Nonaspirin NSAIDs, Cyclooxygenase 2 Inhibitors, and the Risk for Stroke

Nonaspirin NSAIDs, Cyclooxygenase 2 Inhibitors, and the Risk for Stroke Christianne L. Roumie, MD, MPH; Edward F. Mitchel, Jr, MS; Lisa Kaltenbach, MS; Patrick G. Arbogast, PhD; Patricia Gideon, RN; Marie R. Griffin, MD, MPH Background and Purpose There is limited information regarding the cerebrovascular safety of cyclooxygenase 2 inhibitors (coxibs) and noncoxib nonsteroidal antiinflammatory drugs (NSAIDs). We determined whether specific NSAIDs, including coxibs, are associated with risk of stroke. Methods Retrospective cohort study among Tennessee Medicaid enrollees aged 50 to 84 years between January 1, 1999 and December 31, 2004. Noninstitutionalized persons with continuous enrollment in Medicaid and no stroke or other serious medical illness in the year before cohort entry were included. The 7 most common NSAIDs were examined: celecoxib, rofecoxib, valdecoxib, ibuprofen, naproxen, diclofenac, and indomethacin. Nonuse of NSAIDs was the reference group. Because new use is less susceptible to bias, we conducted a similar analysis confined to new users. The outcome was hospitalization for an incident cerebrovascular event: ischemic stroke, intracerebral hemorrhage, and subarachnoid hemorrhage. Results The cohort included 336 906 persons, with 989 826 person-years of follow-up, and 4354 stroke hospitalizations. There were 4.51 strokes per 1000 person years in the nonuse group, 5.15 strokes per 1000 person years (adjusted HR 1.28, 95% CI 1.06, 1.53) with rofecoxib use, and 5.95 strokes per 1000 person years (adjusted HR 1.41, 95% CI 1.04, 1.91) with valdecoxib use. New use of rofecoxib and valdecoxib led to 6.06 (adjusted HR 1.46 95% CI 1.08, 1.98) and 6.19 (adjusted HR 1.39, 95% CI 0.74, 2.59) strokes per 1000 person years respectively. No other NSAID significantly increased the risk of incident stroke. Conclusions Our results indicate an increased risk of stroke with current use of two highly selective coxibs, rofecoxib and valdecoxib, also shown to increase cardiovascular risk. These results also provide some reassurance about other specific NSAIDs regarding stroke risk. (Stroke. 2008;39:2037-2045.) Key Words: epidemiology pharmacology public policy stroke NSAIDs Selective cyclooxygenase 2 inhibitors (coxibs) were initially developed to decrease the gastrointestinal toxicity associated with traditional nonsteroidal antiinflammatory drugs (NSAIDs). However, there is now good evidence that these drugs can increase the risk of cardiovascular events in a dose-dependent manner. 1 4 The excess in cardiovascular risk was first detected with rofecoxib in a trial in which the comparator was naproxen 5 and confirmed in another clinical trial with a placebo comparator. 4 Cardiovascular adverse effects led to the withdrawal from the market of rofecoxib (Vioxx, Merck) in September 2004. Valdecoxib (Bextra, Pfizer) was then withdrawn in April 2005. However, questions remain regarding the safety of celecoxib (Celebrex, Pfizer). The Adenoma Prevention with Celecoxib (APC) study reported an increased cardiovascular risk with higher doses of celecoxib (400 mg/d and 800 mg/d). 6 Uncertainty also remains regarding the safety of nonselective NSAIDs. In December 2004, the NIH halted the Alzheimer Disease AntiInflammatory Prevention Trial (ADAPT), when an interim analysis suggested that naproxen raised the risk of heart attack and stroke in study patients. 7,8 NSAIDs can both inhibit the synthesis of platelet thromboxane, an antithrombotic activity, as well as inhibit the synthesis of prostacylin in endothelial cells, a prothrombotic activity. The balance of these activities or the relative potency of selective cyclooxygenase 2 inhibition may in part determine cardiovascular risk. Other possible mechanisms include elevation of blood pressure or accelerated atherogenesis. 9 11 Most NSAIDs, increase blood pressure in a dose-dependent manner. Both isoforms of cyclooxygenase are present in the Received October 26, 2007; accepted November 22, 2007. From the Veterans Administration, Tennessee Valley Healthcare System (C.L.R., M.R.G.), Tennessee Valley Geriatric Research Education Clinical Center (GRECC); HSR&D Targeted Research Enhancement Program (TREP) Center for Patient Healthcare Behavior (C.L.R., M.R.G.); Tennessee Valley VA Clinical Research Training Center of Excellence (CRCoE) (C.L.R., M.R.G.); the Departments of Medicine (C.L.R., M.R.G.) and Biostatistics (L.K., P.G.A.), Vanderbilt University; the Department of Preventive Medicine (Pharmacoepidemiology), and Center for Education and Research on Therapeutics (E.F.M., P.G.A., P.G., M.R.G.) Vanderbilt University, Nashville, Tenn. Correspondence to Christianne L. Roumie, MD, MPH, Nashville VA Medical Center, GRECC, 1310 24th Ave South, Nashville, TN 37212. E-mail christianne.roumie@vanderbilt.edu 2008 American Heart Association, Inc. Stroke is available at http://stroke.ahajournals.org DOI: 10.1161/STROKEAHA.107.508549 2037

2038 Stroke July 2008 kidney vasculature and are critical to regulation of renal blood flow. 12 It is inhibition of this response that accounts for the most common renal side effects, edema and hypertension, associated with NSAID therapy. 13 Thus, there are several complex mechanisms by which NSAIDs (including coxibs) could have an adverse effect on vascular-associated health outcomes, particularly the risk of stroke. A recent case control study of NSAIDs and stroke found no association between NSAIDs and ischemic or hemorrhagic stroke; however, no coxibs were included in that analysis. 14 Another case control study conducted in the UK General Practice Research Database found an increased risk for ischemic stroke associated with rofecoxib, etoricoxib, and diclofenac use, but not with celecoxib or other NSAIDs. 15 Our aim in this study was to determine whether specific NSAIDs, including coxibs, are associated with an increased risk of ischemic or hemorrhagic stroke. New use and high doses of selective cyclooxygenase 2 inhibitors and nonselective NSAIDs were specifically examined. Methods Study Design and Data Sources We performed a retrospective cohort study to examine the relationship between NSAIDs and stroke among Tennessee Medicaid program (TennCare) enrollees between 1999 and 2004. The Institutional Review Board of Vanderbilt University approved this study. The primary sources of data were the administrative files of TennCare including the enrollment file, hospital and physician files, and the pharmacy files. The enrollment file identifies persons who are eligible to receive benefits, the dates of coverage as well as demographics such as sex, date of birth, race, and county of residence. The hospital and physician files include admission and discharge dates for hospitalizations, office visit encounters, and emergency department visits. During the study period diagnoses associated with health care visits were coded according to the International Classification of Diseases, Ninth Revision; Clinical Modification (ICD9-CM). 16 Population The source population was noninstitutionalized persons aged 50 to 84 years with continuous enrollment in TennCare for the year before study entry. Persons were eligible if they were enrolled between 1/1/1999 and 12/31/2004, had date of birth and gender in the enrollment file, did not reside in a nursing home, and did not have evidence for a prior stroke or other serious medical illness (defined as evidence of any cancer except non melanoma skin cancer, liver failure, end stage renal disease, HIV infection, or organ transplant) in the year before cohort entry. Patients were followed until development of the study outcome (stroke), a serious medical illness including stroke not meeting the study outcome definition, disenrollment, death, or the end of the study (December 31, 2004). Exposure The drugs of interest included the 7 most commonly prescribed NSAIDs: celecoxib, rofecoxib, valdecoxib, ibuprofen, naproxen, indomethacin, and diclofenac. All other NSAIDs, including multiple concurrent NSAID use, were considered together in an other NSAID category. For each of the 7 NSAIDs the daily dose was determined by multiplying the number of pills dispensed by the dose prescribed divided by the recorded days supply. If drug dose was missing or the formulation was for a liquid ( 2%), the prescription was excluded from dose analyses. If duration was missing, it was assumed to be the lesser of twice the number of pills or 30 days. We used the 2 most frequently prescribed doses to define high and low dose, and used the midpoint between these doses to separate daily doses that were not in either dose category. The modal low and high daily dose for each drug was: celecoxib 200 and 400 mg; rofecoxib 25 and 50 mg; valdecoxib 10 and 20 mg; ibuprofen 1800 and 2400 mg; naproxen 750 and 1000 mg; indomethacin 75 and 150 mg; diclofenac 100 and 150 mg. Current use was the period from the date an NSAID prescription was filled through the end of the days of supply. Former use was 365 days after the end of current use. Nonuse included all other person time, that is, person time in which there was no NSAID use in the prior 365 days, and was the reference for all analyses. Episodes of use of a specific NSAID began on the first study day of current use and stopped at time of switching to another NSAID, after 365 days of nonuse, or on exit from the cohort. Current use was also classified by dose (high dose and low dose based on the cut point) and duration ( 90 days of continuous use or 90 days of continuous use). Because extended follow-up data from the APPROVe trial indicate that patients who stopped rofecoxib remain at an increased risk of stroke, we also examined former use of rofecoxib and valdecoxib. 17 Because new users are less susceptible to bias introduced by events that might arise shortly after the onset of therapy or by modification of disease risk factors, such as hypertension, by the exposure of interest, we conducted a similar analysis confined to new users. 18 New users were defined as persons who started an NSAID and had no prescription for any NSAID in the prior 365 days. Outcome The stroke outcome was determined with ICD 9 CM codes, using an algorithm with a high predictive value for identifying new strokes. 19 The study outcome date was their first hospitalization with a primary discharge diagnosis indicating an incident cerebrovascular event. These included fatal and nonfatal ischemic stroke (ICD 9-CM codes 433.x1, 434 [excluding 434.x0], or 436), intracerebral hemorrhage, (431) and subarachnoid hemorrhage. (430) Strokes associated with traumatic brain injury (ICD-9-CM codes 800 to 804, and 850 to 854) or those with rehabilitation care as the primary code (V57) were excluded. Strokes that were coded as a secondary diagnosis did not meet criteria for the main study outcome and were censored. We chose to study ischemic/thrombotic stroke events because inhibition of cyclooxygenase 2 has been postulated to increase platelet thromboxane thereby promoting thrombosis. 9 11,20,21 We also included hemorrhagic strokes because of the epidemiological evidence that use of certain NSAIDs might lead to intracerebral or subarachnoid hemorrhage. 22,23 A systematic sample of 250 charts was reviewed to verify the accuracy of the incident stroke diagnoses. Using our study definition, the positive predictive value for identifying any new stroke was 97%. The predictive value for a true incident event, that is a new stroke in a patient with no prior history of stroke, was 80%. Covariates Important comorbidities were determined using the administrative files to identify persons that had an encounter with the ICD-9-CMcoded comorbidity of interest or prescription for specific medications in the prior 365 days. All covariates were measured at baseline, at the beginning of each episode of NSAID use, and at a transition to nonuse. Covariates included all factors listed in Tables 1 and 2 and included: age, race, sex, calendar year, residence, enrollment category, presence of rheumatologic disease, prescription for aspirin, smoking-related illness, general access to care, unique number of prescription medications used, and the use of medications indicating other comorbidities (anticonvulsants, antimicrobials, bronchodilators, estrogens, and psychotropic drugs). To control for potential differences in vascular disease between study exposures, we calculated a vascular risk score for all study subjects. 24 28 We fit a proportional hazards model relating the 22 indicators of vascular comorbidities and medications (Table 2) at baseline to our study outcome, time-to-stroke. 29 To account for potential confounding by NSAID use as well as possible instabilities attributable to small numbers of events among certain exposure groups, the regression model was restricted to nonusers of NSAIDs at baseline, which is consistent with risk score applications from other published studies. 25 28 We used the estimated coefficients of

Roumie et al Nonaspirin NSAIDs, Cyclooxygenase 2 Inhibitors, and Risk of Stroke 2039 Table 1. Characteristics of Patients on Cohort Entry Nonuser n 242 450 Celecoxib n 7146 Rofecoxib n 6314 Valdecoxib n 1611 Naproxen n 30 447 Ibuprofen n 25 406 Diclofenac n 2056 Indomethacin n 5056 Gender, % Female 59.4 68.2 65.7 68.6 66.9 67.4 67.6 54.9 Age, % 50 to 64 69.0 69.2 78.4 70.2 75.8 74.7 81.5 74.0 65 to 74 19.7 17.9 13.2 16.8 17.0 17.6 12.5 18.4 75 11.3 12.9 8.4 13.0 7.2 7.7 6.0 7.6 Race,* % White 75.0 77.8 81.4 81.9 71.0 66.4 76.7 64.5 Black 16.0 13.1 9.7 10.6 18.2 22.9 13.6 25.4 Other 2.3 3.2 3.0 3.2 2.2 2.2 3.1 2.4 Residence,* % Urban 29.0 28.3 22.3 24.5 27.5 34.9 26.9 30.1 SMSA 26.7 25.0 29.0 29.6 25.2 24.3 20.5 25.0 Rural 43.9 46.4 48.4 45.7 47.2 40.7 52.3 44.7 Enrollment category, % Disabled 43.8 44.6 43.9 42.2 51.7 54.4 46.8 53.2 Uninsured/uninsurable 36.6 30.4 37.6 26.8 33.6 30.7 38.1 32.2 Aged/other 19.6 25.0 18.5 31.0 14.7 14.9 15.1 14.6 Calendar year, % 1999 55.6 10.5 1.8 66.6 67.5 30.0 67.8 2000 11.4 8.1 7.1 10.0 9.4 10.8 9.5 2001 8.9 17.2 24.1 6.7 6.6 8.5 6.8 2002 9.2 25.0 32.4 18.2 6.2 6.3 11.8 6.2 2003 7.6 22.6 25.9 43.3 5.1 5.1 13.3 4.9 2004 7.3 16.6 8.7 38.5 5.4 5.1 25.6 4.8 Mean vascular risk score 3.58 (0, 6) 4.14 (1, 7) 3.78 (1, 6) 4.19 (1, 7) 3.83 (1, 7) 3.93 (1, 7) 3.83 (1, 7) 4.44 (2, 7) (range) Rheumatologic disease, % 1.5 6.9 4.4 5.6 3.2 2.8 7.5 3.6 Anticonvulsants, % 10.7 20.9 19.7 31.8 11.9 12.2 19.5 11.1 Antimicrobials, % 61.6 75.7 74.6 78.7 75.6 72.8 74.9 71.7 Bronchodilators, % 19.7 28.7 28.5 31.6 25.1 25.4 26.4 24.1 Psychotropic drugs, % 46.4 68.3 68.6 73.9 58.3 57.1 64.1 54.2 Replacement estrogens, % 15.9 28.5 23.5 20.4 24.3 23.5 25.1 19.3 Smoking, % 6.8 9.8 10.2 12.4 6.6 7.2 10.8 5.5 Aspirin use, % 9.8 12.5 11.9 12.9 11.6 12.2 11.5 13.8 Prescriptions # 7 (3, 11) 16 (10, 23) 16 (10, 22) 19 (12, 26) 12 (7, 18) 12 (7, 18) 15 (9, 22) 13 (8, 19) (median, IQR) Outpatient visits # 3 (1, 7) 7 (3, 12) 7 (3, 12) 8 (4, 14) 6 (3, 10) 6 (3, 10) 7 (3, 12) 6 (3, 11) (median IQR) Emergency visits # (median IQR) 0 (0, 1) 0 (0, 1) 0 (0, 1) 0 (0, 1) 0 (0, 1) 0 (0, 1) 0 (0, 1) 0 (0, 1) *Proportion does not add up to 100% because of unknown/missing data. SMSA indicates standard metropolitan statistical area. 1 inpatient or emergency room diagnosis, 1 outpatient diagnosis plus 1 prescription for either a disease-modifying antirheumatic drug (DMARD) or an oral corticosteroid, or 2 outpatient diagnoses separated by at least 7 days: Rheumatoid arthritis (ICD 9 codes 714.0, 714.1, 714.2, 714.81); Systemic lupus erythematosis (ICD 9 code 710.0); Other inflammatory arthropathy (ICD 9 codes 696.0, 710.1, 710.3, 710.4, 710.5, 710.8, 710.9, 713.1, 713.7, 714.30-33, 714.4, 714.8, 714.89, 720.0); DMARDS include methotrexate, gold injections, auranofin, hydroxychloroquine, penicillamine, azathioprine, etanercept, infliximab, cyclosporine, sulfasalazine, leflunomide, minocycline, Anakinra, Adalimumab. All antidepressants, Anxiolytics, and antipsychotic medications. Smoking-related illness (ICD 9 codes 305.1, V15.82, 989.84). #Prescriptions, outpatient visits, and ER visits in the year before cohort entry.

2040 Stroke July 2008 Table 2. Cerebrovascular Risk Score and Components of Score Nonuser n 242 450 Celecoxib n 7146 Rofecoxib n 6314 the vascular comorbidities and medications from this regression model to compute the linear predictor for all cohort members. This risk score was then categorized into 10 groups, with the lowest group representing patients with no vascular comorbidities or medications and the remaining 9 groups representing approximate quantiles of risk. The lowest and highest risk score groups had 2.42 and 10.59 strokes per 1000 person-years, respectively, representing a 4.34-fold difference in stroke rate (95% confidence interval [CI] 3.90 to 4.83). Valdecoxib n 1611 Naproxen n 30 447 Ibuprofen n 25 406 Diclofenac n 2056 Indomethacin n 5056 Mean vascular risk score, 3.58 (0, 6) 4.14 (1, 7) 3.78 (1, 6) 4.19 (1, 7) 3.83 (1, 7) 3.93 (1, 7) 3.83 (1, 7) 4.44 (2, 7) range Congestive heart failure* 10.0 11.1 7.6 11.0 8.6 8.7 7.9 13.7 Myocardial infarction 3.6 3.6 3.6 4.2 2.7 3.1 2.5 3.6 Coronary artery disease 15.3 17.5 15.7 16.3 16.0 15.9 14.4 18.8 Transient ischemic attack 1.1 1.4 1.2 1.7 1.1 1.0 0.8 1.0 Peripheral artery disease 3.3 4.7 3.6 5.3 2.9 1.0 3.5 3.6 Rheumatic aortic or mitral 2.5 3.0 2.2 2.6 1.7 1.7 2.2 2.5 valvular disease# Carotid artery disease** 0.2 0.2 0.2 0.1 0.1 0.2 0.1 0.1 Atrial fibrillation 3.7 3.7 2.3 3.0 1.8 1.9 1.8 2.5 Angiotensin converting 30.8 42.9 40.6 46.4 33.4 33.8 38.8 38.8 enzyme inhibitors Angiotensin receptor 3.5 11.1 10.1 17.3 2.8 2.7 8.0 3.1 blockers Beta-blockers 19.0 25.2 23.6 28.6 18.2 18.4 22.9 22.3 Calcium channel blockers 23.1 26.3 22.4 23.5 26.9 27.3 23.8 33.7 Thiazide diuretics 21.6 31.3 29.6 31.8 28.3 29.6 30.2 31.3 Loop diuretics 16.7 24.0 20.0 25.9 20.2 24.6 20.2 27.7 Other antihypertensives 9.4 11.0 9.7 12.0 9.6 10.1 9.4 12.9 Antidiabetic medications 18.8 21.9 20.0 22.8 21.1 22.1 21.8 24.1 Nitrates 13.0 14.7 12.9 14.2 14.0 14.0 12.2 16.5 Antiarrythmics 20.9 29.3 28.3 34.5 21.5 21.2 27.7 25.9 Digoxin/inotropic agents 7.0 5.5 3.8 4.6 4.9 5.2 4.5 8.2 Anticoagulants 5.7 7.3 5.3 7.1 2.6 2.7 3.4 4.2 Platelet inhibitors 3.6 6.7 5.3 9.8 3.0 2.8 5.5 3.4 Lipid-lowering agents 22.9 37.1 35.2 39.8 25.3 24.6 31.9 26.3 *ICD 9 codes (428, 402.02, 402.11, 402.91, 404.01. 404.03, 404.11, 404.13, 404.91, 404.93) or cardiomyopathy (425.xx), or prescription for a loop diuretic and a digitalis glycoside both in past year. ICD 9 codes 410 (acute MI), 412 (old MI), 429.7 (MI sequela). ICD 9 codes for angina (codes 411 or 413), or Coronary artery revascularization procedures, as indicated on any inpatient admission (ICD 36.01, 36.02, 36.03, 36.05, 36.09, 36.10-36.19; CPT codes 33533-36, 33510-23, 30, 92980-82, 84, 92995-6) or prescription for short or long acting nitrate in past year. ICD 9 code 435. ICD 9 codes 38.08-09, 38.18, 38.38, 38.39, 38.48, 38.49, 38.88, 38.89, 39.25, 39.29, 39.5, 44.02, 44.31, 44.39, 44.42 CPT codes 35226, 35256, 35286, 35351, 35355, 35371, 35372, 35381, 35454, 35456, 35459, 35473, 35474, 35482, 35483, 35485, 35492, 35493, 35495, 35546, 35548, 35549, 35551, 35556, 35558, 35563, 35565, 35566, 35571, 35583, 35585, 35587, 35646, 35651, 35654, 35656, 35661, 35663, 35665, 35666, 35671, 34800, 34802-5. #ICD 9 codes 394, 395, 396, 424.0, 424.1. **ICD 9 code 38.12 or CPT code 35301. ICD 9 code 427.3. covariates in Table 1. Measures of association were expressed as hazard ratios (HR) with 95% confidence intervals (95% CIs). To investigate whether there was effect modification by age or vascular risk we conducted stratified analyses with respect to these factors and tested for interaction. These were not statistically significant and were not included in the final model. All analyses were conducted using Stata 9.0 (Stata Corporation) and R Statistical Program (R Foundation, available at: http://www.r-project.org). Statistical Analyses We constructed Kaplan Meier survival curves for each NSAID exposure group accounting for time-dependent exposures. 30 Proportional hazards regression was used to examine the association between NSAID use as a time-dependent covariate and time-tostroke. We derived the final model by determining a priori covariates for inclusion based on clinical significance (potential confounders). The regression model adjusted for vascular risk score and all Results Characteristics There were 336 906 unique persons included in the cohort, with 989 826 person-years of follow-up. Each study member entered the cohort as a current user of one of 7 study NSAIDs (n 78 036), a current user of other NSAIDs or NSAID

Roumie et al Nonaspirin NSAIDs, Cyclooxygenase 2 Inhibitors, and Risk of Stroke 2041 Table 3. Rate of Stroke per 1000 Person Years and Adjusted Proportional Hazards Model for Time to Stroke by NSAID Status and Among New Users of NSAIDs All NSAID Users New User Analysis n Strokes/n Rate per 1000 Adjusted Hazard n Strokes/n Rate per 1000 Adjusted Hazard Exposure Person-Years Person-Years Ratio* (95% CI) Person-Years Person-Years Ratio* (95% CI) Nonuser 2466/546 213 4.51 Reference 2466/546 212.7 4.51 Reference Celecoxib 144/30 869 4.66 1.04 (0.87, 1.23) 43/8087.9 5.32 1.12 (0.83, 1.52) Rofecoxib 131/25 443 5.15 1.28 (1.06, 1.53) 44/7254.9 6.06 1.46 (1.08, 1.98) Valdecoxib 44/7399 5.95 1.41 (1.04, 1.91) 10/1616 6.19 1.39 (0.74, 2.59) Naproxen 161/39 769 4.05 0.94 (0.80, 1.11) 35/8591.9 4.07 1.02 (0.73, 1.42) Ibuprofen 120/30 309 3.96 0.88 (0.73, 1.06) 29/5568.7 5.20 1.26 (0.87, 1.81) Diclofenac 18/5046 3.57 0.94 (0.59, 1.49) 1/846 1.18 0.31 (0.04, 2.18) Indomethacin 33/5883 5.61 1.20 (0.85, 1.69) 5/841.4 5.94 1.29 (0.53, 3.09) Other NSAID 140/32 348 4.33 1.04 (0.87, 1.23) 28/5553.9 5.04 1.23 (0.84, 1.78) Former NSAID 932/230 003 4.05 1.02 (0.94, 1.10) 1097/266 547.8 4.12 1.02 (0.95, 1.10) Former rofecoxib 132/29 411 4.49 1.16 (0.96, 1.39) Former valdecoxib 33/7134 4.63 1.17 (0.83, 1.66) *Adjusted for age, race, sex, calendar year, residence, enrollment category, presence of rheumatologic disease, aspirin, smoking-related illness, No. of outpatient visits, emergency room visits, hospital admissions, unique No. of prescription medications used, adjusted vascular risk score, and the use of the following medications: anticonvulsants, antimicrobials, bronchodilators, estrogens, and psychotropic drugs. combinations (n 16 420), or as a nonuser (n 242 450). The cohort was 64.8% women, 74% white, and 74.1% were less than 65 years of age (Table 1). NSAID users were more likely to be women, filled more prescriptions in the prior year, and had a higher number of outpatient visits than nonusers. Aspirin use and smoking related comorbidity did not differ appreciably between nonusers and NSAID users. Both nonusers and NSAID users had high baseline risk for vascular disease. The mean vascular risk scores ranged from 3.58 among nonusers to 4.44 among indomethacin users. A fourth of the cohort had serious cardiovascular disease in the year before cohort entry. More than 60% were using 1 or more antihypertensive drugs at baseline. A higher proportion of coxib users had a diagnosis of atrial fibrillation, and used anticoagulants or platelet inhibitors than noncoxib NSAID users (Table 2). NSAIDs and Risk of Stroke During the 989 826 person-years of follow-up there were 4354 stroke hospitalizations. There were 4.51 strokes per 1000 person years among the nonusers of NSAIDs, the reference group. The rate was 4.66 strokes per 1000 person years (adjusted HR 1.04, 95% CI 0.87, 1.23) for users of celecoxib, 5.15 strokes per 1000 person years (adjusted HR 1.28, 95% CI 1.06, 1.53) for rofecoxib, and 5.95 (adjusted HR 1.41, 95% CI 1.04, 1.91) for valdecoxib. No noncoxib NSAIDs significantly increased the risk of incident stroke (Table 3 and Figure, panel A). The majority of the strokes were ischemic 3892/4354 (89.3%), and analyses confined to this group alone were almost identical to those presented for all strokes. The risk of hemorrhagic strokes (subarachnoid hemorrhage and intracerebral hemorrhage) did not differ significantly between NSAID exposure groups, but confidence intervals around these estimates were wide (supplementary Table I, available online at http://stroke.ahajournals.org). Effect of Dose and Duration We examined the effect of NSAID dose and duration on the time to incident stroke. The majority of the stroke hospitalizations occurred among users of low dose rofecoxib or valdecoxib. There were 121 stroke events per 220 361.1 person years or 5.49 strokes per 1000 person years (adjusted HR 1.35, 95% CI 1.12, 1.62) among low dose rofecoxib users. There were 25 stroke hospitalizations during the 3253.2 person years or 7.68 strokes per 1000 person years (adjusted HR 1.73, 95% CI 1.16, 2.57) with low dose valdecoxib use. By contrast among users of high dose rofecoxib there were 10 stroke hospitalizations during 3406.8 person years (2.94 strokes per 1000 person years, adjusted HR 0.78, 95% CI 0.42, 1.45) and 19 stroke hospitalizations during 4145.4 person years of high dose valdecoxib use (4.58 strokes per 1000 person years, adjusted HR 1.12, 95% CI 0.72, 1.78). There was no increased risk of stroke among users of either high or low dose celecoxib, ibuprofen, naproxen, diclofenac, or indomethacin. Among users of rofecoxib there were a total of 131 stroke hospitalizations: 68 stroke hospitalizations during 12 702.6 person years of short duration ( 90 days) use and 63 stroke hospitalizations during the 12 740.3 person years of longer duration use (90 days). Thus among users of rofecoxib, there were 5.35 strokes per 1000 person years (adjusted HR 1.34, 95% CI 1.05, 1.72) during the first 90 days of use and 4.94 strokes per 1000 person years (adjusted HR 1.22, 95% CI 0.95, 1.57) during subsequent use. Among users of valdecoxib, there were a total of 44 stroke hospitalizations, 23 during the 4202.3 person years of short duration use (5.47 per 1000 person years) and 21 stroke hospitalizations during 3196.4 person years of long duration use (6.57 per 1000 person years). When adjusted for all covariates the risk of stroke was 1.35 (95% CI 0.89, 2.05) for short duration users of valdecoxib and 1.50 (95% CI 0.97, 2.31) among long duration users of valdecoxib compared to nonusers. Neither

2042 Stroke July 2008 (a) Cumulative Incidence Cumulative Incidence 0.05 0.04 0.03 0.02 0.01 0.00 Non-user Rofecoxib Valdecoxib Celecoxib Other NSAIDs* 0 500 1000 1500 2000 No. At risk: Observation Time (days) Total 320,486 156,242 113,595 78,860 46,297 Nonuser 242,450 127,253 87,663 58,731 33,610 Rofecoxib 6,314 4,035 6,400 4,469 1,942 Valdecoxib 1,611 773 740 1,830 1,769 Celecoxib 7,146 4,676 6,363 5,540 3,710 Other NSAIDs* 62,965 19,505 12,429 8,290 5,266 *Other NSAIDs represents Ibuprofen, Naproxen, Indomethacin, and Diclofenac (b) 0.05 0.04 0.03 0.02 0.01 0.00 Non-user Rofecoxib Valdecoxib Celecoxib Other NSAIDs* 0 500 1000 1500 2000 No. At risk: Observation Time (days) Total 245,305 133,948 94,726 64,160 36,974 Nonuser 242,450 127,253 87,663 58,731 33,610 Rofecoxib 43 1,099 1,951 1,345 553 Valdecoxib 8 151 157 480 423 Celecoxib 47 1,167 1,879 1,475 1,025 Other NSAIDs* 2,757 4,278 3,076 2,129 1,363 *Other NSAIDs represents Ibuprofen, Naproxen, Indomethacin, and Diclofenac Figure. Cumulative incidence for risk of stroke with each NSAID exposure group compared to nonusers. a, Results for full cohort. b, results for new user cohort. short nor long duration use of any of the remaining NSAIDs was associated with a significantly increased risk of incident stroke. We chose the cutoff of 90 days because it maximized the sample size in each group however; we conducted a sensitivity analysis varying the cutoff for long duration of use (90 to 180 days and 180 days). Among rofecoxib users who had duration of use that was 90 to 180 days there were 5.27 strokes per 1000 person years (adjusted HR 1.30, 95% CI 0.87, 1.95). Users of rofecoxib for 180 days had 4.77 strokes per 1000 person years (adjusted HR 1.16 95% CI 0.84, 1.60). Among valdecoxib users who had duration of use that was 90 to 180 days there were 5.71 strokes per 1000 person years (adjusted HR 1.33, 95% CI 0.66, 2.67). Users of valdecoxib for 180 days had 7.24 strokes per 1000 person years (adjusted HR 1.60 95% CI 0.92, 2.76). Effect of Channeling Bias Patients at high risk for stroke may be prescribed medications that are perceived to be safer. The ability of these patient characteristics to influence cardiovascular risk and cause confounding is called channeling bias. We were concerned that older patients and those with preexisting cardiovascular comorbidities would be preferentially prescribed lower doses of the coxibs. We reexamined the patient characteristics to determine absolute percent differences between patients prescribed high and low dose rofecoxib and valdecoxib. Among rofecoxib and valdecoxib users we found that low dose users were more likely to be 75 years and greater (absolute difference 6.47% [95% CI 5.16, 7.78%] and 5.87% [95% CI 2.34, 9.40%], respectively) and female (absolute difference 7.68% [95% CI 4.42, 10.94%] and 6.88% [95% CI 2.08, 11.67], respectively). More low dose rofecoxib users had congestive heart failure (absolute difference 2.5%, 95% CI 0.92, 4.08%) and used digoxin (absolute difference 1.62%, 95% CI 0.54, 2.70%). There were a higher proportion of low dose valdecoxib users who had atrial fibrillation (absolute difference 2.61%, 95% CI 0.79, 4.43%), used antiarrythmics (absolute difference 4.80%, 95% CI -0.15, 9.74%), and were prescribed a beta blocker (absolute difference 5.53%, 95% CI 0.83, 10.24%). Risk of Stroke Among New Users of NSAIDs An analysis of new users, that is, those with no NSAID use in the 365 days prior, yielded similar results. There were 6.06 stroke hospitalizations per 1000 person years for new users of rofecoxib (adjusted HR 1.46 95% CI 1.08, 1.98) and 6.19 stroke hospitalizations per 1000 person years for new users of valdecoxib (adjusted HR 1.39, 95% CI 0.74, 2.59). New use of other NSAIDs was not associated with a statistically significant increase in the risk of stroke (Table 3 and Figure, panel B). Discussion Our results confirm an increased vascular risk among current users of 2 highly selective cyclooxygenase 2 inhibitors, rofecoxib and valdecoxib. In our population, current use of celecoxib was not associated with an increased risk of stroke. The stroke rates were similar among new users of these drugs, but estimates were less precise. We used the new user design because of its methodological superiority to traditional observational study designs that include prevalent users. Inclusion of prevalent users may result in 2 types of bias. 18 First, prevalent users are survivors of the early period of pharmacotherapy. This can introduce bias if the risk of developing the outcome varies with time. Second, the exposure of interest may alter important comorbidities, such as NSAIDs causing hypertension. Adjustment for these factors that are on the causal pathway may obscure a true association. The new-user design eliminates these biases by restricting the analysis to persons under observation at the start of the current treatment and controls for baseline risk factors in the analysis. This design most closely resembles a clinical trial in which persons enter the trial and have their characteristics measured at baseline, before use of study drug, and then are followed forward in time until they develop the outcome of interest. The majority of early randomized trials of coxibs did not report vascular events as primary adverse outcomes. Among those that captured vascular events, most focused on cardio-

Roumie et al Nonaspirin NSAIDs, Cyclooxygenase 2 Inhibitors, and Risk of Stroke 2043 Table 4. Observational Studies of Specific Coxibs/NSAIDs and Risk of Stroke Author Study Design n Strokes Celecoxib Rofecoxib Etoricoxib Valdecoxib Naproxen Andersohn (2006) 15 Case-control, 3094 1.07 (0.79 to 1.44) 1.71 (1.33 to 2.18) 2.38 (1.10 to 5.13) NA 1.16 (0.80 to 1.70) UKGPRD* Bak (2003) 14 ICH Case control, 867 NA NA NA NA 0.8 (0.3 to 2.1) Denmark Bak (2003) 14 ischemic Case control, Denmark 2717 NA NA NA NA 0.7 (0.4 to 1.1) Lee (2007) 32 coronary disease Lee (2007) 32 no coronary disease Solomon (2006) 33 new users Roumie (current study) new users Case control, VA Case control, VA Retrospective cohort, Penn. Retrospective cohort, Tenn. *United Kingdom General Practice Research Database. Veterans Affairs. 2862 1.40 (0.96 to 2.03) 1.09 (0.77 to 1.54) NA NA 1.20 (1.01 to 1.43) 4787 0.97 (0.68 to 1.37) 1.45 (1.10 to 1.92) NA NA 1.15 (1.01 to 1.31) 3552 1.00 (0.92 to 1.09) 1.15 (1.04 to 1.26) NA 0.93 (0.74 to 1.18) 0.83 (0.67 to 1.04) 4354 1.12 (0.83 to 1.52) 1.46 (1.08 to 1.98) NA 1.39 (0.74 to 2.59) 1.02 (0.73 to 1.42) (Continued) vascular events, and few examined cerebrovascular events separately. Kearney et al 31 conducted a meta-analysis of 138 randomized trials of selective coxibs, and evaluated the cardiovascular and cerebrovascular risks of all coxibs. Compared to placebo, allocation to a coxib was associated with an increased risk for vascular events (rate ratio 1.42, [95% CI 1.13, 1.78]). This risk was largely attributable to an increased risk of myocardial infarction (rate ratio 1.86, 95% CI 1.33, 2.59). In this meta-analysis there was no significant association between coxibs and strokes (rate ratio 1.02, 95% CI 0.71, 1.47). The risk of stroke for use of each individual drug, either coxibs or traditional NSAIDs, was not reported. However, there was limited power to evaluate individual drugs or to explore between-drug differences. There are few observational studies that specifically examined the risk of cerebrovascular events among either NSAID or coxib users (Table 4). Our results demonstrating no increased risk for stroke with celecoxib is consistent with the results reported by Andersohn et al, 15 Lee et al, 32 and Solomon et al 33 In addition, the increased risk we observed with current use of rofecoxib is consistent with findings from these three observational studies. We also observed an increased risk for cerebrovascular events with the use of valdecoxib. One other observational study has specifically examined the risk of cerebrovascular effects of valdecoxib. 33 This study found no increased risk of cerebrovascular or cardiovascular events with the use of valdecoxib; however, the increased risk of composite vascular events, particularly myocardial infarction, has been documented with valdecoxib in randomized controlled trials. 34 36 The results regarding the traditional NSAIDs are less consistent. We found no increased risk for either ischemic or hemorrhagic cerebrovascular events with use of any of the traditional NSAIDs, similar to the results reported by Bak and colleagues. 14 Andersohn et al 15 noted an increased risk for ischemic stroke with the use of diclofenac. Finally, in the study of US veterans by Lee et al 32 there was an increased risk for cerebrovascular events among users of naproxen. A strength of our study is the validation of the stroke outcome definition in a systematic sample of charts. Our method uses an algorithm that selects for incident strokes thereby allowing the study of medications that may lead to stroke. Some limitations need to be considered. NSAIDs such as ibuprofen and naproxen are available over the counter and may have been used by the reference group we classified as nonusers. During the time period, NSAIDs could be obtained free from TennCare; therefore there were monetary incentives not to use over the counter agents. In our validation study 2 of the 44 (4.5%) persons we categorized as nonusers were actually using a nonaspirin NSAID at the time of their stroke. Thus, this type of misclassification appears to be small, but could bias our risk estimates toward the null. Another limitation is that potential confounders are not well recorded in the database. Aspirin use and smoking status are probably underestimated, clinical information such as body mass index, control of blood pressure, lipid profiles, glycosylated hemoglobin, and lifestyle factors such as exercise are not recorded. We adjusted for many of these clinical characteristics through use of the vascular risk score which accounts for prescription of cardiovascular medications such as antihypertensives, hypoglycemic medications, and lipid-lowering medications. Whether each of these risk factors was controlled to clinical goal remains unknown, but there is no reason to believe that control of risk factors occurred differentially between users of NSAIDs and nonusers. Finally, we were concerned about channeling bias and confounding by indication. Our analyses controlled for important risk factors such as age, sex, race, and the vascular risk score. We demonstrated that there were differences in the characteristics of persons prescribed high versus low doses of rofecoxib and valdecoxib. These differences in age, sex, and underlying cardiovascular comorbidities were controlled for in our analyses, but other factors associated with both health status and doses prescribed may explain the stronger association between low dose rofecoxib and incident stroke hospitalization.

2044 Stroke July 2008 Table 4. Continued Ibuprofen Diclofenac Indomethacin 1.12 (0.91 to 1.37) 1.32 (1.10 to 1.57) NA 1.3 (0.9 to 2.0) 1.3 (0.6 to 2.8) 1.3 (0.5 to 3.7) 1.3 (1.0 to 1.6) 1.1 (0.7 to 1.7) 1.4 (0.8 to 2.4) 1.08 (0.91 to 1.27) 1.33 (0.92 to 1.91) 1.12 (0.76 to 1.65) 1.11 (0.99 to 1.25) 1.24 (0.95 to 1.63) 1.13 (0.86 to 1.50) 0.95 (0.78 to 1.16) 0.98 (0.75 to 1.29) NA 1.26 (0.87 to 1.81) 0.31 (0.04 to 2.18) 1.29 (0.53 to 3.09) In conclusion, we report that both rofecoxib and valdecoxib were associated with an increased risk of stroke. These 2 medications were commonly prescribed in our population, particularly among those at highest risk for stroke the elderly and those with preexisting hypertension. 37,38 Both rofecoxib and valdecoxib have since been withdrawn from the market based on studies suggesting their association with a higher risk of vascular events than other NSAIDs, including celecoxib. Our data provide some reassurance about the NSAIDs that remain on the market. However, it is likely that small but clinically significant increases in risk would be missed by observational studies. Given these results and prior studies, we suggest that reintroduction of rofecoxib or valdecoxib would be unwise and advise caution for drugs with similar cox 2 selectivity. 39,40 Additional studies that include both cerebrovascular and cardiovascular outcomes may help establish significant between-drug differences in risk. Caution and supervision is needed with use of any of these drugs in populations at high risk for cardiovascular and cerebrovascular outcomes. Acknowledgments We acknowledge the resources and use of facilities at VA Tennessee Valley Healthcare System, Nashville Tennessee. We are indebted to the Tennessee Bureau of TennCare of the Department of Finance and Administration, which provided the data. We acknowledge C. Varas-Lorenzo, J. Castellsague, and S. Perez-Gutthann, MD, MPH, Pfizer for critical review of original protocol and final manuscript. Sources of Funding This material is based upon work supported by the Department of Veterans Affairs-Clinical Research Center of Excellence (C.L.R., M.R.G.) with resources and the use of facilities at VA Tennessee Valley Healthcare System, Nashville. This work was also supported by VA Career Development Award 04-342-2 (C.L.R.); VA Geriatric Research, Education, and Clinical Center (GRECC) (M.R.G., C.L.R.); HSR&D Targeted Research Enhancement Program (TREP) Center for Patient Healthcare Behavior TRP 03-073 (C.L.R., M.R.G.); Vanderbilt Center for Education and Research Vanderbilt Center for Education and Research on Therapeutics, Vanderbilt University School of Medicine (M.R.G., E.F.M., P.S.G.); and Grant support from Pfizer (M.R.G.). Disclosures The funding agency, Pfizer, had no role in the conduct of the study, collection of data, data management, and analysis. Pfizer did have a role in the study design, interpretation of data and final review of the manuscript. Dr Griffin reports receiving consulting fees from Merck Inc. Christianne L. Roumie and Marie R. Griffin had full access to all of the data in the study and take responsibility for the integrity and the analysis of data. References 1. Ray WA, Stein CM, Daugherty JR, Hall K, Arbogast PG, Griffin MR. Cox-2 selective non-steroidal anti-inflammatory drugs and risk of serious coronary heart disease. Lancet. 2002;360:1071 1073. 2. Zhang J, Ding EL, Song Y. Adverse effects of cyclooxygenase 2 inhibitors on renal and arrhythmia events: Meta-analysis of randomized trials. JAMA. 2006;296:1619 1632. 3. McGettigan P, Henry D. Cardiovascular risk and inhibition of cyclooxygenase: A systematic review of the observational studies of selective and nonselective inhibitors of cyclooxygenase 2. JAMA. 2006;296: 1633 1644. 4. Bresalier RS, Sandler RS, Quan H, Bolognese JA, Oxenius B, Horgan K, Lines C, Riddell R, Morton D, Lanas A, Konstam MA, Baron JA. Cardiovascular events associated with rofecoxib in a colorectal adenoma chemoprevention trial. N Engl J Med. 2005;352:1092 1102. 5. Bombardier C, Laine L, Reicin A, Shapiro D, Burgos-Vargas R, Davis B, Day R, Ferraz MB, Hawkey CJ, Hochberg MC, Kvien TK, Schnitzer TJ. Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. Vigor study group. N Engl J Med. 2000;343:1520 1528. 6. Solomon SD, McMurray JJ, Pfeffer MA, Wittes J, Fowler R, Finn P, Anderson WF, Zauber A, Hawk E, Bertagnolli M. Cardiovascular risk associated with celecoxib in a clinical trial for colorectal adenoma prevention. N Engl J Med. 2005;352:1071 1080. 7. John Burklow, Don Ralbovsky. Use of non-steroidal anti-inflammatory drugs suspended in large alzheimer s disease prevention trial. 2004; Available at: http://www.nih.gov/news/pr/dec2004/od%2d20.htm Accessed: December 12, 2006. 8. Cardiovascular and cerebrovascular events in the randomized, controlled alzheimer s disease anti-inflammatory prevention trial (adapt). PLoS Clin Trials. 2006;1:e33. 9. Fitzgerald GA. Prostaglandins: Modulators of inflammation and cardiovascular risk. J Clin Rheumatol. 2004;10:S12 17. 10. FitzGerald GA. Cardiovascular pharmacology of nonselective nonsteroidal anti-inflammatory drugs and coxibs: Clinical considerations. Am J Cardiol. 2002;89:26D 32D 11. FitzGerald GA, Cheng Y, Austin S. Cox-2 inhibitors and the cardiovascular system. Clin Exp Rheumatol. 2001;19:S31 S36. 12. FitzGerald GA, Patrono C. The coxibs, selective inhibitors of cyclooxygenase-2. N Engl J Med. 2001;345:433 442. 13. Patrono C, Dunn MJ. The clinical significance of inhibition of renal prostaglandin synthesis. Kidney Int. 1987;32:1 12. 14. Bak S, Andersen M, Tsiropoulos I, Garcia Rodriguez LA, Hallas J, Christensen K, Gaist D. Risk of stroke associated with nonsteroidal anti-inflammatory drugs: A nested case-control study. Stroke. 2003;34: 379 386. 15. Andersohn F, Schade R, Suissa S, Garbe E. Cyclooxygenase-2 selective nonsteroidal anti-inflammatory drugs and the risk of ischemic stroke: A nested case-control study. Stroke. 2006;37:1725 1730. 16. International classification of diseases, ninth revision, clinical modification. 1988. 17. Afilalo J, Coussa-Charley MJ, Eisenberg MJ. Long-term risk of ischemic stroke associated with rofecoxib. Cardiovasc Drugs Ther. 2007;21: 117 120. 18. Ray WA. Evaluating medication effects outside of clinical trials: New-user designs. Am J Epidemiol. 2003;158:915 920. 19. Roumie CL, Mitchel EF, Gideon P, Varas-Lorenzo C, Castellsague J, Griffin MR Validation of ICD-9 codes with a high positive predictive value for incident strokes resulting in hospitalization using medicaid health data. Pharmacoepidemiology and Drug Safety. In Press. 20. Antman EM, DeMets D, Loscalzo J. Cyclooxygenase inhibition and cardiovascular risk. Circulation. 2005;112:759 770. 21. Antman EM, Bennett JS, Daugherty A, Furberg C, Roberts H, Taubert KA. Use of nonsteroidal antiinflammatory drugs: An update for clinicians: A scientific statement from the american heart association. Circulation. 2007;115:1634 1642. 22. Saloheimo P, Juvela S, Hillbom M. Use of aspirin, epistaxis, and untreated hypertension as risk factors for primary intracerebral hemorrhage in middle-aged and elderly people. Stroke. 2001;32:399 404.

Roumie et al Nonaspirin NSAIDs, Cyclooxygenase 2 Inhibitors, and Risk of Stroke 2045 23. Saloheimo P, Ahonen M, Juvela S, Pyhtinen J, Savolainen ER, Hillbom M. Regular aspirin-use preceding the onset of primary intracerebral hemorrhage is an independent predictor for death. Stroke. 2006;37:129 133. 24. Miettinen OS. Stratification by a multivariate confounder score. Am J Epidemiol. 1976;104:609 620. 25. Graham DJ, Campen D, Hui R, Spence M, Cheetham C, Levy G, Shoor S, Ray WA. Risk of acute myocardial infarction and sudden cardiac death in patients treated with cyclooxygenase 2 selective and non-selective non-steroidal anti-inflammatory drugs: Nested case-control study. Lancet. 2005;365:475 481. 26. Ray WA, Meredith S, Thapa PB, Meador KG, Hall K, Murray KT. Antipsychotics and the risk of sudden cardiac death. Arch Gen Psychiatry. 2001;58:1161 1167. 27. Ray WA, Murray KT, Meredith S, Narasimhulu SS, Hall K, Stein CM. Oral erythromycin and the risk of sudden death from cardiac causes. N Engl J Med. 2004;351:1089 1096. 28. Ray WA, Stein CM, Hall K, Daugherty JR, Griffin MR. Non-steroidal anti-inflammatory drugs and risk of serious coronary heart disease: An observational cohort study. Lancet. 2002;359:118 123. 29. Arbogast P, Kaltenbach L, Ding H, Ray W. Adjustment of multiple cardiovascular risk factors with a summary risk score. Epidemiology. In press. 30. Simon R, Makuch RW. A non-parametric graphical representation of the relationship between survival and the occurrence of an event: Application to responder versus non-responder bias. Stat Med. 1984;3:35 44. 31. Kearney PM, Baigent C, Godwin J, Halls H, Emberson JR, Patrono C. Do selective cyclooxygenase-2 inhibitors and traditional non-steroidal antiinflammatory drugs increase the risk of atherothrombosis? Meta-analysis of randomised trials. BMJ. 2006;332:1302 1308. 32. Lee TA, Bartle B, Weiss KB. Impact of nsaids on mortality and the effect of preexisting coronary artery disease in us veterans. AmJMed. 2007; 120:98 e99 e116. 33. Solomon DH, Avorn J, Sturmer T, Glynn RJ, Mogun H, Schneeweiss S. Cardiovascular outcomes in new users of coxibs and nonsteroidal antiinflammatory drugs: High-risk subgroups and time course of risk. Arthritis Rheum. 2006;54:1378 1389. 34. Ott E, Nussmeier NA, Duke PC, Feneck RO, Alston RP, Snabes MC, Hubbard RC, Hsu PH, Saidman LJ, Mangano DT. Efficacy and safety of the cyclooxygenase 2 inhibitors parecoxib and valdecoxib in patients undergoing coronary artery bypass surgery. J Thorac Cardiovasc Surg. 2003;125:1481 1492. 35. Nussmeier NA, Whelton AA, Brown MT, Joshi GP, Langford RM, Singla NK, Boye ME, Verburg KM. Safety and efficacy of the cyclooxygenase-2 inhibitors parecoxib and valdecoxib after noncardiac surgery. Anesthesiology. 2006;104:518 526. 36. Nussmeier NA, Whelton AA, Brown MT, Langford RM, Hoeft A, Parlow JL, Boyce SW, Verburg KM. Complications of the cox-2 inhibitors parecoxib and valdecoxib after cardiac surgery. N Engl J Med. 2005;352: 1081 1091. 37. Griffin MR, Stein CM, Graham DJ, Daugherty JR, Arbogast PG, Ray WA. High frequency of use of rofecoxib at greater than recommended doses: Cause for concern. Pharmacoepidemiol Drug Saf. 2004;13: 339 343. 38. Roumie CL, Arbogast PG, Mitchel EF Jr, Griffin MR. Prescriptions for chronic high-dose cyclooxygenase-2 inhibitors are often inappropriate and potentially dangerous. J Gen Intern Med. 2005;20:879 883. 39. Cannon CP, Curtis SP, FitzGerald GA, Krum H, Kaur A, Bolognese JA, Reicin AS, Bombardier C, Weinblatt ME, van der Heijde D, Erdmann E, Laine L. Cardiovascular outcomes with etoricoxib and diclofenac in patients with osteoarthritis and rheumatoid arthritis in the multinational etoricoxib and diclofenac arthritis long-term (medal) programme: A randomised comparison. Lancet. 2006;368:1771 1781. 40. Lenzer J. Fda advisers warn: Cox 2 inhibitors increase risk of heart attack and stroke. BMJ. 2005;330:440.