Are selective COX 2 inhibitors superior to traditional non steroidal anti-inflammatory drugs?

Dear Editor

We welcome the opportunity to answer Dr Geis' response to our editorial [1].

According to Dr Geis, CLASS was a single study, and it was presented as such
in the JAMA article, which reported patients to be "randomly assigned on a
2:1:1 basis" [2]. This description is misleading. As acknowledged by Dr
Geis, there were two separate trials with two separate patient recruitment and
randomisation procedures. In this situation, any analysis of the data should
preserve randomisation. However, CLASS' authors combined data from the two
trials by simply adding up numbers [2]. The assumption of this approach is that
the process of allocating patients to the two trials was ruled by chance alone.
We tested this assumption by analysing the 18 patient characteristics at
baseline that were available [3] and could be compared between protocols. If the
allocation to the two trials had been governed by chance alone then we would
have expected about one of the 18 comparisons to be statistically significant at
P<0.05. However, patients in trial # 102 comparing celecoxib versus
diclofenac were older (P=0.015), had higher global assessment scores
(P<0.0001), had lower pain scores (P=0.01), were more frequently white
(P<0.0001), were more likely to have a history of gastrointestinal NSAID
intolerance (P<0.0001) and were more frequently alcohol users (P<0.0001)
than patients in trial # 035 comparing celecoxib against ibuprofen. The
probability that these differences have occurred by chance alone is rather small
(P<1 x 10-19). Considering these differences as well as different follow-up
durations, and different COX-2 selectivities [4] and gastro- intestinal
side-effects [5] of the comparator drugs, simple pooling of the data by adding
up is clearly misleading and potentially drowns important differences in effect
estimates between the two trials. We understand that Pharmacia is also
presenting inappropriately combined short-term results from different protocols
performed on different continents with different comparator drugs [6] within the
framework of CLASS' successor study called SUCCESS-1 [7].

Figure (top) shows the relative risks of all the comparisons that have been
discussed during the current debate according to separate and combined analyses
of the two trials, addressing different follow-up durations (6 months versus
entire follow-up) and different subgroups (aspirin users versus non-users).
While there were some significant differences between celecoxib and ibuprofen,
there were none between celecoxib and diclofenac in any of 18 different analyses
with effect estimates scattered around and confidence intervals overlapping the
null effect line. A comparison between individual and combined results shows
that relevant differences between trials were hidden and spurious statistical
precision was implied through the inappropriate process of pooling. The wide
confidence intervals indicate that the trials were also underpowered. In
addition, they reflect the considerable uncertainty about the gastrointestinal
benefits of celecoxib and underscore the need for an individual patient data
meta-analysis. Figure (top) also shows no consistent pattern distinguishing
aspirin users from non-users, suggesting that CLASS' comparisons were not
obscured by aspirin use. For the time being, there is no evidence to support the
notion that celecoxib is superior to diclofenac, and insufficient evidence
favouring celecoxib over ibuprofen. CLASS' failure to demonstrate superiority of
celecoxib may have to do more with celecoxib's shortcomings as a COX-2 inhibitor
than with CLASS' shortcomings as a trial. Figure (bottom) shows results of an
in-vitro analysis of COX-2 selectivity by Warner et al [4] of the agents
relevant to this debate. According to this analysis, which admittedly is in
contrast to others [8], celecoxib and diclofenac actually have similar COX-2
selectivity with no reason to presume different rates of gastrointestinal ulcer
complications.

Dr Geis states that we have "incomplete understanding" of CLASS'
statistical issues. In our view it is Dr Geis who appears to have misunderstood
the rationale behind CLASS' statistical approaches. It was explicitly stipulated
that first the overall hypothesis had to be tested whether celecoxib and the
pooled NSAIDs were different in terms of the incidence of ulcer complications.
If the test used for this comparison resulted in non-significant P-values, the
null hypothesis (i.e. no difference between celecoxib and NSAIDs) would be
retained and the procedure stopped. Only if the test was significant, would
there be a pairwise comparison of celecoxib with each of the two control drugs,
ibuprofen and diclofenac. According to the protocol celecoxib was claimed to be
different from NSAIDs only if both overall and pairwise comparisons of ulcer
complications showed statistically significant differences [9]. We find it
remarkable that Dr Geis misuses this two-step procedure, which should have been
a safeguard against a type I error, as a justification for obscuring individual
trial results and as a rationale to actually commit a type I error by
inadequately pooling secondary outcomes (see Figure (top) ).

The main argument for only reporting CLASS' six months data, which was
provided by Pharmacia when the study was submitted to the FDA, was that patients
dropping out due to gastrointestinal adverse events were at higher risk to
develop ulcer complications [9]. In their response to criticisms published in
JAMA [10;11] CLASS' authors further elaborated this argument and stated that
those patients with "gastrointestinal intolerance or symptomatic
ulcers" were at "greatest risk" [12]. Now, Dr Geis argues that it
was mainly symptomatic ulcers which were precursors of ulcer complications. To
address this developing argument, we evaluated a narrative review of the 44
patients who experienced ulcer complications in CLASS [3]. Only 11 patients
(25%) developed gastro-intestinal symptoms before the actual ulcer complication
occurred. The remainder either experienced no gastro-intestinal symptoms (39%)
or experienced them at the same day as the ulcer complication occurred (36%).
Notably, none of the 44 patients who experienced ulcer complications had a
symptomatic ulcer as a precursor. So instead of being a pre-event of ulcer
complications, gastro- intestinal adverse-events were at best sentinel symptoms,
having no predictive ability at all [9]. In fact, gastrointestinal adverse
events were even associated with a significantly decreased risk of subsequent
ulcer complications (relative risk 0.28, 95% confidence interval 0.12 to 0.66,
p=0.0018). This makes clinical sense: patients who experience gastrointestinal
adverse events may be monitored and treated more carefully for detrimental
effects than patients without any gastrointestinal symptoms. As a consequence,
ulcer complications may be avoided particularly in symptomatic patients. We
suggest therefore CLASS' 6- to 15-month data to be clinically the most valid
because they represent true clinical decision making in long-term treated
patients.

Finally, Dr Geis states that CLASS' inadequate conclusions as described in
the JAMA article [2] have been accepted by numerous regulatory bodies. Since it
is clear that the FDA's experts [3;9] did not accept these conclusions, we are
worried by this statement. After considering CLASS' results, the FDA actually
refused to change celecoxib's labelling, which still states in its revised
version as of June 7, 2002 that "serious gastrointestinal toxicity such as
bleeding, ulceration, and perforation of the stomach, small intestine or large
intestine, can occur at any time, with or without warning symptoms" [13].

Pharmacia's continuing efforts to defend and disseminate the JAMA publication
worries us. The further we explore it the more disturbed we become about the
irregularities we find.

Peter Jüni MD, Senior Research Fellow in Clinical Epidemiology Departments
of Rheumatology, and Social and Preventive Medicine, University of Berne, 3010
Berne, Switzerland (peter.juni@insel.ch)

Anne W.S. Rutjes MSc, Research Fellow in Clinical Epidemiology Department of
Clinical Epidemiology and Biostatistics, Academic Medical Center, University of
Amsterdam, 1100 DE Amsterdam, Netherlands

Paul Dieppe MD, Professor of Health Services Research MRC Health Services
Research Collaboration, Department of Social Medicine, University of Bristol,
Bristol BS8 2PR

Acknowledgements

We thank Yoav Ben-Shlomo for helpful comments and Ben Calnan for data
management and entering.

[View Larger
Version of this Image (185K JPEG file)]


Figure


Top: Relative risks and 95% confidence intervals calculated from proportions

for primary and secondary outcomes in CLASS study according to separate and

combined analyses of the two trials for different follow up durations and

different groups of patients. Black squares indicate main outcome measures

as prespecified in CLASS protocols, grey squares indicate main outcome

measures as reported.4 Identical event rates were assumed in celecoxib

groups of both trials for secondary outcome because separate trial data were

unavailable. Arrowheads indicate truncation of confidence intervals.


Bottom: Analysis of percentage inhibition of COX 1 when COX 2 is inhibited

by 80%. Horizontal line indicates equiactivity - 80% inhibition of COX 1.

Adapted from Warner et al, 1999 [4].

Reference List

1. Jüni P, Rutjes AW, Dieppe PA. Are selective COX 2 inhibitors superior to
traditional non steroidal anti-inflammatory drugs? BMJ 2002;324:1287-8.

2. Silverstein FE, Faich G, Goldstein JL, Simon LS, Pincus T, Whelton A et
al. Gastrointestinal toxicity with celecoxib vs nonsteroidal anti- inflammatory
drugs for osteoarthritis and rheumatoid arthritis: the CLASS study: A randomized
controlled trial. Celecoxib Long-term Arthritis Safety Study. JAMA
2000;284:1247-55.

3. Witter J. Medical officer review. Available at: http://www.fda.gov/ohrms/dockets/ac/01/briefing/3677b1_03_med.pdf.
[Accessed December 10, 2001.]

4. Warner TD, Giuliano F, Vojnovic I, Bukasa A, Mitchell JA, Vane JR.
Nonsteroid drug selectivities for cyclo-oxygenase-1 rather than cyclo-
oxygenase-2 are associated with human gastrointestinal toxicity: a full in vitro
analysis. Proc Natl Acad Sci U S A 1999;96:7563-8.

5. Henry D, Lim LLY, Garcia Rodriguez LA, Perez Gutthann S, Carson JL,
Griffin M et al. Variability in risk of gastrointestinal complications with
individual non-steroidal anti-inflammatory drugs: results of a collaborative
meta-analysis. BMJ 1996;312:1563-6.

6. Singh G, Goldstein J, Fort J, Bello A, Boots S. SUCCESS-1 in
osteoarthritis: celecoxib demonstrates similar efficacy to the conventional
NSAIDs, diclofenac and naproxen, in patients with osteoarthritis treated in 39
countries in 6 continents. Abstract presented at EULAR 2001, European Congress
of Rheumatology . Available at: http://www.eular.org/eular2001/AbstractsOnline.cfm.
[Accessed June 24 2002]

7. Goldstein J, Eisen G, Bensen W, Agrawal N, Singh G, Pavelka K et al.
SUCCESS in osteoarthritis (OA) trial: celecoxib significantly reduces the risk
of upper gastrointestinal (UGI) hospitalizations compared to diclofenac and
naproxen in 13,274 randomized patients with OA. Abstract presented at EULAR
2001, European Congress of Rheumatology . Available at: http://www.eular.org/eular2001/AbstractsOnline.cfm.
[Accessed June 24 2002]

8. Gierse JK, Koboldt CM, Walker MC, Seibert K, Isakson PC. Kinetic basis for
selective inhibition of cyclo-oxygenases. Biochem J 1999;339 ( Pt 3):607-14.

9. Lu HL. Statistical reviewer briefing document for the advisory committee.
Available at: http://www.fda.gov/ohrms/dockets/ac/01/briefing/3677b1_04_stats.doc.
[Accessed December 10, 2001.]

10. Berg Hrachovec J,.Mora M. Reporting of 6-month vs 12-month data in a
clinical trial of celecoxib. JAMA 2001;286:2398.

11. Wright JM, Perry TL, Bassett KL, Chambers KG. Reporting of 6- month vs
12-month data in a clinical trial of celecoxib. JAMA 2001;286:2398-9.

12. Silverstein F, Simon L, Faich G. Reporting of 6-month vs 12- month data
in a clinical trial of celecoxib. In reply. JAMA 2001;286:2399- 400.

13. Celebrex labeling, revised June 7, 2002. Available at: http://www.fda.gov/cder/foi/label/2002/20998s009lbl.pdf
[accessed June 28, 2002]

Dear Dr. Smith:

We appreciate the opportunity to correct the inaccuracies in the BMJ
editorial,1 which appear to be based on an incomplete understanding of the
CLASS design, data and statistical issues. Contrary to the BMJ editorial
assertions, there were no post-hoc changes in the study design or the
outcome definitions. The data, analyses and conclusions as described in
the JAMA publication2 have been reviewed and accepted by numerous
regulatory bodies.

The CLASS authors extensively reviewed all of the data and its
adherence to underlying assumptions for analytical validity and made the
decision that the 6-month analyses were most valid for the initial CLASS
publication. More recently, FDA approved a US celecoxib label change using
all the results from the CLASS study.3 The FDA chose the 9-month data as
most appropriate for describing the key information and included these
analyses in the label. Even with the differing medical judgment as to
which time interval best reflected the data, similar conclusions were
reached.

The CLASS design was completed with collaboration from investigators
and FDA representatives. The endpoints, duration and statistical analyses
were all prespecified. The trial was a single study with the objective of
comparing the rate of ulcer complications with celecoxib versus
traditional NSAIDs. The need for two protocols was prespecified to ensure
blinding of study medications. One protocol included celecoxib 400 mg BID
and ibuprofen 800 mg TID; the other included celecoxib 400 mg BID and
diclofenac 75 mg BID. Aside from this difference and the inclusion of
quality of life measures in one protocol, the protocols were identical.
The statistical plan stated that the data from the two protocols would be
analyzed as a single study. The trial was designed to mimic usual
medical practice in order to generate the most clinically meaningful data.
Adult OA and RA patients were enrolled irrespective of age, disease
severity and co-morbidities. Patients taking low-dose aspirin for
cardiovascular prophylaxis, taking corticosteroids and those with a
history of peptic ulcers were also included. Minimum duration of
participation was 6 months unless the patient was withdrawn early for an
adverse event or insufficient efficacy. Two important assumptions of the
design were: 1) a constant rate of ulcer complications in the NSAID group4
and 2) an enrollment of about 11% of patients using low-dose aspirin.5

The prespecified primary endpoint was ulcer complications (bleeding,
perforation and outlet obstruction); an analysis of symptomatic ulcers was
also prespecified. A “traditional” definition of ulcer complications6 was
used which was consistent with the definition used for the ulcer
complication meta-analysis submitted in the NDA.5 Each ulcer
complication required a confirmatory UGI endoscopy or contrast x-ray. Of
importance, the protocols mandated study withdrawal if a patient was found
to have a non-bleeding ulcer (i.e., symptomatic ulcer).

The predefined statistical plan identified the primary comparison to
be the rate of ulcer complications (traditional definition) with celecoxib
versus the combined NSAID group (ibuprofen plus diclofenac) and only if
the difference were statistically significant would comparisons of
celecoxib to each of the individual NSAIDs be performed. This was
specified in order to control the overall alpha-level. The plan included
an analysis of the effects of risk factors for ulcer complications on the
results of the different treatments. Since low-dose aspirin is a risk
factor for serious GI events,7 an analysis of the cohort of patients not
taking aspirin was planned.

The comparison of ulcer complication rates did not show a
statistically significant difference between celecoxib and the NSAID
group. However, the analysis of the combined endpoint of symptomatic
ulcers/complicated ulcers showed a statistically significant lower rate in
the celecoxib group versus the NSAID group. Since the primary endpoint
was not reached the comparisons of celecoxib to the individual NSAIDs were
not considered valid.

Once the study blind was broken, it was clear that important
assumptions made in designing the trial had not proved true. First, the
rate of ulcer complications for the NSAID group did not remain constant
over time but, rather, decreased (Fig. 1). Also, the withdrawal rate of
patients due to symptomatic ulcers was statistically greater in the NSAID
group versus the celecoxib group and the difference was most apparent
after the first 6 months of the study (Fig. 2). Since symptomatic ulcers
are precursors of ulcer complications, clearly high-risk patients were
being depleted from the NSAID group more quickly than from the celecoxib
group. This differential withdrawal of NSAID patients at risk of ulcer
complications led to progressive study bias, which caused the analyses to
become less valid statistically and medically with time. As described by
other authors of the JAMA paper,8 after extensive review, the CLASS
oversight committees judged the analyses of the 6-month data to be the
most scientifically and clinically valid since: 1) 6 months was the
minimum duration of study participation; 2) more than 50% of patients were
continuing in the study at 6 months (median duration of exposure: 9 months
for celecoxib and diclofenac; 6 months for ibuprofen) and 3) the impact of
differential withdrawal of patients at risk was most apparent beyond this
time point. The CLASS oversight committees reported that “the data after
6 months were so confounded as to be difficult to interpret for assessing
a drug-related causal GI toxicity”.8

The second assumption that was not met was the incidence of aspirin
use. Approximately 22% of patients in each treatment group took low-dose
aspirin rather than the expected 11%. Further, analyses showed that low-
dose aspirin use in CLASS was a risk factor for ulcer complications for
celecoxib-treated patients but not for the NSAID-treated patients. This
is not unexpected because aspirin is an inhibitor of both COX-I and COX-
II, as are traditional NSAIDs. These factors impacted the comparisons of
ulcer complication rates between celecoxib versus NSAIDs. Based on the 6-
month analysis for the all-patient cohort, ulcer complication rates were
not significantly different between the celecoxib and the NSAID groups.
However, the cohort of patients not using aspirin showed a statistically
significant lower rate of ulcer complications with celecoxib versus the
NSAIDs. Recently, in describing the CLASS trial the FDA stated “…the use
of aspirin (a drug known to cause stomach ulcers and bleeding) have
obscured the ability to accurately compare the GI safety of Celebrex to
other nonsteroidal anti-inflammatory drugs”.9 Based on the CLASS data,
the US label describes a 4-fold increase in the 9-month rate for
complicated ulcers for patients taking celecoxib plus aspirin compared to
patients taking celecoxib alone. A near 3-fold increase in the rate of
the combined endpoint of symptomatic ulcers/complicated ulcers is shown in
the label for the celecoxib plus aspirin group versus the group taking
celecoxib alone.

Additional points made in the BMJ editorial warrant comment. The
editorial describes the JAMA paper as “overoptimistic”, using “post hoc
changes to the protocol while omitting disappointing longer term data”.1
The JAMA manuscript clearly acknowledges that the primary endpoint of the
study was not reached.2 There were no post hoc changes to the protocol
and the analyses of the longer-term data, although complicated by the
differential dropouts do not differ substantially from the 6-month
analyses.10

We continue to stand behind the CLASS design, analyses and
conclusions as stated in JAMA and invite discussions that will ensure an
understanding of the facts and assist in clarifying the safety profile of
celecoxib.

Respectfully yours,

G. STEVEN GEIS, Ph.D., M.D.

Group Vice President, Clinical Research

Pharmacia

5200 Old Orchard Road,
Skokie, Illinois 60077


george.s.geis@pharmacia.com

References

1. Juni P, Rutjes AWS, Dieppe PA. Are selective COX 2 inhibitors
superior to traditional non-steroidal anti-inflammatory drugs? BMJ
2002;324:1287-8.

2. Silverstein FE, Faich G, Goldstein JL, Simon LS, Pincus T. Whelton
A, et al. Gastrointestinal toxicity with celecoxib vs nonsteroidal anti-
inflammatory drugs for osteoarthritis and rheumatoid arthritis: the CLASS
study: a randomized controlled trial. Celecoxib Long-term Arthritis
Safety Study. JAMA 2000;284:1247-55

3. Celecoxib Label, June, 2002

4. MacDonald TM, Morant SV, Robinson GC, Shield MJ, McGilchrist MM,
Murray FE, McDevitt DG. Association of upper gastrointestinal toxicity of
nonsteroidal anti-inflammatory drugs with continued exposure: cohort
study. BMJ 1997;315:1333-7

5. Celecoxib NDA, 1998

6. Silverstein FE, Graham DU, Senior JR, Davies HW, Struthers BJ,
Bittman RM, Geis GS. Misoprostol reduces serious gastrointestinal
complications in patients with rheumatoid arthritis receiving non-
steroidal anti-inflammatory drugs: a randomized, double-blind, placebo-
controlled trial. Ann Intern Med 1995;123:241-9

7. Weil J. Colin JD, Langman M, David Lawson, Richard Logan, Michael
Murphy, Michael Rawlins, et al. Prophylactic aspirin and risk of peptic
ulcer bleeding. BMJ 1995;310:827-30

8. Silverstein F, Simon, L. Faich G. Reporting of 6-month vs 12-month
data in a clinical trial of celecoxib. In reply. JAMA 2001;286:2399-400

9. FDA Press Release, June, 2002; www.FDA.gov

10. Goldstein JL, for the CLASS investigators. Gastrointestinal (GI)
event rates in the CLASS study: six months vs longer-term follow-up
analyses. Gastroenterology 2002;122 (4) Suppl. 1:A-469

Figure 1. Ulcer complication rates expressed as annual incidence as
calculated using the cumulative number of events and duration of treatment
exposure in the celecoxib and NSAID groups over the first 6 months of the
study (L) and over period of maximum treatment.

Figure 2. Kaplan-Meier estimates for withdrawals due to symptomatic
ulcers. No withdrawal due to a symptomatic ulcer occurred after final
time points.

We agree that the wide dissemination of the misleading results of the
CLASS trial must be counterbalanced by the equally wide dissemination of
the findings of the reanalysis, according to the original protocol.1
However, misleading promotional activities of Pharmacia, celecoxib’s
manufacturer, are not limited to distribution of the CLASS trial. Current
advertising for celecoxib (Celebrex) in Australia shows a blood pack with
the headlines “this is the price thousands of patients are paying for
arthritis relief” and “if you think this risk is too high, Celebrex lowers
the risk of upper GI ulcer complications by 8 times compared with
NSAIDs”.2 This latter claim is based on a flawed meta-analysis supported
by Searle, Pharmacia’s parent company, which does not include long-term
clinical data reported to the FDA.3

Worldwide, Searle and Pharmacia’s promotional materials for celecoxib
have repeatedly been found misleading, including in the USA and
Australia.4,5 However, the absence of sanctions, and imposition of
derisory fines (in Australia Aus$10 000, equivalent to £3 700) are
insufficient to deter these companies from continuing their misleading
promotion to boost sales.

Up to April 2002, sales of celecoxib cost Australian taxpayers more
than Aus $270m through the Pharmaceutical Benefits Scheme (which provides
subsidised access to medicines in Australia), approximately at least twice
the cost of using other NSAIDs.6

We believe that serious sanctions should be taken to halt misleading
promotional practices. It might be suspension of the drug’s marketing
approval and/or fines, but of a size comparable to the manufacturer’s
unjust enrichment.

Agnes I Vitry

Senior Editor, Australian Medicines Handbook

Lecturer, School of Pharmaceutical Molecular and Biomedical Sciences,
North Terrace, Adelaide SA 5000

agnes.vitry@unisa.edu.au

Eve Hurley

Senior Editor, Australian Medicines Handbook

PO Box 240, Rundle Mall, Adelaide SA 5000

eve.hurley@adelaide.edu.au

1. Jüni P, Rutjes A and Dieppe P. Are selective COX 2 inhibitors
superior to traditional non steroidal anti-inflammatory drugs? BMJ 2002;
324: 1287-8.

2. Australian Pharmacist 2002; 21: 256-7.

3. Goldstein J, Silverstein F, Agrawal N, Hubbard R, Kaiser J,
Maurath C et al. Reduced risk of upper gastrointestinal ulcer
complications with celecoxib, a novel COX-2 inhibitor. Am J Gastroenterol
1999; 95: 1681-90.

4. Food and Drug Administration . Warning letter sent to Pharmacia
Corporation on February 1, 2001 and warning letter sent to G.D. Searle
&Go on April 6, 2000. www.fda.gov/cder/warn/index.htm [Accessed 4 June
2002].

5. Australian Pharmaceutical Manufacturers Association. Code of
Conduct Annual Report 2000.

6. Health Insurance Commission. Pharmaceutical Benefits Schedule
Statistics. http://www.hic.gov.au/prof/ [Accessed 4 June 2002].

To the Editor:

The authors are to be congratulated for their excellent analysis of the CLASS trial, which nicely demonstrates no decrease in gastrointestinal complications for celecoxib as compared to diclofenac or ibuprofen [1]. I also commend them for calling for an independent adjudication of all the long-term trials of COX 2 inhibitors. We also asked for a full review of the CLASS data in our letter to JAMA [2], but it was removed by the JAMA letter editor along with our serious adverse event data analysis.

A meta-analysis of the FDA data from the CLASS and VIGOR trials demonstrates a statistically significant increase in serious adverse events for COX 2 inhibitors as compared to the non selective NSAIDs (RR = 1.19 [1.06 – 1.32], p = 0.002)[3,4]. This increase exists despite a reduction in complicated ulcers (one category of serious adverse events) by rofecoxib in the VIGOR trial. This analysis strongly suggests that COX 2 inhibitors are inferiour overall to non-selective NSAIDs and underscores the need for the independent individual patient meta-analysis.

James M Wright, Professor

Departments of Pharmacology & Therapeutics and Medicine
University of British Columbia

References:

1. Juni P, Rutjes AWS, Dieppe PA. Are selective COX-2 inhibitors superior to traditional non steroidal anti-inflammatory drugs? Adequate analysis of the CLASS trial indicates that this may not be the case. Brit Med J 2002;324:1287-8.

2. Wright JM, Perry TL, Bassett Kl, Chambers KB. REporting of 6-month vs 12-month data in a clinical trial of celecoxib. JAMA 2001;286:2398-9.

3. US Food and Drug Administratio. Transcript of the arthritis advisory committee. www.fda.gov/ohrms/dockets/ac/01/transcripts/3677t1.rtf [Accessed 10 december 2001]

4. Therapeutics Letter #43. COX-2 inhibitors update: Do journal publications tell the full story? www.ti.ubc.ca [accessed 7 June 2002].