Education and debate

CLASS clarification: reaffirms the medical importance of the analyses and results

G Steven Geis, group vice president, clinical research. 

Pharmacia 5200 Old Orchard Rd Skokie, IL 60077

Correspondence to: G S Geis george.s.geis{at}pharmacia.com

From BMJ USA 2002;September:522

We appreciate the opportunity to correct the inaccuracies in the article by Jüni et al,¹ which appear to be based on an incomplete understanding of the CLASS design, data, and statistical issues. Contrary to the editorial's assertions, there were no post hoc changes in the study design or the outcome definitions. The CLASS authors reviewed all the data and decided the 6-month analyses were most appropriate for initial publication,² while the FDA chose 9-month data as most appropriate for a recent label change.³ Despite differing medical judgment for the time interval that best reflected the data, and contrary to the BMJ editorial allegations, the conclusions were similar.

CLASS 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. Low-dose aspirin was allowed and the minimum expected duration of study participation was 6 months. Two important assumptions of the design were: 1) a constant rate of ulcer complications in the NSAID group⁴ and 2) that about 11% of enrolled patients would use low-dose aspirin.⁵

The primary end point was ulcer complications (bleeding, perforation and outlet obstruction) verified by endoscopy or contrast x ray, but analysis of symptomatic ulcers was also prespecified. The protocols mandated study withdrawal if a patient was found to have a non-bleeding ulcer (ie, symptomatic ulcer).

The primary analysis was a comparison of ulcer complications (traditional definition)⁶ with celecoxib versus the combined NSAID group (ibuprofen plus diclofenac) and only if the differences 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 (eg, low-dose aspirin use) on the results of the different treatments.

Ulcer complication rates were not significantly different for celecoxib versus the NSAID group. However, the rate of the combined end point of symptomatic/complicated ulcers was significantly lower with celecoxib. Since the primary analysis was not significant, comparisons to the individual NSAIDs were not valid.

View larger version (42K): [in this window] [in a new window]   Fig 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. Arrow notes the reduction in NSAID rate over time.

Once the study blind was broken, it was clear that important assumptions made in designing the trial had not proved true. NSAID ulcer complication rates decreased over time instead of remaining constant (figure 1). Moreover, 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 (figure 2). Since symptomatic ulcers are precursors of ulcer complications, high-risk patients were being depleted from the NSAID group more quickly than from the celecoxib group. This differential withdrawal rate introduced study bias, which caused the analyses to become less valid with time.

View larger version (43K): [in this window] [in a new window]   Fig 2.   Kaplan-Meier estimates for withdrawals due to symptomatic ulcers. No withdrawal due to a symptomatic ulcer occurred after final time points.

As described by other authors of the JAMA paper,⁷ 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."⁷

The second unmet assumption in the CLASS trial was that approximately 22% of patients in each treatment group took low-dose aspirin rather than the expected 11%. 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. 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. The FDA noted ". . . the use of aspirin . . . may have obscured the ability to accurately compare the GI safety of Celebrex to other nonsteroidal anti-inflammatory drugs."⁸

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 end point of the study was not reached.² There were no post hoc changes to the protocol. The analyses of the longer-term data, although complicated by the differential dropouts, do not differ substantially from the 6-month analyses.⁹

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.

References

1.	Jüni P, Rutjes AWS, Dieppe PA. Are selective COX 2 inhibitors superior to traditional non steroidal anti-inflammatory drugs? BMJ 2002; 324: 1287-1288[Free Full Text].
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: 1247-1255[Abstract/Free Full Text].
3.	FDA-Approved Celebrex Approved Prescribing Information (June 7, 2002): http://www.fda.gov/cder/foi/label/1998/20998lbl.pdf.
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: 1333-1337[Abstract/Free Full Text].
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: 241-249[Abstract/Free Full Text].
7.	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: 2399-2400.
8.	FDA Talk Paper: http://www.fda.gov/bbs/topics/ANSWERS/2002/ANS01151.html.
9.	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.