Treatment with non-steroidal anti-inflammatory drugs (NSAIDs) is known to be complicated by gastrointestinal toxicity. NSAIDs impair prostaglandin-dependent gastric mucosal protective mechanisms. When these defences have been breached, a second wave of injury caused by luminal gastric acid may facilitate deeper ulceration 1. Prevention of gastroduodenal ulcers attributable to the use of NSAIDs may target the inhibition of gastric acid secretion with histamine-2 receptor antagonists (H2RAs) or proton-pump inhibitors (PPIs). Alternatively, locally depleted endogenous cytoprotective prostaglandins may be replaced by the administration of prostaglandin E₁ analogues, such as misoprostol. Several studies have evaluated and compared these strategies 2. High-dose misoprostol is effective in the primary prevention of endoscopic NSAID ulcers and also NSAID ulcer complications, such as bleeding and perforation, but is often poorly tolerated because of diarrhoea and abdominal discomfort 3. Elevation of the intragastric pH by PPIs and high-dose H2RAs reduces the risk of endoscopic NSAID ulcers 2. In direct comparison, PPIs show an efficacy comparable to that of misoprostol, but they are better tolerated 4. Furthermore, PPIs are more effective in the prevention of NSAID ulcers than low-dose H2RAs 5. However, the efficacy of PPIs and H2RAs in the primary prevention of clinically relevant endpoints, such as bleeding and perforated NSAID ulcers, remains unproven.

The discovery of the isoenzymes cyclooxygenase (COX)-1 and COX-2 made it possible to develop highly selective COX-2 inhibitors 6. The hypothesis is that COX-1 is expressed constitutively and regulates normal physiology, such as the maintenance of gastric mucosal integrity. Conversely, COX-2 is expressed selectively after exposure to inflammatory mediators or trauma, and has a role in inflammation and pain 7. In randomised controlled clinical trials, selective COX-2 inhibitors have demonstrated a decreased risk for NSAID ulcers and also ulcer complications 891011. Furthermore, in elderly patients with a recent history of bleeding NSAID ulcers, secondary prevention (preventing recurrent bleeding) with a selective COX-2 inhibitor seems comparable to combining a non-selective NSAID with a PPI, although in that study the number of cases was small and neither strategy provided adequate protection 12.

Because of their relatively low incidence, severe gastrointestinal ulcer complications such as bleeding and perforated ulcers can be evaluated most effectively in large observational studies 13. Randomised controlled clinical trials are designed to evaluate the efficacy of a certain strategy, and despite including thousands of patients they may fail to detect infrequent or long-term complications or side effects. Furthermore, rigorous inclusion and exclusion criteria are maintained, and those at high risk for drug side effects or complications are usually excluded. Conversely, in daily clinical practice, it is especially at-risk patients who are likely to be treated with these new strategies under the assumption of safe, evidence-based pharmacotherapy. Although observational studies are subject to possible bias, they best reflect daily clinical practice and are well suited to study infrequent and long-term complications and side effects. Therefore, to determine the characteristics of patients who are especially at risk for serious NSAID ulcer complications and to compare the effectiveness of different preventive strategies in daily clinical practice, we conducted a large nested case-control study.

This nested case-control study was performed within the government-initiated healthcare region of the city of Enschede in The Netherlands. On 31 December 2003 the population consisted of 152,989 persons living in a well-defined geographically isolated area largely bordering on Germany. All in-patient healthcare is provided by a single teaching hospital, supplied with all diagnostic and therapeutic facilities. All drug prescriptions are registered in electronic prescription records of 14 local pharmacies. Most drugs, including NSAIDs, are provided by the patient's own pharmacy, directly reimbursed by the healthcare system. A cohort of NSAID users can be identified continuously from the electronic prescription records.

Serious NSAID ulcer complications were defined as ulcerations of the stomach or proximal duodenum causing perforation, obstruction or bleeding that occurred during the use of NSAIDs, necessitating hospitalisation of the patient.

Over the 26-month prospective observational period the cohort of NSAID users contained 51,903 NSAID users with 10,402 patient years of NSAID exposure. From this cohort, 104 cases were hospitalised with serious NSAID ulcer complications. Because of the geographically isolated position, referral to other hospitals, especially for acute gastrointestinal events, is extremely rare. Therefore, in this population the incidence of hospitalisation due to serious NSAID ulcer complications can be reliably calculated at 1% per year of NSAID use.

Table 1 shows demographic characteristics and co-morbidities. The typical case is an elderly patient, age at diagnosis 70.4 ± 16.7 years (mean ± SD; range 22 to 98 years), 55.8% were female. Many patients reported concurrent disease or previous medical events suggesting serious, especially cardiovascular, co-morbidity. This self-reported co-morbidity was supported by the concomitant medication used (Table 2). The 104 cases together used 12 different NSAIDs (Table 2). The duration of NSAID use before the gastrointestinal event varied; the median was 1.13 months (interquartile range 10 days to 12 months). Most patients did not exceed their prescribed maximum daily dose. However, occasional use of more than one NSAID simultaneously was reported by 12 patients (11.5%).

In most cases (80 patients, 76.9%), serious NSAID ulcer complication was the reason for presentation and hospitalisation. In the remainder a serious NSAID ulcer complication took place during hospitalisation for another reason. Characteristics of the gastrointestinal events are presented in Table 3. No diagnostic procedure was performed in only six (5.8%) patients, because of co-morbidity or advanced age. The mean haemoglobin level at presentation was 6.1 ± 1.9 mmol/l (mean ± SD; range 1.8 to 9.8). In those using coumarin, the international normalized ratio (INR) at presentation was 4.87 ± 1.41 (mean ± SD) but the mean haemoglobin level at presentation did not differ from that in patients not taking coumarin, and neither did the number of units of blood administered during hospitalisation.

Mortality due to serious NSAID ulcer complications was high: 11 patients (10.6%) died in hospital, and another 4 (3.8%) died within 3 months of the diagnosis. The incidence of in-hospital mortality due to serious NSAID ulcer complications can be calculated at 21.2 per 100,000 NSAID users.

For 104 cases, 757 controls were selected from the remaining cohort of NSAID users. On receiving the first questionnaire 225 controls responded, of whom 203 were included. On receiving a second questionnaire, a further 123 responded, of whom 81 were included. From the 64 excluded responders, 18 questionnaires were returned by someone other than the selected control, 15 denied taking NSAIDs, 17 refused, 1 had been hospitalised in a psychiatric hospital, 1 was a case who had already been included as such, and for 12 controls relatives informed us that the selected person had died. In the group of 20 randomly selected non-responders who were telephoned, no bias for non-responding was found.

In total 284 controls, frequency matched for age and sex, with NSAID use on the index date were included. Demographic characteristics, co-morbidities and current medication use are summarised in Tables 1 and 2. The mean age was slightly lower for the controls than for the cases because insufficient numbers of controls could be found for some of the extremely elderly cases.

In this nested case-control study, the concomitant use of proton-pump inhibitors was associated with a two-thirds reduction in the risk for serious NSAID ulcer complications. The efficacy of PPIs in the primary prevention of NSAID-associated gastropathy has so far only been proven for subjective symptoms and surrogate endpoints, such as dyspepsia and endoscopic ulcers, and in the secondary prevention of serious NSAID ulcer complications, PPIs do not seem to prevent recurrence 121415. Our data suggest that PPIs may be effective in the primary prevention of clinically relevant bleeding and perforated NSAID ulcers, confirming other recent observational studies 161718. However, randomised controlled trials powered on these hard endpoints need to be conducted to prove efficacy.

It is noteworthy that in this study the use of selective COX-2 inhibitors was not associated with protection for serious NSAID ulcer complications. Lack of protection from selective COX-2 inhibitors could not be explained by confounders such as concomitant use of aspirin, coumarin, heparin or steroids or by ulcer history. Previous studies demonstrating the efficacy of selective COX-2 inhibitors in the primary prevention of NSAID ulcer complications largely excluded high-risk patients, whereas in high-risk patients selective COX-2 inhibitors may fail to prevent the recurrence of NSAID ulcer bleeding 121415. Although neither selective COX-2 inhibitors nor concomitant PPIs seem to be entirely effective in preventing the recurrence of ulcer complications, our data suggest that PPIs may be superior to selective COX-2 inhibitors in the primary prevention of NSAID ulcer complications.

Cases used coumarin more often than controls (adjusted odds ratio 2.09; 95% confidence interval 0.93 to 4.70; p = 0.075). Furthermore, in those cases using coumarin, the mean INR at presentation was 4.87 ± 1.41 (mean ± SD) and one-third (5 patients) had an INR greater than 6.5. Although no INR was measured in the controls, it is possible that this elevated INR contributed to these patients developing serious NSAID ulcer bleeding.

Cases used low-molecular-mass heparin significantly more often than controls (adjusted odds ratio 17.33; 95% confidence interval 3.71 to 80.95; p < 0.001). In addition, cases used acetaminophen significantly more often than controls (adjusted odds ratio 2.80; 95% confidence interval 1.64 to 4.79; p < 0.001). It is possible that these differences reflect in-hospital treatment protocols rather than a true elevated risk for serious NSAID ulcer complications. However, an increased risk for NSAID ulcers with concomitant high-dose acetaminophen has been reported previously 13. Exclusion of patients with in-hospital NSAID ulcer complications truncated the odds ratio for low-molecular-mass heparin (adjusted odds ratio 6.06; 95% confidence interval 0.91 to 40.60; p = 0.06; Table 5).

Cases reported a history of heart failure significantly more often than controls (adjusted odds ratio 2.44; 95% confidence interval 1.28 to 4.66; p = 0.007). The association between heart failure and risk for NSAID-associated gastropathy has previously been demonstrated, but a credible causational mechanism remains to be identified 19.

One of the strengths of this study is that it reflects daily clinical practice. The large randomised controlled clinical trials that demonstrated the efficacy and safety of selective COX-2 inhibitors were conducted in relatively young, healthy subjects. Our study suggests that these may not be the patients who are especially at risk for serious NSAID ulcer complications and confirms another recently conducted large nested case-control study that also found no evidence for enhanced gastrointestinal safety with selective COX-2 inhibitors 20. Another strength of our study lies in the robustness of the data. Gastrointestinal events in cases and controls were verified, as were data on actual medication used. Other groups have studied populations of up to several thousand patients, but associations were derived by coupling databases and the validity of the data was not always verified 2122.

The local infrastructure makes it unlikely that many cases were missed. However, one weakness of this study is that underestimation of the number of events might still have occurred. Another weakness of this study, as in any case-control study, is the possibility of selection bias. Although we have controlled for all known possible confounders, selection by indication or an unknown confounding mechanism cannot be excluded with certainty.

Serious NSAID ulcer complications have a significant mortality rate: 10.6% die in hospital and 14.4% within 3 months of the event. At risk are especially elderly patients with cardiovascular co-morbidity. In daily clinical practice, the concomitant use of PPIs is associated with a two-thirds reduction in the risk for serious NSAID ulcer complications.

COX = cyclooxygenase; H2RAs = histamine-2 receptor antagonists; INR = international normalized ratio; NSAIDs = non-steroidal anti-inflammatory drugs; PPIs = proton-pump inhibitors.

The authors declare that they have no competing interests.

All authors contributed significantly to writing the article. HEV and RWF also contributed to the selection and inclusion of cases and controls. JvdP also contributed to the statistical analysis of the data. All authors read and approved the final manuscript.