To the Editors:
The recent report by Spiegel and colleagues concludes that generic, nonselective nonsteroidal antiinflammatory drugs (NSAIDs) are more cost-effective than nonselective NSAIDs plus proton pump inhibitors (PPIs) or cyclooxygenase 2-selective inhibitors (coxibs) alone for treating arthritis in average or low-risk patients ages 60 years who require daily, long-term NSAIDs (1). Those authors also conclude that nonselective NSAIDs plus PPIs may be more cost-effective in patients who are receiving low-dose aspirin (ASA), or in those with an elevated risk of upper gastrointestinal (GI) or cardiovascular adverse events.
Several of those authors' critical ulcer and cardiovascular adverse event probabilities are inconsistent with the best available evidence or are lacking in factual support. In addition, their model and accompanying narrative apply to probabilities of side-effects for a hybrid of celecoxib and rofecoxib, which is now withdrawn, and trials involving rheumatoid arthritis (RA) and osteoarthritis (OA) patients who are receiving supratherapeutic doses of celecoxib. It should be noted that in the management of patients with arthritis, celecoxib is mainly used to treat OA. Also, the report does not evaluate celecoxib plus PPI cotherapy, which is often used by high-risk patients in the community.
Those authors estimate annual risks for their composite ulcer end point (symptomatic ulcers, bleeding ulcers, perforations, and obstructions [PUBs]) in users and nonusers of low-dose ASA as follows: nonselective NSAIDs, 7.5%/3.0%; nonselective NSAIDs plus PPIs, 2.45%/1.5%; and coxibs composite, 7.5%/1.5%. Their assumption that 20% of patients are receiving low-dose ASA implies all-patient base-case probabilities of a PUB of 3.90%, 1.69%, and 2.70%, respectively.
A main tenet of their report is that ASA raises the risk of having a PUB much more in patients taking coxibs (1.5–7.5%, relative risk [RR] 5.0) than in those taking nonselective NSAIDs (3.0–7.5%, RR 2.5). However, this assumption is not supported by robust evidence and conflicts with OA clinical trial data. The single quantitative reference in the study by Spiegel et al for the assumed relative risk of combining ASA with coxibs contains a general formula whose “anticoagulant” relative risk was based on journal references published prior to the introduction of coxibs (2). Further, application of the indirect formula to coxibs with Spiegel et al's base-case nonselective NSAID monotherapy risk of 3.0% would imply a coxib plus ASA probability of 3.75% (3.0% × 0.5 × 2.5) and a relative risk versus coxib alone of 2.5, rather than Spiegel et al's 7.5% and 5.0, respectively. In the Successive Celecoxib Efficacy and Safety Study (SUCCESS), the crude relative risk of PUBs for ASA plus celecoxib versus celecoxib (2.67) is lower than that for nonselective NSAIDs with and without ASA (3.88), contradicting the authors' inferences from the Celecoxib Long-term Arthritis Safety Study (CLASS) (3, 4). The point estimate PUB rates with ASA of 2.4% for celecoxib versus 6.6% (RR 0.36) for nonselective NSAIDs in SUCCESS are a departure from the assumed rate of 7.5% for both in the model by Spiegel et al. These SUCCESS findings are more relevant than those from CLASS because they were based on an OA population taking celecoxib dosages of 200–400 mg/day, whereas, in the CLASS trial, OA and RA patients received dosages of 800 mg/day, which is quadruple the normal OA dose. But, even at 800 mg/day, the 6-month CLASS point estimate of PUB risk with ASA was 22% lower with celecoxib compared with nonselective NSAIDs, a finding that is also in conflict with the model (5). Additionally, a large observational study found a relative risk reduction of almost 50% in rates of hospitalization for ulcer perforations and bleeds in patients using coxibs plus ASA compared with those using nonspecific NSAIDs plus ASA (6).
In non-ASA using patients, those authors assume that PPIs reduce the 3.0% annual PUB risk of using nonselective NSAIDs to 1.5%, which is identical to the coxib risk. However, the purportedly “conservative” 0.50 RR was based on 3 endoscopic studies, 2 involving only high-risk patients (7). Coxib versus nonselective NSAID studies have shown that endoscopic ulcers provide poor estimates of clinically significant ulcers and their relative risks (8). Spiegel et al contend that the only fully published head-to-head prospective outcomes trial then available of clinically significant ulcer events in celecoxib and nonselective NSAID plus PPI patients supports their model's assumed equality of PUB risks in non-ASA patients (9). However, this trial has both a different population (previous bleeding ulcer) and end point (recurrent bleeding). In addition, 6-month, all-patient rates of recurrent bleeds were 4.9% for celecoxib versus 6.4% for diclofenac plus PPI. Although the difference in this small trial was not significant, it might be more appropriate in modeling to use a 0.77 base-case RR for celecoxib versus nonselective NSAID plus PPI, especially for bleeding in high-risk cohorts, rather than assuming equal risks, given the absence of stronger evidence (10). The results of a trial with the same primary endpoint published later in 2005 provide additional support for assuming a RR of 0.77 (or lower) for celecoxib versus nonspecific NSAID plus PPI (11). In patients using ASA, the authors abandon the 0.50 RR assumption with PPIs (and associated 3.75% PUB risk) instead they use a Delphi panel's estimate that the nonselective NSAID plus PPI plus ASA PUB risk is 2.45% for a RR of 0.33 compared with nonselective NSAID plus ASA alone. Therefore, the stratified nonselective NSAID plus PPI probabilities of PUB by ASA status lack evidentiary support.
The Spiegel et al model's myocardial infarction (MI) event rate of 0.77% for coxibs is based on annualized rofecoxib and celecoxib probabilities from the Vioxx Gastrointestinal Outcomes Research study (12) and CLASS, and its probability of 0.4% for nonselective NSAIDs is assumed. It would have been more appropriate to 1) use a cardiovascular thrombotic indicator encompassing both MI and stroke risks; 2) model for differences in risk between rofecoxib and celecoxib; and 3) estimate risks from studies using OA doses of celecoxib. A meta-analysis and a pooled analysis of clinical trials comparing celecoxib with nonselective NSAIDs are consistent with the conclusion that celecoxib does not raise the risk of serious cardiovascular thrombotic events (MI, stroke, or death) (13, 14). The relative risk point estimates for celecoxib in these analyses, which include some high-dose celecoxib trials, are 0.86 and 1.06, whereas the implied RR for MI is 1.93 in the evaluation by Spiegel et al.
In conclusion, several critical ulcer and cardiovascular adverse event probabilities driving the model's results are not well supported by current evidence. Further, the sensitivity analyses conducted by the authors do not adequately address these biases.
Michael Loyd & Associates, Ltd received consultancies >$10,000 from Pfizer, and Mr. Loyd and spouse own Pfizer stock. Dr. Jacobs received honoraria <$10,000 from Pfizer, and received consultancies >$10,000 from Michael Loyd & Associates, Ltd. Dr. Rublee owns Pfizer stock and options.