Rheumatology 2003;42:1354–1364

doi:10.1093/rheumatology/keg401, available online at www.rheumatology.oupjournals.org

Advance Access publication 16 July 2003

Comparison of the incidence rates of

thromboembolic events reported for patients
prescribed celecoxib and meloxicam in general
practice in England using Prescription-Event
Monitoring (PEM) data
Deborah Layton1,2, Kerry Hughes1, Scott Harris1,3 and
Saad A.W. Shakir1,2

Background. Celecoxib and meloxicam are classified as cyclooxygenase (COX)-2

selective inhibitors. The Drug Safety Research Unit monitored the post marketing
safety of these drugs in England using the non-interventional observational cohort
technique of Prescription-Event Monitoring (PEM).
Objectives. To compare the incidence rates of selected thromboembolic (TE)
(cardiovascular, cerebrovascular and peripheral venous thrombotic) events
reported for patients prescribed celecoxib and meloxicam in general practice.
Methods. Patients were identified from dispensed prescriptions written by general
practitioners (GPs) for meloxicam (December 1996–March 1997) and celecoxib
(May and December 2000). Simple questionnaires requesting details of events
occurring during/after treatment, indication and potential risk factors (including
age, sex and whether NSAIDs had been prescribed within 3 months of treatment)
were posted to prescribing GPs at least 6 months after the first prescription for
each patient. Incidence rates of the first event were calculated; crude and adjusted
rate ratios (RRs) were obtained using Poisson regression modelling.
Results. During the 9 months after starting treatment, 28 (0.16%) and 19 (0.10%)
of patients were reported to have experienced cardiovascular TE events, 68
(0.39%) and 52 (0.27%) cerebrovascular TE events, and 17 (0.10%) and 20
(0.10%) experienced peripheral venous thrombotic events for celecoxib and
meloxicam, respectively. Regarding time to first event, there was a persistent
divergence between the two drugs from 30 days after the start of treatment for
both the cardiovascular TE event group (log rank test P ¼ 0.0153) and
cerebrovascular TE event group (log rank test P ¼ 0.0055). Indication and use
of an NSAID within 3 months prior to starting treatment had no effect on the
relative risk estimates of the event groups and was excluded in subsequent
analyses. Adjusting for the two identified risk factors of age (age2) and sex, the
cerebrovascular TE event group rate was higher for celecoxib than for meloxicam,
RR 1.66 (95% CI 1.10–2.51), over the study period and no different for the
cardiovascular TE event group, RR 1.72 (95% CI 0.87–3.40) or peripheral venous
thrombotic group, RR 1.06 (95% CI 0.51–2.19).
Conclusions. This study reports a relative increase in the rate of cerebrovascular
TE events in users of celecoxib compared to meloxicam. There was no difference
in the rate of cardiovascular TE events or peripheral venous thrombotic events

1
Drug Safety Research Unit, Bursledon Hall, Blundell Lane, Southampton, 2University of Portsmouth and 3University of Southampton, UK.

Submitted 6 January 2003; revised version accepted 8 April 2003.

Correspondence to: D. Layton, Drug Safety Research Unit, Southampton SO31 1AA, UK. E-mail: deborah.layton@dsru.org

1354
Rheumatology Vol. 42 No. 11 ß British Society for Rheumatology 2003; all rights reserved
 Thromboembolic events after taking celecoxib and meloxicam
between users of these two drugs. The incidence of these three groups of events
reported in each of these two drug cohorts was low (<0.5%), therefore the
relevance of our findings need to be taken into consideration with other clinical and
pharmacoepidemiological studies.
KEY WORDS: Adverse events, Celecoxib, COX-2 selective inhibitors, Drug safety, Meloxicam,
Prescription-Event Monitoring.
Cyclooxygenase (COX)-2 isoenzyme inhibitors were
developed with the aim of reducing gastrointestinal
(GI) adverse reactions compared to non-selective non-
steroidal anti-inflammatory drugs (NSAIDs) [1–6].
However, while emerging information suggests that use
of such drugs may contribute to an increased risk of
adverse vascular events [7], this is yet to be confirmed by
a sufficient number of studies. Furthermore, it is unclear
whether the higher risk applies to all thromboembolic
(TE) events and whether it applies to all COX-2
inhibitors at all doses, or to some products at specific
doses or dose ranges [8].
The pharmacology of NSAIDs appears to be well
described [9, 10]. However, accumulating evidence
regarding the relationship between COX-1 mediated
platelet-derived thromboxane-A2 and COX-2-mediated
macrovascular endothelial-cell-derived prostacyclin
[11–16] suggests that vascular haemostasis may be
impaired in circumstances where blockade of COX-2-
induced prostacyclin, unopposed by COX-1-induced
platelet aggregation, may result in an increased risk of
TE events in susceptible individuals [17, 18].
Celecoxib (CelebrexÕ ), launched in May 2000 was
the second COX-2-specific isoenzyme inhibitor to be
marketed in the UK, and was indicated for the
symptomatic relief of osteoarthritis (OA) or rheumatoid
arthritis (RA). As reported for other COX-2 selective
agents, celecoxib (600 mg b.d. for 10 days) does not
inhibit platelet aggregation or prolong bleeding time in
studies in healthy volunteers [19]. However, elevated
prothrombin times and bleeding episodes have been
observed with concomitant use of celecoxib and warfarin
[20], and there have been four patients with connective
tissue disorders that developed ischaemic complications
associated with thrombosis after receiving celecoxib
[21–23]. Meloxicam (MobicÕ ), launched in the UK in
December 1996, is also considered to be a COX-2
selective inhibitor [24, 25], and was indicated for relief of
pain and inflammation in rheumatic disease, in exacer-
bations of osteoarthritic pain and ankylosing spondylitis
at launch. As described previously [26], meloxicam does
not appear to affect COX-1-dependent platelet throm-
boxane formation or platelet aggregation [27–29], and
reports of serious cardiovascular events have not been
thought attributable to treatment [30].
Among the randomized, controlled trials with the
COX-2 inhibitor rofecoxib, one study demonstrated a
significant difference between rofecoxib and its NSAID
comparator (naproxen) in the risk of cardiovascular
thrombotic events [4]. Yet the evidence from other
1355
studies of rofecoxib is conflicting [31, 32]. In contrast the
CLASS (celecoxib long-term arthritis safety study) trial,
which involved 8059 patients with OA or RA, and
compared celecoxib (400 mg b.d.) with NSAIDs (ibu-
profen 800 mg t.d.s or diclofenac 75 mg b.d.), demon-
strated no excess of serious TE cardiovascular events [6].
For both drugs, these large-scale trials were designed to
demonstrate that gastrointestinal safety was superior to
that of traditional NSAIDs in clinical practice, but were
not sufficiently powered to detect differences of TE
events against the background cardiovascular event rates
in the placebo groups. A review of four randomized trials
[including the CLASS trial and VIGOR (Vioxx
Gastrointestinal Outcomes Trial) study] that was
conducted to determine whether COX-2 inhibitors are
associated with a protective or hazardous effect on the
risk of cardiovascular events reported a potential
increase in cardiovascular event rates for users of
COX-2 inhibitors [31]. However, these trials were
different in several ways and the results were not directly
comparable.
Monitoring for adverse effects in the post-marketing
phase forms an important part of a drug’s safety profile.
Post-marketing data derived from spontaneous reports
suggest that the risks of renal and cardiovascular ad-
verse events associated with the use of rofecoxib are
significantly higher than those of celecoxib and NSAIDs
(diclofenac and ibuprofen) [33]. Prescription-Event
Monitoring (PEM) studies of newly marketed drugs
during their immediate post-marketing period in
England, provide complementary data on safety issues
in addition to randomized-controlled trials and sponta-
neous reporting schemes. PEM uses a non-interventional
observational cohort technique with a systematic
approach to data collection; the methodology has been
described in detail elsewhere [34–38]. As part of its
monitoring program, the Drug Safety Research Unit
(DSRU) has carried out individual PEM studies of
meloxicam [39] and celecoxib [40]. This paper reports
the results of a study to examine and compare the
cardiovascular risk of these two COX-2 selective
inhibitors. An identical study was conducted using
rofecoxib and meloxicam [26]. These studies, using data
collected via the observational technique of PEM, did
not require reference to an ethics committee or patient
consent.
The aim of this second study was to retrospectively
investigate, using large cohorts from the general
population of England, whether there is a difference
in the type and incidence of TE cardiovascular events
1356 D. Layton et al.

reported during routine clinical use in general practice TABLE 1. Thromboembolic event groups
of meloxicam and celecoxib. As before, our objectives
Cardiovascular Cerebrovascular Peripheral venous
were: to calculate and compare rates for TE events thrombotic
occurring within the first 9 months after starting
treatment with celecoxib or meloxicam; to determine Cardiac arresta Amaurosis fugax Deep vein thrombosis
relative risks, or rate ratios (RRs) separately for CVS not specifieda Aphasia Embolus pulmonary
Myocardial Cerebrovascular Infarction pulmonary
cardiovascular and cerebrovascular TE events and infarction accident
peripheral venous thrombotic events, adjusted for the Dysarthria
possible confounders of age and sex [41–43], and Dysphagiaa
whether other oral NSAIDs had been prescribed in the Dysphasia
Embolus cerebral
3 months prior to starting the drug [15, 42, 43]; and to Embolus mesenteric
calculate and compare the time to first event within Hemianopia
each TE event group for each cohort. Hemiparesis
Hemiplegia
Hypoaesthesiaa
Methods Paralysis facial
Paralysis pseudobulbar
The PEM study conducted for celecoxib was conducted as Paralysis ocular
described previously for rofecoxib [44]. For this study, Paraesthesiaa
exposure data were obtained from green forms received for Paresisa
patients identified from NHS prescriptions written by GPs in Retinal thrombosis
England for meloxicam between December 1996 and March artery
Retinal thrombosis vein
1997 (n ¼ 19 087) and for celecoxib between May and
Slurred speach
December 2000 (n ¼ 17 458). For comparative purposes the Thrombosis cerebral
exposed are those patients prescribed celecoxib and the Transient ischaemic
unexposed are those patients prescribed meloxicam. As attack
described previously [26], the event terms for this study were Visual disturbancea
selected by medical practitioners from the DSRU dictionary
prior to the analysis and aggregated into the three TE groups During the 9 month study period, one report of embolus artery
(cardiovascular, cerebrovascular and peripheral venous was recorded for celecoxib; no reports were recorded for the
thrombotic) events (Table 1). The data were subject to the following miscellaneous TE event terms: infarction gastrointestinal,
same inclusion and exclusion criteria as specified previously, thrombosis mesenteric, thrombosis spinal, thrombosis artery, infarc-
for calculation of person-time exposed (pte) [26, 44, 45]. tion renal.
a
Non-specific terms evaluated by clinician and relevant lowest level
(doctor terms) included in the analysis.
Sample size CVS, cardiovascular system.
The sample size calculation for PEM studies is described
previously [44]. This study has a 95% chance of observing a
statistically significant relative difference in rates of 10%
tively [28.1 vs 23.2%, 2 P < 0.0001], with similar
between the drugs for each event group, if such an underlying
background relative difference exists [26]. proportions of users prescribed either celecoxib or
meloxicam for treatment of symptoms of RA, respec-
Analysis tively [6.5% (1128/17 458) vs 6.6% (1253/19 087), 2
P ¼ 0.3520]. Where answers to the additional questions
Data analysis was conducted in an identical manner to that
described previously [44]. The unadjusted RRs, as well as were given, significantly more celecoxib users than
ratios adjusted for the selected risk factors were calculated meloxicam users had been prescribed an NSAID
and examined using both univariate and multivariate Poisson within the 3 months prior to starting treatment [49.4%
regression modelling. Evidence of effect modification by the (7006/14 195) vs 48.0% (7978/16 634), 2 P ¼ 0.014].
selected risk factors on the estimates of relative risk were also During the 9 months after starting treatment with
investigated, and the estimate of time to first event for each celecoxib and meloxicam, 28 (0.16%) and 19 (0.10%) of
group for each cohort calculated and compared as stated patients were reported to have had cardiovascular TE
previously [26, 44, 46]. events, 68 (0.39%) and 52 (0.27%) were reported to have
had cerebrovascular TE events, and 17 (0.10%) and 20
Results (0.10%) were reported to have had peripheral venous
thrombotic events, respectively. The proportion of events
The characteristics of both study cohorts are presented in excluded from the study as defined in the methods for
Table 2. As described previously [26], where reported both drugs were similar [celecoxib 56.8% (149/262)
celecoxib users were more likely than meloxicam users to versus meloxicam 51.0% (95/186); 2 P ¼ 0.225].
be aged 60 yr or more [59.7% (6378/10 727) vs 55.0% Regarding the time to first event, the crude estimate of
(9280/16 877), 2 P < 0.0001] and be female [68.3% time to first event of both cohorts for each event group
(11 928/17 455) vs 67.1% (12 590/18 763), 2 P ¼ 0.012]. separately is presented in the form of Kaplen–Meier
OA was the most frequently reported indication for both survival curves in Figures 1–3. There was a significant
celecoxib and meloxicam, although the proportion was difference in the estimate of time to first event over the
higher for celecoxib compared to meloxicam, respec- study period in the cardiovascular TE event group for
 Thromboembolic events after taking celecoxib and meloxicam 1357

TABLE 2. Characteristics of study cohort

Drug

Risk factor Celecoxib (N ¼ 17 458) Meloxicam (N ¼ 19 087) 2 P valuea

Age (yr)

39 871 (5.0) 1852 (9.7) <0.0001
40–49 1281 (7.3) 2297 (12.0)
50–59 2197 (12.6) 3448 (18.1)
60–69 2582 (14.8) 3947 (20.1)
70–79 2467 (14.1) 3457 (18.1)
80þ 1329 (7.6) 1876 (9.8)
Not known 6731 (38.6) 2210 (11.6)
Sex
Male 5527 (31.7) 6173 (32.3) 0.012
Female 11 928 (68.3) 12 590 (67.1)
Sex not known 3 (<0.1) 324 (1.7)
Indication
Osteoarthritis 4905 (28.1) 4434 (23.2) <0.0001
Others 12 553 (71.9) 14 653 (76.8)
NSAID prescribed within 3 months prior to starting drug
Yes 7006 (40.1) 7978 (41.8) 0.014
No 7189 (41.2) 8658 (45.4)
Not known 3263 (18.7) 2451 (12.9)

All values are n (%).

a
Excludes values not known.

FIG. 1. Kaplan–Meier survival estimates for cardiovascular

TE events, between celecoxib and meloxicam cohorts. FIG. 3. Kaplan–Meier survival estimates for peripheral
venous thrombotic events, between celecoxib and meloxicam
cohorts.

celecoxib compared to meloxicam [median pte 75.5 days

FIG. 2. Kaplan–Meier survival estimates for cerebrovascular
TE events between celecoxib and meloxicam cohorts.
(interquartile range (IQR), 39.5–145.5) and 95 days
(IQR, 38–108), respectively, log rank test P ¼ 0.0153;
Fig. 1], with the curves separating 30 days after starting
treatment and no significant convergence after that time.
There was also a significant difference in the time to first
cerebrovascular TE event for celecoxib compared to
meloxicam [median pte 105 days (IQR 40–171.5) and 100
days (IQR 23.5–140.5), log rank test P ¼ 0.0055; Fig. 2],
with the survival curves separating 30 days after starting
treatment and no significant convergence after that time.
There was no difference observed in the time to first
peripheral venous thrombotic event between the study
drugs [median pte 103 days (IQR 28–186) and 118.5 days
(IQR 48.5–178), log rank test P ¼ 0.7930; Fig. 3], which is
1358 D. Layton et al.
reflected by the survival curves, one almost superimposed
upon the other.
Cross-tabulation of risk factors with event groups
suggested a significant association between age and
experiencing cardiovascular TE (2 P < 0.0001), cere-
brovascular TE (2 P < 0.0001) and peripheral venous
thrombotic events (2 P ¼ 0.017); between sex and
experiencing cardiovascular TE events (2 P < 0.0001)
and between indication and experiencing cerebrovas-
cular TE events (2 P ¼ 0.017) and peripheral venous
thrombotic events (2 P ¼ 0.037). Use of a NSAID
within the 3 months prior to starting treatment was not
associated with any of the event groups.
At launch and at the time of the PEM studies, the
recommended dose range of celecoxib was 100–400
mg/day, whilst meloxicam only had two doses licensed
(7.5 or 15 mg/day). As reported previously [26], limited
information on dose was provided on the green forms
for meloxicam. For celecoxib, information on starting
dose was available for 84.4% (n ¼ 14 726). Dose at
event was provided for 57.1% (16/28) of patients
reported to have cardiovascular TE events, 75.0%
(51/68) of patients reported to have cerebrovascular TE
events and 90.9% (10/17) of patients reported to have
peripheral venous thrombotic events. Of these, 15
(93.8%), 49 (96.1%) and seven (70.0%) patients,
respectively were taking 200 mg/day or less. As the
reporting of dose data was low, it was not adjusted for
in the multivariate analysis.
Table 3 shows crude event rates per 1000 person-
years (pyr) for both drug cohorts over the first 9
months of treatment and RRs for each risk factor
category. We reported earlier that age is associated
with each of these event groups. Patients from the
celecoxib cohort aged 80 yr or more have the highest
(but not statistically significantly different) rate of
experiencing cardiovascular or cerebrovascular TE
events compared to the younger age groups (age 80
yr or more treated as the reference group), and the
lowest (but not statistically significantly different) rate
of peripheral venous thrombotic events. Conversely,
patients from the meloxicam cohort aged 80 yr or more
had the lowest rates (but not statistically significantly
different) of cardiovascular TE events than the younger
age groups. As reported previously [26], the 95% CIs
for this event group for meloxicam were wide, thus one
cannot exclude the possibility of a similar relationship
to that observed for celecoxib. The age-specific
estimates of RRs obtained via stepwise comparison of
the age-specific rates indicated that these relationships
were not linear for either drug, with no systematic
difference in the age-specific rates for each of the
three event groups (tests for effect modification, all 2
P > 0.1493). A between-drug comparison revealed that
for cerebrovascular TE events, celecoxib users aged 80
yr or more were more likely to have these events than
meloxicam users of the same age [RR 2.59 (95%CI
1.17–5.70)].
Females tended towards a lower rate of experiencing
the selected TE events than males (treated as the
reference group), although statistical significance was
only observed for users of either drug experiencing
cardiovascular TE events. There was no evidence of a
sex–drug interaction for any of the event groups (tests
for effect modification, all 2 P > 0.063). Examination
of drug-indication specific rates revealed no statistically
significant effect on the rate of any of the event groups
within each cohort, nor did a prescription of an
NSAID within 3 months of starting treatment
(compared to ‘none’ treated as the reference group).
Furthermore, there was no evidence of a drug–risk
factor interaction (tests for effect modification, all
2 P > 0.2032).
The crude and adjusted RRs are presented in Table
4. As reported previously [26], indication and prescrip-
tion of an NSAID within 3 months of starting
treatment were initially regarded as important risk
factors in this study; however, adjusting for these
variables made no statistically significant difference to
the RR estimates. Thus age and gender were the two
risk factor variables included in the final Poisson
regression model. Adjusting for these two risk factors
of age (also as a quadratic variable age2) and sex
suggests that a difference exists between subjects
prescribed either of the two drugs and the rate of
experiencing cerebrovascular TE events; the adjusted
rate was higher for celecoxib than meloxicam [RR 1.66
(95% CI 1.10–2.51)]. With regard to cardiovascular TE
events and peripheral venous thrombotic events the
difference did not achieve statistical significance.
Evidence of effect modification was further examined
by inclusion of interaction terms within the final
Poisson models for the three groups. An age–drug
interaction was identified in the model predicting the
estimate of peripheral venous thrombotic events
(likelihood ratio test, 2 P ¼ 0.0128). The age and sex
adjusted estimate for celecoxib users aged 50–59 yr was
higher compared to meloxicam users of the same age,
but this did not achieve statistical significance [RR 3.35
(95% CI 0.30–37.00)]. Conversely celecoxib users aged
60–69, 70–79 or 80 yr or more had a lower relative risk
of these events than meloxicam users of the same age
categories, but again these were not statistically
different [RR 0.76 (95% CI 0.23–2.46), RR 0.79
(95% CI 0.20–3.17) and RR 0.55 (95% CI 0.06–5.26),
respectively].
The effect of the 25.2% reduction in the total
number of observations when fitting the final Poisson
model [n ¼ 36 545 vs 27 329] was also examined, where
statistically significant differences were found. The
relative risk estimates for each group calculated with
removal of subjects with missing values for the
adjusting variables sex and age are also shown in
Table 4. The unadjusted model for cardiovascular TE
events fitted using the full model was of borderline
statistical significance and fitting the model to the
reduced dataset [n ¼ 27 329] led to a non-significant
difference. However, this change was unlikely to have
an impact on our findings. Furthermore, adjusting for
age and sex had no statistically significant effect on
TABLE 3. Crude ratesa and RRsa of thromboembolic (cardiovascular, cerebrovascular and peripheral venous thrombotic) events, per 1000 pyr by risk factor

Celecoxib (n ¼ 17 358) Meloxicam (n ¼ 19 087)

Cardiovascular (n ¼ 28) Cerebrovascular (n ¼ 68) Peripheral venous (n ¼ 17) Cardiovascular (n ¼ 19) Cerebrovascular (n ¼ 52) Peripheral venous (n ¼ 20)

Rate Rate ratio Rate Rate ratio Rate Rate ratio Rate Rate ratio Rate Rate ratio Rate Rate ratio
Risk factor n (95% CI) (95% CI) n (95% CI) (95% CI) n (95% CI) (95% CI) n (95% CI) (95% CI) n (95% CI) (95% CI) n (95% CI) (95% CI)

Age (yr)

39 0 – – 0 – – 2 8.3 3.8 0 – – 4 6.5 0.5 – –
(2.1, 33.2) (0.2, 223.2) (2.4, 17.2) (0.2, 1.5)
40–49 0 – – 1 2.5 0.1 0 – – 0 – – 2 2.3 0.2 – –
(0.4, 18.0) (0.0, 0.5) (0.6, 9.2) (0.0, 0.8)
50–59 1 1.3 0.2 4 5.3 0.1 2 2.8 1.2 0 – – 2 1.5 0.1 1 0.7 0.2
(0.2, 9.4) (0.0, 1.5) (2.0, 14.1) (0.0, 0.5) (0.7, 10.5) (0.1, 70.8) (0.4, 5.8) (0.0, 0.5) (0.1, 5.2) (0.0, 1.7)
60–69 5 5.5 0.6 8 8.8 0.2 4 4.4 2.0 9 5.6 4.2 9 5.7 0.4 9 5.7 1.4
(2.3, 13.2) (0.1, 3.2) (4.4, 17.6) (0.1, 0.6) (1.6, 1.17) (0.2, 98.3) (2.9, 10.9) (0.5, 32.7) (2.9, 10.9) (0.2, 1.0) (2.9, 10.9) (0.4, 5.1)
70–79 7 8.0 0.9 16 18.2 0.5 3 3.4 1.5 6 4.4 3.2 18 13.1 1.0 6 4.4 1.1
(3.8, 16.7) (0.2, 4.2) (11.2, 29.7) (0.2, 1.0) (1.1, 10.6) (0.1, 81.3) (2.0, 9.7) (0.4, 26.7) (8.3, 20.9) (0.4, 2.1) (2.0, 9.7) (0.3, 4.3)
80þ 4 8.8 1.0 16 35.3 1.0 1 2.2 1.0 1 1.4 1.0 10 13.6 1.0 3 4.1 1.0
(3.3, 23.4) (21.6, 57.6) (0.3, 15.6) (0.2, 9.7) (7.3, 25.4) (1.3, 12.7)
Not known 11 – – 23 – – 5 – – 3 – – 7 – – 1 – –
Sex
Male 15 8.1 1.0 22 11.8 1.0 6 3.2 1.0 13 5.3 1.0 20 8.2 1.0 9 3.7 1.0
(4.9, 13.4) (7.8, 18.0) (1.4, 7.2) (3.1, 9.2) (5.3, 12.7) (1.9, 7.1)
Female 13 3.2 0.4 46 11.35 0.9 11 2.7 0.8 6 1.2 0.2 32 6.5 0.8 11 2.2 0.6
(1.9, 5.5) (0.2, 0.8) (8.5, 15.2) (0.6, 1.3) (1.5, 4.9) (0.3, 2.3) (0.5, 2.7) (0.1, 0.6) (4.6, 9.2) (0.5, 1.4) (1.2, 4.0) (0.3, 1.5)
Not known 0 – – 0 – – 0 – – 0 – – 0 – – 0 – –
Indication
Other 17 4.2 1.0 39 9.6 1.0 10 2.5 1.0 16 2.8 1.0 39 6.9 1.0 12 2.1 1.0
(2.6, 6.7) (7.0, 13.2) (1.3, 4.6) (1.7, 4.6) (5.1, 9.5) (1.2, 3.8)
Osteoarthritis 11 5.9 1.4 29 15.6 1.4 7 3.8 1.5 3 1.6 0.6 13 6.9 1.0 8 4.3 2.0
(3.3, 10.7) (0.7, 3.0) (10.9, 22.5) (1.0, 2.0) (1.8, 7.9) (0.6, 4.0) (0.5, 5.0) (0.2, 2.0) (4.0, 11.9) (0.5, 1.9) (2.1, 8.5) (0.8, 4.9)
NSAIDb
No 11 5.0 1.0 28 12.8 1.0 4 1.8 1.0 10 3.3 1.0 22 7.3 1.0 8 2.7 1.0
Thromboembolic events after taking celecoxib and meloxicam

(2.8, 9.1) (8.9, 18.6) (0.7, 4.9) (1.8, 6.2) (4.8, 11.1) (0.3, 5.3)
Yes 14 5.2 1.0 30 11.2 0.9 11 4.1 2.2 6 1.8 0.5 23 3.4 0.9 10 2.9 1.1
(3.8, 8.8) (0.5, 2.3) (7.8, 15.9) (0.6, 1.3) (2.3, 7.4) (0.7, 7.0) (0.8, 3.9) (0.2, 1.5) (6.7, 4.5) (0.5, 1.7) (1.6, 5.4)
Not known 3 – – 10 – – 2 – – 3 – – 7 – – 3 – –

a
Rates and rate ratio calculated using Poisson regression modelling.
b
Prescribed
3 months to starting drug.
1359
1360 D. Layton et al.

1.66 (1.10, 2.51)

1.72 (0.87, 3.40)

1.06 (0.51, 2.19)

Adjusted RRc
cardiovascular risk in this study. The estimate for

(95% CI)
cerebrovascular TE event was inflated slightly after
fitting the model using the reduced dataset, but would
not be sufficient to account for the relative rate
increase observed for cerebrovascular TE events with
TABLE 4. Crude and adjusted RRs of thromboembolic cardiovascular, cerebrovascular and peripheral venous thrombotic events in users of celecoxib compared to meloxicam

celecoxib compared to meloxicam.

The effect of treatment duration was also examined
by restricting the study period to the first 90 days after
Unadjusted RRb

1.77 (1.18, 2.69)

1.89 (0.95, 3.74)

1.12 (0.54, 2.31)

starting either drug for each event group. During this
(95% CI)

period, 15 (0.09%) and nine (0.05%) patients were

reported to have cardiovascular TE events, 31 (0.18%)
and 22 (0.12%) were reported to have cerebrovascular
TE events and seven (0.04%) and nine (0.0.05%) were
reported to have peripheral venous thrombotic events
for celecoxib and meloxicam, respectively. For cardio-
Unadjusted RRa

1.66 (1.16, 2.38)

1.87 (1.04, 3.35)

1.08 (0.56, 2.06)

vascular TE events, the final adjusted estimate to 90

(95% CI)

days was RR 1.60 (95% CI 0.64–4.10), respectively, for

celecoxib compared to meloxicam. The corresponding
estimates for cerebrovascular TE events and peripheral
venous thrombotic events were RR 1.72 (95% CI
0.95–3.13) and RR 1.11 (95% CI 0.38–3.11), respec-
tively. For these two groups the 95% CI reported for
the adjusted RR did not change to such a degree that
0.69 (0.52, 0.90)
2.53 (1.61, 3.97)

2.66 (1.72, 4.13)

might indicate that time may contribute a confounding

(95% CI)

effect.
Rate

Discussion
Meloxicam

The PEM observational studies of celecoxib and

meloxicam enable a retrospective comparison between
1000 pyr exposure

a highly selective and a partially selective COX-2

No events

inhibitor type of NSAID to be performed under general

Poisson regression model excluding patients where age and sex not known (n ¼ 9216).
52/7.50
19/7.50

20/7.51

practice conditions. Data was collected for over 36 000

patients prescribed meloxicam and/or celecoxib under
routine clinical practice conditions. In this study, a
statistically significant relative rate increase was observed
for cerebrovascular TE events 1.66 (95% CI 1.10–2.51)
for celecoxib compared to meloxicam, after adjustment
11.51 (9.07, 14.59)

for age (also as a quadratic variable age2) and sex. This

4.73 (3.26, 6.85)

2.87 (1.79, 4.62)

relative difference (of the same magnitude) was also

(95% CI)
Rate

observed in the previous study [26]. A reduction in the

Poisson regression model adjusted for age (age2) and sex.

sample size (occurring as a result of including variables

with missing values in the statistical model) is unlikely to
account for this relative rate increase alone. As observed
Celecoxib

in the previous study [26], restriction of the study period

to the first 90 days after starting treatment revealed no
Poisson regression model (whole data set).
1000 pyr exposure

statistically significant difference in the relative risk

estimates for each of the three event groups. Examina-
No events/

68/5.91
28/5.92

17/5.92

tion of the effect of time suggested that differences

between celecoxib and meloxicam for cardiovascular or
cerebrovascular TE events became apparent after 30 days
of starting treatment, which persisted for the subsequent
study period. Unlike the previous study there was no
difference in the time to first peripheral venous
thrombotic event between celecoxib and meloxicam,
Peripheral venous
Cerebrovascular
Cardiovascular

and no difference in relative risk difference was observed

[RR 1.06 (95% CI 0.51–2.19)]. Clearly the temporal
relationships need to be examined further.
Event

In both of these studies, indication and recent use of

b
a

NSAIDs had no important effect on the relative rates

 Thromboembolic events after taking celecoxib and meloxicam
of any of the event groups. We acknowledge that
information on co-morbidity, such as past medical
history of cardiovascular disease, recent surgery and
lifestyle factors (e.g. smoking) is important. However,
the effect of these and other risk factors for TE events
were not controlled for in this study. As reported
previously [26], age and sex are known to be strong
confounders [41, 43, 47]. Both of these studies support
this relationship, and again reports that older patients
are at greater risk of cerebrovascular TE events.
Regarding peripheral venous thrombotic events, there
was evidence of an age–drug interaction on the relative
risk estimate, in that women aged 50–59 yr were at
higher but non-significant risk of these events if
prescribed celecoxib rather than meloxicam, but a
lower (non-significant) risk if aged 60 yr or more. We
discussed previously [26] that the risk factors for
peripheral venous thrombotic events are different to
those for cardiovascular or cerebrovascular events, but
because the number of events recorded for each groups
is small it is possible that this finding may have
occurred by chance.
A strength of this study is that the comparison of
celecoxib with meloxicam takes account of the similar
baseline risk of adverse gastrointestinal events of
the two cohorts who may have been preferentially
prescribed these drugs because of their reported
improved gastrointestinal (GI) tolerability [44, 48]. As
for the first study [26], a channelling effect of past users
of NSAIDs onto celecoxib exists, in that more patients
within this cohort had been prescribed an NSAID
within the 3 months prior to starting treatment. Such
channelling effects of groups at high risk of GI adverse
events have been reported previously for both these
agents [48, 49]. We stated in the comparison between
rofecoxib and meloxicam examining TE cardiovascular
risk [26] that we did not examine the effect of this
phenomenon, but we acknowledge that there may also
be channelling of patients at high-risk of cardiovas-
cular events. In PEM, incomplete information on risk
factors predisposing patients to these types of events,
such as past history of cardiovascular disease or
lifestyle factors, is available and thus the confounding
effect of these and other risk factors could not be
controlled for. It is important to stress that the quality
of the data in PEM is dependent on the precision and
completeness of the form-filling by GPs.
We highlighted many of the limitations of PEM in the
previous paper [26]. While the collection of data in PEM
is of a systematic and prospective nature, it often results
in incomplete information on concomitant medication
that may later prove to be important risk factors for
selected events [26, 44]. In our studies, information on
concomitant aspirin use was incomplete and thus the
confounding effect of aspirin could not be controlled for.
The VIGOR study [4] and CLASS trial [6] differed in
several aspects [50]; importantly, in the CLASS trial
patients were permitted to take prophylactic low dose
aspirin (<325 mg/day) or other antiplatelet agents [51].
Furthermore, only patients with RA were enrolled in
1361
the VIGOR study, whereas in the CLASS trial the
proportions of RA and OA patients were 28 and 72%,
respectively, leading to differences in baseline risk of
patients. Data from CLASS suggested no evidence of
signals of any increased risk of cardiovascular events,
including myocardial infarction (MI) and angina for
celecoxib users. Separate analyses were performed for all
patients and those not taking aspirin. The incidence rate
of serious cardiovascular TE events out of 3987 patients
taking celecoxib was reported as: MI (0.8 per 100 pyr),
cerebrovascular accident (CVA) (0.2 per 100 pyr). There
was evidence of important differences among this group
and the other treatment groups. The relative risks for any
serious cardiovascular TE event were 1.1 (95% CI
0.7–1.6) for all patients and 1.1 (95% CI 0.6–1.9) for the
subgroup not taking aspirin, for celecoxib versus
NSAID comparator drugs. Furthermore, no difference
was observed when the TE events were aggregated into
cardiac (fatal/non-fatal) or peripheral vascular events
(fatal/non-fatal), but celecoxib was associated with fewer
CVA than diclofenac/ibuprofen (0.2 vs 0.5%, respec-
tively, P < 0.05). No difference was reported in the
subgroups not taking aspirin. Thus, these analyses
demonstrated no increased risk of serious cardiovascular
TE events associated with celecoxib compared to
conventional NSAIDs.
Our justification for choice of event terms for this
comparative study has been presented previously [26].
Our study reflects the findings from CLASS, with the
exception of the increased relative risk of cerebrovascular
events for celecoxib compared to meloxicam. However, it
is noteworthy that while the adjusted RR for cerebro-
vascular TE events in our study was 1.66, the lower end
of the 95% CI was 1.10, marginally greater than the null
estimate which could have resulted from bias or chance.
One cannot exclude the possibility that the variation in
COX-1/COX-2 selectivity of the two drugs at the clinical
dosing regimes used may have undetermined effects on
vascular haemostasis, however this is beyond the scope of
this study.
The restrospective, observational cohort study by Ray
et al. [52], used data collected from the expanded
Tennesse Medicaid programme, TennCare, to investi-
gate the occurrence of serious coronary heart disease in
non-users (n ¼ 202 916) and users of rofecoxib
(n ¼ 24 132), celecoxib (n ¼ 22 337) and other NSAIDs
(n ¼ 129 391), aged 50–84 yr, who lived in the community
and had no life-threatening non-cardiovascular illness.
New users of high-dose rofecoxib (>25 mg/day) had a
rate of serious coronary heart disease (CHD) events
(hospital admission for acute MI or death from CHD) of
24.0 per 1000 pyr, new users of low-dose rofecoxib
(
25 mg/day) had a rate of 13.7, new users of celecoxib
had a rate of 12.2, and non-users had a rate of 13.0. The
adjusted incidence RR for high-dose rofecoxib was 2.20
(95% CI 1.17–4.10) compared to celecoxib, 1.93 (95% CI
1.09–3.43) compared to non-users. The corresponding
adjusted estimate for celecoxib compared to non-users
was 0.88 (95% CI 0.67–1.16). In our study the rate of
cardiovascular events (MI, cardiac arrest and relevant
1362 D. Layton et al.
non-specific cardiovascular events) was lower than the
observed rate in the study by Ray et al. [52], but this is to
be expected since this study uses event data reported by
GPs. While a large number of events had been followed
up requesting further information, it is possible that
some events may either not have been reported or been
reported incorrectly. Nevertheless, we have no reason to
believe that there was differential under-reporting
between the two products. We acknowledge that the
two PEM studies were conducted during different
calendar periods, and the study for celecoxib was
initiated at the same time as the results of the VIGOR
study were published. It is possible that publication bias
might have influenced the reporting of such events
between the two drugs, and this issue requires further
investigation.
By May 2002, four reports of MI (one fatal), one
report of pulmonary embolism (PE) (fatal) and two
reports of CVA had been reported for celecoxib to the
Medicines Control Agency in the UK via the sponta-
neous reporting system of suspected adverse drug
reactions. No reports had been received for deep vein
thrombosis (DVT) by that time. Information available
from 51 case histories of TE events (CVA, MI, DVT and
PE) reported in the celecoxib PEM study [40], showed
that 76.1% (35/46 where age was specified) were >65 yr,
86.6% (39/45) had risk factors for thromboembolism/
CHD and 43.5% (20/46) were on concomitant aspirin
or other anticoagulant/antiplatelet agents. There is no
evidence from the PEM data currently available to
suggest that any cardiovascular deaths were attributable
to celecoxib. Interestingly, there are published reports
of venous thrombosis, possible PE and arterial throm-
bosis in four patients with connective tissue disorders
receiving celecoxib, which suggests that patients with
diseases that predispose to thrombosis may be at greater
risk of peripheral vascular events [21]. As mentioned
previously, in PEM there is incomplete information
available on risk factors predisposing patients to these
types of events.
Conclusion
Our understanding of the clinical effects of COX-2
inhibitors is still evolving. Our observational study has
demonstrated that the age- and sex-adjusted relative rate
of cerebrovascular TE events for celecoxib compared to
meloxicam in this cohort was 1.66 (95% CI 1.10–2.51)
over the study period of 270 days. There was no
significant difference in the rate of cardiovascular TE
events or peripheral venous thrombotic events.
The debate as to whether the association between an
increased risk of thrombosis and the use of COX-2
inhibitors is a ‘class effect’, continues. It is possible that
COX-2 inhibitors may possess different pharmacological
characteristics and some of these differences may be dose
related, however, the implications of these differences
remain unclear. With regard to the clinical implications
of our findings, the results of our study are only useful
when considered together with other studies seeking to
determine the association between the use of COX-2
inhibitors and TE events.
Conflict of interest
The DSRU is an independent charity, which works in
association with the University of Portsmouth. It
receives unconditional donations from pharmaceutical
companies. The companies have no control on the
conduct or the publication of the studies conducted by
the DSRU. The DSRU has received such funds from
the manufacturers of celecoxib and meloxicam. S. A.
W. Shakir has received support from Pfizer to attend
scientific meetings.
References
1. Jackson LM, Hawkey CJ. COX-2 selective nonsteroidal
anti-inflammatory drugs: do they really offer any
advantages? Drugs 2000;59:1207–16.
2. Hawkey C, Laine L, Simon T et al. Comparison of the
effect of rofecoxib (a cyclooxygenase 2 inhibitor),
ibuprofen, and placebo on the gastroduodenal mucosa
of patients with osteoarthritis: a randomized, double-
blind, placebo-controlled trial. The Rofecoxib
Osteoarthritis Endoscopy Multinational Study Group.
Arthritis Rheum 2000;43:370–7.
3. Laine L, Harper S, Simon T et al. A randomized trial
comparing the effect of rofecoxib, a cyclooxygenase 2-
specific inhibitor, with that of ibuprofen on the
gastroduodenal mucosa of patients with osteoarthritis.
Rofecoxib Osteoarthritis Endoscopy Study Group.
Gastroenterology 1999;117:776–83.
4. Bombardier C, Laine L, Reicin A et al. Comparison of
upper gastrointestinal toxicity of rofecoxib and naproxen
in patients with rheumatoid arthritis. VIGOR Study
Group. N Engl J Med 2000;343:1520–8. [Plus 2 pages
following 1528.]
5. Schoenfeld P. Gastrointestinal safety profile of melox-
icam: a meta-analysis and systematic review of random-
ized controlled trials. Am J Med 1999;107:48S–54S.
6. Silverstein FE, Faich G, Goldstein JL et al.
Gastrointestinal toxicity with celecoxib vs nonsteroidal
anti-inflammatory drugs for osteoarthritis and rheuma-
toid arthritis: the CLASS study: A randomized controlled
trial. Celecoxib Long-term Arthritis Safety Study. J Am
Med Assoc 2000;284:1247–55.
7. FitzGerald GA, Cheng Y, Austin S. COX-2 inhibitors
and the cardiovascular system. Clin Exp Rheumatol
2001;19 (6 Suppl. 25):S31–6.
8. Food and Drug Administration advisory committee.
Cardiovascular safety review of rofecoxib. 2001: http:/
www.fda.gov/ohrms/dockets/ac/01/briefing/3677b2_06_
cardio.pdf.
9. Seibert K, Masferrer JL, Fu JY, Honda A, Raz A,
Needleman P. The biochemical and pharmacological
manipulation of cellular cyclooxygenase (COX) activity.
Adv Prostaglandin Thromboxane Leukot Res
1991;21A:45–51.
10. Brooks P, Emery P, Evans JF et al. Interpreting the
clinical significance of the differential inhibition of
 Thromboembolic events after taking celecoxib and meloxicam
cyclooxygenase-1 and cyclooxygenase-2. Rheumatology
1999;38:779–88.
11. Lipsky PE, Abramson SB, Breedveld FC et al. Analysis
of the effect of COX-2 specific inhibitors and
recommendations for their use in clinical practice.
J Rheumatol 2000;27:1338–40.
12. Catella-Lawson F. Vascular biology of thrombosis:
platelet-vessel wall interactions and aspirin effects.
Neurology 2001;57 (5 Suppl. 2):S5–7.
13. Kam PC, See AU. Cyclo-oxygenase isoenzymes: physio-
logical and pharmacological role. Anaesthesia
2000;55:442–9.
14. Brochier ML. Evaluation of flurbiprofen for prevention of
reinfarction and reocclusion after successful thrombolysis
or angioplasty in acute myocardial infarction. The
Flurbiprofen French Trial. Eur Heart J 1993;14: 951–7.
15. Ray WA, Stein CM, Hall K, Daugherty JR, Griffin MR.
Non-steroidal anti-inflammatory drugs and risk of
serious coronary heart disease: an observational cohort
study. Lancet 2002;359:118–23.
16. Garcia Rodriguez LA, Varas C, Patrono C. Differential
effects of aspirin and non-aspirin nonsteroidal antiin-
flammatory drugs in the primary prevention of myocar-
dial infarction in postmenopausal women. Epidemiology
2000;11:382–7.
17. Freston JW. Rationalizing cyclooxygenase (COX) inhibi-
tion for maximal efficacy and minimal adverse events.
Am J Med 1999;107:78S–88S.
18. McAdam BF, Catella-Lawson F, Mardini IA, Kapoor S,
Lawson JA, FitzGerald GA. Systemic biosynthesis of
prostacyclin by cyclooxygenase (COX)-2: the human
pharmacology of a selective inhibitor of COX-2. Proc
Natl Acad Sci USA 1999;96:272–7.
19. Leese PT, Hubbard RC, Karim A, Isakson PC, Yu SS,
Geis GS. Effects of celecoxib, a novel cyclooxygenase-2
inhibitor, on platelet function in healthy adults: a
randomized, controlled trial. J Clin Pharmacol
2000;40:124–32.
20. Linder JD, Monkemuller KE, Davis JV, Wilcox CM.
Cyclooxygenase-2 inhibitor celecoxib: a possible cause of
gastropathy and hypoprothrombinemia. South Med J
2000;93:930–2.
21. Crofford LJ, Oates JC, McCune WJ et al. Thrombosis in
patients with connective tissue diseases treated with
specific cyclooxygenase 2 inhibitors. A report of four
cases. Arthritis Rheum 2000;43:1891–6.
22. Haase KK, Rojas-Fernandez CH, Lane L, Frank DA.
Potential interaction between celecoxib and warfarin.
Ann Pharmacother 2000;34:666–7.
23. Anon. Celecoxib: early Australian reporting experience.
Aust Adv Drug Reactions Bull 2000;19:6–7.
24. Noble S, Balfour JA. Meloxicam. Drugs 1996;51:424–30.
25. Panara MR, Renda G, Sciulli MG et al. Dose-dependent
inhibition of platelet cyclooxygenase-1 and monocyte
cyclooxygenase-2 by meloxicam in healthy subjects.
J Pharmacol Exp Ther 1999;290:276–80.
26. Layton D, Heeley E, Hughes K, Shakir SA. Comparison
of the incidence rates of thromboembolic events reported
for patients prescribed rofecoxib and meloxicam in
general practice in England using prescription-event
monitoring data. Rheumatology 2003;42:622–31.
27. Kaplan-Machlis B, Klostermeyer BS. The cyclooxygen-
ase-2 inhibitors: safety and effectiveness. Ann
Pharmacother 1999;33:979–88.
28. Wojtulewski JA, Schattenkirchner M, Barcelo P et al.
A six-month double-blind trial to compare the efficacy
1363
and safety of meloxicam 7.5 mg daily and naproxen
750 mg daily in patients with rheumatoid arthritis. Br J
Rheumatol 1996;35(Suppl. 1):22–8.
29. Lemmel EM, Bolten W, Burgos-Vargas R et al. Efficacy
and safety of meloxicam in patients with rheumatoid
arthritis. J Rheumatol 1997;24:282–90.
30. Distel M, Mueller C, Bluhmki E, Fries J. Safety of
meloxicam: a global analysis of clinical trials. Br J
Rheumatol 1996;35(Suppl. 1):68–77.
31. Mukherjee D, Nissen SE, Topol EJ. Risk of cardiovas-
cular events associated with selective COX-2 inhibitors.
J Am Med Assoc 2001;286:954–9.
32. Konstam MA, Weir MR, Reicin A et al. Cardiovascular
thrombotic events in controlled, clinical trials of
rofecoxib. Circulation 2001;104:2280–8.
33. Zhao SZ, Reynolds MW, Lejkowith J, Whelton A,
Arellano FM. A comparison of renal-related ad-
verse drug reactions between rofecoxib and celecoxib,
based on the World Health Organization/Uppsala
Monitoring Centre safety database. Clin Ther 2001;
23:1478–91.
34. Mann RD. Prescription-event monitoring–recent progress
and future horizons. Br J Clin Pharmacol 1998;46:
195–201.
35. Shakir SA. PEM in the UK. In: Mann RD, Andrews E,
eds. Pharmacovigilance, Chichester, UK: John Wiley &
Sons, 2002:333–44.
36. Legemaate J. The CIOMS guidelines for biomedical
research involving human subjects. Eur J Health Law
1994;1:161–5.
37. Royal College of Physicians of London. Guidelines on
the practice of Ethical Committees in Medical Research
involving Human Subjects. 1996.
38. Multi-Centre Research Ethics Committees Guidance
Notes. Examples of enquiries and surveys in the public
interest where no reference to a Research Ethics
Committee is necessary. Appendix C, 2000. http://www.
corec.org.uk/wordDocs/Guidenotes.doc
39. Martin RM, Biswas P, Mann RD. The incidence of
adverse events and risk factors for upper gastrointestinal
disorders associated with meloxicam use amongst 19,087
patients in general practice in England: cohort study. Br J
Clin Pharmacol 2000;50:35–42.
40. Layton D, Wilton LW, Shakir SA. Celecoxib. PEM
report No.74. 2002.
41. Mann RD, Rawlins MD, Fletcher P. Age and the
spontaneous reporting of adverse reactions in the
United Kingdom. Pharmacoepidemiol Drug Saf
1992;1:19–23.
42. Peterson S, Rayner M. Coronary heart disease statistics.
London: British Heart Foundation, 2002.
43. Laine L. Approaches to nonsteroidal anti-inflammatory
drug use in the high-risk patient. Gastroenterology
2001;120:594–606.
44. Layton D, Heeley E, Hughes K, Shakir S. Comparison of
the incidence rates of selected gastrointestinal events
reported for patients prescribed rofecoxib and
meloxicam in general practice in England using prescrip-
tion-event monitoring (PEM) data. Rheumatology 2003;
Advance Access publication: June 30, 2003, DOI:
10.1093/rheumatology/keg379.
45. Martindale 32nd edition. London: Pharmaceutical Press,
2002.
46. Layton D, Hughes K, Harris S, Shakir SA. Comparison
of the incidence rates of selected gastrointestinal
events reported for patients prescribed celecoxib and
1364 D. Layton et al.
meloxicam in general practice in England using prescrip-
tion-event monitoring (PEM) data. Rheumatology 2003;
Advance Access publication: June 16, 2003, DOI:
10.1093/rheumatology/keg376.
47. Raskin JB. Gastrointestinal effects of nonsteroidal anti-
inflammatory therapy. Am J Med 1999;106:3S–12S.
48. Lanes S, Garcia Rodriguez LA, Hwang E. Baseline risk
of gastrointestinal disorders among new users of melox-
icam, ibuprofen, diclofenac, naproxen and indomethacin.
Pharmacoepidemiol Drug Saf 2002;9:113–17.
49. Layton D, Riley J, Wilton LW, Shakir S. Safety profile
of rofecoxib as used in general practice in England:
results of a Prescription-Event Monitoring study. Br J
Clin Pharmacol 2002;55:166–74.
50. Strand V, Hochberg MC. The risk of cardiovascular
thrombotic events with selective cyclooxygenase-2 inhibi-
tors. Arthritis Rheum 2002;47:349–55.
51. The physicians’ health study: aspirin for the primary
prevention of myocardial infarction. N Engl J Med
1988;318:924–6.
52. Ray W, Stein C, Daugherty J, Hall K, Arbogast P,
Griffin M. COX-2 selective non-steroidal anti-inflamma-
tory drugs and risk of serious coronary heart disease.
Lancet 2002;360:1071.