Data derived from references1, 2. Class = Celecoxib Long-term Arthritis Safety Study; VIGOR = Vioxx Gastrointestinal Outcomes Research; OA = osteoarthritis; RA = rheumatoid arthritis; QD = daily; BID = twice daily; NSAIDs = nonsteroidal antiinflammatory drugs; TID = three times daily; OTC = over the counter; GI = gastrointestinal; CV = cardiovascular; APTC = Antiplatelet Trialists' Collaboration.
The risk of cardiovascular thrombotic events with selective cyclooxygenase-2 inhibitors
Article first published online: 12 AUG 2002
Copyright © 2002 by the American College of Rheumatology
Arthritis Care & Research
Volume 47, Issue 4, pages 349–355, 15 August 2002
How to Cite
Strand, V. and Hochberg, M. C. (2002), The risk of cardiovascular thrombotic events with selective cyclooxygenase-2 inhibitors. Arthritis & Rheumatism, 47: 349–355. doi: 10.1002/art.10560
- Issue published online: 12 AUG 2002
- Article first published online: 12 AUG 2002
- Manuscript Accepted: 12 FEB 2002
- Manuscript Received: 8 NOV 2001
The selective cyclooxygenase-2 (COX-2) inhibitors celecoxib and rofecoxib are now widely prescribed instead of the traditional nonsteroidal antiinflammatory agents (NSAIDs). Short-term randomized controlled trials (RCTs) documented that both agents were superior to placebo, with comparable efficacy and better upper gastrointestinal (GI) adverse-event profiles than traditional NSAIDs, including ibuprofen, naproxen, and diclofenac. However, these RCTs were not powered to demonstrate GI tolerability equivalent to placebo, and consequently the GI warning common to the NSAID class remained in both celecoxib and rofecoxib labels. The Celecoxib Long-term Arthritis Safety Study (CLASS) and the Vioxx Gastrointestinal Outcomes Research (VIGOR) trials were designed to demonstrate GI safety superior to that of traditional NSAIDs, as used in regular clinical practice, assessed by “clinically meaningful” upper GI events, at twice recommended clinical doses for long-term use (celecoxib 400 mg twice daily; rofecoxib 50 mg daily) (1, 2).
Although both large outcome trials shared similar designs, they differed in several important aspects (Table 1). Comparator NSAIDs in CLASS were ibuprofen and diclofenac, and in VIGOR it was naproxen. Only patients with rheumatoid arthritis (RA) were enrolled in VIGOR; in CLASS, RA patients represented 28% of the study population, and osteoarthritis (OA) patients represented the other 72%. Enrolled patient numbers were similar: in CLASS, 3,987 patients took celecoxib, 1,996 patients took diclofenac (75 mg twice daily), and 1,985 patients took ibuprofen (800 mg 3 times daily); in VIGOR 4,047 patients took rofecoxib and 4,029 took naproxen (500 mg twice daily). Patient populations differed at baseline predominantly by risk factors for cardiovascular (CV) and GI events (Table 2). Enrollment of patients taking prophylactic low-dose aspirin (≤325 mg/day) or other antiplatelet agents (such as ticlopidine) was permitted in CLASS and proscribed in VIGOR. Patients with angina or congestive heart failure with symptoms at rest or with minimal activity were excluded from participation in VIGOR. The population enrolled in CLASS included more patients at risk for thromboembolic CV disease. During presentations at the FDA Arthritis Advisory Committee meetings, February 7–8, 2001, where results of the CLASS and the VIGOR trial were reviewed, CV safety issues potentially associated with administration of COX-2 selective agents were raised (3, 4). This editorial will discuss these data and possible explanations for the findings.
|Parameter||VIGOR (n = 8,076)||CLASS (n = 7,968)|
|Drug/dosage||Rofecoxib 50 mg QD (2× max OA chronic dose)||Celecoxib 400 mg BID (4× maximum OA chronic dose; 2× maximum RA chronic dose)|
|Patients, %||RA†||OA (72), RA (28)|
|Comparator NSAIDs||Naproxen 500 mg BID||Ibuprofen 800 mg TID Diclofenac 75 mg BID|
|Low-dose aspirin, %||No||Yes (22)|
|Anti-ulcerant allowed?||Antacids and OTC H2-receptor antagonists||Only antacids (≤2 day)|
|Duration, months||Median 9; maximum 13||Median 9; maximum 13|
|Analysis||Intent to treat||Intent to treat; excludes events at days 0–2|
|Primary end point||Clinical upper GI events||Ulcer complications|
|Secondary end point||Complicated upper GI events||Symptomatic ulcers|
|CV analysis||Adjudicated by Vascular Events Committee||Investigator reported|
|Secondary analyses||APTC definitions|
|VIGOR (n = 4,047)||CLASS (n = 3,987)||CLASS non-ASA (n = 3,105)|
|Age ≥ 65||24.0||37.9||34.8|
|Age ≥ 75||5.1||12.2||10.2|
|History of GI disease||8.0||15.3||14.6|
|History of CV disease||46.5||40.2||34.7|
|≥1 risk factor for coronary disease||50.1||NR||NR|
|History of a CV event and NOT receiving ASA||4.0||3.9||5.0|
|Total exposure, patient-years||3,947||2,320||1,803|
Although repeatedly demonstrated in in vitro, ex vivo models of inflammatory disease, expression of COX-2 predominantly mediates inflammation; constitutive production of COX-2 is limited, predominantly to maintain homeostatic regulation in brain, spinal cord, reproductive organs, kidney, and vasculature (5–7). In the absence of effects on platelet thromboxane production, selective COX-2 inhibition may decrease vascular prostacyclin (PGI2) synthesis, thereby promoting prothrombotic events. However, data to support this occurrence in vivo are lacking; measured urinary excretion of COX metabolites may be affected by PGI2 production in other cells, such as bladder epithelium (8). Inhibition of COX-2 production prevents development of ischemic preconditioning in conscious rabbits, although relevance of this model to human disease is unclear (9). In general, any increased risk for thromboembolic CV events associated with administration of COX-2 selective agents would be expected to be small, in view of other endothelial-derived protective mechanisms, such as nitric oxide production (10). Nonetheless, risk factors for CV disease may alter the benefit/risk profile of these agents, as exemplified by reported thromboses in 4 patients with connective tissue disease and antiphospholipid antibodies who received celecoxib (11).
Incidence of CV events in CLASS and VIGOR
Adverse events reported in the VIGOR trial indicated that the cumulative incidence of investigator-reported serious thromboembolic CV events was significantly lower in naproxen-treated than rofecoxib-treated patients (12, 13). Commensurate with reported differences in thromboembolic CV events, serious CV adverse events occurred more frequently in rofecoxib than naproxen treatment groups (2.5% and 1.1%, respectively) and similarly for treatment-associated discontinuations and hospitalizations due to CV adverse events (2.7% and 0.8%, representing 65 and 24 patients, respectively). Although there was no difference in deaths attributed to CV events (0.2% in each group), a statistically significantly higher incidence of myocardial infarctions (MIs) was evident in rofecoxib-treated patients compared with naproxen-treated patients: 0.5% versus 0.1%.A post-hoc analysis identified 321 patients (4%) enrolled in VIGOR with a prior history of a CV event at baseline that was an indication for low-dose aspirin prophylaxis (12). All of the 8 MIs that occurred in these 321 patients were in patients receiving rofecoxib; none occurred in patients randomized to the naproxen group. In the remaining 96% of the population, the incidence of MIs was 0.2% in the rofecoxib group and 0.1% in the naproxen treatment group.
A Vascular Events Committee, in place before initiation of patient enrollment in VIGOR, adjudicated all CV events in the study, discarding approximately 33% of investigator-reported adverse events due to incomplete documentation (13). Serious thromboembolic CV events were reported by investigators in 64 (1.6%) versus 32 (0.8%) patients in the rofecoxib versus naproxen treatment groups, representing 2.37 and 1.19 events per 100 patient-years, respectively. The number (incidence) of adjudicated events was 45 (1.1%) versus 19 (0.5%), or 1.67 versus 0.70 events per 100 patient-years in the 2 treatment groups. Adjudicated MI rates per 100 patient years were 0.74 in the rofecoxib group and 0.15 in the naproxen group. At the FDA Arthritis Advisory Committee meeting in February 2000, it was queried as to why enrollment in VIGOR was not terminated early due to thromboembolic CV events, because Kaplan-Meier incidences in the 2 treatment groups separated early (within 45–60 days) and remained different thereafter in the trial (Figure 1) (12). The statistician member of the data safety monitoring board answered that more time was required to complete adjudication of CV events than was required for GI events; therefore the VIGOR study was continued until the primary (GI) end point was reached (3).
Three plausible hypotheses may explain differences between reported CV adverse events in CLASS and VIGOR: naproxen administration offers an antithrombotic effect; rofecoxib administration, without inhibition of platelet aggregation, results in a prothrombotic effect; or the results occurred by chance (10). Each will be discussed in turn.
Recent studies show that naproxen, when given twice daily, inhibits platelet aggregation over 24 hours, in comparison with ibuprofen and diclofenac (14). However, in vitro data do not clarify whether transient inhibition of platelet aggregation by naproxen is pharmacodynamically similar to the irreversible effects of aspirin. To date, no RCTs support the conclusion that naproxen is “cardioprotective” and offers either primary or secondary prophylaxis for MIs. A meta-analysis performed by the Antiplatelet Trialists' Collaboration (APTC) in 1994 showed only aspirin effective for secondary prevention of MIs, reducing risk by approximately 25% (15). Subsequent RCTs and meta-analyses reported in the last 2 years have confirmed the efficacy of aspirin in primary prevention of MIs, reducing risk by approximately 30% (16–19). Finally, the recent Primary Prevention Project trial was stopped early for ethical reasons: patients not receiving aspirin had a 29% higher risk for CV events over 3.5 years (20). Several RCTs have reported benefit in secondary prevention of MIs with traditional NSAIDs, flurbiprofen and indobufen. However, RCTs in these small patient populations have failed to demonstrate differences in reocclusion and/or mortality rates; most importantly, they did not include aspirin comparison groups (21, 22).
In the absence of RCTs, epidemiologic studies may help to clarify the possible relationship between NSAID use and risk of MI and other CV events. One epidemiologic analysis in postmenopausal women from the UK General Practice Research Database documented significant risk reductions for CV thromboembolic events with low-dose aspirin, without any evident reduction in risk following administration of conventional NSAIDs (23).
Recent publications remain divided on this issue. Garcia-Rodriguez extended his epidemiologic analysis showing that the relative risk for MIs did not vary by plasma half-life or chemical class of NSAID administered (24). A prospective observational cohort study by Ray et al, utilizing data from the Tennessee Medicaid Program between 1987 and 1998, estimated relative risks for MI in users of naproxen and ibuprofen compared with non-NSAID users (25). In a population with an approximately 4-fold higher incidence of coronary artery disease than patients enrolled in VIGOR, no benefit was evident with regular use of NSAIDs, including naproxen. Although multivariate-adjusted rate ratios for naproxen were “slightly lower than for all NSAIDs,” it was of insufficient magnitude to explain observed differences in CV thromboembolic events between treatment groups in VIGOR. Three publications in the Archives of Internal Medicine support a potential cardioprotective effect for naproxen. A case-control study in the New Jersey Medicare and Medicaid programs, prior to introduction of the COX-2 selective agents, indicated reduced risk associated with naproxen use in the 6 months preceding an MI (26). Inexplicably, the relative risk was less when naproxen was utilized from days 180 to 61 days prior to the event, than within 60 days of or covering the index date. A population based, case-control study of acute MIs in the elderly in Quebec from 1992 to 1994 suggested that only long-term, regular use of naproxen was associated with a protective CV effect (27). And, as presented at the FDA meeting in February, analyses in RA patients ages 40–79 years in the UK General Practice Research Database showed a lower risk for acute CV thromboembolic events associated with naproxen use in the previous year (28). An interesting perspective regarding these observations is offered in an editorial accompanying the article by Ray et al, arguing that long-term NSAID use may, in fact, have a deleterious effect (29). These observations may or may not be related to data reported by Catella-Lawson et al, showing that prior administration of ibuprofen antagonizes the well-known inhibition of thromboxane B2 formation and platelet aggregation by aspirin (30). Finally, Cleland, in the editorial, argues that aspirin treatment may not be cardioprotective (29).
If differences in thromboembolic CV events are to be attributed to selective COX-2 inhibition by rofecoxib, an obvious question is whether data from other RCTs with rofecoxib suggest an excess of thromboembolic CV events. At the advisory panel meeting reviewing the VIGOR trial, the FDA reviewer presented data from a 12-week RCT in OA, comparing rofecoxib 25 mg/day (n = 2,785) with naproxen 500 mg twice daily (n = 2,771), in which 12.1–12.8% of patients were taking low-dose aspirin (12). Serious CV events were reported in 10 rofecoxib- and 7 naproxen-treated patients, with more cardiac events and MIs in the rofecoxib group (7 versus 1) and more ischemic strokes in the naproxen group (0 versus 6).
A recently published meta-analysis of controlled clinical trials with rofecoxib included more than 28,000 patients (31). Using the APTC definitions of thromboembolic CV events, it showed no significant difference in risk between rofecoxib and placebo (relative risk [RR] = 0.84, 95% confidence interval [95% CI] 0.51–1.38) and between rofecoxib and non-naproxen NSAIDs (RR = 0.79, 95% CI 0.40–1.55), and a difference between rofecoxib and naproxen (RR = 1.69, 95% CI 1.07–2.69) that was dominated by the results in RA patients enrolled in VIGOR. However, it should be noted that these end points have previously only been utilized by the APTC group in meta-analyses specifically performed by them, independent of sponsors who conducted the trials. Thromboembolic CV events were more broadly defined in CLASS, including transient ischemic attacks (TIAs). Using this definition, in contrast to VIGOR, there was no excess of serious thromboembolic CV events in any treatment group in CLASS: 52 (1.3%) in the celecoxib group versus 49 (1.2%) in the NSAID group (diclofenac 28 [1.4%] and ibuprofen 21 [1.1%]), representing incidence rates per 100 patient years of 2.24 for celecoxib and 2.22 for combined NSAIDs (32). Similarly, there were no differences in the rates of MIs: 19 (0.5%) for the celecoxib group versus 4 (0.2%) for the diclofenac group and 9 (0.5%) for the ibuprofen group, or 0.82, 0.46, and 0.80 events per 100 patient years, respectively. There was no significant difference in risk for serious CV events between celecoxib and NSAIDs (RR = 1.1, 95% CI 0.7–1.6) (33).
The best direct comparison with VIGOR is the population in CLASS not taking aspirin (78% of patients; Table 2), in which serious thromboembolic CV events were reported in 25 (0.8%) celecoxib-treated patients versus 23 (0.7%) NSAID-treated patients (diclofenac 16 [1.0%], ibuprofen 7 [0.4%]), representing incidence rates per 100 patient-years of 1.39 and 1.34, respectively (25). Crude rates for MIs were 6 (0.2%) for the celecoxib group versus 2 (0.1%) for diclofenac and 2 (0.1%) for ibuprofen, or 0.33, 0.24, and 0.23 events per 100 patient-years, respectively. There were no significant differences in risk for serious CV events between celecoxib and NSAID treatment groups (RR = 1.1, 95% CI 0.6–1.9) (34). A Kaplan-Meier curve of percent annualized incidence for thromboembolic CV events in CLASS revealed celecoxib to be intermediate between diclofenac and ibuprofen, and not significantly different (Figure 2). These incidences are based on investigator-reported adverse events without any adjudication procedure on the part of the sponsor.
The majority of serious thromboembolic CV events in the CLASS trial did not occur in the subpopulation receiving aspirin prophylaxis. In fact, the number of events per 100 patient-years in this “high-risk” population, 21–22% of all enrolled patients in CLASS (n = 412–882), ranged from 3.35 to 5.62. In the 3.9% of patients not taking aspirin for whom aspirin was indicated for secondary prophylaxis, 2 of 156 (1.3%) receiving celecoxib and 1 of 139 (0.7%) receiving NSAIDs had an MI. A review of all RCTs in both OA and RA included in the celecoxib New Drug Application failed to indicate an excess of thromboembolic CV events in the COX-2 selective treatment groups (33, 34).
The CLASS and the VIGOR trial were neither designed nor powered to demonstrate prothrombotic effects of selective COX-2 agents, which would require treatment groups approximating 20,000 (as argued by Dr. J. Faich at the February 7, 2001, FDA advisory committee meeting) (8). The observed differences in thromboembolic CV events between treatment groups in VIGOR require confirmation in another large, adequately powered RCT before these findings can be accepted.
Although discussed publicly at both FDA advisory panel meetings in February 2001, when sponsor and agency briefing documents were made available on the FDA and the Center for Drugs Evaluation and Research web sites, broad recognition and discussion of these potential CV safety issues did not occur until the recent publication of an article by Mukherjee et al in JAMA (35). Time to thromboembolic CV events in the VIGOR trial are presented in the Mukherjee article in Figure 1, yet similar data from the CLASS trial are not graphically shown; instead, incidence rates, without correction for patient exposure and therefore not directly comparable with data from VIGOR, are presented in Figure 2 of the JAMA article, segregated by use/no use of low-dose aspirin prophylaxis. As presented at the FDA meeting, the Kaplan-Meier curve for thromboembolic CV events in the nonaspirin population in CLASS is a more appropriate comparison (Figure 2 compared with Figure 1 in this editorial).
Comparing a meta-analysis of aspirin for primary prevention of coronary artery disease to CV events in CLASS and VIGOR, the report by Mukherjee et al suggests that treatment with selective COX-2 agents predisposes patients to thromboembolic CV adverse events (35). It included the US Physicians' Health Study, the UK Doctors Study, the Thrombosis Prevention Trial (TPT), and the Hypertension Optimal Treatment (HOT) trial; in total, 25,133 patients received aspirin and 23,407 received placebo (36–39). This meta-analysis, designed to determine the benefit/risk profile of aspirin for primary prophylaxis, concluded that the decision to prescribe aspirin required a formal and accurate estimation of an individual's risk for coronary events. Comparing results of the JAMA meta-analysis with those from CLASS and VIGOR is of dubious validity. First, this is not a prospective, randomized comparison and, therefore, relies on numerous assumptions that are not valid in this instance. There are clearly differences in the patient populations enrolled in each trial, with different underlying risks for CV events. Indeed, it has been reported that patients with RA have an increased rate of MI compared with not only patients with OA but also the general population (40–42). There are differences in the methodology of ascertainment, validation, and classification of end points between the studies. Finally, indications for treatment, as well as size and duration of each trial, differ. Applicability of such a meta-analysis to comparisons with CV events in CLASS and VIGOR should be questioned, as should 2 smaller RCTs comparing rofecoxib (at a lower dose of 12.5 mg/day) to nabumetone, as presented in the JAMA article. All patients in 3 of the trials, and 53% in the fourth RCT included in the primary prevention meta-analysis, were men, compared with only 31% and 20% in CLASS and VIGOR, respectively (Table 3). Their mean age was younger, presence of RA or OA was unknown, comorbidities were variable, and sample sizes were significantly larger in 2 RCTs. Trial durations ranged from 46 to 82 months, compared with mean durations of 9 months. Finally Mukherjee et al make no distinction between primary and secondary prevention. This is particularly relevant when retrospective analyses by both sponsors indicated that, in CLASS, 70% of patients receiving prophylactic aspirin and 50% of those not taking aspirin had risk factors for CV disease; in VIGOR, despite exclusion of patients with clinically significant CV disease, 4% of patients should have been receiving prophylactic aspirin, and almost 50% had risk factors for CV disease.
|US n = 2,2071||UK n = 5,139||TPT n = 2,540||HOT n = 18,790||CLASS n = 7,968||VIGOR n = 8,076|
|Mean age, years||–||–||57.5||61.5||60.1||58.2|
|< 60 years, %||75||47||–||–||47||62|
|Duration||5.0 years||6.0 years||6.8 years||3.8 years||9 months||9 months|
The 4 primary prevention studies included in the meta-analysis by Sanmuganathan et al were very heterogeneous, with a 4-fold difference between the lowest and highest incidence of coronary events in the placebo groups (0.36 in TPT versus 1.33 in HOT) (19). Given these extremes, it could be argued that it would be more appropriate to use the mean number of MIs per 100 patient-years from the meta-analysis, 0.77% rather than the weighted mean of 0.52%, as cited by Mukherjee et al (Figure 3) (19, 35). For either comparison, the annualized MI rates in CLASS (0.80%) and VIGOR (0.74%) show no difference from one another and fall within the range of those reported in the individual studies included in the meta-analysis. The more appropriate analysis, however, is to use the CLASS nonaspirin group. The number of MIs per 100 patient-years in this population is 0.33% with celecoxib, 0.24% with diclofenac, and 0.23% with ibuprofen, not different from that with naproxen in VIGOR (0.15%) or the weighted mean rate in the meta-analysis (0.52%), but statistically different from the mean rate (0.77%) in the meta-analysis and the rate for rofecoxib in VIGOR (0.74%) (Figure 4) (33).
The authors of the JAMA article appropriately point out that the cited RCTs did not include sufficient patient numbers to accurately adjudicate CV events, yet raised a clinically significant safety issue. Further, they included in their analysis two 12-week RCTs comparing rofecoxib (at 25% of the dose used in VIGOR) with nabumetone and placebo in OA, completed and reported to the FDA but not yet published, and did not include similar RCTs with celecoxib (34). Although there were no differences in the annualized incidence of MIs in trial 085 (1,042 patients), in trial 090 (978 patients) the MI rate with rofecoxib was 6 (1.5%) compared with 2 (0.5%) with nabumetone and 1 (0.5%) with placebo.
Unfortunately, postmarketing surveillance data derived from the FDA MedWatch reporting system (35) and others such as the World Health Organization/UMC database are not helpful because they do not allow accurate assessment of the incidence of severe or clinically important adverse events (43).
Finally, Mukherjee and colleagues conclude from their analyses that selective COX-2 agents appear to be associated with increased risk for CV events (35). They also conclude with a statement that “aspirin and naproxen … have a cardioprotective effect.” In a letter from the FDA, Division of Drug Marketing, Advertising and Communications to Merck, dated September 17, 2001, the following is stated: “In fact, the situation is not at all clear. There are no adequate and well-controlled studies of naproxen that support [the] assertion that naproxen's inhibition of platelet aggregation is pharmacodynamically comparable to aspirin or clinically effective in decreasing the risk of MIs. … it is also possible that Vioxx has pro-thrombotic properties” (44).
Data from the CLASS and the VIGOR trial demonstrated that patients with arthritis randomized to receive a selective COX-2 inhibitor had a significantly lower risk of developing clinical upper GI ulcers with or without complications than those receiving traditional NSAIDs (1, 2, 45). It has not been conclusively demonstrated that one or both of the currently approved selective COX-2 inhibitors have prothrombotic effects. The comparisons performed by Mukherjee and colleagues are fraught with methodologic limitations, both agents were given at twice the dosage of that approved or recommended for long-term treatment of arthritis, and the trials were neither designed nor powered to demonstrate differences in CV events.
Regardless, guidance is needed for the practitioner prescribing these agents to patients with arthritis. As stated recently by Fitzgerald and Patrono (10), “patients who have had a major cardiovascular event should be treated with low-dose aspirin.” It appears that the preferential GI safety of selective COX-2 inhibitors over traditional NSAIDs is preserved even when used in combination with low-dose aspirin, as indicated in CLASS (46). Because selective COX-2 inhibitors do not inhibit platelet aggregation, patients in need of secondary prophylaxis for MIs or with predisposing risk factors, such as antiphospholipid antibodies, should receive low-dose aspirin when taking these agents. If these patients have identified risk factors for GI adverse events, then concomitant administration of proton pump inhibitors should be seriously considered. For patients who have not had a major cardiovascular event, the practitioner must use his or her clinical judgment whether to prescribe low-dose aspirin for primary prophylaxis, based on the patient's risk factor profile.
- 3Food and Drug Administration, Arthritis Advisory Committee Open Panel Meetings 2/ 7–8/2001: review of CLASS and VIGOR, CLASS and VIGOR Advisory Committee Meetings, Rockville MD, Food and Drug Administration. Accessed October 1, 2001. URL: www.fda.gov/ohrms/dockets/ac/01/briefing/3677 minutes.
- 12Food and Drug Administration, Arthritis Advisory Committee Open Panel Meeting 2/8/ 2001: review of VIGOR, Cardiovascular Safety Review of Rofecoxib Rockville MD, Food and Drug Administration. Accessed October 1, 2001. URL: www.fda.gov/ohrms/dockets/ac/01/briefing/3677b2_06_cardio.pdf
- 13Food and Drug Administration, Arthritis Advisory Committee Open Panel Meeting 2/8/ 2001: review of VIGOR, VIGOR Advisory Committee Briefing Document, Rockville MD, Food and Drug Administration. Accessed October 1, 2001. URL: www.fda.gov/ohrms/dockets/ac/01/briefing/3677b1_merck.pdf.
- 32Food and Drug Administration, Arthritis Advisory Committee Open Panel Meeting 2/7/2001: review of CLASS, CLASS Advisory Committee Briefing Document, Rockville MD, Food and Drug Administration. Accessed October 1, 2001. www.fda.gov/ohrms/dockets/ac/01/briefing/3677 minutes.
- 34Food and Drug Administration, Arthritis Advisory Committee Open Panel Meeting 2/7/2001: review of CLASS, CLASS Advisory Committee Briefing Document, Rockville MD, Food and Drug Administration. Accessed October 1, 2001. www.fda.gov/ohrms/dockets/ac/01/briefing/3677b1_01_searle.pdf.
- 38Medical Research Council's General Practice Research Framework. Thrombosis prevention trial: randomized trial of low intensity oral anticoagulation with warfarin and low dose aspirin in the primary prevention of ischemic heart disease in men at increased risk. Lancet 1998; 351: 233–41.
- 44Letter from the Food and Drug Administration, Division of Drug Marketing, Advertising and Communications [DDMAC] to Merck and Co, September 17, 2001.
- 46Prescribing information: celecoxib capsules. Food and Drug Administration. Revised June 7, 2002.