Langman MJ, Jensen DM, Watson DJ, Harper SE, Zhao P, Quan H, Bolognese JA, Simon TJ. Adverse Upper Gastrointestinal Effects of Rofecoxib Compared With NSAIDs. JAMA. 1999;282(20):1929-1933. doi:10.1001/jama.282.20.1929
Author Affiliations: Department of Medicine, University of Birmingham, Birmingham, England (Dr Langman); Department of Medicine, University of California, Los Angeles Center for the Health Sciences (Dr Jensen); and Merck Research Laboratories, Merck & Co Inc, West Point, Pa (Drs Watson, Harper, Zhao, Quan, Simon, and Mr Bolognese).
Context Nonsteroidal anti-inflammatory drug (NSAID)–induced gastrointestinal
(GI) toxic effects, such as upper GI tract perforations, symptomatic gastroduodenal
ulcers, and upper GI tract bleeding (PUBs), are thought to be attributable
to cyclooxygenase 1 (COX-1) inhibition. Rofecoxib specifically inhibits COX-2
and has demonstrated a low potential for causing upper GI injury.
Objective To compare the incidence of PUBs in patients with osteoarthritis treated
with rofecoxib vs NSAIDs.
Design Prespecified analysis of all 8 double-blind, randomized phase 2b/3 rofecoxib
osteoarthritis trials conducted from December 1996 through March 1998, including
one 6-week dose-ranging study, two 6-week efficacy studies vs ibuprofen and
placebo, two 1-year efficacy studies vs diclofenac, two 6-month endoscopy
studies vs ibuprofen and placebo, and one 6-week efficacy study vs nabumetone
Setting Multinational sites.
Participants Osteoarthritis patients (N = 5435; mean age, 63 years [range, 38-94
years]; 72.9% women).
Interventions Rofecoxib, 12.5, 25, or 50 mg/d (n = 1209, 1603, and 545, respectively,
combined) vs ibuprofen, 800 mg 3 times per day (n = 847), diclofenac, 50 mg
3 times per day (n = 590); or nabumetone, 1500 mg/d (n = 127) (combined).
Main Outcome Measure Cumulative incidence of PUBs for rofecoxib vs NSAIDs, based on survival
analysis of time to first PUB diagnosis, using PUBs that met prespecified
criteria judged by a blinded, external adjudication committee.
Results The incidence of PUBs over 12 months was significantly lower with rofecoxib
vs NSAIDs (12-month cumulative incidence, 1.3% vs 1.8%; P = .046; rate per 100 patient-years, 1.33 vs 2.60; relative risk,
0.51; 95% confidence interval, 0.26-1.00). The cumulative incidence of dyspeptic
GI adverse experiences was also lower with rofecoxib vs NSAIDS over 6 months
(23.5% vs 25.5%; P = .02), after which the incidence
Conclusion In a combined analysis of 8 trials of patients with osteoarthritis,
treatment with rofecoxib was associated with a significantly lower incidence
of PUBs than treatment with NSAIDs.
Two isoforms of cyclooxygenase (COX), COX-1 and COX-2,1- 4
catalyze human prostaglandin synthesis. Almost all currently available nonsteroidal
anti-inflammatory drugs (NSAIDs) inhibit both COX isoforms.5
COX-1 is constitutively expressed and generates prostaglandins believed to
be involved in gastrointestinal (GI) mucosal protection,6
while COX-2 is induced at sites of inflammation throughout the body and generates
prostaglandins that mediate inflammation and pain.7
Therefore, the anti-inflammatory effects of NSAIDs appear to be mediated via
inhibition of COX-2,8 while the deleterious
GI effects, a significant source of morbidity and mortality, are believed
to occur primarily via inhibition of COX-1.5
These GI-related adverse effects (AEs)9- 11
are estimated to be responsible for 107,000 hospitalizations and 16,500 deaths
annually in the United States alone.12
Rofecoxib (Vioxx, Merck & Co, Inc, Whitehouse Station, NJ) specifically
inhibits COX-213,14; at dosages
of up to 375 mg/d for 12 days15 and up to 1000
mg in single doses,13 it had no effect on COX-1
isoenzyme activity. In vitro studies of rofecoxib showed no significant effect
on prostaglandin synthesis in human gastric biopsies.16
In addition, rofecoxib demonstrated low potential for GI injury (at dosages
of 25 and 50 mg/d) as measured by intestinal permeability17
and fecal red blood cell loss,18 and by assessment
of endoscopic gastroduodenal mucosal injury in both healthy volunteers (at
250 mg/d)19 and patients with osteoarthritis
(at 25 and 50 mg/d).20
This study was a planned, blinded, combined analysis of 8 randomized,
double-blind, phase 2b/3 clinical trials performed from December 1996 through
March 1998 of rofecoxib inpatients with osteoarthritis to examine the incidence
of upper GI perforations, symptomatic gastroduodenal ulcers, and upper GI
bleeding (PUBs). We hypothesized that the incidence of PUBs would be lower
with rofecoxib (12.5-, 25-, and 50-mg combined treatment groups) than with
NSAIDs (ibuprofen, diclofenac, and nabumetone combined treatment groups).
The plan used to analyze the incidence of PUBs with rofecoxib compared
with NSAIDs was prespecified. PUBs are rare and large numbers of patients
are necessary to evaluate the rate of incidence with precision; therefore,
in this analysis, we pooled patients from all 8 phase 2b/3 osteoarthritis
trials of rofecoxib and their blinded extensions (Table 1). All patients gave signed informed consent and an institutional
review board approved each study.
The analysis plan prespecified that patients with asymptomatic ulcers
diagnosed within a 7-day window surrounding the scheduled procedure dates
in the 2 surveillance endoscopy studies would be excluded from the analysis.
Eleven additional patients with asymptomatic ulcers confirmed by investigators
to have been detected at surveillance endoscopies scheduled outside the 7-day
window for a variety of reasons were also excluded.
Investigators were instructed to report all laboratory and clinical
AEs, including upper GI dyspepsia and PUBs, that occurred during treatment
and within 14 days of study drug discontinuation. In all the studies, a final
patient contact and/or evaluation was scheduled for the 14th day following
study completion or study drug discontinuation. Patients were followed up
in each study until final evaluation after completion of study therapy or
after early discontinuation for any reason (eg, diagnosis of PUB, other AE,
lack of treatment efficacy, withdrawal of consent), death, or loss to follow-up.
Clinical source documentation for suspected PUBs was reviewed by a blinded,
external adjudication committee. Only PUBs judged as confirmed, according
to prespecified definitions (Table 2),
were analyzed. PUBs that occurred more than 14 days after the last dose of
study drug were not included in the analysis (as prespecified).
Adverse events, including PUBs, were coded blind to treatment with a
standard automated dictionary that classified AEs into broader categories
grouped by body system. We also compared discontinuations due to any GI AE
and discontinuations due to a prespecified subset of GI AEs typical of upper
GI symptoms associated with NSAIDs.21- 23
The latter category (hereafter referred to as "dyspeptic-type GI AEs") consists
of all AEs mapped by the dictionary to the categories of acid reflux, dyspepsia,
epigastric discomfort, heartburn, nausea, or vomiting.
Analyses were based on all patients treated in the 8 trials, with exceptions
as follows. Patients from the placebo and 5-mg rofecoxib groups of Protocol
029 (Table 1) were switched after
6 weeks to diclofenac, 12.5 mg of rofecoxib, or 25 mg of rofecoxib in an extension
phase. Similarly, patients from the placebo group of Protocol 058 were switched
after 6 weeks to nabumetone, 12.5 mg of rofecoxib, or 25 mg of rofecoxib in
an extension phase. To avoid double-counting patients in the analyses, it
was prespecified that the data from patients randomized to the placebo or
5-mg rofecoxib groups in the 6-week placebo-controlled phase of both studies
would be excluded from analyses, while the data from these patients' extension
phase would be included. However, 117 such patients (2.1% of the total population)
did not continue into extensions or did not have extension data at the time
of the analysis and were therefore excluded.
A survival analysis of the time to first PUB diagnosis date was used
for between-treatment comparisons. This method is appropriate because it takes
into account the varying lengths of treatment in the 8 studies. For PUBs diagnosed
in inpatients, the hospital admission date was used as the diagnosis date.
For PUBs diagnosed in outpatients, the date of the diagnostic procedure or
clinical observation was used. The log-rank test was the primary method used
to compare time-to-first-event distributions between groups. The cumulative
incidence difference between rofecoxib and NSAIDs was assessed at 6 weeks
and at 4, 6, 12, and 24 months, using the method of Breslow and Crowley.24 Results are reported at 4 months (maximum duration
of placebo) and 12 months (no PUBs occurred beyond 12 months). Differences
between groups were considered significant when P<.05.
The Cox proportional hazards model was used to estimate overall relative risks
(RRs) and 95% confidence intervals (CIs) of rofecoxib vs NSAIDs. Treatment
by type of protocol interactions were evaluated in the Cox model. Analyses
stratified by type of protocol and analyses in which each type of protocol
was removed from the analysis 1 at a time were performed to assess possible
The analysis included 5435 patients. Of these, 3357 patients were treated
with rofecoxib (1209, 1603, and 545 patients received 12.5, 25, and 50 mg,
respectively, once daily), 1564 patients were treated with NSAIDs (847 received
ibuprofen, 800 mg 3 times daily; 590 received diclofenac, 50 mg 3 times daily;
and 127 received nabumetone, 1500 mg once daily) and 514 patients were treated
with placebo. Total patient-years of exposure were 1428, 615, and 112, in
the rofecoxib, NSAID, and placebo groups, respectively. The average dosage
of rofecoxib was 24.7 mg once daily.
There were no clinically meaningful differences in baseline characteristics
between groups (Table 3). Mean
age overall was 63 years (range, 38-94 years); 45% of patients were 65 years
or older, and 73% were women. Most (90%) patients had previously used NSAIDs
for their osteoarthritis. Approximately 10% of the patients in each group
had a prior medical history of PUB.
Rofecoxib was generally well tolerated and fewer patients overall discontinued
rofecoxib compared with the NSAIDs (Table
4); 9.4% of rofecoxib patients vs 10.7% of NSAID patients discontinued
the study drug because of any clinical AE, and 3.5% of the rofecoxib patients
discontinued the study drug due to a GI AE, compared with 4.8% of NSAID patients
(Table 4). The cumulative incidence
of dyspeptic-type GI AEs up to 6 months was significantly lower with rofecoxib
than with NSAIDs (23.5% vs 25.5%; P = .02), after
which the incidence rates converged. The 12-month cumulative incidences of
study drug discontinuation due to GI AEs were 5.7% vs 7.8% for the rofecoxib
and NSAID groups. The difference was significantly lower (P = .02) in the rofecoxib group compared with the NSAID group over
12 months (8.2 vs 12.0 per 100 patient-years, respectively; RR = 0.70; 95%
Forty-nine potential PUBs were submitted by investigators and adjudicated;
5 (3 in the rofecoxib and 2 in the NSAID group) did not meet the prespecified
case definition because they occurred more than 14 days after study drug discontinuation.
Of the remaining 44 cases, 38 (19 in the rofecoxib, 16 in the NSAID, 3 in
the placebo group) were adjudicated as having at least 1 confirmed PUB, and
6 (all in the NSAID group) were adjudicated as having unconfirmed PUBs. Nine
patients had more than 1 PUB (2, 3, and 4 in the placebo, placebo, refecoxib,
and NSAID groups respectively); all of these patients had 2 PUBs each, except
for 1 patient (NSAID group) with 3 PUBs. Only the first PUB in a given patient
was included in the analysis. There were no PUBs during the small amount of
patient exposure (25 patient-years) to nabumetone in the 1 study that included
The cumulative incidence of confirmed PUBs over 12 months with rofecoxib
was significantly lower (1.3% vs 1.8%, P = .046)
than with NSAIDs (Figure 1). The
rates per 100 patient-years over 12 months were 1.33 and 2.60 for rofecoxib
and NSAIDs, respectively. The overall RR over 12 months for rofecoxib vs NSAIDs
was 0.51 (95% CI, 0.26-1.00). Analyses stratified by type of protocol yielded
very similar results to the unstratified analyses (RR = 0.53; 95% CI, 0.27-1.03; P =.06), and analyses that sequentially removed each protocol
type also yielded consistent results, demonstrating a lack of confounding
by type of protocol. No statistically significant treatment by type of protocol
interaction was present (P>.10) in the unstratified
Cox model. The difference between rofecoxib and NSAIDs in the incidence of
confirmed PUBs became statistically significant as early as 6 weeks (RR for
rofecoxib vs NSAIDs, 0.21; 95% CI, 0.06-0.67; P =
.004), and remained so up to 12 months.
In analyses confined to placebo-controlled protocols, which were up
to 4 months in duration, the cumulative incidence rates of confirmed PUBs
were 0.9%, 0.9%, and 1.6% for placebo, rofecoxib, and NSAIDs, respectively.
The corresponding rates per 100 patient-years over 4 months were 2.68, 2.50,
and 7.21. The RR over 4 months for NSAIDs vs placebo was 2.50 (95% CI, 0.68-9.24),
while that for rofecoxib vs placebo was 0.94 (95% CI, 0.25-3.60).
Because PUBs are rare, we carried out a prespecified analysis of data
obtained from 8 randomized, double-blind phase 2b/3 osteoarthritis trials
with rofecoxib to determine whether COX-2 specific inhibition would be associated
with a lower incidence of PUBs than nonspecific COX inhibition. The combined
population included patients at high risk of PUBs (eg, those with a history
of PUB and/or age ≥65 years), as reported previously.25
The analysis demonstrated a significantly lower incidence of confirmed PUBs
with rofecoxib than with NSAIDs over 12 months. In addition, the total rate
of study drug discontinuations due to GI AEs was significantly lower in the
group receiving rofecoxib than in the group receiving NSAIDs.
The analysis has particular strengths. It was prospectively designed
and used prespecified adjudication criteria, hypotheses, and data analysis
plans, and an external, blinded committee adjudicated all reported PUBs. It
included all phase 2b/3 osteoarthritis trials of rofecoxib and a broad range
of patients, including those with prior PUB history. The primary hypothesis
involved a biologically meaningful comparison of GI safety with COX-2 specific
inhibition vs nonspecific COX-1/COX-2 inhibition, and the doses of NSAID comparators
for the included studies were chosen to be within the clinical dose range
for treatment of osteoarthritis. The average dosage of rofecoxib in this study
was 24.7 mg/d, which corresponds to the highest recommended daily dose (25
mg) for osteoarthritis. Both the 12.5-mg and 25-mg doses of rofecoxib have
been shown, using prespecified criteria, to be therapeutically equivalent,
in terms of osteoarthritis symptom relief, to the doses of the main comparator
NSAIDs (diclofenac and ibuprofen) used in this study.26,27
The analytic methods took account of the varying lengths of the studies. Analyses
stratified by type of protocol demonstrated that difference in study design
was not a confounding variable in the assessment of PUB risk with treatment.
Statistical tests of treatment by type of protocol interaction in the unstratified
COX model indicated that the assumption of a common RR was appropriate. Because
PUBs are uncommon, there were generally too few within any 1 study to provide
precise study-specific RR estimates. However, the variation that did occur
was compatible with the assumption of a common, study-specific incidence rate
ratio. Analyses in which each type of protocol was removed from the analysis
1 at a time showed consistent results.
The analysis also had limitations. First, there were only enough patient-years
of exposure to compare pooled rofecoxib and NSAID comparators, rather than
individual treatments. Second, inclusion of studies with scheduled endoscopies
mandated in their protocols may have caused a bias against rofecoxib. Patients
in these studies were systematically discontinued from treatment when they
developed endoscopically evident gastroduodenal ulcers 3 mm or more in diameter,
and a much higher rate of endoscopically detected ulceration was observed
in the ibuprofen than in the rofecoxib group.21
If patients discontinued from the study had greater potential to develop a
PUB (eg, because of an endoscopic ulcer or a prior history of PUB), then the
inclusion of these 2 studies may have reduced the observed incidence of PUBs
in the ibuprofen groups.
Our analysis shows that COX-2 specific inhibition with rofecoxib was
associated with a significantly lower risk of PUBs relative to NSAIDs. These
findings are consistent with the results of studies of intestinal permeability,17 fecal red blood cell loss,18
and upper GI endoscopy19,20 with
rofecoxib and indicate that risks of GI toxic effects associated with NSAIDs
can be reduced by COX-2 specific inhibition.