been observed in East Asia, this association has not been apparent in Western populations. This can probably be attributed to a higher prevalence of severe H. pylori-related corpus gastritis with reduced gastric acid secretory capacity and resultant hypochlorhydria in the Asian populations (44). An American study (45) found that the adjusted relative risk for predictors of suspected BE for patients undergoing endoscopy was 0.27 in Black non-Hispanic, 0.38 in Hispanic, 0.34 in Asian Pacific, and 0.35 in non-Hispanic native American populations using the non-Hispanic population as a reference. Another American study (46) found that the incidence of BE was different in White (8.9%), Black American (2.4%), West Asian (4.8%), and East Asian patients (0). Therefore, geographic and racial differences may be one of the sources of heterogeneity between studies. More studies are needed to confirm that H. pylori infection may play a different role in BE patients in different geographic regions and in different ethnic groups. Most included studies are small with less than 50 BE cases in studies with normal endoscopy controls. The largest study was from Chile (21) with 100 BE cases and 190 endoscopically normal controls and interestingly shows an OR of 1. When this large study was included or excluded by sensitivity analysis it did not change the effect size estimates (OR 0.50 and OR 0.45, respectively). Considering that centers with a low case load are likely to be less experienced with the handling of biopsies from columnar lined epithelium than large centers, we The American Journal of GASTROENTEROLOGY

497

REVIEW

A Systematic Review and Meta-Analysis

REVIEW

498

Wang et al.

need more large studies to confirm the nonsignificant effect shown by this study to minimize this type of selection bias, or, when more large studies are published, the conclusion of this meta-analysis may be changed accordingly to favor a non-significant conclusion. According to the Montreal global consensus, the term BE is variably interpreted at the present time and lacks the clarity needed for clinical and scientific communication about columnar metaplasia of the esophageal mucosa (47). When biopsies of endoscopically suspected esophageal metaplasia show columnar epithelium it should be called BE and the presence or absence of intestinal-type metaplasia specified. In patients in whom endoscopy initially suggested BE but the biopsy was not confirmatory of intestinal metaplasia, 42% of patients did not have BE at endoscopy or on histology at a subsequent examination. In our meta-analysis, we noticed the inconsistency of the BE definition in the included studies which might have impacted on the estimated effect size. A distinction between micro-, short-segment, and long-segment BE should be made; however, most studies did not provide us with the detailed diagnostic information such as length of BE, and only two studies (26,29) reported the prevalence of H. pylori infection in longsegment BE as well as short-segment BE. Therefore, we could not perform further subgroup analyses. These factors may also contribute to heterogeneity between the studies. A standardized definition and diagnostic methods for BE are needed in future clinical trials. We also noted different subject enrollment methods. Six studies prospectively enrolled patients, which should provide for a higher quality in outcome assessment with a standard H. pylori diagnostic protocol. Three studies retrospectively collected cases and controls, and therefore the methods for BE diagnosis and H. pylori detection were not standardized and thus study quality was lower. Future studies should enroll patients prospectively to ensure standard BE diagnostic methods, H. pylori detection methods, and overall high study quality. Furthermore, we noted that only 1 case–control study was age-matched between BE cases and controls with normal endoscopy (20). It is known that both H. pylori infection and BE prevalence increase with age and therefore age is very possibly a confounder in the case–control studies. Age was a significant risk factor, with a 5% (95% CI 1–9%) increase per year in BE prevalence (48). In current meta-analysis, the age of cases in six studies was older in BE patients than in controls and no study provided data on the age stratified H. pylori infection rate; whereas the only age-matched endoscopically normal study showed that H. pylori infection was increased in the BE group (20). The difference in the infection rate between BE and controls may further decrease or even disappear if all studies were to use age-matched controls. It is difficult to explain the reasons why H. pylori infection was significantly increased in BE cases as shown in the three age-matched blood donor studies. More age-matched studies or adjustment for age in the studies in the future may influence the direction of the relationship between BE and healthy controls. The American Journal of GASTROENTEROLOGY

Therefore, a reliable conclusion could not be drawn from the current study designs with their limited evidence and the variable and weak methodologies of the available studies. These observational studies only provided the prevalence of H. pylori infection in BE and controls, without data allowing further determination of gastric acid secretion and for comparing the incidence of BE in patients with or without H. pylori-associated corpus-predominant gastritis or pangastritis. In theory, low gastric acid output resulting from H. pylori-related corpus gastritis or pangastritis is the main “protective” factor in the development of BE or GERD. It has been suggested that the presence of H. pylori infection enhanced the healing of erosive lesions with proton pump inhibitor treatment. This positive effect of H. pylori infection on healing was only observed in erosive reflux disease in patients with concurrent BE (49). A possible explanation for this selective effect of H. pylori may be that BE patients represent the most severe stage of GERD and therefore obtain the greatest benefit from the most profound acid suppression during short-term treatment. Our results show that the prevalence of H. pylori infection was significantly lower in BE than in endoscopically normal controls, but further good quality studies in well-matched homogenous patient populations and controls are needed before we can conclude that there is a possible association, if any, between H. pylori infection and BE. In our current study, we also analyzed the prevalence of H. pylori infection at different sites of gastric colonization in BE cases. Only three studies (20–22) reported the prevalence of H. pylori infection in the esophagus and stomach in the same study. We extracted the raw data for esophageal H. pylori colonization according to the authors’ description. In most cases the biopsy sites were likely correct because the protocol of biopsy for BE was standardized, although sometimes it was hard to distinguish clearly between the distal esophagus and the cardia on biopsy. The pooled prevalence of H. pylori infection was significantly lower in the esophagus than in stomach. This is likely because H. pylori only colonizes columnar epithelium and there is no evidence of a direct causal link between H. pylori infection and BE. Our results indicate that the prevalence of H. pylori decreased from the antrum, through the corpus to the fundus (24.2%, 21.3%, and 6.4%, respectively) and no available data were reported for the cardia. We took the data for H. pylori infection in the antrum rather than corpus for analysis from three studies that provided infection rates at both sites. We believe, as do others, that the shifting colonization of H. pylori from antrum to corpus results in decreasing of H + concentration, and that particularly CagA + strains may result in the greatest decrease in gastric acidity by causing more severe inflammation and subsequent atrophy of the gastric corpus. These changes may thus protect the esophagus from the damage of an acidic gastric refluxate. Therefore, biopsy in other sites alone rather than the antrum may underestimate the H. pylori infection rate. In our study, we could not find enough available data on H. pylori infection in the corpus or at the cardia to analyze the possible contribution to VOLUME 104 | FEBRUARY 2009 www.amjgastro.com

causality of H. pylori-associated corpus gastritis in the pathogenesis of BE. Future studies are needed to investigate the prevalence of H. pylori infection at the corpus and at the cardia, respectively, and the possible mechanisms of H. pylori infection in the development of BE. There is no clear evidence to support the idea that eradicating H. pylori consistently worsens or improves GERD symptoms. Treatment of H. pylori infection should not be withheld for concerns of creating or worsening GERD (50). A recent systematic review including 27 studies concluded that the available evidence does not support an association between H. pylori infection and the development of reflux esophagitis or worsening of heartburn in patients with a duodenal ulcer (51). Moreover, there is little information about the possible outcome of H. pylori eradication in BE patients. To date, only one study (52) reported that short-segment BE developed at 24 months after H. pylori eradication in 6 of 58 patients who did not have BE before the therapy, whereas the condition did not improve in 24 patients with BE before eradication. More prospective, well-controlled, long-term studies are needed and should be designed specifically to clarify the relationship of H. pylori eradication and any possible development of GERD and BE in the future. There are several limitations to our meta-analysis. First, the quality of a meta-analysis depends on the quality of the original studies, particularly the study design and reporting. As discussed above, significant heterogeneity is seen before excluding the only Asian study and the characteristics of the controls varied across studies, for instance, most controls presented with some form of GI symptoms and were diagnosed as “patients” before screening although they had a normal upper GI endoscopy. Moreover, the controls were allowed to be symptomatic in some studies, whereas they were required to be asymptomatic in others (Table 1); eight of nine case–control studies with endoscopically normal controls were not well matched for age, and age is likely to be a confounder in this kind of study. Matched population controls with a standard protocol would be the best approach when prospective studies such as cohort studies or randomized control trials are not practical for a long-term disease such as BE. However, population controls are unrealistic because it is hard to get a normal population to undergo endoscopic examination. When case–control studies are still the first choice, carefully selected cases and controls with standardized outcome and exposure assessments using a protocol based on high quality methodology are needed. Second, to correct reporting bias from original studies is difficult and requires the collaboration of the investigators involved. We failed to obtain detailed information on the data from several studies. In addition, most studies did not provide the exact biopsy sites in the stomach, which we expected such as the corpus or cardia alone, which possibly resulted in our analyzing and interpreting the current results incorrectly. Third, most of our included studies were of small sample size, which produced small statistical power, and therefore, large sample studies or meta-analysis with more homogenous studies would be © 2009 by the American College of Gastroenterology

helpful for drawing reliable conclusions. Fourth, more studies are needed to investigate the prevalence of H. pylori infection in BE in Asian and American populations. In conclusion, H. pylori infection and BE are inversely related when compared with endoscopically normal controls but not blood donor controls, suggesting that there is no clear relationship between H. pylori infection and BE. Whether H. pylori infection “protects” against the development of BE could not be answered by the current study with the limited evidence available. To determine more accurately the effect size of H. pylori infection in BE, studies in age-matched or adjusted for age, endoscopic normal controls, prospectively enrolled patients, are needed in high quality case-controlled studies. ACKNOWLEDGMENTS

We thank Dr. Laheij, Dr. Anderson, Dr. Corley, and Dr. Kuipers for providing raw data or detailed information on our studies. CONFLICT OF INTEREST

Guarantor of the article: Richard H. Hunt, MD, FRCP, FRCPC, FACG, AGAF. Specific author contributions: Changcheng Wang had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis; study concept and design: Changcheng Wang and Richard H. Hunt; independent data extraction: Changcheng Wang and Yuhong Yuan; analysis and interpretation of data: Changcheng Wang, Yuhong Yuan, and Richard H. Hunt; drafting of the paper: Changcheng Wang; substantial paper revisions: Changcheng Wang, Yuhong Yuan, and Richard H. Hunt; statistical analysis: Changcheng Wang; study supervision: Richard H. Hunt who is also the guarantor and takes overall responsibility for all aspects of the study. Financial support: None. Potential competing interest: None. REFERENCES 1. Haggitt RC. Adenocarcinoma in Barrett’s esophagus: a new epidemic? Hum Pathol 1992;23:475–6. 2. Cameron AJ, Lomboy CT, Pera M et al. Adenocarcinoma of the esophagogastric junction and Barrett’s esophagus. Gastroenterology 1995;109:1541–6. 3. Cameron AJ, Ott BJ, Payne WS. The incidence of adenocarcinoma in columnar-lined (Barrett’s) esophagus. N Engl J Med 1985;313:857–9. 4. Robertson CS, Mayberry JF, Nicholson DA et al. Value of endoscopic surveillance in the detection of neoplastic change in Barrett’s oesophagus. Br J Surg 1988;75:760–3. 5. Hameeteman W, Tytgat GN, Houthoff HJ et al. Barrett’s esophagus: development of dysplasia and adenocarcinoma. Gastroenterology 1989;96:1249–56. 6. Lord RVN, Frommer DJ, Inder S et al. Prevalence of Helicobacter pylori infection in 160 patients with Barrett’s oesophagus or Barrett’s adenocarcinoma. Aust NZ J Surg 2000;70:26–33. 7. Schenk BE, Kuipers EJ, Klinkenberg-Knol EC et al. Helicobacter pylori and the efficacy of omeprazole therapy for gastroesophageal reflux disease. Am J Gastroenterol 1999;94:884–7. 8. Loffeld RJ, Werdmuller BF, Kusters JG et al. Colonization with cagA-positive Helicobacter pylori strains inversely associated with reflux esophagitis and Barrett’s esophagus. Digestion 2000;62:95–9.

The American Journal of GASTROENTEROLOGY

499

REVIEW

A Systematic Review and Meta-Analysis

REVIEW

500

Wang et al.

9. Vaezi MF, Falk GW, Peek RM et al. CagA-positive strains of Helicobacter pylori may protect against Barrett’s esophagus. Am J Gastroenterol 2000;95:2206–11. 10. Kiltz U, Baier J, Schmidt WE et al. Barrett’s metaplasia and Helicobacter pylori infection. Am J Gastroenterol 1999;94:1985–6. 11. Abbas Z, Hussainy AS, Ibrahim F et al. Barrett’s oesophagus and Helicobacter pylori. J Gastroenterol Hepatol 1995;10:331–3. 12. Houck JA, Lucas JG. Absence of Campylobacter-like organisms in Barrett’s esophagus. Arch Pathol Lab Med 1989;113:470–2. 13. Quddus MR, Henley JD, Sulaiman RA et al. Helicobacter pylori infection and adenocarcinoma arising in Barrett’s esophagus. Hum Pathol 1997;28:1007–9. 14. Rokkas T, Pistiolas D, Sechopoulos P et al. Relationship between Helicobacter pylori infection and esophageal neoplasia: a meta-analysis. Clin Gastroenterol Hepatol 2007;5:1413–7. 15. Gisbert JP, Pajares JM. Prevalence of Helicobacter pylori infection in gastroesophageal reflux disease and Barretts esophagus. Med Clin (Barc) 2002;119:217–23. 16. Des Jarlais DC, Lyles C, Crepaz N, the TREND Group. Improving the reporting quality of nonrandomized evaluations of behavioral and public health interventions: the TREND statement. Am J Public Health 2004;94:361–6. 17. MOOSE group. Meta-analysis of observational studies in epidemiology (MOOSE) group. JAMA 2000;283:2008–12. 18. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986;7:177–88. 19. Loffeld RJ, van der Putten AB. Helicobacter pylori and gastro-oesophageal reflux disease: a cross-sectional epidemiological study. Neth J Med 2004;62:188–91. 20. Sirigu F, Capeccioni S, Dessi A et al. Helicobacter pylori in Barrett’s esophagus and in normal or inflamed esophageal mucosa: a retrospective study. Eur Rev Med Pharmacol Sci 1994;16:131–4. 21. Csendes A, Smok G, Cerda G et al. Prevalence of Helicobacter pylori infection in 190 control subjects and in 236 patients with gastroesophageal reflux, erosive esophagitis or Barrett’s esophagus. Dis Esophagus 1997;10:38–42. 22. Newton M, Bryan R, Burnham WR et al. Evaluation of Helicobacter pylori in reflux oesophagitis and Barrett’s oesophagus. Gut 1997;40:9–13. 23. Werdmuller BFM, Loffeld RJLF. Helicobacter pylori infection has no role in the pathogenesis of reflux esophagitis. Dig Dis Sci 1997;42:103–5. 24. Laheij RJF, Vanrossum LGM, Deboer WA et al. Corpus gastritis in patients with endoscopic diagnosis of reflux oesophagitis and Barrett’s oesophagus. Aliment Pharmacol Ther 2002;16:887–91. 25. Martinek J, Hucl T, Spicak J. The prevalence of Helicobacter pylori infection in some disease of the esophagus, stomach and duodenum—a retrospective analysis. Cesk Slov Gastroenterol Hepatol 2003;57:228–32. 26. Inomata Y, Koike T, Ohara S et al. Preservation of gastric acid secretion may be important for the development of gastroesophageal junction adenocarcinoma in Japanese people, irrespective of the H.pylori infection status. Am J Gastroenterol 2006;101:926–33. 27. Blaser MJ, Perez-Perez GI, Lindenbaum J et al. Association of infection due to Helicobacter pylori with specific upper gastrointestinal pathology. Rev Infect Dis 1991;13 (Suppl 8): S704–8. 28. Loffeld RJLF, Tentije BJ, Arends JW. Prevalence and significance of Helicobacter pylori in patients with Barrett’s esophagus. Am J Gastroenterol 1992;87:1598–600. 29. Ferrandez A, Benito R, Arenas J et al. CagA-positive Helicobacter pylori infection is not associated with decreased risk of Barrett’s esophagus in a population with high H. pylori infection rate. BMC Gastroenterol 2006;6:7–16. 30. Weston AP, Badr AS, Topalovski M et al. Prospective evaluation of the prevalence of gastric Helicobacter pylori infection in patients with GERD, Barrett’s esophagus, Barrett’s dysplasia, and Barrett’s adenocarcinoma. Am J Gastroenterol 2000;95:387–94. 31. Pfaffenbach B, Hullerum J, Orth K-H et al. Bile and acid reflux in long and short segment Barrett’s esophagus, and in reflux disease. Z Gastroenterol 2000;38:565–70.

The American Journal of GASTROENTEROLOGY

32. Vieth M, Masoud B, Meining A et al. Helicobacter pylori infection: protection against Barrett’s mucosa and neoplasia? Digestion 2000;62: 225–31. 33. Rugge M, Russo V, Busatto G et al. The phenotype of gastric mucosa coexisting with Barrett’s oesophagus. J Clin Pathol 2001;54:456–60. 34. Hirota WK, Loughney TM, Lazas DJ et al. Specialized intestinal metaplasia, dysplasia, and cancer of the esophagus and esophagogastric junction: prevalence and clinical data. Gastroenterology 1999;116:277–85. 35. Voutilainen M, Färkkilä M, Mecklin JP et al. Classical Barrett esophagus contrasted with Barrett-type epithelium at normal-appearing esophagogastric junction. Central Finland Endoscopy Study Group. Scand J Gastroenterol 2000;35:2–9. 36. Vicari JJ, Peek RM, Falk GW et al. The seroprevalence of CagA-positive Helicobacter pylori strains in the spectrum of gastroesophageal reflux disease. Gastroenterology 1998;115:50–7. 37. Csendes A, Smok G, Burdiles P et al. ‘Carditis’: an objective histological marker for pathologic gastroesophageal reflux disease. Dis Esophagus 1998;11:101–5. 38. Sorberg M, Nyren O, Granstrom M. Unexpected decrease with age of Helicobacter pylori seroprevalence among Swedish blood donors. J Clin Microbiol 2003;41:4038–42. 39. Ronkainen J, Aro P, Storskrubb T et al. Prevalence of Barrett’s esophagus in the general population: an endoscopic study. Gastroenterology 2005;129:1825–31. 40. Vaira D, Miglioli M, Mule P et al. Prevalence of peptic ulcer in Helicobacter pylori positive blood donors. Gut 1994;35:309–12. 41. Wilhoite SL, Ferguson DA, Soike DR Jr et al. Increased prevalence of Helicobacter pylori antibodies among nurses. Arch Intern Med 1993;153:708–12. 42. Menegatti M, Holton J, Figura N et al. Clinical significance of Helicobacter pylori seropositivity and seronegativity in asymptomatic blood donors. Dig Dis Sci 1998;43:2542–8. 43. Raghunath A, Hungin AP, Wooff D et al. Prevalence of Helicobacter pylori in patients with gastro-oesophageal reflux disease: systematic review. BMJ 2003;326:737. 44. Koike T, Ohara S, Sekine H et al. Helicobacter pylori infection prevents erosive reflux oesophagitis by decreasing gastric acid secretion. Gut 2001;49:330–4. 45. Lieberman D, Fennerty MB, Morris CD et al. Endoscopic evaluation of patients with dyspepsia: results from the national endoscopic data repository. Gastroenterology 2004;127:1067–75. 46. Spechler SJ, Jain SK, Tendler DA et al. Racial differences in the frequency of symptoms and complications of gastro-oesophageal reflux disease. Aliment Pharmacol Ther 2002;16:1795–800. 47. Vakil N, van Zanten SV, Kahrilas P, et al., the Global Consensus Group. The Montreal definition and classification of gastroesophageal reflux disease: a global evidence-based consensus. Am J Gastroenterol 2006;101:1900–20. 48. Johansson J, Hakansson HO, Mellblom L et al. Risk factors for Barrett’s oesophagus: a population-based approach. Scand J Gastroenterol 2007;42:148–56. 49. Malfertheiner P, Lind T, Willich S et al. Prognostic influence of Barrett’s oesophagus and Helicobacter pylori infection on healing of erosive gastrooesophageal reflux disease (GORD) and symptom resolution in non-erosive GORD: report from the proGORD study. Gut 2005;54:746–51. 50. Chey WD, Wong BC, Practice Parameters Committee of the American College of Gastroenterology. American College of Gastroenterology guideline on the management of Helicobacter pylori infection. Am J Gastroenterol 2007;102:1808–25. 51. Raghunath AS, Hungin AP, Wooff D et al. Systematic review: the effect of Helicobacter pylori and its eradication on gastro-oesophageal reflux disease in patients with duodenal ulcers or reflux oesophagitis. Aliment Pharmacol Ther 2004;20:733–44. 52. Yachida S, Saito D, Kozu T et al. Endoscopically demonstrable esophageal changes after Helicobacter pylori eradication in patients with gastric disease. J Gastroenterol Hepatol 2001;16:1346–52.

VOLUME 104 | FEBRUARY 2009 www.amjgastro.com