COX-2-selective inhibitors and the risk of upper gastrointestinal bleeding in high-risk patients with previous gastrointestinal diseases: a population-based case–control study

Authors


Dr B. Nørgård, Department of Clinical Epidemiology, Aarhus University Hospital, Vennelyst Boulevard 6, DK-8000 Aarhus C, Denmark.
E-mail: bn@soci.au.dk

Summary

Background : Clinical trials have suggested that cyclo-oxygenase-2-selective inhibitors are associated with a lower risk of upper gastrointestinal bleeding than are non-selective, non-aspirin, non-steroidal anti-inflammatory drugs. This has not yet been confirmed in studies of patients with an increased susceptibility to upper gastrointestinal bleeding.

Aim : To examine the risk of upper gastrointestinal bleeding in high-risk patients who filled prescriptions for cyclo-oxygenase-2 inhibitors or other non-steroidal anti-inflammatory drugs.

Methods : A population-based case–control study was performed in the Danish county of North Jutland from 1 January 2000 to 31 December 2002. From the County Hospital Discharge Registry and the Civil Registration System, we identified incident cases with upper gastrointestinal bleeding (n = 780) and randomly selected controls (n = 2906), respectively. All cases and controls had previous gastrointestinal diseases. Data on drug exposure were obtained from the countywide Prescription Database.

Results : Thirty-five cases (4.5%) filled prescriptionss for cyclo-oxygenase-2 inhibitors within 30 days of the date of upper gastrointestinal bleeding, compared with 79 controls (2.7%). Adjusted odds ratios for upper gastrointestinal bleeding according to prescription for celecoxib, rofecoxib and non-steroidal anti-inflammatory drugs were 1.3 [95% confidence interval (CI), 0.7–2.8], 2.1 (95% CI, 1.2–3.5) and 3.3 (95% CI, 2.4–4.4), respectively.

Conclusions : In patients with increased susceptibility to gastrointestinal adverse events, a lower risk of upper gastrointestinal bleeding was observed in users of cyclo-oxygenase-2 inhibitors compared with users of other non-aspirin, non-steroidal anti-inflammatory drugs.

Introduction

There is a well-documented association between the use of non-aspirin, non-steroidal anti-inflammatory drugs [NSAIDs, inhibitors of both cyclo-oxygenase-1 (COX-1) and cyclo-oxygenase-2 (COX-2) isoforms] and gastro-duodenal injury, ranging from dyspepsia to fatal upper gastrointestinal tract bleeding and perforation.1–5 There is substantial evidence that the pathophysiological mechanism for the gastrointestinal effects of NSAIDs is due to the inhibition of cytoprotective mucosal prostaglandin production, primarily mediated by COX-1 activity.3, 6–10 Therefore, with the design of drugs with selective affinity for COX-2, an improved gastric tolerability might be expected.3, 6, 7 Two COX-2-selective inhibitors are available in Denmark: rofecoxib since November 1999 and celecoxib since May 2000.

Data from pre- and early post-marketing trials on patients with rheumatological diseases — the largest of which are the Vioxx Gastrointestinal Outcomes Research (VIGOR) trial1 and the Celecoxib Long-Term Arthritis Safety Study (CLASS) trial11— have suggested that COX-2-selective inhibitors are associated with fewer clinically symptomatic ulcers and ulcer complications and less faecal blood loss than are non-selective NSAIDs.1, 8, 11–18 Although the CLASS trial has become the subject of criticism on the grounds of study design, data analysis and selective presentation of the results,19–23 data from these randomized studies support the hypothesis of a lower risk of gastrointestinal problems in users of COX-2-selective inhibitors compared with users of non-selective NSAIDs.16, 17

To different extents, the two largest trials have excluded patients who might have a particularly increased risk of upper gastrointestinal bleeding, such as those with active gastrointestinal disorders11 and recent oesophageal or gastro-duodenal ulcers,11 and users of anticoagulants, proton pump inhibitors and H2-receptor antagonists in prescription-strength doses.1 However, both studies have included patients with a history of gastrointestinal events and elderly patients. Comprehensive exclusion criteria in clinical trials might limit the generalization of the randomized trial results to community medical practice and to patients who might be at high risk of upper gastrointestinal bleeding.

The only observational study to date has reported that the short-term risk of hospitalization for upper gastrointestinal bleeding is lower in users of COX-2-selective inhibitors than in users of non-selective NSAIDs amongst the population in Ontario, Canada, aged ≥ 66 years. In that study, celecoxib was associated with a lower risk of bleeding than rofecoxib.24

Thus, so far, no studies have specifically examined the safety of COX-2-selective drugs in patients with previous upper gastrointestinal disorders. Therefore, in a population-based case–control study in Denmark, we examined the risk of first-time upper gastrointestinal bleeding in high-risk patients who filled prescriptionss for COX-2-selective inhibitors or non-selective, non-aspirin NSAIDs, taking other known risk factors, such as a history of non-bleeding gastrointestinal events, age, gender and co-morbidity index, into account.

Methods

This population-based case–control study was conducted in the Danish county of North Jutland (about 493 000 inhabitants, approximately 9% of the total Danish population), where incident cases of upper gastrointestinal bleeding were identified during the period from 1 January 2000 to 31 December 2002. The National Health Service provides tax-supported health care for all inhabitants, allowing free access to general practitioners and hospitals, and refunds a variable proportion of the costs of prescribed drugs. Using the unique 10-digit civil registry number assigned to all Danish citizens shortly after birth, a complete hospital discharge and prescription history can be established for each individual, and valid linkage between population-based registries can be performed.

Identification of cases with upper gastrointestinal bleeding

The County Hospital Discharge Registry of North Jutland County contains information on all patients discharged from somatic hospitals in the county since 1977, and includes patients from out-patient clinics since 1994.25 Data in this registry include dates of admission and discharge, surgical procedures performed and up to 20 discharge diagnoses, classified according to the Danish version of the International Classification of Diseases, 8th revision (ICD-8), until the end of 1993 and ICD-10 thereafter.26, 27 (ICD-9 has never been used in Denmark.)

Based on the hospital discharge history from 1977, we included only first incident cases of upper gastrointestinal bleeding, i.e. patients who were hospitalized for the first time with a discharge diagnosis of upper gastrointestinal bleeding, between 1 January 2000 and 31 December 2002. Upper gastrointestinal bleeding diagnostic codes from ICD-10 included: bleeding from gastritis (K29.0); bleeding from ulcers in the stomach (K25.0, K25.2, K25.4, K25.6); bleeding from ulcers in the duodenum (K26.0, K26.2, K26.4, K26.6); bleeding from gastro-duodenal ulcers (K27.0, K27.2, K27.4, K27.6); bleeding from anastomotic ulcers, i.e. gastro-jejunal ulcers (K28.0, K28.2, K28.4, K28.6); haematemesis (K92.0); melaena (K92.1); and gastrointestinal haemorrhage without specification (K92.2).

We excluded patients who had been hospitalized for any of the above-mentioned bleeding conditions prior to the study period (n = 259), based on the hospital discharge history for each individual (using ICD-8 and ICD-10 codes) beginning in 1977. From the remaining total (n = 1891), we further excluded patients with an address outside the county and an age of less than 18 years or over 90 years (n = 177).

Within the total of 1714 eligible cases with a first incident episode of upper gastrointestinal bleeding, we identified four high-risk groups of patients with previous diseases in the gastrointestinal tract and diseases predisposing to gastrointestinal bleeding: (i) patients with a discharge history of non-bleeding ulcer before case status; (ii) patients with a discharge history of oesophagitis, gastritis, duodenitis or Mallory–Weiss lesions before case status; (iii) users of proton pump inhibitors and/or H2-antagonists within 2 years before case status; and (iv) a mixed group consisting of patients with a discharge history of alcoholism, chronic liver diseases (including cirrhosis) and oesophageal varices before case status. Patients could belong to more than one sub-group.

A total of 780 cases of upper gastrointestinal bleeding met the criteria to be in at least one of the four high-risk groups of patients.

Selection of controls

The Civil Registration System has maintained electronic records on all changes in vital status (including change of address, date of emigration and date of death) for the entire Danish population since 1968. From this registry, we randomly selected controls (individuals with no hospitalization for upper gastrointestinal bleeding during the study period) residing in North Jutland County. For the identification of controls, we utilized the same four categories as in the identification of cases, i.e. increased susceptibility to upper gastrointestinal bleeding based on past disease and prescription history. Thus, all controls, like all cases, had increased susceptibility to upper gastrointestinal bleeding based on their medical histories. As with cases, we excluded controls who had previous hospitalizations for a bleeding episode in the upper gastrointestinal tract. Controls were selected using the incidence density sampling technique,28 i.e. they had to be alive and at risk for a first episode of upper gastrointestinal bleeding at the time the corresponding case was diagnosed.

A total of 2906 high-risk controls were available for analyses, 1–8 controls per case, and a mean of four controls per case.

Data on drug use (COX-2-selective inhibitors and non-aspirin NSAIDs)

The population-based Pharmaco-Epidemiological Prescription Database of North Jutland, established in 1989, was used to identify all prescriptions filed for COX-2-selective inhibitors and non-selective, non-aspirin NSAIDs. The county is served by pharmacies equipped with computerized accounting systems through which data are sent to the Danish National Health Service and to the Prescription Database, with key information on prescriptions for refundable drugs. Thus, the database includes information on the patient's civil registry number, the type of drug prescribed, according to the Anatomical Therapeutical Chemical (ATC) classification system, and the date the prescription was filed.

The present study identified all prescriptions with the ATC codes M01A H02 (rofecoxib) and M01A H01 (celecoxib); both drugs are available only on prescription. For prescriptions of other non-aspirin NSAIDs, the ATC code M01A was used. Patients were classified as exposed if they obtained prescriptions for one or more COX-2-selective inhibitors or other non-aspirin NSAIDs within 30 days before case status or the index date for controls.

Data on possible confounders

Data on potential confounders were collected from the County Hospital Discharge Registry and the Pharmaco-Epidemiological Prescription Database. Information from the County Hospital Discharge Registry included previous hospitalizations (since 1977) for alcoholism (ICD-8 code, 303; ICD-10 code, F10), non-bleeding ulcer diagnosis (ICD-8 codes, 53100, 53101, 53108, 53109, 53191, 53193, 53194, 53196, 53198, 53199, 53209, 53291, 53299, 53309, 53391, 53399, 53409, 53491, 53499; ICD-10 codes, K25.1, K25.3, K25.5, K25.7, K25.9, K26.1, K26.3, K26.5, K26.7, K26.9, K27.1, K27.3, K27.5, K27.7, K27.9, K28.1, K28.3, K28.5, K28.7, K28.9) and for non-bleeding conditions from oesophagitis, gastritis, duodenitis or Mallory–Weiss lesions (ICD-8 codes, 530, 535; ICD-10 codes, K20.9, K21, K22, K23, K29.1–K29.9). Furthermore, information from this registry was used to assess co-morbidity using the validated Charlson index (a standard co-morbidity scale with chronic diseases graded for severity of disease, also including cancer diagnoses).29

Through the Pharmaco-Epidemiological Prescription Database, we obtained information on other possible confounders, including prescriptions for oral anticoagulants (B01A A03, B01A A04), low-dose aspirin (75–150 mg) (B01A C06), high-dose aspirin (100–500 mg) (N02B A01, N02B A51), glucocorticoids (oral or injection) (H02A B), selective serotonin re-uptake inhibitors (N06A B), organic nitrates (C01D), calcium antagonists (C08), proton pump inhibitors (A02B C05, A02B C03, A02B C01, A02B C02, A02B C04) and H2-antagonists (A02B A01, A02B A03, A02B A04, A02B A02, A02B A07).

Statistical analysis

Contingency tables were constructed for the main study variables and odds ratios (OR) with 95% confidence intervals (CI) for upper gastrointestinal bleeding were estimated. Unconditional logistic regression analyses were performed to estimate the relative risk for upper gastrointestinal bleeding amongst users of COX-2-selective inhibitors compared with non-users. By including the use of non-aspirin NSAIDs in the model, we were able to compare the risk of upper gastrointestinal bleeding amongst COX-2 vs. NSAID users. The model mutually adjusted for prescriptions of rofecoxib and/or celecoxib and/or non-aspirin NSAIDs occurring within the time window of 30 days. We took the following variables into consideration in the model: gender; age; history of alcoholism; oesophagitis; non-bleeding gastritis or duodenitis; Mallory–Weiss lesions; non-bleeding ulcer diagnosis; co-morbidity index (calculated as the sum of the weighted index for each subject included);29 and prescriptions obtained for oral anticoagulants, low-dose aspirin, high-dose aspirin, glucocorticoids, selective serotonin re-uptake inhibitors, proton pump inhibitors and H2-antagonists within 90 days of the index date. These factors were included in the models as dichotomous variables, except for age and co-morbidity index, which were included as continuous variables. In the model, the number of prescriptions for COX-2-selective inhibitors for each individual did not change our estimates, nor did prescriptions for organic nitrates or calcium antagonists. Therefore, these were excluded from the final model.

Analyses using the full model were conducted separately for three of the four groups of high-risk patients with previous gastrointestinal diseases: (i) discharge history of non-bleeding ulcer before case status or index date for controls; (ii) discharge history of oesophagitis, gastritis, duodenitis or Mallory–Weiss lesions before case status or index date for controls; and (iii) users of proton pump inhibitors and/or H2-antagonists within 2 years before case status or index date for controls. Due to the sparse data, the full model could not be fitted to estimate the risk according to the use of COX-2 inhibitors in the fourth high-risk group of patients. However, this group of patients was included in an analysis in which we combined all four high-risk sub-groups.

Finally, we stratified according to gender, and estimated the risk of upper gastrointestinal bleeding in elderly patients (≥ 65 years).

All analyses were performed using SAS version 8.02 (SAS Institute Inc., Cary, NC, USA). The study was approved by the Danish Registry Board (No. 2002-3311-0094).

Results

Table 1 shows the characteristics of the 780 cases (belonging to the four categories of high-risk patients) with first hospitalization for upper gastrointestinal bleeding and the 2906 controls. Thirty-five cases (4.5%) filled prescriptionss for COX-2-selective inhibitors within 30 days of the date of diagnosis of upper gastrointestinal bleeding [12 (1.5%) filled prescriptionss for celecoxib and 23 (3.0%) for rofecoxib], compared with 79 (2.7%) in the control group [33 (1.1%) filled prescriptionss for celecoxib and 46 (1.6%) for rofecoxib]. A higher proportion of cases than controls filled prescriptionss for other non-aspirin NSAIDs: 12.4% and 4.1%, respectively. Similarly, a higher proportion of cases than controls filled prescriptionss for drugs such as oral anticoagulants, selective serotonin re-uptake inhibitors, glucocorticoids, proton pump inhibitors and high- and low-dose aspirin.

Table 1.  Descriptive characteristics of cases with first upper gastrointestinal bleeding episodes and controls (from the North Jutland County, Denmark, 1 January 2000 to 31 December 2002)
  Cases (n = 780)Controls (n = 2906)
  • COX-2, cyclo-oxygenase-2; NSAID, non-steroidal anti-inflammatory drug; PPI, proton pump inhibitor.

  • *

     Prescription filed within 30 days of the date of hospital admission for upper gastrointestinal bleeding (or corresponding index date for controls).

  •  Prescription filed within 90 days of the date of hospital admission for upper gastrointestinal bleeding (or corresponding index date for controls).

  •  Based on information from the Hospital Discharge Registry from 1977.

  • §

     Cases and controls can belong to more than one sub-group.

Gender, n (%)
 Female335 (42.9)1363 (46.9)
 Male445 (57.1)1543 (53.1)
Age (years)
 Mean (s.e.)66.8 (15.9)72.5 (13.5)
 Range18–8918–89
Patients exposed to COX-2-selective inhibitors and other non-aspirin NSAIDs, n (%)*
 Celecoxib12 (1.5)33 (1.1)
 Rofecoxib23 (3.0)46 (1.6)
 Other NSAIDs97 (12.4)120 (4.1)
Prescription, n (%)
 Oral anticoagulants38 (4.9)81 (2.8)
 Selective serotonin re-uptake inhibitors98 (12.6)245 (8.4)
 Glucocorticoids63 (8.1)178 (6.1)
 High-dose aspirin85 (10.9)259 (8.9)
 Low-dose aspirin111 (14.2)335 (11.5)
 PPI286 (36.7)737 (25.4)
 H2-antagonists69 (8.9)259 (8.9)
Previous diagnosis of alcoholism, n (%)132 (16.9)200 (6.9)
Co-morbidity index, n (%)
 0–2591 (75.8)2495 (85.9)
 > 2189 (24.2)411 (14.1)
Groups of patients with previous gastrointestinal diseases, n (%)§
 (i) Non-bleeding ulcers214 (27.4)618 (21.3)
 (ii) Oesphagitis, gastritis, duodenitis and Mallory–Weiss lesions229 (29.4)893 (30.7)
 (iii) Users of PPI and/or H2-antagonists within 2 years before index date550 (70.5)2028 (69.8)
 (iv) Alcoholism, chronic liver disease and oesophageal varices168 (29.4)893 (30.7)

Table 2 shows the crude and adjusted ORs for upper gastrointestinal bleeding according to prescription for rofecoxib, celecoxib and other non-aspirin NSAIDs within 30 days of the date of hospital diagnosis for: (i) patients with a discharge history of non-bleeding ulcer before case status or index date for controls; (ii) patients with a discharge history of oesophagitis, gastritis, duodenitis or Mallory–Weiss lesions before case status or index date for controls; (iii) users of proton pump inhibitors and/or H2-antagonists within 2 years before case status or index date for controls; and (iv) patients with a discharge history of alcoholism, chronic liver diseases and oesophageal varices before case status. The risk of upper gastrointestinal bleeding according to prescription for rofecoxib, celecoxib and non-aspirin NSAIDs was rather uniform across these groups of high-risk patients. Compared with the risk of upper gastrointestinal bleeding after non-aspirin NSAID use (OR ranged from 2.5 to 4.7), the risk for COX-2-selective inhibitors was lower in all groups of patients, with the lowest risk for celecoxib in the first and third groups [OR = 0.9 (95% CI, 0.2–3.5) and OR = 1.3 (95% CI, 0.6–2.9), respectively] (Table 2).

Table 2.  Crude and adjusted odds ratios (ORs) for upper gastrointestinal bleeding (UGIB) episode according to prescription for rofecoxib, celecoxib and other non-aspirin, non-steroidal anti-inflammatory drugs (NSAIDs) within 30 days of the date of hospital admission in four high-risk groups of patients. Within each high-risk group, crude and adjusted ORs are given (adjusted for each other and for other possible confounders in the same model)
 Cases of UGIB
exposed/total (%)
Controls
exposed/total (%)
Crude OR
(95% CI)
Adjusted OR
(95% CI)
  1. CI, confidence interval; PPI, proton pump inhibitor.

  2. * Adjusted for discharge diagnoses of alcoholism, oesophagitis, gastritis, duodenitis or Mallory–Weiss lesions, co-morbidity index, age, sex and prescriptions for oral anticoagulants, low-dose aspirin, high-dose aspirin, glucocorticoids, selective serotonin re-uptake inhibitors, PPI or H2-antagonist filed within 90 days of the date of admission for upper gastrointestinal bleeding.

  3. † Adjusted for discharge diagnosis of alcoholism, co-morbidity index, age, sex, previous non-bleeding ulcer diagnosis before case status or index date for controls (0–2 years, > 2 years, none) and prescriptions for oral anticoagulants, low-dose aspirin, high-dose aspirin, glucocorticoids, selective serotonin re-uptake inhibitors, PPI or H2-antagonist filed within 90 days of the date of admission for upper gastrointestinal bleeding.

  4. ‡ Adjusted for discharge diagnoses of alcoholism, oesophagitis, gastritis, duodenitis or Mallory–Weiss lesions, co-morbidity index, age, sex, previous non-bleeding ulcer diagnosis before case status or index date for controls (0–2 years, > 2 years, none) and prescriptions for oral anticoagulants, low-dose aspirin, high-dose aspirin, glucocorticoids, selective serotonin re-uptake inhibitors, PPI or H2-antagonist filed within 90 days of the date of admission for upper gastrointestinal bleeding.

Cases and controls with previous non-bleeding ulcer diagnosis
 Rofecoxib8/214 (3.7)9/618 (1.5)2.8 (1.1–7.3)2.6 (0.9–7.2)*
 Celecoxib3/214 (1.4)9/618 (1.5)0.9 (0.2–3.5)0.9 (0.2–3.5)*
 Other NSAIDs18/214 (8.4)19/618 (3.1)3.0 (1.5–5.8)3.6 (1.8–7.3)*
Cases and controls with previous diagnosis of oesophagitis, gastritis, duodenitis or Mallory–Weiss lesions
 Rofecoxib4/229 (1.8)14/893 (1.6)1.1 (0.3–3.4)1.9 (0.6–6.2)
 Celecoxib6/229 (2.6)12/893 (1.3)1.8 (0.7–5.0)2.1 (0.7–6.7)
 Other NSAIDs28/229 (12.2)28/893 (3.1)4.3 (2.5–7.4)4.7 (2.6–8.6)
Users of PPI and/or H2-antagonists within 2 years before index date
 Rofecoxib18/550 (3.3)39/2028 (1.9)1.9 (1.1–3.3)1.8 (1.0–3.2)
 Celecoxib11/550 (2.0)30/2028 (1.5)1.3 (0.6–2.6)1.3 (0.6–2.9)
 Other NSAIDs69/550 (12.6)88/2028 (4.3)3.2 (2.3–4.5)3.1 (2.2–4.4)
Cases and controls with previous diagnosis of alcoholism, chronic liver diseases and oesophageal varices
 Rofecoxib4/168 (2.4)0/266 (0.0)
 Celecoxib1/168 (0.6)0/266 (0.0)
 Other NSAIDs16/168 (9.5)12/266 (4.5)2.2 (1.0–4.8)2.5 (1.1–5.9)

Table 3 shows the crude and adjusted ORs for upper gastrointestinal bleeding according to prescription for rofecoxib, celecoxib and other non-aspirin NSAIDs within 30 days of diagnosis in all four high-risk groups combined. The non-aspirin NSAID-associated adjusted risk of upper gastrointestinal bleeding was OR = 3.3 (95% CI, 2.4–4.4), in contrast with 1.3 (95% CI, 0.7–2.8) and 2.1 (95% CI, 1.2–3.5) for celecoxib and rofecoxib, respectively.

Table 3.  Crude and adjusted (for each other and for other possible confounders in the same model) odds ratios (ORs) for upper gastrointestinal bleeding (UGIB) episode according to prescription for rofecoxib, celecoxib and other non-aspirin, non-steroidal anti-inflammatory drugs (NSAIDs) within 30 days of the date of hospital admission in four high-risk groups of patients with previous gastrointestinal diseases
 Cases of UGIB
exposed/total (%)
Controls
exposed/total (%)
Crude OR
(95% CI)
OR*
(95% CI)
  1. CI, confidence interval.

  2. * Adjusted for discharge diagnoses of alcoholism, oesophagitis, gastritis, duodenitis and Mallory–Weiss lesions, co-morbidity index, age, sex, previous non-bleeding ulcer diagnosis before case status or index date for controls (0–2 years, > 2 years, none) and prescriptions for oral anticoagulants, low-dose aspirin, high-dose aspirin, glucocorticoids, selective serotonin re-uptake inhibitors, proton pump inhibitors or H2-antagonist filed within 90 days of the date of admission for upper gastrointestinal bleeding.

Rofecoxib23/780 (3.0)46/2906 (1.6)2.0 (1.2–3.4)2.1 (1.2–3.5)
Celecoxib12/780 (1.5)33/2906 (1.1)1.3 (0.7–2.5)1.3 (0.7–2.8)
Other NSAIDs97/780 (12.4)120/2906 (4.1)3.5 (2.5–4.4)3.3 (2.4–4.4)

Based on all four groups of high-risk patients, stratified analyses according to gender showed, in men, adjusted ORs of 1.4 (95% CI, 0.5–4.0), 2.1 (95% CI, 1.0–4.6) and 3.0 (95% CI, 2.0–4.6) for celecoxib, rofecoxib and non-aspirin NSAIDs, respectively. In women, the adjusted ORs were 1.4 (95% CI, 0.5–3.6), 2.0 (95% CI, 0.9–4.4) and 3.6 (95% CI, 2.4–5.6) for celecoxib, rofecoxib and non-aspirin NSAIDs, respectively. In patients aged ≥ 65 years, the adjusted risks of upper gastrointestinal bleeding according to prescription for celecoxib, rofecoxib and non-aspirin NSAIDs were OR = 1.7 (95% CI, 0.8–3.6), OR = 2.5 (95% CI, 1.4–4.6) and OR = 3.9 (95% CI, 2.7–5.6), respectively.

Discussion

We found that COX-2-selective inhibitors, especially celecoxib, were associated with a lower risk of upper gastrointestinal bleeding than were non-selective, non-aspirin NSAIDs in patients with previous gastrointestinal diseases or diseases predisposing to upper gastrointestinal bleeding. Similar results were obtained in both men and women, and in patients aged 65 years and over.

The main strengths of our study were its large size, the uniformly organized health care system, allowing a population-based design, and the use of data on exposure and confounders that were collected before the date of admission for upper gastrointestinal bleeding. By using hospital discharge diagnoses to ascertain case status, we relied upon the coding of gastrointestinal diagnoses by hospital doctors, and some coding errors may have occurred. The positive predictive values (PPVs) of recorded gastrointestinal discharge diagnoses have previously been shown to be higher for site-specific codes (duodenal, gastric/gastro-jejunal ulcers: PPV, 61–97%) than for non-specific codes (e.g. gastrointestinal haemorrhage: PPV, 59%).30, 31 Misclassification of case status is unlikely to be related to the use of COX-2-selective inhibitors or NSAIDs before hospitalization, and therefore such non-differential misclassification would lead to an underestimation of our risk estimates.

Recall bias, a potential threat in interview-based case–control studies, was avoided by using a complete prescription history from a database, in which information was collected before the date of admission for upper gastrointestinal bleeding. We have no information on patient compliance, as the filing of a prescription was used as a proxy for actual use of a drug. The patients, however, paid for part of the cost of the drug, increasing the likelihood of compliance. Any patient non-compliance would tend to underestimate our risk estimates. From our database, we were able to extract information on the total amount of COX-2-selective inhibitor prescribed (in milligrams). Unfortunately, we were unable to relate this amount to the number of milligrams prescribed per day or to the length of treatment.

For our evaluation of potential confounders, no data were available on smoking, but, in other studies, smoking has not been reported to vary with NSAID or COX-2 use.11, 32, 33 In our various registries, however, information was available on many of the most important confounders. No data were available on the over-the-counter use of NSAIDs. However, only low-dose ibuprofen (200 mg per tablet) is available over the counter in Denmark, and it constitutes only about 13.5% of total NSAID sales (Janne Kampmann, The Danish Medicines Agency, personal communication, 1995). Moreover, pensioners and regular users of low-dose ibuprofen are typically captured in our pharmaco-epidemiological database, because they receive a 50% refund when a physician advises the use of ibuprofen for any appreciable amount of time. Thus, patients receiving low-dose ibuprofen based on a physician's recommendation have a strong economic incentive to obtain these drugs by prescription. Any possible bias introduced by uncaptured users of low-dose ibuprofen would most likely lead to an underestimation of our risk estimates.34

Confounding due to channelling bias can constitute a severe problem in observational pharmaco-epidemiological studies of upper gastrointestinal bleeding.35 As COX-2-selective inhibitors are more likely to be given to high-risk patients because of the presumed better gastrointestinal tolerability of these drugs, the risk of upper gastrointestinal bleeding associated with COX-2 inhibitors may be overestimated (i.e. individuals prescribed COX-2 inhibitors have a higher underlying risk of upper gastrointestinal bleeding before taking COX-2 drugs). To avoid severe confounding due to channelling bias, we conducted the study exclusively in groups of high-risk subjects. The comparison of high-risk cases with high-risk controls should reduce the impact of channelling bias. Even within the well-defined high-risk groups, however, cases were more likely than controls to have used medications other than analgesics and to have had more co-morbid diseases before the study period (Table 1). When we limited the study to these high-risk groups of patients, we were able (with these register-based data) to record information according to concomitant co-morbid diseases and drug use. Otherwise, confounding due to channelling bias could have been a serious problem, because patients might have been included who filed a prescription for COX-2-selective inhibitors, but for whom no registry-recorded history of former gastrointestinal diseases was available (due to risk assessment and treatment only by general practitioners, from whom no data were available).

The number of medications used and the number of co-morbid diseases are predictive of both COX-2 inhibitor use and upper gastrointestinal bleeding.35, 36 Thus, some bias in the risk estimates for COX-2 inhibitors due to residual confounding cannot be ruled out. The relative risk estimates for non-aspirin NSAIDs in the current study are somewhat lower than those reported previously for Danish and other populations.5, 24, 37 However, those studies did not focus on high-risk groups of patients, and therefore most likely reflect the influence of negative confounding by contraindication in the high-risk subjects studied here.38 Thus, the risk estimates reported in this paper probably underestimate the difference in upper gastrointestinal bleeding risk between COX-2-selective inhibitors and non-selective, non-aspirin NSAIDs.

The first observational study published on this subject reported a lower short-term risk of upper gastrointestinal bleeding amongst users of COX-2-selective inhibitors compared with users of non-selective NSAIDs in Ontario, Canada.24 Our results are consistent with the results of the Canadian study, despite the fact that the study populations differed both with regard to history of gastrointestinal disease and age. Our findings are also consistent with those of pre- and early post-marketing randomized trials, which suggested a better gastrointestinal tolerability for COX-2-selective inhibitors compared with non-selective NSAIDs.

With respect to the risk of upper gastrointestinal bleeding, our results suggest that COX-2-selective inhibitors, especially celecoxib, are better tolerated than non-selective, non-aspirin NSAIDs in patients with previous gastrointestinal diseases or diseases predisposing to upper gastrointestinal bleeding. Because of the dramatic increasing use of COX-2-selective inhibitors, further surveillance and clinical epidemiological studies are warranted.

Acknowledgements

The staff at the Department of Health Insurance and Preventive Medicine and Hospital Discharge Registries in the county of North Jutland (Sygesikringen, Nordjyllands amt) are gratefully thanked for the excellent assistance in preparing the data for analyses. The study was supported by a grant from the Western Danish Research Forum for Health Sciences and funding from the Institute of Cancer Epidemiology of the Danish Cancer Society and the International Epidemiology Institute.

Ancillary