the impact of nsaid or cox-2 inhibitor use on the initiation of antihypertensive therapy

ORIGINAL REPORT

The impact of NSAID or COX-2 inhibitor use on theinitiation of antihypertensive therapyy

Marianne Ulcickas Yood DSc, MPH1,2,3*, Emmeline Watkins DPhil4, Karen Wells BS2,Gena Kucera MPH2, Christine Cole Johnson PhD, MPH2 and Eva Lydick PhD4,5

1Epidemiology and Public Health, Yale University School of Medicine, New Haven, CT, USA2Josephine Ford Cancer Center, Henry Ford Health System, Detroit, MI, USA3Epi Source, Hamden, CT, USA4Epidemiology and Quantitative Decision Sciences, Astra Zeneca, Wilmington, DE, USA5Lovelace Clinic Foundation, Albuquerque, New Mexico, USA

SUMMARY

Purpose The objective of this study was to quantify the associations between NSAIDs and COX-2 inhibitors and risk forinitiating antihypertensive therapy.Methods We conducted a population-based case-control study in a large, integrated health system in the MidwesternUnited States. Cases (N¼ 23 562) were new users of antihypertensive therapy from 1, July 1997, through 31, January 2003.Controls (N¼ 23 562) were randomly selected and matched to cases on age, copay, medical care utilization, sex, and indexdate. The main outcome measures were exposure to NSAIDs and COX-2 inhibitors.Results Recent prescription NSAID use was associated with an increased risk for initiation of antihypertensive therapy(odds ratio (OR)¼ 1.6, 95%CI 1.5, 1.7) as were selective COX-2 inhibitors (OR¼ 1.8, 95%CI 1.6, 2.1). After adjusting forage, sex, co-payment, race, and exposure to other NSAIDs/COX-2, each non-selective NSAID (diclofenac, ibuprofen,indomethacin, naproxen, oxaprozin) was associated with an increased risk of antihypertensive therapy initiation, with ORsranging from 1.4 to 1.8. Recent users of COX-2 inhibitors had an increased risk of initiating antihypertensive therapy,regardless of specific drug (celecoxib adjusted OR¼ 1.7 (95%CI 1.3, 2.1); rofecoxib adjusted OR¼ 1.7 (95%CI 1.4, 1.9)).Conclusions A consistent increased risk of initiation of antihypertensive therapy was observed among recent users ofNSAIDs and COX-2 inhibitors. Unlike previous studies, the results indicate that the effects of rofecoxib and celecoxib areequivalent. Copyright # 2006 John Wiley & Sons, Ltd.

key words—NSAIDs; COX-2 inhibitors; antihypertensive therapy

Received 3 April 2006; Revised 7 October 2006; Accepted 28 November 2006

INTRODUCTION

Non-selective non-steroidal anti-inflammatory drugs(NSAIDs) and selective cyclooxygenase-2 inhibitors(COX-2) are frequently prescribed medications.1 Inaddition to well-documented gastrointestinal side

effects of NSAIDs,2–4 clinical and research evidenceindicate that NSAIDs and COX-2 may impact bloodpressure, directly5–9 or by interaction with anti-hypertensive agents.7,10,11

Clinical studies provide mixed messages, someshowing NSAIDs have no blood pressure effect, withothers indicating that NSAIDs and COX-2 increasemean blood pressure.12,13 A 52-week study of once-daily rofecoxib showed small, dose-related increasesof diastolic and systolic BP (�1mmHg and�3mmHg,

pharmacoepidemiology and drug safety 2006; 15: 852–860Published online 6 October 2006 in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/pds.1327

*Correspondence to: M. Ulcickas Yood, 275 Blue Trail, Suite 3,Hamden, CT 06518, USA. E-mail: [email protected] conflict of interest was declared.

Copyright # 2006 John Wiley & Sons, Ltd.

respectively).14 Head-to-head comparisons of rofe-coxib and celecoxib in hypertensive osteoarthritispatients showed a slight mean systolic blood pressureincrease in the rofecoxib group at 6 weeks, comparedto a slight decrease in the celecoxib group.15,16 In oneof the studies, 10% of both treatment groupsexperienced clinically important changes in systolicblood pressure at any trial visit.16

Population-based studies have shown an associationbetween NSAIDs and hypertension. Gurwitz et al.5

determined that the use of non-selective prescriptionNSAIDs increased the risk for initiation of anti-hypertensive therapy (OR¼ 1.66) in Medicaid enrol-lees over 65 years of age. This effect was highest inhigh-dose, recent NSAID users and was consistent forall non-selective NSAIDs studied except ketoprofen.More recently, two studies conducted using theNurses’ Health Study cohort showed an increasedrisk of hypertension in users of acetaminophen andNSAIDs.6,8

Rofecoxib was first associated with an increasedrisk of acute myocardial infarction (MI) in the VIGORstudy,17 which showed a five-fold increase in the riskof MI compared with naproxen. This finding wasreinforced in several epidemiologic studies18–20 andlater in the APPROVe trial.21 Rofecoxib wasvoluntarily withdrawn from the market in September2004, after APPROVe results indicated that doses of25mg increased the relative risk of confirmedcardiovascular events after 18 months or more ofuse.22 While most studies investigating the associationbetween celecoxib exposure and cardiovascular riskshowed no strong effect20,23 in December 2004, resultsfrom the randomized APC study reported that highdoses of celecoxib resulted in an approximate 2.5-foldincrease in the combined endpoint of fatal or non-fatalcardiovascular events.24

It is important to understand the full range ofcardiovascular risks of the COX-2 selective inhibitorsand the differences between them. This paper assessesthe risk of hypertension in rofecoxib, celecoxib andnon-selective NSAIDs users and provides informationon the level of risk for each of the drugs.

METHODS

Data sources and study population

We conducted a case-control study sampling from apopulation of subjects identified from a largeintegrated health system in the Midwestern UnitedStates. The health system includes an owned multi-specialty, salaried, physician group affiliated with a

nonprofit health plan, including a large mixed-modelhealth maintenance organization (HMO). To optimizethe computerized data available for this study, thepopulation was limited to health plan membersassigned to medical group physicians. During thestudy period, the health system had no formularyrestrictions on NSAIDs or COX-2 selective inhibitors.This study was approved by the medical groupinstitutional review board (IRB).

Computerized data sources and medical records

The health system maintains a centralized system ofcomputerized databases. Data for this study primarilycame from electronic and paper medical record reviewand three computerized databases: (1) the membershipfile; (2) the outpatient pharmacy database; and (3) theencounter and claims databases.The membership database includes information on

coverage dates, benefits, and co-payments. Thepharmacy database contains information on outpatientprescriptions filled by members. To receive payment,contracted pharmacies, including several majorchains, must file a claim for each prescription filled.This database includes information on date memberfilled prescription, drug name, national drug code(NDC), number of pills, dosage dispensed, and dayssupply. Because this database captures all prescrip-tions paid by the plan (regardless of pharmacylocation) and pharmacies will not be paid unless aclaim is submitted, these data are of high quality.The encounter and claims databases include

comprehensive patient data for outpatient, emergencydepartment, and inpatient care delivered. For eachoutpatient encounter, information on date of visit,diagnoses, physician delivering care, proceduresdelivered, and clinic in which the care was deliveredis compiled. Likewise, for each inpatient stay,information on admission and discharge date, diag-noses, and procedures is compiled. Finally, inpatientand outpatient medical records are maintained in bothtraditional paper format and electronic format.

Study population

Using the computerized databases described above,we identified cases (as filling a new anti-hypertensiveprescription) and matched controls from the basepopulation of individuals who were, at one time, non-users of antihypertensive therapy and continuouslyenrolled in the health system for at least 1 yearsometime during the study period 1, July 1996,through 21, January 2003.

Copyright # 2006 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2006; 15: 852–860DOI: 10.1002/pds

nsaids and htn 853

Identification of cases

To identify all incident anti-hypertensive users, weselected all individuals 20 years of age or older whofilled a prescription for an antihypertensive medi-cation (including ACE inhibitors, alpha blockers, betablockers, calcium channel blockers, and diuretics(including loop diuretics)) any time during the studyperiod 1, July 1997, through 31, January 2003, with noprevious antihypertensive filled within 1 year. Thedate that the patient filled the first antihypertensiveprescription was termed the ‘index date’. From thisgroup of new users, we excluded individuals whowerenot continuously enrolled for at least 1 year before theindex date. Because these drugs are also used to treatconditions other than hypertension, we excludedpatients who did not have an ICD-9-coded encounterindicative of hypertension (401–401.9). Therefore,only treated patients with a hypertension diagnosiswere included as cases.

Identification of controls

Controls were randomly selected from the basepopulation of health system members who did notfill an antihypertensive prescription during the studyperiod. One control was sampled for each case,matched to the case on the following variables: age(� 2 years); co-payment; sex; index date (i.e., theindex date for the case was the reference point forthe matched control). In addition, because thehypertension cases encountered the health system atthe time of the index date, controls were required tohave an encounter with the health system within30 days of the case’s index date. We excluded ascontrols individuals who had an ICD-9-coded diag-nosis for hypertension within 1 year before the indexdate. Like the cases, controls not continuouslyenrolled in the health system for at least 1 yearbefore the index date were excluded from the study.

Ascertainment and classification of NSAID andCOX-2 exposure

From the outpatient pharmacy database, we down-loaded all NSAID and COX-2 prescriptions filled. Weclassified a patient as exposed to pharmacotherapyonly if they filled a prescription for at least a 30-daysupply, and formed the following exposure categories:Recent use: NSAID or COX-2 use within the 60-dayperiod before the index date; Former use: NSAID orCOX-2 use within 1 year before the index date, butsupply ended more than 60 days before the index date;

Non-use: less than 30-day supply or no prescriptionsfilled within 1 year before the index date.

Medical record review

We conducted medical record review for a randomsample of 450 patients (225 cases and 225 matchedcontrols). This record review collected data on height,weight, smoking, and blood pressure readingsobtained and recorded during the study period.

Data analysis

We conducted all analyses using SAS statisticalsoftware version 8.1 (SAS Institute, Inc., Cary, NorthCarolina).25 We calculated odds ratios (OR) andcorresponding 95% confidence intervals (CI) toestimate the relative risk of initiation of antihyperten-sive therapy among recent users of NSAIDs/COX-2and former users of NSAIDs/COX-2 compared tononusers. To control confounding, we fit a multiplelogistic regression model controlling for age, co-payment, and number of visits in the year before theindex date (as continuous variables); sex (male/female); and race (African American vs. Caucasian/Other). We also fit a multiple logistic regressionmodel, evaluating the independent effect of individualNSAIDs/COX-2s. In this analysis, in addition to thevariables listed above, we also controlled for exposureto other NSAIDs/COX-2 inhibitors.

We conducted analysis using conditional logisticregression (to account for matching) and found thatthe results and conclusions were the same whenunconditional logistic regression was used. Therefore,we present results from the unconditional model,thereby simplifying the analysis and presentation andallowing evaluation of the matching factors if needed.

RESULTS

We identified 55,337 cases with a filled prescriptionfor an antihypertensive medication during the period1, July 1997, through 31, January 2003, and noprevious antihypertensive prescription filled within1 year. From these cases, we excluded 15,921individuals (29%) whowere not continuously enrolledin the health system for at least 1 year before the indexdate, and also excluded 15,417 patients (28%) who didnot have an ICD-9-coded encounter indicative ofhypertension. For each of these cases, we randomlysampled one control matched on age, sex, co-payment,and encounter with health systemwithin 30 days of theindex date. Among the 23,999 eligible cases, we could


854 m. ulcickas yood ET AL.

not identify a control for 437 patients (1.8%; mainlydue to the visit-related criteria in the older age groupsfor whom limited numbers of eligible controls wereavailable). Table 1 shows the demographics of the finalstudy sample of 47,124 patients (23,562 cases and23,562 controls).

The mean duration of exposure to NSAIDs was 86.0days and the mean duration of exposure to COX-2selective inhibitors was 100.0 days (97.3 days forrofecoxib and 81.0 days for celecoxib). The distri-bution of NSAID and COX-2 exposure among thecases and controls is shown in Table 2. As can be seenin Table 2, after adjusting for age, sex, copayment,race, and number of visits to the health system duringthe year before the index date, recent users of NSAIDsor COX-2 were 1.6 times more likely to initiateantihypertensive therapy when compared to non-users(95%CI 1.5, 1.7). This increased risk was seen for bothrecent users of NSAIDs (OR¼ 1.6 (95%CI 1.5, 1.7)and COX-2s (OR¼ 1.8 (95%CI 1.6, 2.1). Noincreased risk of initiation of antihypertensivepharmacotherapy was seen among former users.

Table 3 presents results according to specificNSAID or COX-2 exposure. Drugs were selectedfor this analysis if more than 300 patients wereexposed to the specific drug during the study period.As can be seen in Table 3, each of the individual studydrugs was independently associated with an increased

risk of antihypertensive therapy initiation, afteradjusting for exposure to other NSAIDs/COX-2s,age, sex, co-payment, and race. Recent users of COX-2 inhibitors had an increased risk of initiatingantihypertensive therapy, regardless of specific drug(celecoxib OR¼ 1.7 (95%CI 1.3, 2.1); rofecoxibOR¼ 1.7 (95%CI 1.4, 1.9)).

Medical record review sub-analysis

We conducted medical record review on a randomsample of 225 cases and their 225 matched controls toobtain data on blood pressure measurements (todetermine the prevalence of unrecorded hypertensionamong controls) and potential confounding factors(height, weight and smoking). Among the 225controls, three patients (1.3%) had at least oneelevated blood pressure measurement (systolic�140 or diastolic �90) within 180 days of the indexdate. These findings indicate that 1.3% of controlswere misclassified in this study.

Potential impact of unidentified hypertensivepatients in control group

Based on data from the medical record review, 1.3% ofpatients in the control group had at least one elevatedblood pressure measurement. In order to evaluate the

Table 1. Characteristics of study population (N¼ 47,124)

Variable N (%)

TOTAL (N¼ 47,124) Cases (N¼ 23,562) Controls (N¼ 23,562)

Age Group20–29 1,116 (2.4%) 523 (2.2%) 593 (2.5%)30–49 16,760 (35.6%) 7,976 (33.9%) 8,784 (37.3%)50–69 19,508 (41.3%) 9,943 (42.2%) 9,565 (40.6%)70þ 9,740 (20.7%) 5,120 (21.7%) 4,620 (19.6%)

Mean Age (� SD) 55.6 years (�14.4) 56.1 years (�14.4) 55.0 years (�14.5)SexMale 20,734 (44.0%) 10,367 (44.0%) 10,367 (44.0%)Female 26,390 (56.0%) 13,195 (56.0%) 13,195 (56.0%)

RaceAfrican American 15,081 (32.0%) 9,535 (40.5%) 5,546 (23.5%)Caucasian 29,528 (62.7%) 13,004 (55.2%) 16,524 (70.1%)Other/Unknown 2,515 (5.3%) 1,023 (4.3%) 1,492 (6.3%)

Co-payment$0 2,988 (6.3%) 1,494 (6.3%) 1,494 (6.3%)$2 13,528 (28.7%) 6,764 (28.7%) 6,764 (28.7%)$3–5 9,440 (20.0%) 4,720 (20.0%) 4,720 (20.0%)$7–8 15,276 (32.5%) 7,638 (32.4%) 7,638 (32.4%)$10 5,882 (12.5%) 2,941 (12.5%) 2,941 (12.5%)$15 10 (0.02%) 5 (0.02%) 5 (0.02%)


nsaids and htn 855

impact of this potential misclassification, we con-ducted an analysis assuming that 1.3% of controls inthis study should be classified as cases, and that theseindividuals were unexposed to NSAIDS or COX-2inhibitors. With this potential, and most extreme,misclassification taken into account, the results are notchanged, with the crude OR for recent NSAID orCOX-2 users¼ 1.7 (95%CI 1.6, 1.8).

In order to assess potential confounding, weevaluated the impact of controlling for body massindex (BMI) in the subset of patients for whom weconducted medical record review. The results indi-cated that the crude effect for recent users (OR¼ 2.8(95%CI 1.4, 5.8)) was not materially different fromthe estimate adjusting for BMI (OR¼ 2.7 (95%CI 1.2,6.0). Likewise, current smoking status did not change

Table 3. Distribution of drug exposure among cases and controls, crude and adjusted odds ratio (OR)

Exposure category� Cases Controls Crude OR (95%CI) Adjusted OR (95%CI)y

Recent useDiclofenac 105 70 1.6 (1.2, 2.2) 1.4 (1.0, 1.9)Ibuprofen 1,487 918 1.7 (1.6, 1.9) 1.5 (1.4, 1.7)Indomethacin 129 58 2.4 (1.7, 3.2) 1.8 (1.3, 2.5)Naproxen 1,024 605 1.8 (1.6, 2.0) 1.6 (1.4, 1.7)Oxaprozin 105 60 1.9 (1.4, 2.6) 1.7 (1.2, 2.4)Other NSAIDz 267 174 1.7 (1.6, 1.8) 1.6 (1.3, 1.9)y

Rofecoxib 528 294 1.9 (1.7, 2.2) 1.7 (1.4, 1.9)Celecoxib 182 102 1.9 (1.5, 2.4) 1.7 (1.3, 2.1)

Former useDiclofenac 94 72 1.4 (1.0, 1.9) 1.0 (0.7, 1.3)Ibuprofen 1,101 1,018 1.2 (1.1, 1.3) 1.0 (0.9, 1.1)Indomethacin 92 78 1.3 (0.9, 1.7) 1.0 (0.7, 1.3)Naproxen 831 783 1.1 (1.0, 1.3) 0.9 (0.8, 1.0)Oxaprozin 104 69 1.6 (1.2, 2.2) 1.2 (0.9, 1.7)Other NSAIDz 102 106 1.1 (1.0, 1.2) 1.0 (0.7, 1.3)y

Rofecoxib 206 197 1.1 (0.9, 1.4) 0.9 (0.7, 1.1)Celecoxib 92 66 1.5 (1.1, 2.0) 1.1 (0.8, 1.5)

No NSAID or COX 2 Use 18,585 19,880 1.0 1.0

�All users must have filled at least one 30-day supply. Recent users defined as individuals whose NSAID/COX-2 supply continued into the60-day period before the index date; Former users defined as individuals who received an NSAID/COX-2 prescription within the 365 daysbefore the index date but supply endedmore than 60 days before the index date. Nonusers defined as subjects with less than 30-days supply orno NSAID/COX-2 prescription within 365 days before the index date.yControlling for age, sex, copay, race, number of visits in year prior to index and exposure to other all other drugs listed in table.zIncludes drugs for which number of users totaled fewer than 300 individuals: aspirin/salicylates, diflunisal, fenoprofen, flurbiprofen,ketorolac, meclofenamate, mefanamic acid, meloxicam, and tolmetin.

Table 2. Distribution of NSAID and COX-2 exposure among cases and controls, crude and adjusted odds ratios (OR)

Exposure category� Cases Controls Crude OR (95%CI) Adjusted OR (95%CI)y

Recent useNSAID or COX-2 3,533 2,183 1.7 (1.6, 1.8) 1.6 (1.5, 1.7)NSAID 2,990 1,866 1.7 (1.6, 1.8) 1.6 (1.5, 1.7)COX-2 691 388 1.9 (1.7, 2.2) 1.8 (1.6, 2.1)

Former useNSAID or COX-2 1,444 1,499 1.0 (1.0, 1.1) 1.0 (0.9, 1.0)NSAID 1,535 1,514 1.1 (1.0, 1.2) 1.0 (0.9, 1.1)COX-2 243 227 1.1 (1.0, 1.4) 1.0 (0.8, 1.2)

No NSAID or COX-2 Use 18,585 19,880 1.0 1.0Total 23,562 23,562 — —

�All users must have filled at least one 30-day supply NSAID/COX-2 prescription. Recent Users defined as individuals whose NSAID/COX-2 supply continued into the 60-day period before the index date; Former Users defined as individuals who received an NSAID/COX-2prescription within the 365 days before the index date but supply endedmore than 60 days before the index date. Nonusers defined as subjectswith less than 30 days supply or no NSAID/COX-2 prescription within 365 days before the index date.yControlling for age, sex, copay, race, and number of visits in year prior to index date.



the results. Therefore, we proceeded with analysesusing automated data (in which BMI and informationon smoking were not available).

Potential impact of pharmacy utilization

The endpoint in this study, initiation of antihyperten-sive pharmacotherapy, requires that a patient fill aprescription and that this prescription is billed to thehealth system. Likewise, the exposures in this study,NSAID and COX-2 use, require that patients fill aprescription that is billed to the health system. In orderto evaluate whether the results of this study were dueto the possibility that NSAID exposurewas a surrogatefor pharmacy utilization, we conducted a sub-analysislimiting the population to individuals who filled atleast one prescription for any medication within 1 yearbefore the index date (excluding the antihypertensiveprescription among cases). This analysis (N¼ 18,122cases, 16,458 controls) provided results similar tothose found for the entire study population (crude ORfor recent NSAID or COX-2 use¼ 1.5 (95%CI 1.4,1.6).

Potential impact of over-the-counter (OTC)NSAID use

Because NSAIDs are available OTC, we evaluatedthe potential effect that misclassification of OTCNSAID use may have on the results of this study.In previous work using data from patient interview, wedetermined that up to 15% of the health systempopulation uses OTC NSAIDs.26 Assuming thisextreme distribution (i.e., that 15% of cases and15% of controls classified as non-users in this studywere actually OTC NSAID users), NSAIDs remainassociated with initiation of antihypertensive therapy(crude OR¼ 1.3, 95%CI 1.25, 1.37).

CONCLUSIONS

We conducted a population-based study in a real-world setting to evaluate the effect of NSAIDs andCOX-2 inhibitors on initiation of antihypertensivetherapy. Our findings indicate a strong and consistentassociation between recent exposure to NSAIDs orCOX-2 inhibitors and initiation of antihypertensivetherapy. Among the non-selective NSAIDs, the effectwas lowest for recent users of diclofenac (OR¼ 1.4,95%CI 1.0, 1.9) and highest among recent users ofindomethacin (OR¼ 1.8, 95%CI 1.3, 2.5). Among theselective COX-2 inhibitors, the magnitude of the

effect in our study was identical for recent users ofcelecoxib (OR¼ 1.7, 95%CI 1.3, 2.1) and rofecoxib(OR¼ 1.7, 95%CI 1.4, 1.9), after adjusting for age,sex, copayment, race, number of visits, and exposureto other NSAIDs/COX-2 inhibitors.Most information regarding the potential effect of

NSAIDs and COX-2 inhibitors is derived from clinicaltrials with limited numbers of patients, and subsequentmeta- analyses of these trials.9,13 Some population-based epidemiologic studies have been conducted onnon-selective NSAIDs using automated data5 andquestionnaires.6,8 Although the literature to date is notdefinitive, most studies indicate, in general, that non-selective NSAIDs are associated with increased bloodpressure and increased risk of hypertension. However,much less information is known about selective COX-2 inhibitors. Many studies have concluded that, whencompared to celecoxib, rofecoxib is more likely toincrease blood pressure or lead to initiation oftherapy,7,15,27 while others propose that these differ-ences between selective COX-2 inhibitors may be dueto underlying characteristics of the study populationsor suboptimal study methods.12

In terms of rofecoxib use, our findings are similar tothose of Solomon et al.,27 who found that, in a low-to-moderate income elderly population, rofecoxib usershad an increased risk of initiating antihypertensivetherapy compared to non-users (OR¼ 1.6, 95%CI1.3–2.0). However, our results differ for celecoxibusers; Solomon et al. found no difference in risk forhypertension comparing celecoxib users to patientsunexposed to NSAIDs or COX-2 selective inhibitors(OR¼ 1.0, 95%CI 0.9–1.2). One possible explanationfor this discrepancy is the fact that our study includedpatients whowere exposed to both NSAID and COX-2selective inhibitors during the study period, andadjusted for previous exposure in the analysis. Patientsexposed to both classes were excluded from theSolomon study. To evaluate the potential impact thatthis difference may have, we briefly reviewed our dataand found that, in our population, 55% of rofecoxibusers and 64% of celecoxib users were exposed to anNSAID or a different COX-2 selective inhibitor duringthe study period. As a result, the base population ofexposed patients in our study differed from thatof Solomon et al., with potentially a larger number ofcelecoxib users excluded from the Solomon study dueto crossover. Another difference between our studies isthat we required at least 30 days exposure to anNSAID or COX-2, while no minimum exposure wasrequired by Solomon et al.A recent randomized trial in osteoarthritis

patients with type 2 diabetes receiving treatment for


nsaids and htn 857

hypertension conducted by Sowers et al. compared theeffect of 200mg/day celecoxib, 25mg/day rofecoxiband 1000mg/day naproxen. This study found that after12 weeks, rofecoxib exposed patients experiencedmean 24-hour systolic blood pressure increases(3.7mmHg increase), while celecoxib and naproxenexposed patients did not experience increases in meanblood pressure compared to baseline.28 This study alsofound that after 6 weeks of treatment, 30.1% ofrofecoxib-exposed patients who were normotensive atbaseline became hypertensive at 6 weeks, comparedwith 16.2% of celecoxib patients and 19.3% ofnaproxen patients. While the results demonstrated thatrofecoxib users were almost two times more likely todevelop hypertension compared with celecoxib users,the results also indicated that 16% of celecoxib-exposed patients became hypertensive.Potential mechanisms by which non-selective

NSAIDs may affect blood pressure have beenproposed and discussed. As described by Frishman,12

NSAIDs may lead to changes in blood pressure byinhibiting prostaglandin synthesis, and the relativecontribution of COX-1 and COX-2 in this process isunknown. Our results suggest a pharmacologic effectof both selective NSAIDs and non-selective COX-2based upon the consistency of the findings amongcurrent users and the lack of effect among former users(i.e., patients with NSAID or COX-2 exposure endingat least 60 days before the index date).This observational study is potentially limited by

several sources of bias that must be addressed. First,this study relied upon prescription claims data tomeasure exposure to NSAIDs and COX-2 inhibitors aswell as exposure to antihypertensive therapy. There-fore, although we know from these data that the patientfilled a prescription, we do not know for a fact thatthey took the prescribed medication. However, studieshave demonstrated that patients are likely to take amedication after a prescription is filled.29 Second,because this study relied on prescription claims tomeasure both exposure and antihypertensive outcome,we evaluated the hypothesis that the findings in thisstudy were not due to a pharmacologic effect butresulted because both exposure and outcome requiredpharmacy utilization. However, when we conductedan analysis limiting the study population to patientswho filled at least one prescription for any medicationduring the year before the index date, the results didnot substantially change. In this study, we were unableto evaluate the potential limitation that health careproviders may be more likely to monitor patientsexposed to NSAIDs and, therefore, treat the conditionsooner. However, given the limited and inconclusive

literature on this topic, this possibility should notexplain our findings.

An additional limitation of this study is the fact thatcertain non-selective NSAIDs are available OTC. Inour study, individuals who use OTC NSAIDs will beclassified as unexposed to NSAIDs because their usewill not be recorded in the claims database. Usingresults from a patient interview during a previousstudy of NSAID use in this study population,20 weapplied the estimated 15% OTC exposure prevalenceto the study results. We found that missed OTCexposure would diminish the overall effect estimates,although an overall increased risk for NSAIDs wouldremain. In addition, it is important to note that three ofthe NSAIDs highlighted in this study (indomethacin,oxaprozin, and diclofenac) are not available OTC, norare the COX-2 inhibitors. Another issue to consider isthat COX-2 selective inhibitors were not clinicallyavailable in 1997 and 1998. To evaluate whether thisfact potentially biased our results, we conducted asubanalysis, limiting the study period to 1999–2003.The results of this subanalysis and the conclusionsremained the same.

In order to identify newly diagnosed hypertensivecases, this study relied upon ICD-9 coded diagnoses ofhypertension in addition to claims-based pharma-cotherapy data. As a result, the control group(classified as non-hypertensive) may have elevatedblood pressure clinically definable as hypertensionand not treated with pharmacotherapy (becausetherapy was not prescribed or a prescription wasnot filled). We conducted medical record review toevaluate the impact of this potential bias and found nosubstantial difference in the effect estimate. Becausewe required an ICD-9 coded diagnosis of hypertensionin addition to receipt of pharmacotherapy, 28% ofpatients receiving pharmacotherapy did not meet thehypertension case definition. If these patients actuallyhad hypertension and have a distribution of NSAIDand COX-2 selective inhibitor exposure different fromthat found in the hypertension cases included in thisstudy, our results may be biased.

Obesity is a potentially important source of bias inthis study because these patients may be more likely tobe exposed to the analgesics in this study and are alsomore likely to develop hypertension. However, whenwe adjusted for body mass index in our sub-sample,the results did not change.

In summary, because NSAIDs and COX-2 inhibi-tors are frequently used medications, particularly inthe elderly population in which the baseline risk ofcardiovascular events is increased, potential adverseeffects of these therapies may have considerable



consequences. Our results indicate that users ofNSAIDs and selective COX-2 inhibitors are morelikely to initiate antihypertensive therapy whencompared to non-users or previous users. We foundno difference in risk between celecoxib and rofecoxib.The effect of these medications appears to diminishwhen the drug is discontinued, with no increased riskfound in former users. Because NSAIDs and COX-2inhibitors often represent chronic treatment, cliniciansshould be aware of the potential risks and appro-priately monitor their patients. In light of the recentwithdrawal of rofecoxib from worldwide markets dueto serious cardiovascular events,30 it is critical tocompletely understand whether these outcomes arerestricted to a particular drug, the COX-2 class, or allNSAIDs.

ACKNOWLEDGEMENTS

The research was funded by Astra Zeneca.

REFERENCES

1. Prescription Drug Expenditure Trends in 2000. The UpwardTrend Continues. Washington, DC: National Institute for HealthCare Management Research and Educational Foundation; May2001.

2. Griffin MR. Epidemiology of nonsteroidal anti-inflammatorydrug-associated gastrointestinal injury. Am J Med 1998; 104:23S–29S.

3. Smalley WE, Ray WA, Daugherty JR, Griffin MR. Non-steroidal anti-inflammatory drugs and the incidence of hospi-talization for peptic ulcer disease in elderly persons. Am JEpidemiol 1995; 141: 539–545.

4. Smalley WE, Griffin MR, Fought RL, Ray WA. Excess costsfrom gastrointestinal disease associated with nonsteroidalanti-inflammatory drugs. J Gen Intern Med 1996; 11: 461–469.

5. Gurwitz JH, Avorn J, Bohn RL, Glynn RJ, Monane M, Mogun H.Initiation of antihypertensive treatment during nonsteroidalanti-inflammatory drug therapy. JAMA 1994; 272: 781–786.

6. Curhan GC, Willett WC, Rosner B, Stampfer MJ. Frequency ofanalgesic use and risk of hypertension in younger women. ArchIntern Med 2002; 162: 2204–2208.

7. Armstrong EP, Malone DC. The impact of nonsteroidal anti-inflammatory drugs on blood pressure, with an emphasis onnewer agents. Clin Ther 2003; 25: 1–18.

8. Dedier J, Stampfer MJ, Hankinson SE, Willett WC, Speizer FE,Curhan GC. Nonnarcotic analgesic use and the risk of hyper-tension in US women. Hypertension 2002; 40: 604–608.

9. Pope JE, Anderson JJ, Felson DT. A meta-analysis of the effectsof nonsteroidal anti-inflammatory drugs on blood pressure.ArchIntern Med 1993; 153: 477–484.

10. Patrono C, Dunn MJ. The clinical significance of inhibition ofrenal prostaglandin synthesis. Kidney Int 1987; 32: 1–12.

11. Oates JA, Fitzgerald GA, Branch RA, Jackson EK, Knapp HR,Roberts LJ. Clinical implications of prostaglandin andthromboxane A2 formation(2). N Engl J Med 1988; 319:761–767.

12. Frishman WH. Effects of nonsteroidal anti-inflammatory drugtherapy on blood pressure and peripheral edema. Am J Cardiol2002; 90 ( Suppl): 18D–25D.

13. Johnson AG, Nguyen TV, Day RO. Do nonsteroidal anti-inflammatory drugs affect blood pressure? Ann Intern Med1994; 121: 289–300.

14. Brater DC, Harris C, Redfen JS, Gertz BJ. Renal effects ofCOX-2-selective inhibitors. Am J Nephrol 2001; 21: 1–15.

15. Whelton A, White WB, Bello AE, Puma JA, Fort JG, SUC-CESS-VII Investigators. Effects of celecoxib and rofecoxib onblood pressure and edema in patients ‡65 years with systemichypertension and osteoarthritis. Am J Cardiol 2002; 90: 959–963.

16. Whelton A, Fort JG, Puma JA, et al. Cyclooxygenase-2-specificinhibitors and cardiorenal function: a randomized, controlledtrial of celecoxib and rofecoxib in older hypertensive osteoar-thritis patients. Am J Ther 2001; 8(2): 85–95.

17. Bombardier C, Lain L, Reicin A, et al. Comparison of uppergastrointestinal toxicity of rofecoxib and naproxen in patientswith rheumatoid arthritis. N Engl J Med 2000; 343: 1520–1528.

18. RayWA, Stein CM,Daugherty JR, Hall K, Arbogast PG, GriffinM.COX-2 selective non-steroidal anti-inflammatory drugs andthe risk of coronary heart disease. Lancet 2002; 360: 1071–1073.

19. SolomonDH, Schneweis S, Glynn RJ, et al. Relationship betweenselective cyclo-oxygenase-2 inhibitors and acute myocardialinfarction in older adults. Circulation 2004; 109: 2068–2073.

20. GrahamDJ, Campen DH, Hui R, et al. Risk of acute myocardialinfarction and sudden cardiac death in patients treated withcyclo-oxygenase 2 selective and non-selective non-steroidalanti-inflammatory drugs: nested case-control study. Lancet2005; 365: 475–481.

21. Bresalier R, Sandler RS, Quan H et al. Cardiovascular eventsassociated with rofecoxib in a colorectal adenoma chemopre-vention trial. N Engl J Med 2005; 352: 1092–1102.

22. USDepartment of Health and Human Services. NIH halts use ofCOX-2 inhibitor in large cancer prevention trial. NIH News.December 17, 2004. Available at: http://www.nih.gov/news/pr/dec2004/od-17.htm. Accessed February 23, 2005.

23. Silverstein FE, Faich G, Goldstein JL, et al. Gastrointestinaltoxicity with celecoxib vs nonsteroidal anti-inflammatory drugsfor osteoarthritis and rheumatoid arthritis The CLASS study: arandomized controlled trial. JAMA 2000; 284: 1247–1255.

24. Solomon SD, McMurray JJV, Pfeffer MA, et al. Cardiovascularrisk associated with celecoxib in a clinical trial for colorectaladenoma prevention. N Engl J Med 2005; 352: 1071–1080.

25. SAS/STAT User’s Guide, Version 8. Cary, NC: SAS Insti-tute;1999: 1–3884.

KEY POINTS

� Recent users of NSAIDs and COX-2 inhibitorshad an increased risk of initiation of antihyper-tensive therapy in this study.

� The results of this research indicate that theeffects of rofecoxib and celecoxib are equivalent,in contrast to results of previous studies.


nsaids and htn 859

26. Ulcickas Yood M, Rothman KJ, Johnson CC, Jick S, Lang J,Wells K, Jick H. Using prescription claims data for drugsavailable over the counter (OTC). Pharmacoepidemiol DrugSaf 2000; 9 : S37.

27. Solomon DH, Schneeweiss S, Levin R, Avorn J. Relationshipbetween COX-2 specific inhibitors and hypertension.Hypertension 2004; 44: 140–145.

28. Sowers JR, White WB, Pitt B, et al. The effects of cycloox-ygenase-2 inhibitors and nonsteroidal anti-inflammatorytherapy on 24-hour blood pressure in patients with hyperten-

sion, osteoarthritis, and type 2 diabetes Mellitus. Arch InternMed 2005; 165: 161–168.

29. West SL, Savitz DA, Koch G, Strom BL, Guess HA, HartzemaA. Recall accuracy for prescription medications: self reportcompared with database information. Am J Epidemiol 1995;142: 1103–1112.

30. FDA issues public health advisory on Vioxx as its manufacturervoluntarily withdraws the product. http://www.fda.gov/bbs/topics/news/2004/NEW01122.html. Accessed October 5,2004.



the impact of nsaid or cox-2 inhibitor use on the initiation of antihypertensive therapy

Documents