To investigate the role of Helicobacter pylori, expressing the virulence marker CAGA (cytotoxin associated gene product A) in ulcer complications and its interaction with nonsteroidal anti-inflammatory drugs (NSAIDs) and other risk factors.
To investigate the role of Helicobacter pylori, expressing the virulence marker CAGA (cytotoxin associated gene product A) in ulcer complications and its interaction with nonsteroidal anti-inflammatory drugs (NSAIDs) and other risk factors.
Case control study using conditional logistic regression analysis.
University and City Hospitals, Nottingham.
203 consecutive patients with ulcer bleeding and 203 age- and sex-matched controls.
Ulcer bleeding was more likely with positive H. pylori serology (odds ratio = 3.3, 95% CI: 1.7–6.6 for CagA positive, but only OR = 1.6, 95% CI: 0.7–3.7 for CagA negative serology), current smoking (OR 2.2, 95% CI: 1.04–4.7), aspirin ≤ 300 mg daily (OR 7.7, 95% CI: 2.8–20.6), all other nonsteroidal anti-inflammatory drugs (NSAIDs: OR 10.6, 95% CI: 3.1–35.7 for ≤ 1 defined daily dose lower and OR 22.6, 95% CI: 6.2–82.0 for higher doses) and past ulcer history (OR 5.6, 95% CI: 2.3–14.1). Aspirin ≤ 300 mg daily was used by 25.1% of patients vs. 7.4% of controls. Smoking only enhanced risk in the presence of H. pylori, with a synergistic interaction (interaction odds ratio = 4.9, 2.4–9.9, P=0.002). Conversely, risks with non-aspirin NSAIDs were reduced in the presence of H. pylori, particularly if CagA-positive (interaction odds ratio=0.21, 0.05–0.9, P=0.03).
CagA positive H. pylori infection is associated with an increased risk of ulcer bleeding. The risk from non-aspirin NSAIDs is even higher, but is less in H. pylori infected people. Low-dose aspirin is now commonly associated with ulcer bleeding.
The most important causes of peptic ulcer disease and its complications are Helicobacter pylori1 and NSAIDs,2–12 but how they interact remains unclear. In a previous study, H. pylori infection did not appear to increase the risks of nonsteroidal anti-inflammatory drug (NSAID) associated ulcer bleeding.12 However, the collection of drug and H. pylori data in that study was not simultaneous and it did not investigate low dose aspirin use, which has since become more widespread. Moreover, it did not differentiate between more virulent H. pylori expressing the cytotoxin associated gene product13 (CagA positive) and CagA negative bacteria. We therefore conducted a case control study, carried out prospectively, into the relative contribution of CagA positive and CagA negative H. pylori infection to peptic ulcer bleeding and its interactions with other risk factors. A second goal of the study was to determine the extent to which the growing use of low-dose aspirin for cardiovascular prophylaxis now contributes to peptic ulcer bleeding.
We studied all patients admitted to the University and City Hospitals, Nottingham, UK as a result of a clinical diagnosis of acute upper gastrointestinal bleeding with haematemesis and/or melaena, over a 24-month period. Strenuous efforts were made to identify all such patients by a daily review of ward and high dependency unit admissions, accident and emergency records, endoscopy lists and blood cross-match requests. All cases where an ulcer (gastric, duodenal or both) was found at endoscopy formed the index case cohort. Hospital patients were used as controls in order to facilitate the efficient and reliable collection of samples for serology from large numbers of individuals. For each case, the control was the next patient of the same sex and of similar age (± 5 years) admitted to the medical wards, within a maximum of 3 months of the index case admission. Patients admitted for reasons that would have had a major effect on use of aspirin or non-aspirin NSAIDs (peptic ulcer disease or previously recognized ischaemic heart disease or arthritis) were ineligible to be controls, although patients admitted for other reasons who coincidentally had these conditions were eligible. Previous studies have shown that hospital controls selected in this way yield similar results to community controls.2, 7
Each bleeding ulcer case and control was interviewed by a trained research associate (G.M.H.) who was not aware of the specific study hypotheses. Structured questionnaires were used to seek details of all NSAID use over the past year, whether prescribed or self-administered. Patients were also questioned as to the type, dose, duration and indication for NSAID use. Current NSAID use was defined as any use within the week prior to admission. Where patients reported a past history of peptic ulcer, evidence for endoscopic confirmation of this was sought from hospital or general practice records. Serum samples were taken from each case and control subject and stored (−20 °C) for H. pylori and CagA serology.
The study was approved by the University Hospital, Nottingham, Ethical Committee and conducted by staff established within Nottingham University Medical School who received no specific funding and had no competing interests.
H. pylori status was determined by the presence of IgG antibodies to H. pylori using a Premier ELISA kit (Meridian Diagnostics, data on file). This ELISA uses a sonicated H. pylori bacterial cell lysate absorbed to plastic microwells. Numerous publications have shown that this and other serological tests distinguish between those who are currently infected with H. pylori and those who are not.14, 15 Thus, when validated against the rapid urease test (CLO-Test, Delta West) and histology in 462 patients, the Premier ELISA had a 98.8% (95% confidence interval: 96.7–99.6%) sensitivity and 95.5% (95% CI: 91.7%–97.6%) specificity (Meridian Diagnostics, data on file) and when validated against histology and urea breath test in 84 dyspepsia patients it had a 100% sensitivity and specificity (Meridian Diagnostics, data on file).
The presence of anti-CagA antibodies was determined by an in-house ELISA modified from Blaser et al. 1995.16 The recombinant CagA antigen orv 220 (a gift from Orvax Inc., Cambridge, MA) was loaded into 96-well microtitre plates at 1 μg per well. Serum for testing was diluted 1 : 100. Binding was detected using a horseradish peroxidase-conjugated goat antihuman IgG (Sigma) at a dilution of 1 : 1000. An assay range was established and standardized using 162 H. pylori-negative samples by taking the mean H. pylori-negative CagA optical density plus three standard deviations as the threshold for designating a sample CagA positive.16 To establish sensitivity and specificity, H. pylori was cultured from a gastric biopsy samples of 36 patients and the presence of the cagA gene established by colony hybridization.16 Independent, blinded anti-CagA serotyping was positive in all of the 24 strains that were cagA positive by DNA hybridization (ELISA sensitivity 100%) and negative in 8 of the 10 stains that were cagA negative by DNA hybridization (specificity 80%). Since `false positive' ELISA results can arise because of co-infection with an un-isolated cagA positive strain,17 the two `false positive' serum samples, along with 14 others, were compared with Western blotting (Helicoblot 2.0, Genelabs Diagnostics, Geneva), resulting in complete agreement (sensitivity and specificity 100%).
Our primary hypotheses were that CagA positive, but not CagA negative H. pylori would enhance the risk of ulcer bleeding and that, overall, NSAIDs and H. pylori would not enhance each other's risks. We hypothesized that this overall null effect would arise because increased risks with pathogenic CagA positive H. pylori in NSAID users would be counter-balanced by a protective effect of less virulent CagA negative strains (via inflammation-induced increased prostaglandin synthesis18). We also hypothesized that smoking, past history, use of anticoagulants and use of low dose aspirin would be associated with increased risks. In the absence of prior information, we had no hypothesis about the interaction between H. pylori and low dose aspirin. Because we wished to study the unconfounded effects of these factors and their interactions, we used a univariate conditional logistic regression analysis to confirm their association with ulcer bleeding and then entered the terms into a multivariate model that also included two levels of NSAID dose. Dose comparisons were established by converting all the NSAID doses to defined daily doses (DDDs).19 For non-aspirin NSAIDs, lower doses were defined as ≤ 1 DDD per day; higher doses were > 1 DDD per day. For aspirin, lower doses were ≤ 0.4 DDD (≤ 300 mg) per day; higher doses were> 0.4 DDD (> 300 mg) per day. Since there were extremely wide confidence intervals around the estimate of risk with anticoagulants, this factor was not entered into the multivariate model. Possible interactions between individual determinants were investigated by a series of individual bivariate logistic regression analyses. The odds ratios for each possible combination of the two potentially interacting factors were also calculated for the purposes of illustration.
In order to identify factors which specifically associated with gastric ulcer or duodenal ulcer bleeding, similar analyses, but with past history specified as gastric ulcer or duodenal ulcer, were carried out for patients with gastric ulcer and their matched controls or for patients with duodenal ulcer and their matched controls. Because the numbers were too small, interactions were not investigated.
Comparisons of proportions were made using χ2 analysis (if a logistical regression analysis had identified the factor as a significant influence). T-testing was used for comparisons of means. In all cases, a P-value of 0.05 was taken as significant. Statistical analysis was performed using S+2000 and SPSS version 8.1.
Details of patients and controls are shown in Table 1. All cases were admitted because of bleeding gastric or duodenal ulcers. Reasons for admission in the controls are shown in Table 2. One hundred and twenty-three of the cases had a duodenal ulcer (61%) and 80 had a gastric ulcer (39%) as the primary cause of the bleed. Cases and controls were appropriately matched for age and sex but cases were significantly more likely to be seropositive for H. pylori, to be using NSAIDs, non-aspirin NSAIDs and aspirin, to have a past history of peptic ulcer, to be a current smoker or to be taking anticoagulants (Table 1). There was also a trend for higher alcohol consumption in cases than controls (Table 1).
In the 7 days prior to admission, 59 cases (29.1%) had taken aspirin compared to 15 (7.4%) controls (P < 0.0001). Fifty-one cases (25.1% of all cases, 86.4% of aspirin users) were taking aspirin ≤ 300 mg daily (10 with a NANSAID) compared to 15 (7.4%) of controls (P < 0.0001). In turn, most of the aspirin ≤ 300 mg daily was for 150 mg daily (20 cases, 6 controls) or 75 mg (22 cases, 8 controls). Most of the cases and controls were using aspirin for cardiovascular prophylaxis. The use of a NANSAID was seen in 22.2% of cases and 6.9% of controls (P < 0.0001). The main non-aspirin NSAIDs used were diclofenac, ibuprofen, naproxen, indometacin or ketoprofen. The most common reason for use was a musculoskeletal problem. Co-use of aspirin and NANSAIDs was seen in 10 and 1 cases and controls, respectively.
Information on H. pylori was available in all cases and controls. Information on CagA status was incomplete for 35 pairs of cases and controls because the samples were lost during a freezer breakdown. Cases were more likely than controls to have a positive serology for both H. pylori (73.9% vs. 53.7%, P < 0.0001) and for CagA (51.6% vs. 28.5%, P < 0.0001). The proportion of cases and controls who were seropositive for H. pylori but seronegative for CagA was not significantly different (19.2% in cases vs. 21.0% in controls, P=0.70). Thus, the increase in positive serology for H. pylori in cases over controls was entirely attributable to an excess of CagA positive infection (Figure 1). Twenty four cases and 22 controls reported antibiotic use. None was for more than one antibiotic and none reported having received H. pylori eradication in the previous year.
The excess prevalence of positive H. pylori and positive CagA serology was restricted to patients not taking NSAIDs. Amongst those currently using NSAIDs, 65.6% of cases were H. pylori positive vs. 60.1% of controls (P=0.66) and CagA serology was positive in 41.4% of cases vs. 42.3% of controls (P=1.0).
For comparison with previous studies that did not take account of H. pylori status, we calculated univariate odds ratios for aspirin and non-aspirin NSAIDs. The odds ratio for aspirin ≤ 300 mg was 3.5 (1.8–6.8). For all other NSAIDs, it was 4.1 (1.5–10.9) for ≤ 1 DDD and 9.9 (3.3–30.0) for > 1 DDD. H. pylori was associated with an odds ratio of 2.9 (1.7–4.9) for CagA positive, P=0.001, and 1.5 (0.78–2.9) for CagA negative strains, P=0.213.
Estimates of risk were considerably higher than in the univariate analyses, since estimates were not blunted by the effect of unanalysed risk factors. Overall, positive H. pylori serology, current NSAID use, past history of peptic ulcer disease and current smoking were associated with an increased risk of bleeding peptic ulcer (Figure 1). Current anticoagulant use was more common in cases (11.3%) than controls (1.5%, P < 0.0001), but was removed from the multivariate analysis because small numbers led to a very large confidence interval, making it impossible to analyse their influence. With H. pylori, only patients who were seropositive for CagA had a significantly increased risk (odds ratio=3.3, 95% CI: 1.7–6.6 vs. OR 1.6, 95% CI: 0.7–3.7). Risks were significantly dose-dependent for non-aspirin NSAIDs (Figure 1). The risk with aspirin ≤ 300 mg daily was 7.7 (95% CI: 2.8–20.6), similar to that with non-aspirin NSAIDs ≤ 1 DDD. None of the controls was taking aspirin > 300 mg daily.
We primarily investigated interactions with CagA positive H. pylori infection, reflecting our initial hypotheses, since CagA negative infection was not increased in cases compared to controls or associated with an increased odds ratio vs. no infection, whilst the odds ratio for CagA positive H. pylori was significantly increased compared to no infection, and near significantly vs. CagA negative infection. However a broadly similar pattern of results was seen if interactions with all H. pylori positive infection were investigated. In general, the risk of ulcer bleeding in patients who were infected with CagA positive H. pylori and who had other risk factors were higher than for either factor alone, although for most there was no significant interaction, indicating that the effects of both factors together were no more than additive (Figure 3).
There was however, a significantly synergistic interaction between smoking and H. pylori (whether for all cases or only those that were CagA positive). The interaction odds ratio for smoking with H. pylori was 4.9 (95% CI: 2.4–9.9) and for CagA positive serology 3.0 (95% CI: 1.04–8.5). Smoking had no effect in the absence of H. pylori (odds ratio 0.60, 95% CI: 0.27–1.29).
There was a highly significant interaction between other NSAIDs and H. pylori, but in the opposite direction, with reduced risks in H. pylori positive vs. negative users of NSAIDs. The interaction term for CagA positive H. pylori and all NSAIDs other than low dose aspirin was 0.21 (95% CI: 0.05–0.89, P=0.03). All six possible interactions between all H. pylori or CagA positive H. pylori and all NSAIDs, non-aspirin NSAIDs or all NSAIDs other than low dose aspirin showed a negative interaction of a similar magnitude. By contrast, there was no interaction between CagA negative H. pylori and other NSAIDs, non-aspirin NSAIDs or all NSAIDs. The effects of aspirin ≤ 300 mg and H. pylori (all or CagA positive) were compatible with an additive effect.
All other interactions (e.g. between past history and smoking or between current use of aspirin or non-aspirin NSAIDs) were additive (Figure 2).
Overall, similar patterns were seen for gastric and duodenal ulcer, analysed separately. The reported site of a past history was associated with whether the patient presented with a gastric or duodenal ulcer (Figure 3).
H. pylori infection1 smoking8, 10, 20–21 use of NSAIDs or aspirin2–13, 23, 24 past history4, 6–8, 10, 11 and anticoagulants25, 26 have all previously been associated with bleeding peptic ulcer, although not uniformly in all studies. However, they have often been investigated separately and their precise influence and relationships to each other have been uncertain. The effect of H. pylori has not been related to CagA status and the relative role of aspirin ≤ 300 mg has not been studied extensively. The approach we have taken has allowed many outstanding questions to be addressed.
There are five main conclusions from our study. First, whilst H. pylori is associated with bleeding gastric and duodenal ulcer this is only significant for CagA positive H. pylori, as assessed by serology. Less virulent CagA negative H. pylori were not significantly associated with either bleeding duodenal or gastric ulcer. Second, whilst the use of non-aspirin NSAIDs remains a major cause of bleeding peptic ulcer (with higher odds ratios in our study than for H. pylori), low dose aspirin, which is now widely used for cardiovascular prophylaxis, was associated in our study with as much ulcer bleeding as all non-aspirin NSAIDs together. Third, the increased risks associated with a past ulcer history appear to be site preferential, as previously reported.27, 28 They are independent of, and cannot be attributed to, general risk factors such as H. pylori and may therefore be mediated by a local mucosal consequence of prior ulceration.28, 29 Fourth, the effects of smoking were confined to patients infected with H. pylori where their interaction appeared to be synergistic. Similar results have been reported for endoscopic ulcers.22 Our data are compatible with those of Weil et al. who, in a study not assessing H. pylori, found a small overall effect of smoking.10 In that study, risks were higher with a combination of NSAIDs and smoking. Finally, whilst risks tended to be higher with two compared to one risk factor, this was not the case for the interaction between H. pylori and non-aspirin NSAID use. Patients using NSAIDs other than low dose aspirin who were sero-positive for H. pylori (particularly CagA positive strains) had a significantly reduced risk compared to uninfected individuals.
Several aspects of our study design enabled us, for the first time, to simultaneously collect systematic data about H. pylori infection, CagA status, patient demography and drug use in patients presenting with peptic ulcer bleeding. We used serotyping because this is highly sensitive and specific for current infection, applicable to all patients (including those that die and undergo urgent surgery) and does not have the inaccuracy of endoscopy-based methods in the presence of gastrointestinal bleeding.30 Whilst it takes up to a year after successful treatment for H. pylori antibodies to decline to levels indistinguishable from patients that have never been infected,31, 32 antibiotic use did not differ between patients and controls in our study and none had previous specific H. pylori eradication. Serology in such populations is a reliable index of current infection.14–17
Control selection is an important potential source of bias. Community controls may be subject to Berkson's bias,33 whilst the reason for admission may be a distortion with hospital controls. Overall community controls are more favoured than hospital controls, and the use of hospital controls may be a weakness of our study. However, previous studies have shown that hospital controls selected as in this study yield results similar to those seen with community controls.2, 7 The main advantage of hospital controls for our study was that serology could be guaranteed, whilst our previous study had shown that distortions were likely to occur with community controls, which were an unreliable source.12 Drug history is potentially more accurate with hospital controls. Moreover, aspirin and non-aspirin use by controls ≥ 65 (20% for all NSAIDs, 11% for NANSAIDs) was similar to those recorded by similar controls in our local community.2, 7 Furthermore, for H. pylori, NSAIDs, aspirin, smoking and alcohol, the use of hospital controls is likely, if anything, to be conservative, although the potential for distortion with other factors is less certain.
In our study, the ability to accurately define H. pylori status resulted in an increased precision in assessing other simultaneously analysed factors. Our data comparing odds ratios derived from univariate and multivariate analyses emphasize the extent to which previous studies have underestimated the risks associated with non-aspirin NSAIDs, by not allowing for unmeasured H. pylori-associated risk in controls. Our data show that, amongst patients without H. pylori infection or other risk factors, odds ratios are increased by nearly eightfold, even for low doses of aspirin, with even higher (and dose dependent) risks for non-aspirin NSAIDs. Odds ratios calculated without reference to H. pylori, which represent overall population risks, were about half these values in our study, as in previous studies.2–13
Our data, showing an association between ulcer bleeding and H. pylori that was significant only where there was serological evidence of infection with a CagA positive strain, is in line with previous endoscopic studies where the dominant ulcer association is with virulent strains.34 It is unclear whether CagA negative H. pylori strains are entirely nonpathogenic in endoscopic studies.35 In our study, any associations between CagA negative H. pylori and ulcer bleeding were not statistically significant, although we cannot altogether rule out the possibility of some association. The mechanism of enhanced pathogenicity associated with CagA positive strains is incompletely understood.13 CagA production is an important marker for the presence of a collection of genes comprising the cag pathogenicity island, many of which are important for inducing inflammation.13 Recent evidence also suggests that CagA itself may be important in signalling host epithelial changes.36
A major stimulus for our study was our previous observation that there was no overall enhancement of risk in NSAID users who were H. pylori positive.12 For the present study, we hypothesized that virulent CagA positive strains would synergise with NSAID use to enhance risk, whilst less virulent, possibly symbiotic CagA negative strains would have a predominantly benign influence via inflammation-induced enhanced prostaglandin synthesis.18 This hypothesis turned out to be wrong, since NANSAID risks were reduced rather than enhanced in patients with CagA positive H. pylori infection.
This is now one of seven studies that have reported on the interaction between NSAIDs and H. pylori presenting with bleeding peptic ulcer.12, 37–41 One of these studies suggested that, for patients using NSAIDs, risk is increased with H. pylori infection,37 three have shown no effect12, 38, 39 and three, including the present study, have suggested that risks may be reduced in some circumstances.40, 41 Whilst these may be chance variations around a null effect, there are grounds to believe that H. pylori could have the protective effect in NANSAID, at least in some circumstances, that our study implies. It has been the only study specifically designed to investigate the interaction between H. pylori and NSAIDs in an unselected patient population and to relate findings to the CagA virulence marker. In addition, other studies have either not distinguished between aspirin and non-aspirin NSAIDs or have amalgamated data to achieve statistical significance.12, 37–41 It is possible that host or bacterial factors may vary from one population to another, so that H. pylori could have a net protective effect under some circumstances and a net harmful effect under others. In our study, only 8% of cases (vs. 3% of controls) were both CagA positive and using non-aspirin NSAIDs. Taken with the results of other studies, it is clear that such a combination is rarely associated with ulcer bleeding and that the potential for any reduction in the risk of ulcer bleeding from H. pylori eradication in NSAID users is small or non-existent. The results of studies that tested this strategy directly favour this interpretation for existing NSAIDs users,42–44 although other studies raise the possibility that H. pylori eradication prior to NSAID use may offer protection, at least in the short term.45–47
As use of low dose aspirin for cardiovascular prophylaxis becomes more widespread, the results of our study highlight its growing importance as a potential cause of ulcer complications. This was first pointed out by Langman and his colleagues.23, 24 Our estimates of risk with aspirin ≤ 300 mg are of a very similar order of magnitude to those which they reported23, 24 when no account is taken of the influence of H. pylori, but rise to around sixfold when its effects are included in analysis. In future, it will be important to disentangle risks due to aspirin from those attributable to cardiovascular disease for which it is used26, 48, 49 and to determine the relative contribution of the ulcerogenic and anti-haemostatic actions of aspirin to this risk.49 Recent data indicate that use of COX-2 inhibitors can substantially reduce the risk of ulcer complications50, 51 compared to nonselective NSAIDs, but that this advantage may be reduced or lost in patients that take low dose aspirin.50 It is ironic, just as new such safer non-aspirin NSAIDs are becoming available, that aspirin has re-emerged as a major factor associated with peptic ulcer bleeding. Identifying optimal strategies to reduce risks associated with low dose aspirin has arguably become the most pressing issue in this area.49