Selective cyclooxygenase 2 inhibitors and cardiovascular events


  • Daniel H. Solomon

    Corresponding author
    1. Brigham and Women's Hospital, Boston, Massachusetts
    • Division of Pharmacoepidemiology, Division of Rheumatology, Immunology, and Allergy, Department of Medicine, Brigham and Women's Hospital, 1620 Tremont Street, Suite 3030, Boston MA 02120
    Search for more papers by this author


Just when rheumatologists thought we had more specific and less toxic drugs for our patients, we learned that selective cyclooxygenase 2 (COX-2) inhibitors (coxibs) may not provide a true overall safety advantage compared with traditional nonselective nonsteroidal antiinflammatory drugs (NSAIDs). A report in the lay press from the National Institutes of Health (1) even called into question the cardiovascular safety of naproxen, one of the “workhorses” in a rheumatologist's armamentarium. Naproxen has been considered safe enough to be sold in the US without a prescription.

While NSAIDs and coxibs provide important analgesic and antiinflammatory benefits to millions of patients, the cardiovascular safety concerns have led many patients to discontinue their use. Reports about potential links between these agents and adverse cardiovascular outcomes have been presented by the media before publication in the peer-reviewed literature and have left treating doctors unsure whether yesterday's recommended drug might be on today's “blacklist.” The alphabet soup of trials—APPROVe, APC, PreSAP, ADAPT, and others—is dizzying, and without reporting of important details from these trials, rhetoric and finger-pointing have been more common than scientific debate.

Reviewing a subject that is a “moving target” is not without hazard. With this caveat in mind, this review will focus on the basic and clinical science underlying the current controversy regarding the cardiovascular safety of coxibs. First, I will summarize the proposed mechanisms underlying the relationship between thrombotic cardiovascular events and these medications. Second, I will examine the key randomized controlled trials that have been published or presented publicly and the relevant cardiovascular safety information. Finally, I will discuss the data from published observational epidemiologic studies. This review will concentrate on thrombotic cardiovascular outcomes, mainly acute myocardial infarction (MI) and ischemic cerebrovascular accident (CVA). Data on congestive heart failure will be noted when available, but few studies have addressed this outcome. There are scant data available on the cardiovascular safety of nonselective NSAIDs, and thus they will not be the focus of this review. Etoricoxib will not be included because safety data on this agent have not yet been published in the peer-reviewed literature.

Biologic plausibility

Early during the development and testing of the coxibs, studies in humans suggested that these agents may be associated with adverse cardiovascular events. Selective COX-2 inhibition had been heralded as a means of deriving analgesic and antiinflammatory benefit without the risk of bleeding. This assertion was based on the understanding that platelet-derived COX-1 is the major source of thromboxane (2). However, the origins of circulating prostacyclin, the other important vasoactive eicosanoid, were not as clear. The importance of prostacyclin lies partly in its reciprocal relationship with thromboxane—thromboxane critical for platelet aggregation and vasoconstriction, and prostacyclin for dampening the effects of thromboxane through fibrinolysis and vasodilatation.

Studies of healthy volunteers showed that urine levels of a prostacyclin metabolite were reduced in subjects taking rofecoxib, suggesting that endovascular production of prostacyclin may be COX-2 mediated (3). Additional concerns were raised by studies in animal models showing that COX-2 is up-regulated in vascular segments under conditions of increased shear stress (4). One potential implication of these observations is that COX-2 plays a beneficial role in vascular health and that its inhibition may create an imbalance between thromboxane and prostacyclin, thereby favoring thrombosis and vasoconstriction. The potential risk of thrombosis based on these mechanisms was articulated in an early review of coxibs (5). Furthermore, this risk was highlighted in a report of a case series describing 4 patients with connective tissue disease who developed thrombotic complications (6). While the prothrombotic potential of coxibs does not seem to be relevant in most patients, results of a series of elegant animal studies using knockout mice suggest that an imbalance between prostacyclin and thromboxane may account for some of the cardiovascular risk associated with these agents (7, 8). It is unclear whether this imbalance may also link NSAIDs with cardiovascular outcomes. Even nonselective NSAIDs are differentially selective for COX-1 and COX-2 (9) (Figure 1).

Figure 1.

Concentrations of selected nonsteroidal antiinflammatory drugs and coxibs required to inhibit 50% of the cyclooxygenase 1 (COX-1) and COX-2 enzymatic reactions in assays with whole human blood (50% inhibition concentrations [IC50]). Reproduced from ref. 9.

The vasoreactive potential of coxibs was further demonstrated in several studies in which flow-mediated dilatation in humans was measured. This test assesses the release of nitric oxide from the endothelium and has been established as an important surrogate measure of cardiovascular health, with reduced flow-mediated dilatation associated with a greater number of future cardiovascular events (10). Separate studies have tested the effects of rofecoxib and celecoxib on flow-mediated dilatation. Two small studies with celecoxib were both ≤2 weeks in duration (11, 12); subjects were hypertensive patients without known coronary artery disease in 1 study and men with stable coronary artery disease in the other. Both of these studies demonstrated an improvement in flow-mediated dilatation while patients were taking celecoxib 200 mg twice daily compared with placebo. Two studies investigating effects of rofecoxib 25 mg daily in patients with known coronary artery disease showed no change in flow-mediated dilatation compared with placebo over the 2–6-month study periods (13, 14).

Other possible mechanisms linking certain coxibs (and NSAIDs) to long-term cardiovascular outcomes include changes in blood pressure and possible oxidative modification of biologic lipids. The hypertensive potential of NSAIDs was recognized more than a decade ago, with some suggestion that certain agents had greater effects on blood pressure than others (15, 16). Celecoxib and rofecoxib have been noted to have differing effects on blood pressure. Findings of a recent meta-analysis suggest that both agents are associated with elevations in blood pressure and that rofecoxib is associated with a 2.8–mm Hg greater rise in systolic blood pressure compared with celecoxib, but with no relative elevation in diastolic blood pressure (17). Much less is known about the effect of different agents on oxidative modification of lipids leading to atherosclerosis. Findings of 1 study have suggested that rofecoxib and etoricoxib, agents with a sulfone moiety, exhibit pro-oxidant activity leading to F2-isoprostane formation; this activity is not associated with their COX activity and was not observed with other coxibs and NSAIDs (18).

No one mechanism linking coxibs (or NSAIDs) to cardiovascular events has been proven. However, several compelling possibilities have been described: thrombosis and vasoconstriction associated with an imbalance between thromboxane and prostacyclin, hypertension from inhibition of prostaglandin-dependent counterregulatory mechanisms, and COX-independent oxidative stress. These potential mechanisms could explain both short-term and long-term risks, similar to those suggested by the clinical trial and observational data.

Randomized clinical trial data

It is difficult to gain a complete picture of the cardiovascular safety of coxibs (or NSAIDs) based on the existing randomized clinical trial data. Several factors contribute to this characterization. Few trials were long in duration or enrolled enough subjects to observe adequate numbers of cardiovascular events. When cardiovascular events were recorded and evaluated, definitions of these outcomes were often created retrospectively and differed across studies. Except for several prevention studies, most recent trials have compared a coxib with an NSAID in patients with arthritis. While it is difficult to imagine conducting a placebo-controlled trial in patients with arthritis, the use of active NSAID comparators has created complexity in terms of interpreting the relative cardiovascular event rates, because these agents may also affect cardiovascular outcomes. Comparison of one coxib trial with another is also fraught with difficulty because of the different populations recruited across studies (19). Finally, none of the trials have been large enough to determine whether the risk of cardiovascular events associated with coxibs is concentrated in certain subgroups of patients. Of course, patients with many cardiovascular risk factors have larger cardiovascular event rates, but the relative risk of cardiovascular events may be higher in coxib users with fewer risk factors. The current data do not address the critical question of whether coxibs are “safe” from a cardiovascular standpoint in certain subgroups of patients—perhaps, for example, younger persons without known cardiovascular risk factors. However, patients in this subgroup rarely “require” the potential gastrointestinal (GI) safety afforded by coxibs.


Rofecoxib was withdrawn from the market by Merck on September 30, 2004, after it was determined in an adenomatous polyp prevention trial that long-term treatment with the drug at 25 mg/day was associated with a doubling of the risk for thrombotic cardiovascular events. However, the potential for an elevated risk of thrombosis with rofecoxib treatment was widely recognized 4 years earlier, with publication of the Vioxx Gastrointestinal Outcomes Research (VIGOR) trial (20). In that trial, event rates diverged after ∼6 weeks, and the hazard ratio (the event rate in the rofecoxib users compared with the event rate in the naproxen users) did not appear proportional over the followup period (21) (Figure 2).

Figure 2.

Thrombotic cardiovascular events experienced by patients in the VIGOR (Vioxx Gastrointestinal Outcomes Research) trial. Note that the event curves diverge after ∼6 weeks. Vertical bars show the 95% confidence intervals (95% CIs). Reproduced from ref. 21.

The Adenomatous Polyp Prevention on Vioxx (APPROVe) trial tested rofecoxib 25 mg daily against placebo among 2,586 patients with a prior adenomatous polyp (22) (Table 1). Seventeen percent reported daily aspirin use at the time of study enrollment. Patients in the rofecoxib arm experienced a 4.61-fold increase (95% confidence interval [95% CI] 1.50–18.83) in the risk of congestive heart failure, which became clear early during followup (Figure 3). Severe thrombotic cardiovascular events (acute MI, CVA, unstable angina, sudden cardiac death, transient ischemic attack, peripheral arterial thrombosis, peripheral venous thrombosis, and pulmonary embolus) were approximately twice as frequent in the rofecoxib arm than in the placebo arm. Divergence in these curves became apparent after 18 months of followup (Figure 4).

Table 1. Cardiovascular events observed in long-term trials of coxibs and NSAIDs*
Trial (ref.)Patient populationMedian followup, monthsAspirin, %Coxib armComparator armRelative risk (95% CI)
  • *

    Some of the cardiovascular events were not analyzed in the references noted. When post-publication analysis changed numbers, the revised numbers from the Food and Drug Administration reviewer's reports ( were used. NSAIDs = nonsteroidal antiinflammatory drugs; 95% CI = 95% confidence interval; VIGOR = Vioxx Gastrointestinal Outcomes Research study; RA = rheumatoid arthritis; bid = twice a day; APPROVe = Adenomatous Polyp Prevention on Vioxx trial; CLASS = Celecoxib Long-term Arthritis Safety Study; OA = osteoarthritis; tid = 3 times a day; APC = Adenoma Prevention with Celecoxib trial; PreSAP = Prevention of Spontaneous Adenomatous Polyps study; AD = Alzheimer's disease; NA = not available; CABG = coronary artery bypass grafting; TARGET = Therapeutic Arthritis Research and Gastrointestinal Event Trial.

  • Number of cardiovascular events per 100 person-years. The definition of events differed by study (see text).

  • When not available, the relative risks were calculated as the crude event rate in coxib users divided by the crude event rate in the comparator group.

  • §

    Same comparator group as shown immediately above for celecoxib 200 mg bid.

  • Medications were administered intravenously (parecoxib) for the first 3 days and then orally for 11 more days in CABG I and for 7 more days in CABG II.

  • #

    Same comparator group as shown immediately above for parecoxib/valdecoxib 20 mg bid.

 VIGOR (20)RA904,047Rofecoxib 50 mg/day451.674,029Naproxen 500 mg bid190.702.38 (1.39–4.00)
 APPROVe (22)Adenomatous polyp30161,287Rofecoxib 25 mg/day461.501,299Placebo260.781.92 (1.19–3.11)
 CLASS (24)OA and RA9223,987Celecoxib 400 mg bid341.51,985Ibuprofen 800 mg tid201.80.83
        1,996Diclofenac 75 mg bid151.41.07
 APC (26)Adenomatous polyp≥3330685Celecoxib 200 mg bid160.78679Placebo70.342.3 (0.9–5.5)
    671Celecoxib 400 mg bid231.14§§§§3.4 (1.4–2.8)
 PreSAP (28)Adenomatous polyp≥3316933Celecoxib 400 mg/day200.72628Placebo120.641.1 (0.6–2.3)
 AD 97-02-001 (28)Mild-to-moderate AD12NA285Celecoxib 200 mg bid11NA140Placebo3NA1.80
 CABG I (29)Post-CABG1.5100311Parecoxib/valdecoxib 20 mg bid24NA151Placebo/placebo4NA2.91
 CABG II (30)Post-CABG1.3100555Parecoxib/valdecoxib 20 mg bid11NA560Placebo/placebo3NA2.0 (0.5–8.1)
    556Placebo/valdecoxib 20 mg bid6NA####3.7 (1.0–13.5)
 TARGET (27)OA12224,376Lumiracoxib 400 mg/day190.594,397Ibuprofen 800 mg tid230.740.76 (0.41–1.40)
   254,741Lumiracoxib 400 mg/day401.104,730Naproxen 500 mg bid270.761.46 (0.89–2.37)
Figure 3.

Congestive heart failure (CHF), pulmonary edema (PE), and cardiac failure (CF) events experienced by patients in the APPROVe (Adenomatous Polyp Prevention on Vioxx) trial. Note that the event curves diverge after ∼6 weeks. Vertical bars show the 95% confidence intervals. Reproduced, with permission, from ref. 22 (copyright 2005, Massachusetts Medical Society).

Figure 4.

Thrombotic cardiovascular events experienced by patients in the APPROVe (Adenomatous Polyp Prevention on Vioxx) trial. Note that the event curves appear to diverge after ∼18 months; however, the 95% confidence intervals (vertical bars) are wide. Reproduced, with permission, from ref. 22 (copyright 2005, Massachusetts Medical Society).

The VIGOR trial compared 50 mg daily of rofecoxib versus 500 mg twice daily of naproxen in 8,076 patients with rheumatoid arthritis (Table 1). Aspirin users were excluded from this trial. While the patients in the rofecoxib arm experienced fewer adverse GI outcomes, they experienced acute MIs 5 times more frequently, and the relative risk for severe thrombotic cardiovascular events (acute MI, CVA, unstable angina, sudden cardiac death, transient ischemic attack, peripheral arterial thrombosis, peripheral venous thrombosis, and pulmonary embolus) was doubled. Several other trials with rofecoxib 25 mg had not demonstrated an increased cardiovascular risk, although these were generally of shorter duration (23).


The CLASS (Celecoxib Long-term Arthritis Safety Study) compared celecoxib 400 mg twice daily with ibuprofen 800 mg 3 times daily or diclofenac 75 mg twice daily (24) (Table 1). Patients with osteoarthritis or rheumatoid arthritis were enrolled. At trial enrollment, 22% of the patients reported regular aspirin use. Rates of cardiovascular events (acute MI, CVA, or cardiovascular death) did not differ significantly between the celecoxib users and the combined NSAID user group. Some have suggested that cardiovascular event rates among the non–aspirin users were higher among celecoxib users than the NSAID comparators; however, this difference has not been shown to be statistically significant (25).

The Adenoma Prevention with Celecoxib (APC) trial demonstrated that high-dose, long-term celecoxib use was associated with an increase in the risk of thrombotic cardiovascular events (acute MI, CVA, or cardiovascular death) (26) (Table 1). In this chemoprevention trial, 2,035 patients with prior adenomatous polyps were randomized to receive either celecoxib 200 mg twice daily, celecoxib 400 mg twice daily, or placebo. At trial entry, 30% of patients reported aspirin use. Thirty-three months after its initiation the trial was halted by the National Cancer Institute because of an increase in the risk of cardiovascular events observed in both celecoxib arms (Table 1 and Figure 5). There did appear to be a dose-response effect, with the higher dosage of celecoxib associated with more frequent cardiovascular events. Congestive heart failure occurred in 2 subjects (0.3%) in the placebo arm versus 5 (0.4%) in both celecoxib arms.

Figure 5.

Thrombotic cardiovascular events experienced by patients in the APC (Adenoma Prevention with Celecoxib) trial. Reproduced, with permission, from ref. 26 (copyright 2005, Massachusetts Medical Society).

A number of post hoc analyses of the APC data examined whether patient factors could identify those at particularly high or low risk for thrombotic cardiovascular events. While results of these analyses suggest that older age, male sex, and the presence of cardiovascular risk factors may identify different groups with a higher likelihood of cardiovascular events, none of the interaction terms (between treatment arm and patient factor) reached statistical significance. Importantly, the aspirin user subgroup (hazard ratio 3.8, 95% CI 0.9–16.6) did not experience a lower risk of cardiovascular events than those not using aspirin (hazard ratio 2.4, 95% CI 0.9–6.4). The wide and overlapping CIs around these estimates demonstrate that we have too little data to clarify whether aspirin will have a role as cotherapy with coxibs for cardioprotection. However, we do know that patients taking aspirin do not experience a reduced risk of GI toxicity while taking coxibs compared with nonselective NSAIDs (25, 27).

A second adenomatous polyp prevention trial, the PreSAP (Prevention of Spontaneous Adenomatous Polyps) study (28), also tested celecoxib versus placebo (Table 1). In that yet-to-be-published trial, 1,561 subjects were randomized to receive once-daily (not twice) dosing of celecoxib at 400 mg or placebo. After 33 months of followup, there was no increase in cardiovascular events (acute MI, CVA, or cardiovascular death) observed in the celecoxib arm versus the placebo arm.

The last celecoxib trial to be discussed was completed several years ago and has not been reported in the peer-reviewed literature, but it was presented at the Food and Drug Administration (FDA) Arthritis and Drug Safety and Risk Management Advisory Committee meeting in February 2005 (28) (Table 1). This trial enrolled 425 patients with mild-to-moderate Alzheimer's disease; 285 received celecoxib 200 mg twice daily and 140 received placebo. Over the 52-week study, there was an increase in the occurrence of thrombotic cardiovascular events (acute MI, CVA, and peripheral thrombotic events) among patients taking celecoxib compared with placebo. However, the findings of the trial are difficult to interpret in light of substantial baseline imbalance between the 2 randomized groups: the rate of coronary artery disease at baseline was 4 times higher among patients in the celecoxib group than among those in the placebo group.


There have been several short-term studies of valdecoxib in patients with arthritis, but none have had adequate numbers of events to enable one to deem the medication safe or unsafe. However, the results of trials in patients undergoing coronary artery bypass grafting (CABG), the CABG I (29) and CABG II (30), have shown an increased risk of cardiovascular events in patients receiving valdecoxib or its intravenous formulation, parecoxib (Table 1). These trials were designed to determine whether valdecoxib might have a role in postoperative pain management. In CABG I, 462 patients were enrolled, in a 2:1 active treatment:placebo ratio; those in the valdecoxib arm (n = 311) received the intravenous formulation within 30 minutes after extubation. Intravenous or oral valdecoxib 20 mg twice daily was continued for 14 days. All types of serious adverse events occurred twice as frequently among patients in the valdecoxib arm (19%) than among those in the placebo arm (9.9%). Cardiovascular events (acute MI, CVA, congestive heart failure, thrombophlebitis, death) were also more common with valdecoxib treatment than with placebo.

In CABG II 1,671 patients were assigned to 1 of 3 treatment arms: intravenous parecoxib followed by valdecoxib, intravenous placebo followed by valdecoxib, or intravenous placebo followed by oral placebo. Intravenous treatment was continued for 3 days and then oral treatment for 7 days, both at 20 mg twice daily. The risk ratio for confirmed cardiovascular events (acute MI, unstable angina, sudden death) was 3.7 when the parecoxib/valdecoxib group was compared with the double-placebo group (Table 1).


Lumiracoxib, a coxib not currently approved for use in the US, has a structure different from that of the other agents. It has a very short half-life but can be administered once daily because of its lipophilic nature. It has been approved by the European Medicines Agency for use at 60–120 mg daily by patients with osteoarthritis, rheumatoid arthritis, or acute gouty arthritis. The TARGET (Therapeutic Arthritis Research and Gastrointestinal Event Trial) compared 1-year therapy with lumiracoxib 400 mg once daily versus naproxen 500 mg twice daily or ibuprofen 800 mg 3 times daily (27). Osteoarthritis patients ages 50 and older were randomized and stratified by age and use of low-dose aspirin. The rates for the primary cardiovascular end point (acute MI, CVA, and cardiovascular death) were similar among lumiracoxib and NSAID users (Table 1). The comparisons between lumiracoxib and each NSAID (ibuprofen and naproxen) individually suggested some possible differences, but none were statistically significant.

Observational data

While randomized clinical trial data are the gold standard for judging a drug's benefits, in many trials participants are not studied for more than several weeks, and relatively healthy subjects are preferentially enrolled. Thus, a drug's true side effect profile is often not well understood prior to its marketing. After the coxibs were approved for marketing and their use began to grow, large-scale trials that would enroll enough patients at risk for cardiovascular events were slow to be organized. Eventually, such trials were mounted to test whether these drugs would be effective for other indications, such as adenomatous polyp prevention and Alzheimer's disease. As discussed above, the cardiovascular adverse event analyses from trials such as APPROVe and APC enhanced our understanding of the coxibs' cardiovascular risk profile.

Epidemiologic analyses of postmarketing data can also assist in detecting a potential adverse event “signal” and can be used to study specific questions that would be difficult to examine in trials. For example, rare events that might develop only in specific subgroups of patients (those taking concomitant medications or with specific comorbid conditions) will always be difficult to detect in trials because of limited recruitment. Moreover, well-conducted epidemiologic studies with typical patients may reflect the actual (or “true”) adverse event experience more closely than the somewhat contrived setting of a randomized clinical trial. These pharmacoepidemiologic analyses would include patients with the full range of comorbid conditions, taking concomitant medications, and using the drug in question in typical dosages at usual intervals.

Several pharmacoepidemiologic analyses on coxibs and cardiovascular disease that have been published (Table 2). Ray and colleagues examined celecoxib and rofecoxib use among Medicaid beneficiaries in Tennessee (31). Subjects were new users of these agents and were compared with nonusers. Adjusted models used Poisson regression, which assumes a constant risk of outcomes over the entire period of followup; this is an assumption that appears not to be true based on the event curves from the VIGOR and APPROVe trials. The end point of interest was a composite of acute MI or cardiac death. The study revealed no increased risk with celecoxib and no increased risk with rofecoxib at ≤25 mg/day, but a near-doubling of the risk with rofecoxib at >25 mg/day.

Table 2. Relative risk of cardiovascular events observed in epidemiologic studies of coxibs*
Author, year (ref.)Patient population (mean age, years)CoxibComparatorRelative risk (95% CI)
Agent, daily dosenEventsGroupnEvents
  • *

    HMO = health maintenance organization (see Table 1 for other definitions).

  • See text for study-specific definitions of cardiovascular events.

  • Some of the studies included multiple comparator groups. To simplify comparisons across studies, the number of comparators has been limited.

  • §

    Nonuse refers to no NSAID or coxib use.

  • Same comparator group as shown above for the given study.

Ray et al, 2002 (31)Medicaid (62)Rofecoxib, ≤25 mg20,24555Nonuse§202,9163,0851.01 (0.77–1.33)
  Rofecoxib, >25 mg3,887131.70 (0.98–2.95)
  Celecoxib, any dose22,337740.96 (0.76–1.21)
Mamdani et al, 2003 (36)Ontario (76)Rofecoxib, any dose12,15658Nonuse100,0004181.0 (0.8–1.4)
 Celecoxib, any dose15,271750.9 (0.7–1.2)
Solomon et al, 2004 (32)Medicare (79)Rofecoxib, ≤25 mg876202Nonuse49,0449,7931.21 (1.01–1.44)
 Rofecoxib, >25 mg65231.70 (1.07–2.71)
  Celecoxib, ≤200 mg1,7673410.92 (0.83–1.03)
  Celecoxib, >200 mg373740.94 (0.74–1.19)
Graham et al, 2005 (34)HMO (68)Rofecoxib, ≤25 mg24658Remote use23,3784,6581.23 (0.89–1.71)
 Rofecoxib, >25 mg18103.00 (1.09–8.31)
  Celecoxib, any dose6171260.84 (0.67–1.04)
Kimmel et al, 2005 (35)Community (55)Rofecoxib, any dose10527Nonuse5,8451,3541.16 (0.70–1.93)
 Celecoxib, any dose105180.43 (0.23–0.79)
Shaya et al, 2005 (37)Medicaid (NA)Rofecoxib, any dose497NAOther NSAIDs5,245NA0.99 (0.76–1.30)
Celecoxib, any dose507NA1.19 (0.93–1.51)

My colleagues and I conducted a similar analysis using data on low-income subjects from Pennsylvania or New Jersey who were Medicare beneficiaries and also had a drug benefit (32). We drew from a source population of ∼500,000 persons. The primary analysis used 1999–2000 information. All analyses were prespecified, with the end point being acute MI. We compared coxib users with nonusers, as well as with each other and with NSAID users. Multivariate models included information on potential confounders that preceded use of the agents of interest. The primary analysis revealed an increased risk of acute MI among rofecoxib users compared with nonusers (Table 2) and with celecoxib users. There was a trend toward an increased risk with rofecoxib compared with NSAIDs, but this did not reach statistical significance. Prespecified dosage analyses with all comparators revealed a dose-response relationship between rofecoxib and acute MI. In addition, secondary analysis among new users revealed that the first 90 days comprised the period of highest risk associated with rofecoxib use (odds ratio 1.39), compared with later than 90 days (odds ratio 0.96). Celecoxib was not associated with an increased risk of acute MI in any of these analyses. While the potential effect of confounders was deemed minimal based on examination of an external data set (33), our analysis was limited by lack of information on some cardiovascular risk factors.

An FDA-sponsored epidemiologic analysis examined members of a large health maintenance organization (34). This analysis compared current use of coxibs and NSAIDs with remote use of these agents (>60 days prior). The end points included acute MI and sudden cardiac death. The primary analysis showed a statistically significant increase in risk with rofecoxib at daily dosages >25 mg (Table 2). At dosages of ≤25 mg/day, the risk was elevated but did not reach statistical significance. Celecoxib was not associated with an increase in cardiovascular risk. Naproxen use was associated with a slight increase in risk. An important strength of this study was the use of patient surveys to augment cardiovascular risk factor data.

Kimmel and colleagues conducted an epidemiologic analysis in 36 hospitals located in 5 counties (35). They selected cases of acute MI from hospital discharge information. They then attempted to enroll patients who had had acute MI and controls from the same county. The response rate was 50–55% for each group. Drug exposure information was gathered as part of a detailed telephone survey conducted by trained interviewers. Modest response rates and potential recall bias limit this study, but the detailed collection of cardiovascular risk factor data and aspirin information is an important strength. The results of this study are consistent with those of other analyses (Table 2).

Two observational epidemiologic studies, from Ontario and Maryland, have shown no increased risk with any coxib (36, 37) (Table 2). Each of these studies had potential limitations. The Canadian study (36) restricted the analysis to subjects who filled multiple prescriptions. Thus, events that occur with a first and only prescription may have been missed. In addition, these analyses may have overcontrolled for potential mediators of the coxibs' effects on the cardiovascular system, since they included hypertension, congestive heart failure, and angina diagnosed after the initiation of treatment as adjusters in multivariate models. The methods used by the Maryland investigators (37) are complex, but a key question is whether use of coxibs and/or NSAIDs was required to overlap with the event dates. If drug exposure on the event date was not confirmed, it is impossible to draw any meaningful conclusions about the potential association between coxibs and cardiovascular events.

All epidemiologic studies have limitations that need to be described and their implications understood. A presentation at the American College of Rheumatology 2004 annual scientific meeting described these potential methodologic problems (38). They include the study of prevalent (versus new) users (39), overcontrolling for factors on the causal pathway, and proportional hazards violations or assumptions that hazards were constant over time. Some of these issues may have been important limitations of the epidemiologic studies of coxibs. However, it is interesting to ponder whether the “signal” of adverse cardiovascular events found in several epidemiologic studies should have raised concerns sooner than September 2004, when the APPROVe trial results were announced.

Coxibs: are they all the same?

As long-term trial data have become available, the similarity in cardiovascular risk between agents has emerged. However, there may be a gradient of cardiovascular risk across coxibs. It is not uncommon for one member of a therapeutic class to show an increased risk of a specific adverse event. Two notable examples of this include bromfenac, an NSAID that was briefly marketed in the mid 1990s but withdrawn because of several early cases of liver failure (40), and cerivastatin, a lipid-lowering drug removed from the market because of an increased risk of rhabdomyolysis (41).

Coxibs in the US have different molecular structures and different potency for COX-2 inhibition. As noted above, there are data suggesting that they may interact with the vascular system in different ways. This may result in distinct patterns of nitric oxide release, blood pressure regulation, and oxidative stress. However, it may be that at certain daily or cumulative doses in certain people (due to interindividual pharmacogenomic differences), the agents may be more similar than different. These issues are summarized in Table 3.

Table 3. Are all coxibs associated with the same cardiovascular risk?
  1. * COX-2 = cyclooxygenase 2; FMD = flow-mediated dilatation; RCTs = randomized controlled trials; ADAPT = Alzheimer's Disease Antiinflammatory Prevention Trial (see Table 1 for other definitions).

• Same mechanism of action across coxibs• Different potency for COX-2 inhibition
 • Different level of blood pressure elevation
 • Possible differences in FMD
 • Possible differences in oxidative stress
• Many RCTs, across coxibs, show an increased cardiovascular risk (e.g., VIGOR, APPROVe, APC, CABG-I, CABG-II, AD 97-02-001)• But not all RCTs show an increased risk (e.g., CLASS, PreSAP, ADAPT, many short-term studies)
• Observational data consistently show differences between coxibs

The discrepancy between findings in the APC trial, in which a significantly increased cardiovascular risk caused by celecoxib was found (Table 1), and the observational data, with at least 6 studies showing no cardiovascular risk with this agent (Table 2), warrants inspection. While this difference may be due to methodologic limitations of all of the observational studies, it is more likely that real differences between the APC trial and the observational studies explain the variation. In the APC trial, subjects took celecoxib in high doses twice daily for many months. In typical practice, some patients use similar dosing schedules, but this is probably the exception. Most patients take no more than 200 mg per day, many take the drug once daily as opposed to twice, and few continue the drug as long as patients in the APC trial did. Thus, there may be “safer” coxib dosing schedules for the small group of people in whom therapy with these drugs is needed on a continuing basis.

Patient subgroups: identifying “low-risk” patients

The cardiovascular risk data (as well as the economic cost) suggest that coxib use should be limited to certain patient groups. It may be that coxibs are appropriate for a small group of patients based on risk/benefit considerations. How might we identify that group? It would be ideal from a clinical standpoint if patients who are at low risk, or alternatively, at high risk, for coxib-associated cardiovascular events could be identified. This would allow clinicians and patients to weigh the risk of GI toxicity from NSAIDs against the risk of cardiovascular events from coxibs.

Factors that might help to risk-stratify patients include clinical, demographic, and genetic data (42). The genetic markers may not be available immediately, but perhaps clinical and demographic factors can be used to risk-stratify. Indices for predicting cardiovascular risk, such as the Framingham Risk Score (43), are well accepted. Known cardiovascular risk factors can help identify patients at increased absolute risk of cardiovascular events. However, it is unclear whether these factors would help to identify groups of patients at an increased relative risk for coxib-associated events. Many investigators have suggested this, but currently, the value of risk-stratifying coxib users is unproven.


Development of safer analgesic treatments is important for patients. Investigators in the Nurses' Health Study found that among women 52–77 years of age, 10.8% report using NSAIDs ≥6 days per week, and 26.7% on at least 1 day per week (44). While there has been recent evidence for a reduction in NSAID-associated GI morbidity, the Arthritis, Rheumatism, and Aging Medical Information System investigators suggested that ∼16,500 deaths and 107,000 hospitalizations annually appear to be related to NSAID-associated GI toxicity (45, 46). Since many rheumatic disease patients are older or have systemic inflammatory conditions that already place them at an increased risk for adverse cardiovascular outcomes, it may be possible to “trade” an increased risk of cardiovascular events for the potential of reduced GI toxicity.

The mechanisms underlying the increased risk of cardiovascular events with coxibs are not clear. An apparent short-term risk in some studies and a long-term risk in others is difficult to explain. However, it may be that at different dosages, with different dosing frequencies, in different populations, the dominant mechanisms linking coxibs to cardiovascular outcomes differ. Moreover, the degree of COX-2 selectivity that confers risk is unclear. As noted in Figure 1, the nonselective NSAIDs differentially inhibit COX-1 and COX-2. Diclofenac inhibits COX-2 almost to the same degree that celecoxib does. In several trials, the rate of cardiovascular events in the NSAID comparator arms was similar to that in the coxib arm (Table 1). Thus, an important question is whether some or all of the nonselective NSAIDs also confer cardiovascular risk.

The subgroup analyses from the APC trial suggest that aspirin may not abrogate the potential cardiovascular harm of coxibs. This point is also underscored by the fact that valdecoxib with concomitant aspirin treatment in the CABG I and CABG II trials was associated with an increased cardiovascular risk compared with placebo. Both of these findings provide evidence that the mechanism underlying the cardiovascular risk associated with coxibs is more complicated than a simple imbalance between COX-1 and COX-2 inhibition.

This review has not focused on the policy implications of the recent events regarding coxibs and NSAIDs, but several lessons are apparent. Politicians, regulators, physicians, and patients have realized that we often know less about the safety of marketed medications than we would suppose. This is not only true for coxibs and NSAIDs, but extends to other rheumatic disease drugs (e.g., biologic disease-modifying antirheumatic drugs) and more broadly. The recent FDA and European Medicines Agency proceedings have facilitated a more complete public disclosure of important data and focused attention on the postmarketing surveillance process. The options for analgesia are many, but coxibs and NSAIDs may have important roles for certain patient groups. These subgroups cannot be well defined based on the available data. Thus, as we grapple to formulate appropriate drug safety policy, research on patient safety must accelerate.


On April 7, 2005 (after the current report was accepted for publication), the FDA requested that valdecoxib's manufacturer withdraw the drug from the US market based on insufficent long-term cardiovascular safety data, evidence of an increased cardiovascular risk in short-term studies among patients undergoing heart surgery, the risk of life-threatening skin reactions, and no proven advantage over other NSAIDs. The manufacturer complied, and valdecoxib is no longer available in the US. In addition, the FDA has asked the manufacturers of all NSAIDs, including celecoxib, to include a “boxed warning” highlighting the potential cardiovascular risks of these agents (; accessed April 12, 2005).