Study population

All patients participating in the study gave written informed consent and the study was conducted in conformance with applicable country or local ethics requirements. The study was conducted at 48 sites in 18 countries. Patients between the ages of 21 and 85 years were enrolled, with a confirmed clinical diagnosis of RA and a requirement of at least three months of NSAID therapy. Patients were excluded if they had an oesophageal, gastric, or duodenal ulcer, pyloric obstruction, or erosive oesophagitis at baseline endoscopy.

Patients were also excluded if they had any of the following: creatinine level >2.0 mg/dl; creatinine clearance ≤30 ml/min; bleeding diathesis; requirement for anticoagulants, low dose aspirin, ticlopidine, or clopidogrel; unstable medical disease including current angina or congestive heart failure; previous upper gastrointestinal surgery; faecal occult blood; history of inflammatory bowel disease, history of myocardial infarction, coronary angioplasty, or coronary artery bypass graft within one year, cerebrovascular event or active hepatic disease within the past two years, or malignancy within the past five years. The study population included patients infected with Helicobacter pylori, a prior history of a gastroduodenal PUB, or the presence of gastroduodenal erosions at baseline endoscopy.

Design

This was a randomised, double blind, placebo controlled, parallel group study.

Procedures

Following a two week washout period from non-selective NSAIDs and any antisecretory or cytoprotective drugs, patients who met entry criteria underwent baseline upper gastrointestinal endoscopy and determination of H pylori status by gastric biopsy with a rapid urease test (CLOtest) and by histology. Patients were then randomised to one of three treatment groups: rofecoxib 50 mg once daily, naproxen 500 mg twice daily, or placebo, for 12 weeks. Matching placebo for each medication was used to maintain blinding. Treatment allocation was stratified by history of significant upper gastrointestinal disease (gastroduodenal ulcer or upper gastrointestinal haemorrhage or perforation), and each stratum was further stratified according to concomitant use of oral corticosteroids during the study. Stable doses of non-study antirheumatic medications other than NSAIDs and antisecretory or cytoprotective drugs were permitted, as were injectable or oral corticosteroids (≤10 mg prednisone or equivalent daily). Rescue medications (acetaminophen/paracetamol for pain and Gelusil aluminum hydroxide/magnesium hydroxide/simethicone (Warner-Lambert) for minor dyspepsia) were provided. Patients were prohibited from using the following medications throughout the study: H₂ receptor antagonists, proton pump inhibitors, prostaglandin analogues, other gastroprotective agents, calcium containing antacids, anticoagulants, antiplatelet therapy, and cyclosporin.

Patients were evaluated clinically at study weeks 3, 6, 9, and 12. Endoscopy was performed at baseline, at study weeks six and 12, at unscheduled discontinuation, and when deemed clinically necessary by the investigator. At endoscopy, the number and size of gastric and/or duodenal ulcers (defined as mucosal breaks with an unequivocal depth of at least 3 mm in the longest dimension) were recorded, and gastroduodenal mucosal erosions were counted. If an ulcer was detected, the patient was immediately discontinued from the study and underwent discontinuation procedures and ulcer treatment. Safety assessments also included physical examination and vital signs, laboratory parameters, and monitoring of clinical and laboratory adverse events throughout the study.

Statistical analysis

The cumulative incidence of gastric and/or duodenal ulcers ≥3 mm over 12 weeks was the primary end point of the study. A survival analysis was used to analyse time to event data for ulcer incidence, and the log rank test was used to compare the cumulative incidence of ulcers among treatment groups. Twelve week cumulative rates were estimated via Kaplan-Meier, and the Breslow-Crowley method²⁵ was used to calculate 95% confidence intervals (CIs); 95% CIs were also calculated for between treatment differences and the ratio of the 12 week cumulative rates for the prespecified comparisons. Although the study was not designed to compare rofecoxib with placebo, p values were calculated for comparisons with placebo.

Twelve week cumulative ulcer incidence rates of 7.5% for rofecoxib, 7.5% for placebo, and 25% for naproxen were assumed. The study had 98% power to detect such differences between rofecoxib or placebo versus naproxen, based on a two sided test with α=0.05.

Assessment of treatment effect on ulcer incidence was made for patient subgroups to explore the consistency of effects across subgroups. These classifications were defined at baseline by the following: age (<65 or ≥65 years), sex, race (White or non-White), history of symptomatic PUB, H pylori status, baseline number of gastric and/or duodenal erosions (=0 or ≥0), prior NSAID use, tobacco use, corticosteroid use, and geographical region (USA or international). A Cox proportional hazards model was used to test qualitative treatment by subgroup interactions.

Cumulative incidence of ulcers ≥5 mm over 12 weeks, and mean change from baseline number of erosions over 12 weeks were assessed as secondary end points. Continuous variables were analysed with an analysis of covariance model (with prespecified factors of treatment, gastrointestinal history, corticosteroid use, and baseline covariate where appropriate). Adverse event data were analysed by Fisher’s exact test.

Patient characteristics

Figure 1⇓ shows the disposition of patients throughout the study. Of the 660 patients enrolled at the randomisation visit, 509 completed the study. Nine, 11, and 22 patients discontinued due to adverse experiences on placebo, rofecoxib, and naproxen, respectively, while 11, four, and two patients, respectively, discontinued due to lack of efficacy. As shown in table 1⇓, there were no clinically meaningful differences among groups with regard to age, sex, race, history of gastrointestinal events, H pylori status, tobacco use, presence of gastroduodenal erosions, or prior NSAID use.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Study flow chart.

Incidence of ulcers and erosions

The six and 12 week incidences of ulcers ≥3 mm are shown in fig 2⇓. The 12 week cumulative incidence of gastroduodenal ulcers ≥3 mm in patients on naproxen 500 mg twice daily (25.5%) was significantly greater than that in patients on rofecoxib 50 mg (6.8%; between treatment difference 18.7% (95% CI 11.7%, 25.7%); p≤0.001) or placebo (2.9%; difference 22.6% (95% CI 16.1%, 29.1%); p<0.001) whereas the incidence in patients on rofecoxib did not differ significantly from that in patients on placebo (2.9%; p=0.066). Findings for ulcers ≥5 mm were similar and are shown in fig 3⇓. The 12 week cumulative incidence of ulcers ≥5 mm was 2.9%, 5.3%, and 17.1% in patients on placebo, rofecoxib, and naproxen, respectively (difference for naproxen v rofecoxib (95% CI) 11.8% (5.7%, 18.0%), p≤0.001; difference for naproxen v placebo 14.2%, p≤0.001). The incidence in the rofecoxib group did not differ significantly from that in the placebo group (2.9%; p=0.210). As shown in fig 4⇓, the least squares mean change from baseline in the number of erosions was significantly higher in patients on naproxen versus those on rofecoxib (difference (95% CI) 4.05 (93.37, 4.73); p<0.001) and in patients on naproxen versus patients on placebo (difference (95% CI) 4.41 (3.73, 5.08); p≤0.001). There was no significant difference between patients on placebo and patients on rofecoxib.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Cumulative incidence rate (%) of gastroduodenal ulcers ≥3 mm (intention to treat) in the placebo, rofecoxib 50 mg, and naproxen 500 mg twice daily groups. ***p<0.001 for naproxen versus rofecoxib or placebo.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Cumulative incidence rate (%) of gastroduodenal ulcers ≥5 mm at week 12 (intention to treat) in the three groups. ***p<0.001 for naproxen versus rofecoxib or placebo.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Least squares mean change from baseline number of gastroduodenal erosions at week 12 (intention to treat) in the three groups. ***p<0.001 for naproxen versus rofecoxib or placebo.

A similar pattern of treatment effects was seen when gastric and duodenal ulcers were examined separately. The cumulative incidence of gastric ulcers ≥3 mm at 12 weeks was 2.0% for placebo, 5.9% for rofecoxib, and 22.1% for naproxen. The cumulative incidence of duodenal ulcers ≥3 mm at 12 weeks was 0.9% for placebo, 1.0% for rofecoxib, and 5.2% for naproxen.

Risk modifiers

Between treatment comparisons for ulcer (≥3 mm) incidence in subgroups defined by baseline patient characteristics were found to be generally consistent across all levels of all subgroups. Age ≥65 years, White race, and presence of prior gastrointestinal history were risk factors for gastroduodenal ulcers, as demonstrated by statistically significant main effects. There was no significant main effect of H pylori status (p=0.077) and no significant interaction between H pylori status and treatment (p=0.500). The cumulative incidence of ulcers ≥3 mm at 12 weeks in H pylori negative patients was 3.6% for placebo, 3.9% for rofecoxib, and 33.3% for naproxen. In H pylori positive patients, the cumulative incidence of ulcers ≥3 mm at 12 weeks was 2.3% for placebo, 8.3% for rofecoxib, and 17.9% for naproxen.

Tolerability

Table 2⇓ summarises the tolerability assessments. There was no significant difference among treatment groups in the overall incidence of adverse events, although the incidence of drug related adverse events was higher in patients on naproxen and those on rofecoxib compared with those on placebo (p=0.002 for naproxen v placebo; p=0.030 for rofecoxib v placebo). Patients taking naproxen had a significantly higher rate of discontinuations due to an adverse event compared with the placebo group (p=0.036).

Comparison of the incidence of serious clinical adverse experiences among treatment groups showed no significant differences (2.7% for placebo, 1.8% for rofecoxib, and 4.1% for naproxen) (table 2⇑). One death was reported among patients taking naproxen (hepatic and multiple organ failure in a patient discovered to be seropositive for hepatitis B). This was considered definitely not related to the study drug by the investigator. No serious laboratory adverse events were reported, and rates of all laboratory adverse experiences were generally similar among treatment groups (12.7 % in patients on placebo, 14.2% in patients on rofecoxib, and 18.3% in patients on naproxen) except for slightly more numerous reports of decreases in haematocrit and haemoglobin in patients taking naproxen compared with the two other groups.

The three adverse events most commonly reported over the entire study were epigastric discomfort, dyspepsia, and heartburn, each of which was reported in at least 5% of patients in each treatment group. No significant difference was observed among treatment groups for the incidence of dyspepsia (6.3% on placebo, 10.0% on rofecoxib, and 7.3% on naproxen) or heartburn (5.0% on placebo, 9.1% on rofecoxib, and 7.3% on naproxen); however, the rofecoxib group (9.1%) and the placebo group (8.6%) both had significantly lower incidences of epigastric discomfort compared with the naproxen group (17.3%; p≤0.016 for rofecoxib or placebo v naproxen). The percentages of patients who discontinued because of an adverse event involving the digestive system or abdominal pain were 0.9% on placebo, 3.7% on rofecoxib, and 6.8% on naproxen. The difference between the naproxen and placebo groups was statistically significant (5.9%; p=0.001) but the difference between the rofecoxib and placebo groups was not (p=0.062). There were seven PUBs (one patient on placebo, two patients on rofecoxib, and four patients on naproxen).

Because COX inhibition has known renal and vascular effects, analyses were prespecified for comparisons between groups for adverse experiences involving hypertension, oedema, or congestive heart failure (table 3⇓). The reported incidences of hypertension were 2.3% for the placebo group, 6.4% for the rofecoxib group, and 0.9% for the naproxen group, and lower extremity oedema was reported for 1.8% of patients on placebo, 1.4% of patients on rofecoxib, and 0% of patients on naproxen. No patient discontinued treatment with the study drug due to either hypertension or lower extremity oedema. Likewise, there were no between treatment group differences observed for the incidences of congestive heart failure adverse events, which were reported for one patient in the study (rofecoxib group). This patient’s event was considered not serious and not related to study drug by the investigator.