Helicobacter pylori infection and the risk of upper gastrointestinal bleeding in low dose aspirin users: systematic review and meta-analysis

Low dose aspirin — that is, at a daily dosage of 325 mg or less — is widely used as an antiplatelet agent for reducing cardio- and neurovascular morbidity and mortality. This definition of “low dose aspirin”, widely applied since the earliest investigations of its effectiveness in preventing stroke and myocardial infarction,1 is based on the North American formulation of single analgesic-strength tablets, but in many countries the typical strength is 300 mg. Lower dose formulations have since become widely available, and most clinicians recommend an 80 or 100 mg tablet or half an analgesic-strength tablet as the maximum daily dosage for vascular protection. The increasing use of low dose aspirin, however, has been accompanied by a rise in the number of hospitalisations for aspirin-associated upper gastrointestinal (UGI) bleeding.2,3 This is important because mortality rates among aspirin users who present with UGI haemorrhage are high, despite advances in endoscopic and surgical techniques for controlling UGI bleeding.4

Helicobacter pylori infection and aspirin are independent risk factors for UGI haemorrhage, and their interaction as risk factors is unclear.5 From a pathophysiological perspective, the interaction could conceivably be antagonistic rather than additive or synergistic; for instance, H. pylori infection stimulates the production of gastric mucosal prostaglandins, which may counter the depletion resulting from inhibition of mucosal cyclooxygenase-1 by aspirin.6

European and American guidelines recommend a test-and-treat approach to H. pylori infections in aspirin users who are at high risk of peptic ulcer bleeding.7,8 However, the evidence for this recommendation is based largely on a meta-analysis of studies in which patients were treated with conventional non-steroidal anti-inflammatory drugs (NSAIDs); these results may not be generalisable to aspirin, as this drug also causes UGI bleeding by sustained inhibition of platelet aggregation.9

Information on the effects of H. pylori infection on UGI bleeding in patients taking aspirin is conflicting and indirect. In a randomised controlled trial involving H. pylori-infected aspirin users who had experienced peptic ulcer bleeding, the incidence of recurrent ulcer bleeding was similar for patients in whom H. pylori was eradicated and for those treated with a proton pump inhibitor.10 A prospective study found a low incidence of recurrent peptic ulcer bleeding in H. pylori-infected aspirin users after eradicating the infection.11 However, another randomised controlled trial found that 15% of aspirin users had recurrent peptic ulcer bleeding during 12 months’ follow-up, despite patients receiving treatment for H. pylori infection.12

The authors of a 2010 systematic review of studies of H. pylori-infected aspirin users who had experienced UGI bleeding were unable to retrieve sufficient comparable data to perform a meta-analysis.6 We therefore undertook an updated systematic review with the aim of identifying publications suitable for meta-analysis that had reported UGI bleeding in low dose aspirin users of known H. pylori status.

Methods

Search strategy

We searched MEDLINE, PubMed, Embase, Scopus, and the Cochrane Library databases for relevant articles published to October 2017. In MEDLINE, PubMed, and the Cochrane Library, we searched for the MeSH heading “peptic ulcer hemorrhage” or the text words “peptic ulcer hemorrhage”, “peptic ulcer bleed*”, “upper gastrointestinal hemorrhage”, “upper gastrointestinal bleed*”, “stomach hemorrhage”, “stomach bleed*”, “small intestine hemorrhage”, “small intestine bleed*”, “duodenum hemorrhage”, or “duodenum bleed*”. The search results were filtered for articles indexed with the MeSH heading “aspirin” or including the text words “aspirin” or “acetylsalicylic acid”. The search was restricted to studies in humans, without language restrictions. The search strategy for the Scopus database and Embase (with Emtree phrases) was similar. Embase includes abstracts from major conferences as well as full articles. H. pylori was not used as a search term because we were aware that at least one article with data suitable for meta-analysis was not indexed with this term in the databases. Consequently, a large number of articles needed to be searched individually.

Selection criteria

Articles included in the meta-analysis described original studies that reported the frequency of UGI haemorrhage separately for aspirin users who were H. pylori-positive or -negative. Endoscopic documentation of UGI bleeding was preferred but not essential. We excluded articles without original data (reviews, editorials, guidelines); articles describing bleeding from sources other than peptic ulcers, such as gastric tumours or haemorrhagic gastritis, or concomitant treatment with NSAIDs; articles without an abstract; articles published before 1989 (when H. pylori was named);13 and studies for which the full article was not available online.

Data extraction

The title and abstract of an article were screened for relevance before retrieving the article; its full text was then analysed to determine whether it contained data that could be pooled for meta-analysis according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines,14 including the number of aspirin users with known H. pylori status who had or had not experienced a UGI haemorrhage, allowing the odds ratio (OR) for the risk of UGI haemorrhage in aspirin users who were H. pylori-positive or -negative to be calculated.

Quality assessment

The quality of case–control studies was assessed with the Newcastle–Ottawa Scale.15 This tool assesses three components of a case–control study to rank its quality on a nine-point scale: selection criteria for cases and controls; comparability of cases and controls; and methods used to assess exposure to the investigated risk factor. Articles were assessed independently by each investigator, discrepancies in individual scores discussed, and the mean score calculated. Studies scoring seven or more were deemed to be high quality studies.

Statistical analysis

The key outcome was the summary OR (with 95% confidence interval [CI]) for UGI bleeding in H. pylori-infected v uninfected patients, pooled across the included studies. The ORs for individual studies were calculated from the published raw data; when the required raw data were not explicitly reported, the total numbers of cases and controls who were H. pylori-positive or -negative were calculated from other reported data for the study.

Meta-analysis summary statistics and forest plots were generated in a random effects model with MIX 2.0 Pro (https://www.meta-analysis-made-easy.com). Heterogeneity was estimated as the Q, I² and τ² statistics. Separate sensitivity analyses excluded studies deemed to be outliers or low quality studies. Publication bias was assessed by generating a trim-and-fill plot. Number needed to treat was calculated as the reciprocal of the absolute odds reduction.

Results

Publications

Our search initially retrieved 7599 records. After excluding duplicates and articles published before 1989, the titles and abstracts of 3932 articles were screened; after excluding articles without original data and those that clearly did not meet our inclusion criteria (eg, single case reports, outcomes other than UGI bleeding, no data on H. pylori), 225 articles were reviewed in detail, seven of which were eligible for the meta-analysis (Box 1).16-22 The PRISMA flowchart of the selection process is provided in the online Appendix, figure 1.

Study characteristics

The seven included studies were all case–control studies. They included a total of 1172 patients, 1132 of whom took low dose aspirin, defined as ≤ 325 mg/day; the dosage for 40 patients was not reported. Cases were patients who had presented with UGI bleeding. Five of the seven studies required that UGI bleeding had been endoscopically confirmed; two did not state whether there was endoscopic evidence of UGI haemorrhage, but the context made it highly likely.16,21 Controls were age- and sex-matched, with no clinical signs of UGI haemorrhage at the time of interview. The participants were drawn from Spain, Denmark, Finland, the United Kingdom, and Japan.

Infection with H. pylori was assessed in cases by a variety of methods (Box 1). In all studies, controls were tested with the ¹³C-urea breath test. The pooled prevalence of H. pylori infection was 61.1%.

Quality assessment of studies (Newcastle–Ottawa Scale)

Four of the seven studies were of high quality (7 or more points on the Newcastle–Ottawa Scale) (Box 1). The main reasons for three studies rating lower than 7 were limited representativeness of the cases with respect to the overall population and the absence of community-based controls. There was excellent agreement between the scores given to each study by the two investigators (for discrimination of high quality studies: κ = 1.00; r = 0.93, P = 0.002).

Meta-analysis

All studies reported sufficient data to allow ORs to be calculated for UGI haemorrhage in aspirin users according to H. pylori status. Six of the seven provided raw numbers for each category;16,18-22 one reported only the crude OR for association between H. pylori infection and bleeding peptic ulcers in aspirin users,17 so we calculated numbers of cases and controls from the published OR and the total numbers in each group.

The pooled effect size in the meta-analysis indicated that UGI bleeding was almost two-and-one-half times as frequent in H. pylori-positive patients taking aspirin as in H. pylori-negative patients (OR, 2.32; 95% CI, 1.25–4.33; P = 0.008) (Box 2). The heterogeneity of the included studies was significant (Q = 19.3, P = 0.004; I² = 68.9% [95% CI, 31.5–85.9%]; τ² = 0.43 [95% CI, 0.09–1.19]). Two studies were deemed outliers20,21 (Box 3). A Baujat plot (online Appendix, figure 2) indicated that the Shiotani study21 made the greatest contribution to the Q statistic and summary OR.

A trim-and-fill plot generated to assess the likelihood of publication bias suggested that three negative studies might be missing, one of lower precision and two studies of higher precision (online Appendix, figure 3).

Sensitivity analyses

In the first sensitivity analysis, we recalculated the pooled effect size for meta-analyses from which the two outliers in the heterogeneity plot were sequentially removed. Removing the Udd study20 resulted in a pooled OR of 1.94 (95% CI, 1.13–3.35, P = 0.016) and less heterogeneity (Q = 13.0, P = 0.024; I² = 61.4% [95% CI, 5.9–84.2%]; τ² = 0.27 [95% CI, 0.01–0.89]). Removing the Shiotani study21 resulted in a pooled OR of 2.81 (95% CI, 1.59–4.97; P < 0.001) and heterogeneity was no longer statistically significant (Q = 9.85, P = 0.08; I² = 49.2% [95% CI, 0–79.8%]; τ² = 0.23 [95% CI, 0–0.93]). Removing both studies produced a pooled OR of 2.34 (95% CI, 1.56–3.53; P < 0.001), and heterogeneity was not statistically significant (Q = 4.63, P = 0.33; I² = 13.6% [95% CI, 0–82.0%]; τ² = 0.03 [95% CI, 0–0.96]).

The second sensitivity analysis included only higher quality studies (7 or more points on the Newcastle–Ottawa scale); the summary OR was 2.41 (95% CI, 1.46–3.96; P < 0.001), similar to the result when all studies were pooled.

Number needed to treat

The incidence of UGI bleeding in people taking low dose aspirin is 0.1–1 events per 100 patient-years.23-25 Applying our finding that the incidence of bleeding among patients with H. pylori infections was approximately double that for uninfected people, the number needed to treat to prevent one bleeding event annually ranges between 100 and more than 1000.

Discussion

Our meta-analysis indicates that the proportion of H. pylori-positive patients taking low dose aspirin who have UGI haemorrhages is greater than that for low dose aspirin users who are H. pylori-negative. For only one of the eligible studies was the OR less than 1, and its 95% CI included unity; the other six studies had ORs between 1.6 and 40. Two sensitivity analyses (omitting two outlier studies in the heterogeneity plot or three studies of lower quality on the Newcastle–Ottawa Scale) each yielded similar estimates of the increased frequency of UGI bleeding associated with H. pylori infection. The magnitude of the effect is moderate (OR, 2.32), slightly larger than the OR of 1.67 (95% CI, 1.02–2.72) estimated by a meta-analysis of the effects of H. pylori infection on the risk of ulcer bleeding in people taking NSAIDs other than aspirin.26

Does an increase of this magnitude in the risk of UGI bleeding warrant a test-and-treat approach before starting treatment with low dose aspirin or NSAIDs, as recommended by international guidelines? Our analysis provides a numerical value that physicians and patients can discuss when aspirin treatment is being considered. The test-and-treat approach may seem straightforward, but involves time and costs for both the H. pylori tests and the antibiotic regimens. Further, microbial antibiotic resistance means that success rates for the initial course of treatment are often lower than 80%.8 Data on compliance with the guidelines for aspirin treatment have not been published, but several studies undertaken since the first of the major guidelines was published have found that fewer than one in five patients taking non-aspirin NSAIDs had been tested for H. pylori infection.27-29 A further complication is that the specialists who most often institute aspirin therapy — cardiologists and neurologists — may not be confident about managing H. pylori infection, although they can request the referring general practitioner to assume this role.

As the number needed to treat to prevent one extra bleed annually is estimated to be quite high, and given the poor adherence to guidelines for low dose aspirin treatment, it is reasonable to suggest that the patients receiving aspirin who should be more strongly considered for a test-and-treat approach are those at greatest risk of ulcer bleeding. A wealth of epidemiological data indicate that this group includes older patients, smokers, patients also taking non-aspirin NSAIDs, and those with a history of peptic ulceration.30,31 At greatest risk are patients who have already had a bleeding ulcer while taking aspirin and have resumed taking aspirin after the ulcer healed. Such patients remain at high risk of UGI bleeding despite eradication of H. pylori12 and should probably always receive a proton pump inhibitor while taking aspirin.

One further important factor is that patients who are prescribed low dose aspirin will often be at increased risk of serious consequences were they to have a major UGI haemorrhage. Many have underlying cardiovascular disease and are more likely to sustain myocardial injury should bleeding lead to hypotension.32

Strengths and limitations of our analysis

The main strength of our study is that it is the first meta-analysis to investigate the contribution of H. pylori infection to rates of UGI bleeding in aspirin users. Most of the articles providing relevant data did not pursue this question as their primary aim, and the data often needed to be extracted from their results sections, perhaps explaining why a meta-analysis was not undertaken earlier. However, the most relevant studies were of high quality. A further strength is that the results of our sensitivity analyses increased our confidence in those of the main analysis.

Limitations include the fact that positive H. pylori serology results do not necessarily indicate a current infection. The antibody titre falls gradually after successful treatment; about 50% of treated patients will return a negative test result by 12 months, but in 50% the antibody titre falls more slowly.33-35 If patients in the included studies had been treated for H. pylori infection very recently, they may have been incorrectly assigned to H. pylori-positive groups, and we will have underestimated the strength of the association between H. pylori infection and UGI bleeding. Another limitation is that the data were derived from observational case–control studies, so that comparability of groups cannot be assured and unrecognised confounders may have influenced outcomes. A controlled trial in which patients infected with H. pylori are randomised to treatment or placebo before commencing aspirin would minimise this risk. Trials of this type have not been reported, and may even be unethical, as there is evidence that H. pylori is a class I carcinogen36 and causes non-NSAID-linked peptic ulcers; most practitioners would institute treatment as soon as a patient was known to carry the organism.

Conclusion

Despite these limitations, the consistency of our results (after sensitivity analyses) indicates that the likelihood of UGI haemorrhage in patients taking low dose aspirin is more than twice as high for those with H. pylori infections. The cost–benefit balance of testing for and treating the bacterium may be insufficient to permit recommending this approach for all patients receiving aspirin on a long term basis. However, the evidence is sufficient to warrant considering eradication of the infection in patients who are at high risk of ulcer complications because of comorbid conditions.

Box 2 – Forest plot of studies comparing upper gastrointestinal bleeding in Helicobacter pylori-positive and -negative patients taking low dose aspirin

---

The area of each blue box is proportional to the weighting of the study. The summary odds ratio and its 95% confidence interval, generated in a random effect model, are depicted by the red diamond.

Box 3 – Heterogeneity funnel plot, with the fixed effect pooled estimate shown by the vertical line, and pseudo-95% confidence limits by the curved boundary lines

---

OR = odds ratio. The notable outliers are the studies by Udd et al.20 (log_e(OR) > 3.5) and Shiotani et al.21 (log_e(OR) < 0). Heterogeneity statistics: Q = 19.3 (P = 0.004); I² = 68.9% (95% CI, 31.5–85.9%); τ² = 0.43 (95% CI, 0.09–1.19).