Systematic review of celecoxib for osteoarthritis and rheumatoid arthritisBMJ 2003; 326 doi: https://doi.org/10.1136/bmj.326.7384.334 (Published 08 February 2003) Cite this as: BMJ 2003;326:334
Problems compromise review's validity
- Peter Jüni (firstname.lastname@example.org), senior research fellow in clinical epidemiology,
- Rebekka Sterchi, research associate,
- Paul Dieppe, professor of health services research
- Department of Rheumatology, University of Berne, 3010 Berne, Switzerland
- MRC Health Services Research Collaboration, Department of Social Medicine, University of Bristol, Bristol BS8 2PR
- Pharmaceutical Management Agency (PHARMAC), Level 1 Old Bank Chambers, 98 Customhouse Quay, PO Box 10 254, Wellington 6001, New Zealand
- Centre for Statistics in Medicine, Institute of Health Sciences, Oxford OX7 3LF
- Pfizer Global Research and Development, Sandwich, Kent CT13 9NJ
EDITOR—Deeks et al say that celecoxib has improved gastrointestinal safety and tolerability compared with traditional non-steroidal anti-inflammatory drugs (NSAIDs).1 We have several concerns.
Firstly, Deeks et al reported the papers by Bensen et al, Zhao et al (1999), Simon et al, and Zhao et al (2000) as if they referred to four different trials.1 The papers by Bensen et al and Zhao et al (1999) were, however, merely duplicate reports of one trial, whereas the papers by Simon et al and Zhao et al (2000) reported in duplicate on another trial. Deeks et al either included the same data more than once or mixed up unpublished data with unrelated publications.
Secondly, Deeks et al report similar relative risks for ulcer complications observed after six months in CLASS's two trials2: 0.54 (95% confidence interval 0.20 to 1.47) for study 035 (celecoxib v ibuprofen) and 0.56 (0.19 to 1.66) for study 102 (celecoxib v diclofenac), implying that it is appropriate to pool two trials by using comparator drugs of different cyclo-oxygenase-2 selectivity. According to the Food and Drug Administration (http://www.fda.gov/), however, four events occurred in the celecoxib group and 11 in the ibuprofen group in study 035 (0.36, 0.12 to 1.14), whereas seven events occurred in the celecoxib group and nine in the diclofenac group in study 102 (0.78, 0.29 to 2.08).3 This implies that pooling these trials may be inappropriate.
Thirdly, Deeks et al's justification for considering only CLASS's six month results is problematic.4 Admittedly, data available from the FDA indicate that rates of patient withdrawal were different in the celecoxib and ibuprofen groups, implying that results for study 035 were unreliable at all time points. In accordance with Deeks et al, this trial should therefore have been excluded from all analyses. Contrary to Deeks …