Upper gastrointestinal complications (UGIC) are the major risk associated with aspirin use [1,2,3,4]. As with non-aspirin non-steroidal antiinflammatory drugs (NSAIDs), dose appears to be a major determinant, but even with low (300/325 mg) or very low doses (75 mg) a residual amount of risk seems to persist [3,4].

It has been postulated that enteric-coated formulations of aspirin, designed to resist the disintegration in acid environment and pass by the stomach without dissolution, may have an impact in reducing the risk [ 5]. The idea behind this strategy is grounded on the belief that gastric damage caused by aspirin is mainly due to a local effect [ 6]. Results from several endoscopic studies carried out in healthy volunteers supported this hypothesis showing that enteric-coated aspirin caused less gastric erosion and microbleeding than regular formulations [7,8,9]. However, it is known that these lesions are not good predictors of major upper gastrointestinal complications. From an epidemiological point of view the question is a matter of controversy as well. Two case-control studies have reported data on the effect of enteric-coated preparations on upper gastrointestinal bleeding, yielding to apparently opposite results. Weil et al [3] concluded that "enteric-coated aspirin ... may be free of risk" after estimating an odds ratio of 1.6 but with a wide 95% confidence interval (0.5-4.9). On the other hand, Kelly et al [ 4] found a risk of 2.6 (95% confidence interval, 1.4-5.3) which was essentially similar to the one estimated for plain aspirin when the effect of dose was taken into account.

In order to shed some light onto this issue and to estimate the specific aspirin-related risk of upper gastrointestinal bleeding and perforation, its relation with dose and duration of treatment, as well as the potential interaction between aspirin and other drugs (particularly NSAIDs), we used data from a case-control study designed to estimate risks of UGIC associated with drugs and other factors [10].

A case-control study was carried out using data from the UK-based General Practice Research Database. This database has been described extensively elsewhere [11]. Briefly, it contains details of patients' demographics, medical diagnosis, referrals to consultants and hospitals, and prescriptions. The accuracy and completeness of these data have been validated in previous studies [12].

We identified 1,833 cases of upper gastrointestinal bleeding and 272 cases of perforation. Cases and controls did not differ by age group and sex (table 1). Among bleeding cases, the lesion was located at the stomach in 764 (42%) and the duodenum in 811 (44%), remaining unspecified in 258 (14%); 4.1% of bleeding cases were fatal. Among perforation cases, 36 (13%) were located in the stomach and 236 (87%) in the duodenum with a case-fatality rate of 21.0 %. Two hundred eighty seven cases (13.6%) of upper gastrointestinal bleeding (n= 248; 13.5%) or perforation (n= 39; 14.3%) were current users of aspirin, as compared to 837 current users among controls (7.3%) yielding to an adjusted RR of 2.0 (95% CI: 1.7-2.3) for all cases (table 2). No material difference was found when upper gastrointestinal bleeding (2.0, 1.7-2.4) and perforation (1.7, 1.2-2.6) were considered as separate outcomes, or when location was taken into account (gastric 2.0, 1.6-2.5; duodenal 1.7, 1.4-2.1). The risk was essentially the same either for fatal (1.8, 1.1-2.9) or nonfatal (2.0, 1.7-2.4) cases. No modification of the relative risk estimates was observed by sex (males, 2.0 (1.6-2.4); females, 2.0 (1.5-2.6). The relative risks by age group were as follows: 40-59 years, 2.4 (1.3-4.4); 60-69 years, 2.3 (1.8-3.1); 70-79 years, 1.8 (1.4-2.2).

The indications for aspirin among a random sample of 100 cases were secondary prevention of coronary heart disease (47), secondary prevention of cerebrovascular ischaemic events (36), peripheral vascular disease (5), other cardiovascular disease (2), and analgesia (2). For the remainder (8) the indication was not unequivocally identified from the patient profiles. The distribution of indications in a random sample of 100 controls was 50, 30, 5, 3, 2 and 10, respectively.

Seventy-five mg was the daily dose most frequently prescribed (48%) in controls, followed by 150 mg (27%), 300/325 mg (19%), and more than 325 mg (6%). No dose-effect relation was observed with aspirin within the 75-300 mg range according to the instructions written by the general practitioner (table 3). This remained true when the analysis was restricted to a specific lesion (either bleeding or perforation) or to a specific location (either gastric or duodenal) (see additional material: Appendix 1). Results did not materially change when only patients without any antecedents of gastrointestinal disorders (including dyspepsia) were considered (see additional material: Appendix 2), or when we controlled for use of antiulcer drugs. For 149 patients (45 (16%) cases and 104 (12%) controls) among current users the average daily dose estimated through the number of tablets supplied exceeded the cut-off values established to separate different dose categories (112, 225 and 400 mg). The replacement of the "instructed dose" by the average dose did not make a significant change (table 3). Nor did the restriction of the analysis to patients with consistent data for both methods of aspirin dose assessment.

Enteric-coated aspirin was associated with a risk of UGIC (RR of 2.3, 1.6-3.2) similar to the one of plain aspirin (1.9, 1.6-2.3). Estimates hardly changed when analysing separately bleeding and perforation, or gastric and duodenal sites (table 4). As enteric-coated aspirin could have been prescribed preferentially for those patients at greater risk of gastrointestinal damage, we performed a sub-analysis including only individuals without any antecedent of upper gastrointestinal disorder: the relative risk estimate for enteric-coated aspirin was 2.7 (1.7-4.2) and the one for plain aspirin was 2.2 (1.8-2.8). The results did not change when additional adjustment for antiulcer drugs use was performed. Estimates of risk associated with aspirin formulation and stratified by dose are shown in table 5. We observed a dose-response starting at 300 mg with plain aspirin but could not find a similar pattern with enteric-coated.

As shown in table 6, the duration of treatment had an impact upon the risk of upper gastrointestinal complication, being the first two months of use the period of greater risk (4.5, 2.9-7.1). Such a pattern was observed across different doses (table 7) or formulations of aspirin (see additional material: Appendix 3), though it was less sharp within the 75 mg dose and for enteric-coated preparations. The indication for aspirin in those patients with shorter treatment duration (less than 2 months) was related to cardiovascular disorders in 88% of cases and 89 % of controls. Among short-term users, those who used aspirin for the first time appeared to be at the same relative risk as those with a previous non-consecutive prescription of aspirin ("distant use") (table 6). The initial increased risk was clearly observed for bleeding lesions and less markedly for perforation (see additional material: Appendix 4).

The concomitant use of aspirin together with other non-aspirin NSAIDs at high dose increased substantially the risk of major UGIC (relative risk 13.3, 8.5-20.9), far beyond the sum of their respective independent effects (RERI=7.9, 2.0-13.8), while no interaction was apparent when low-dose aspirin was taken together with NSAIDs at low-medium dose (RR of 2.2, 1.0-4.6) (table 8). The interaction appeared to be greater for perforation (RR=30.3, 15.0-61.3; RERI=23.1, 2.2-44.2) than for bleeding (RR=10.8, 6.6-17.5; RERI= 5.7, 0.5-10.9). Among bleeding lesions no difference was found between the relative risk of the combination for those located at the stomach (RR=12.2, 6.8-21.9) and those located at the duodenum (14.3, 8.3-24.5). Both formulations of aspirin presented this interaction with high-dose NSAIDs (plain, 12.7 (7.7-20.8); enteric-coated, 16.6 (5.9-46.3). No relevant interaction beyond the sum of the independent effects was observed between aspirin and high-dose paracetamol (4.5, 1.9-10.9) or aspirin and steroids (1.9, 0.9-3.8). A small interaction was observed between low-dose aspirin and anticoagulants (5.5, 2.4-12.7).

The results of the present study show that the use of low-dose aspirin for prophylaxis of cardiovascular disorders conveys a twofold increased risk of upper gastrointestinal bleeding and perforation in the general population. This increased risk is found over different categories of age and sex. The effect of aspirin applies equally to gastric and duodenal sites.

We did not find a clear relation with aspirin daily dose within the range 75-300 mg. Such a result did not change when we restricted the analysis to patients without any antecedent of upper gastrointestinal disorder or when we adjusted for use of antiulcer drugs. This reasonably excludes a potential selection of patients with a greater risk of UGIC to lower doses of aspirin. Looking for other sources of bias, we identified that in 13% of patients the dose instructed by the general practitioner was not consistent with the average daily dose calculated by counting the number of tablets supplied. We tried to correct this by restricting the analysis to the population having consistent data with both methods and by replacing the instructed dose by the average. Neither of those corrections yielded different results. Nevertheless, a suggestion of a greater risk associated with aspirin at higher doses within the range 75-300 mg versus lower doses was found among users of the plain formulation and among short-term users (first 60 days of treatment). Finally, although the confidence intervals for lower doses are reasonably narrow, some statistical uncertainty remains and a dose-effect cannot be excluded; therefore, the use of the lowest dose shown effective is still the best strategy to minimise risk.

The enteric-coated formulations appear not to protect from the aspirin-induced risk of upper gastrointestinal bleeding, irrespective of the dose, neither for gastric nor for duodenal lesions. This observation adds up to previous results from other investigators [4]. Assuming that the coating of aspirin effectively spares the stomach from its topical adverse effect, such results cast doubts on the overall contribution of the local effect in serious bleeding lesions observed among aspirin users. Instead, our findings support the hypothesis that the effect of aspirin, both plain and enteric-coated, on upper gastrointestinal bleeding is mainly systemic mediated through inhibition of platelet thromboxane A₂ synthesis [19] and/or inhibition of protective prostaglandin synthesis in the gastroduodenal mucosa [20]. A local effect, however, cannot be ruled out as a relevant contributor for aspirin-induced perforation, as most cases were located in duodenum, and no difference or, perhaps, a slightly greater risk was found with the coated formulation. A preferential prescription of enteric-coated aspirin to patients with antecedents of gastrointestinal disorders does not explain our results, since enteric-coated aspirin users with no such antecedents present a risk similar to the one associated with plain aspirin.

Patients in the first two months after starting treatment with aspirin are at the greatest risk of presenting a major UGIC. Thereafter, the risk goes down and reaches a plateau at around six months. This phenomenon has been observed by other researchers for aspirin [3] but remains controversial for other NSAIDs [10, 15]. It has been suggested that this high hazard rate at the beginning of treatment could be an artefact induced by the use of aspirin for early symptoms of peptic ulcer. This possibility can be reasonably ruled out in our study because almost 90% of cases who experienced bleeding or perforation early after starting treatment with aspirin were using it for prevention of cardiovascular disorders, a pattern of use identical to the one found for new users among controls. Another explanation argued for the decreasing risk with the continuation of treatment has been the "early withdrawal of population susceptible to damage" [3]. In order to explore this possibility we divided short-term users of aspirin (< 2 months) into "distant users" and "first-ever users", depending on the recording or not of a remote non-consecutive prescription for aspirin. We did not find any apparent difference in risk between these two groups. So, patients who stopped treatment with aspirin for whatever reason and then started again after a washout period of at least four months, presented a similar risk during the first two months of treatment as first-ever users. By exclusion, the duration response could be explained by the third mechanism postulated so far: the existence of an adaptation of the gastrointestinal mucosa to the toxic action of aspirin. According to our results, this adaptation process would take at least 2 months to result into clinical benefits. In striking contrast, a constant risk pattern has been found with non-aspirin NSAIDs [10,15]. The reasons for such discrepancy are unknown and should be further studied. In our view, the explanation should be searched in the pharmacological differences of aspirin as compared to non-aspirin NSAIDs. It is well known that aspirin has a much greater topical adverse effect on the stomach lining than non-aspirin NSAIDs [21] and perhaps the adaptive process concerns mainly this mechanism of action.

The concurrent use of low-dose aspirin with high-dose NSAIDs increases the risk of UGIC by a factor of 13. Similar findings have been obtained by others for UGIB [3]. Our results strongly suggest that the combination may have a particular impact on the risk of perforation (lower confidence interval of 15). It is important to note that the joint use of low-dose aspirin with high-dose NSAIDs is not unusual. In our random sample of 11,500 controls, the prevalence of this combination was 3 per 1,000 subjects, a small figure in relative terms but large from a public health perspective. On the other hand, it is reassuring that no interaction beyond the sum of their respective effects was observed between aspirin and low-medium NSAIDs, aspirin and steroids, or aspirin and paracetamol (even when this latter was taken at doses higher than 2 grams a day). Although we detected a slight interaction of low-dose aspirin and anticoagulants, the confidence interval of the estimate is wide and precludes any firm conclusion.

The present study has some limitations. Aspirin prescription is systematically recorded in the GPRD, but its over-the-counter use is only recorded occasionally and this could lead to a misclassification of aspirin exposure. Although misclassification of exposures collected prospectively before the occurrence of the event of interest is usually close to non-differential between cases and controls, we evaluated the extent of the under-recording and the type of misclassification using external data from a recent study that analyzed the risk of peptic ulcer bleeding associated with prophylactic aspirin and was also performed in the UK [3]. The authors interviewed patients aged 60 and over and asked, among other things, information about all prescribed and self-administered drug intake. They found that 27.1 % of cases and 15.9% of controls reported any use of aspirin in the month before (16.5% and 8.3% respectively reported a regular use). In our study population, the prevalence of aspirin use in the month before was 13.6% and 7.3% (18.4% and 10.3% among the subjects aged 60 years or older). Interestingly, the case-control ratio of aspirin exposure for the reference study is 1.70, while the corresponding ratio is 1.86 in our study (1.79 for the older group). These figures suggest that the under-recording of OTC aspirin use in our study is to a major extent non-differential with respect to case status. We performed a sensitivity analysis assuming the worst case of a 50% misclassification of aspirin users. The "true" crude OR would be 2.3 instead of the 2.1 observed. Therefore, it seems unlikely that the underascertainment of OTC aspirin use could have greatly distorted the estimate of aspirin effect in our study. Other authors have recently reached the same conclusion for OTC NSAIDs [22]. It is conceivable that the potential misclassification of enteric-coated preparations due to their higher price were less important than for the plain ones. This means that the distortion to the null would be weaker for the former preparations, but assuming that the misclassification was non-differential with respect to case status, the impact of such difference would be negligible. We also examined the effects of a differential misclassification with respect to case status and only under the unrealistic assumption of a strong differential misclassification selectively affecting the higher doses within the range 75-300 mg of plain preparations, could our results of lack of dose-effect and lack of protective effect of enteric-coated preparations be explained.

We tried to control for other potential sources of bias, including channelling of patients with greater risk to lower-dose or enteric-coated aspirin. However, as with all observational studies, unmeasured or inaccurately measured factors may lead to residual confounding.

Aspirin used at doses as low as 75 mg is still associated with a moderate risk of developing serious upper gastrointestinal complications. The coating of the active principle in order to spare the stomach does not reduce the risk of upper gastrointestinal complications, neither for the stomach nor for the duodenum. The first two months of treatment seem to be the period of greater risk, regardless the patient is first-ever user or not. Patients using concomitantly low-dose aspirin and high-dose nonaspirin NSAIDs are a subgroup of patients with a major increased risk of upper gastrointestinal complications.

UGIB: Upper gastrointestinal bleeding; UGIC: Upper gastrointestinal complications; NSAIDs: Non-steroidal and anti-inflammatory drugs; RR; Relative risk; GPRD: General Practice Research Database.