To evaluate the effects of cardiovascular comorbidities and aspirin coprescription on cyclooxygenase (COX)-2 inhibitor (coxib) prescribing patterns among rheumatologists.
To evaluate the effects of cardiovascular comorbidities and aspirin coprescription on cyclooxygenase (COX)-2 inhibitor (coxib) prescribing patterns among rheumatologists.
A prospective cohort study was carried out with rheumatoid arthritis and osteoarthritis patients in the Consortium of Rheumatology Researchers of North America registry. Medication and comorbidity data were obtained prospectively from physician and patient questionnaires between March 2002 and September 2003. Multivariate adjusted associations between coxib use and specific cardiovascular variables, including aspirin use, were examined.
A total of 3,522 arthritis patients were included. COX inhibitors, including coxibs, nonselective nonsteroidal antiinflammatory drugs (NSAIDs), and meloxicam, were prescribed to a larger proportion of osteoarthritis patients (68.4%) than rheumatoid arthritis patients (47.1%) in our study (P < 0.001). COX inhibitors were prescribed to the majority of aspirin users (51.5%) and a similar proportion of nonusers (49.8%). In multivariate analyses, independent predictors of coxib use versus nonselective NSAID use included diagnoses of osteoarthritis (odds ratio [OR] 2.52, 95% confidence interval [95% CI] 1.81–3.52) and diabetes (OR 1.63, 95% CI 1.06–2.51). Conversely, aspirin use independently predicted selection of a nonselective NSAID rather than a coxib (OR 0.73, 95% CI 0.55–0.98). Neither a history of myocardial infarction nor stroke predicted utilization of a coxib. Similarly, cardiovascular variables did not predict the use of rofecoxib versus celecoxib.
Our data indicate that COX inhibitor coprescription among aspirin users is frequent. Despite cardiovascular concerns regarding the coxibs, our data suggest that aspirin use, but not cardiovascular comorbidities, predicted the selection of nonselective NSAIDs over coxibs.
Nonsteroidal antiinflammatory drugs (NSAIDs) have been one of the cornerstones of therapy for the treatment of arthritis-related pain and inflammation. The gastrointestinal toxicities of nonselective NSAIDs, which inhibit both cyclooxygenase (COX)-1 and COX-2, have been well documented in both clinical trials and epidemiologic studies. Since 1998, 3 COX-2–specific inhibitors of the coxib class have been approved by the Food and Drug Administration, including celecoxib, rofecoxib, and valdecoxib. Two additional compounds are currently under review. Two large studies, Celecoxib Long-term Arthritis Safety Study (CLASS) and Vioxx Gastrointestinal Outcomes Research Study (VIGOR), were undertaken to determine the gastrointestinal safety of supratherapeutic doses of celecoxib and rofecoxib, respectively (1–3). In the VIGOR trial, an unexpected 5-fold increase in myocardial infarctions was noted, although cardiovascular outcomes were not the primary endpoint. No increase in cardiovascular events was noted in the CLASS study, although the study design and patient populations of CLASS and VIGOR were markedly different. Subsequent retrospective outcome studies have examined the effect of individual coxibs on hypertension, myocardial infarction, and death, with somewhat conflicting results (4–10).
Previous studies of nonselective NSAIDs and coxib prescribing patterns have shown that coxib users tend to have increased gastrointestinal and cardiovascular risk factors (11–14). However, these studies were mostly conducted prior to the publication of the VIGOR trial and subsequent cardiovascular pooled analyses and outcome studies; thus, the results of previous studies of coxib prescribing patterns may not reflect the growing concern of the cardiovascular effects of the coxib class. For patients at increased risk of myocardial infarction, we hypothesized that rheumatologists would preferentially avoid prescribing coxibs in favor of nonselective NSAIDs.
Controversies have also emerged regarding the relative risks and benefits of aspirin coprescription with coxib therapy. Post-hoc analysis of the VIGOR trial indicated a disproportionate incidence of myocardial infarction among patients for whom aspirin was indicated but not prescribed (3). However, randomized controlled trials have demonstrated that concomitant aspirin use attenuates the gastrointestinal benefits of coxibs, including both endoscopic ulcers and ulcer complications (1, 15, 16). Moreover, whether or not aspirin coprescription would have prevented the increased incidence of myocardial infarction in the VIGOR trial remains speculative. Given the complexity and incomplete nature of the evidence, we sought to examine the prescribing patterns of nonselective NSAIDs and coxibs with particular emphasis on the effect of cardiovascular comorbidities and concomitant aspirin utilization.
Between March 2002 and September 2003, a total of 36 rheumatology practices enrolled patients in the Consortium of Rheumatology Researchers of North America (CORRONA) registry. The CORRONA registry is a prospective study of rheumatology patients that includes patients with rheumatoid arthritis (RA) and osteoarthritis (OA), among other conditions. Beginning in March 2002, both academic rheumatologists and private practice rheumatologists were invited to join the consortium. Recruitment of physicians was limited to rheumatologists practicing in Canada or the United States. Comorbidity and medication information is prospectively collected by patient and physician questionnaires filled out during routine clinical encounters. Completed questionnaires are sent to a central processing site. All CORRONA registry patients with a physician-reported clinical diagnosis of RA or OA were included. Patients with overlapping diagnoses (i.e., OA and RA) or OA and other arthritis diagnoses were excluded. Approval for the CORRONA registry was obtained from the respective Institutional Review Boards of participating academic sites and a central private Institutional Review Board for patients from private practice sites.
Data were prospectively collected from patient and physician questionnaires at a baseline visit and every 3–6 months thereafter. For this study, we utilized physician-collected data as our primary data source for comorbidities and medication use. Coxib use was defined as patients receiving celecoxib, rofecoxib, or valdecoxib. Meloxicam was recorded as a distinct category, neither as a coxib nor as a nonselective NSAID. We excluded meloxicam from comparative analyses because meloxicam is somewhat COX-2 selective, yet is not designated as a coxib. Nonselective NSAIDs included all prescription nonselective NSAIDs, other than those available in over-the-counter (OTC) strengths. OTC NSAID use included both branded and generic OTC naproxen and ibuprofen. Acetaminophen use was also recorded. Concomitant OTC NSAID use with a prescription NSAID or coxib was captured as well, and categorized based on the prescription drug. Aspirin use was collected from patient questionnaires administered at each visit. For each patient, the last encounter in the study period was analyzed to determine the most recent NSAID or coxib used. Information on each CORRONA physician site was collected upon site enrollment.
For the comparison of baseline characteristics and prescribing patterns between RA and OA patients, chi-square and Student's t-tests were used to compare categorical and continuous variables, respectively. These tests were also used to compare aspirin users and nonusers. In the subset of the population receiving either coxibs or NSAIDs, we then determined univariate predictors of coxib (versus NSAID) use and calculated odds ratios (ORs) with 95% confidence intervals (95% CIs). We also constructed a multivariate logistic regression model to determine independent predictors of coxib versus NSAID use. Both forward and backward stepwise regressions resulted in the same model. Further subsetting the population to patients receiving either celecoxib or rofecoxib, we used the same analytic techniques to determine predictors of celecoxib versus rofecoxib use. All analyses were performed using SAS version 8.2 (SAS Inc., Cary, NC).
Among 36 CORRONA sites with enrolled patients during the study period, we found that 27 (75%) of 36 were private practice sites versus 9 (25%) of 36 academic sites. There was a mean of 2.2 (range 1–7) physicians per site. Among the 80 participating physicians, 55 (69%) were men. Analysis of the geographic location of CORRONA sites indicated that 15 (42%) sites were located in the Northeast, 11 (30%) in the Midwest, 8 (22%) in the South, and 2 (6%) in the West.
Our study sample of 3,522 patients consisted of 3,029 (83.7%) RA patients and 493 (16.3%) OA patients. A number of significant differences in patient demographics and comorbidities were noted between the OA and RA cohorts (Table 1). Although no significant difference in sex was found, the OA cohort had an increased percentage of patients older than 65 years (48.5%) compared with the RA cohort (27.5%). Among cardiovascular comorbidities, hypertension was more common in the OA cohort (47.1%) than the RA cohort (30.1%). Similarly, the prevalence of diabetes was increased in the OA cohort (12.4%) versus the RA cohort (7.5%). In contrast, liver disease and peptic ulcer disease were increased in prevalence in the RA compared with the OA cohort. No statistically significant differences were noted among other comorbidities.
|Variable||RA cohort no. (%)||OA cohort no. (%)||P|
|Age >65 years||834 (27.5)||239 (48.5)||< 0.001|
|Male||781 (25.8)||122 (24.7)||0.657|
|Hypertension||912 (30.1)||232 (47.1)||< 0.001|
|Diabetes||226 (7.5)||61 (12.4)||< 0.001|
|Congestive heart failure||25 (0.8)||0 (0.0)||0.040|
|Myocardial infarction||153 (5.1)||24 (4.9)||1.000|
|Stroke||91 (3.0)||12 (2.4)||0.566|
|Other thrombosis||36 (1.2)||0 (0.0)||0.007|
|Gastroesophageal reflux disease||551 (18.2)||93 (18.9)||0.707|
|Peptic ulcer disease||331 (10.9)||37 (7.5)||0.021|
|Liver disease||131 (4.3)||8 (1.6)||0.003|
|Renal disease||144 (4.8)||8 (1.6)||0.001|
The prescribing patterns of COX inhibitors are summarized in Table 2. Overall, a prescription COX inhibitor was prescribed to 68.4% of the OA cohort and 47.1% of the RA cohort. Coxibs as a class were more frequently prescribed to OA patients (47.7%) than RA patients (26.6%). Nonselective NSAIDs were prescribed to 21.9% of RA patients and 21.3% of OA patients. The use of acetaminophen was frequent in both cohorts, but increased among OA patients (46.4%) relative to RA patients (31.3%). Use of OTC NSAIDs was moderate in both the OA (12.8%) and RA (17.3%) cohorts.
|Medication||RA cohort no. (%)||OA cohort no. (%)||P|
|Prescription COX inhibitors||1,427 (47.1)||337 (68.4)||< 0.001|
|Nonselective NSAIDs||664 (21.9)||105 (21.3)||0.814|
|Meloxicam||75 (2.5)||36 (7.3)||< 0.001|
|Coxib||805 (26.6)||235 (47.7)||< 0.001|
|Celecoxib||385 (12.7)||79 (16.0)||0.052|
|Rofecoxib||245 (8.1)||63 (12.8)||0.001|
|Valdecoxib||177 (5.8)||94 (19.1)||< 0.001|
|OTC NSAIDs||524 (17.3)||63 (12.8)||0.013|
|Acetaminophen||919 (31.3)||227 (46.4)||< 0.001|
|Aspirin||502 (16.6)||114 (23.1)||0.001|
Table 3 describes the results of stratification of COX inhibitor prescribing by aspirin use. In our mixed OA/RA cohort, prescription COX inhibitors were prescribed to the majority of aspirin users (51.5%) and a similar proportion of nonusers (49.8%). Coxibs were prescribed to comparable numbers of aspirin users (28.4%) and nonusers (29.8%). There was a trend toward increased nonselective NSAID use among aspirin users compared with nonusers (P = 0.086). We also found moderate use of OTC NSAIDs among both aspirin users (13.6%) and nonusers (17.3%).
|Medication||Aspirin users no. (%)||Aspirin nonusers no. (%)||P|
|Prescription COX inhibitors||317 (51.5)||1,447 (49.8)||0.478|
|Nonselective NSAIDs||129 (20.9)||520 (17.9)||0.086|
|Meloxicam||28 (4.5)||83 (2.9)||0.041|
|Coxib||175 (28.4)||865 (29.8)||0.527|
|Celecoxib||76 (12.3)||388 (13.4)||0.555|
|Rofecoxib||55 (8.9)||253 (8.7)||0.875|
|Valdecoxib||44 (7.1)||227 (7.8)||0.618|
|OTC NSAIDs||84 (13.6)||503 (17.3)||0.028|
|Acetaminophen||209 (33.9)||937 (33.4)||0.814|
When comparing the prescription of a coxib versus a nonselective NSAID in bivariate analyses, a number of demographic and clinical predictors were found (Table 4). The unadjusted OR of prescribing a coxib was increased for age >65 years (OR 1.30, 95% CI 1.04–1.30), a diagnosis of OA (OR 2.71, 95% CI 2.00–3.67), and diabetes (OR 1.54, 95% CI 1.04–2.29). A trend favoring coxib use was found for peptic ulcer disease (OR 1.29, 95% CI 0.91–1.85) as well. Among patients with liver disease (OR 0.53, 95% CI 0.30–0.92), the reduced OR indicated a reduced likelihood of coxib prescription but an increased likelihood of nonselective NSAID prescription. Similarly, a trend toward reduced coxib use was found for aspirin (OR 0.84, 95% CI 0.65–1.08). In contrast, no significant difference in coxib use was found for patients with a history of myocardial infarction (OR 1.01, 95% CI 0.60–1.70), stroke (OR 1.56, 95% CI 0.77–3.16), or other thrombosis (OR 0.78, 95% CI 0.29–2.10).
|Variable||Coxib versus nonselective NSAID||Celecoxib versus rofecoxib|
|Unadjusted OR||95% CI||P||Unadjusted OR||95% CI||P|
|Age >65 years||1.30||1.04–1.63||0.022||1.14||0.91–1.91||0.146|
|OA diagnosis||2.71||2.00–3.67||< 0.001||1.11||0.83–1.47||0.490|
|Congestive heart failure||0.53||0.19–1.46||0.220||3.32||0.39–28.58||0.274|
|Gastroesophageal reflux disease||0.97||0.75–1.25||0.790||1.10||0.77–1.57||0.595|
|Peptic ulcer disease||1.29||0.91–1.85||0.156||1.36||0.84–2.21||0.214|
In multivariate analysis (Table 5), independent predictors of coxib use versus nonselective NSAID use included a diagnosis of OA (OR 2.52, 95% CI 1.81–3.52) and diabetes (OR 1.63, 95% CI 1.06–2.51). Conversely, aspirin use independently predicted selection of a nonselective NSAID rather than a coxib, as reflected by its adjusted OR (0.73, 95% CI 0.55–0.98). The adjusted OR among patients with a history of peptic ulcer disease (OR 1.38, 95% CI 0.94–2.02) demonstrated a trend favoring coxib prescription.
|Variable||Adjusted OR||95% CI||P|
|OA diagnosis||2.52||1.81–3.52||< 0.001|
|Peptic ulcer disease||1.38||0.94–2.02||0.103|
A secondary analysis of celecoxib versus rofecoxib prescription is found in Table 4. Neither cardiovascular nor gastrointestinal patient characteristics predicted the selection of celecoxib versus rofecoxib. Specifically, myocardial infarction (OR 0.74, 95% CI 0.36–1.50) was not associated with selection of celecoxib versus rofecoxib. There was no difference in predictors of rofecoxib versus celecoxib use for any of the other cardiovascular conditions, including hypertension, diabetes, stroke, congestive heart failure, and other thrombosis. Similarly, concomitant aspirin use was not associated with selection of celecoxib versus rofecoxib (OR 0.89, 95% CI 0.62–1.28).
In our study, we examined the effect of cardiovascular comorbidities and aspirin coprescription on COX inhibitor prescribing patterns among 29 rheumatology practices using the CORRONA registry. Studies of COX inhibitor prescribing patterns conducted prior to publication of the VIGOR trial indicated that coxibs were preferentially prescribed to patients at increased cardiovascular risk. One of these earlier studies from a managed care database specifically investigated cardiorenal predictors of coxib use from October 1999 through September 2000, predating the publication of the VIGOR trial. The managed care study reported an adjusted OR of 1.60 (95% CI 1.20–2.13) for patients with preexisting ischemic heart disease. In contrast, our study examined prescribing patterns among rheumatologists over an 18-month period that began >1 year after the publication of the VIGOR trial. Our data suggests no evidence of preferential use of coxibs versus NSAIDs for patients with a history of myocardial infarction or stroke.
Additionally, our results indicate that aspirin is frequently coprescribed with both nonselective NSAIDs and coxibs. Unlike previous studies that either did not examine aspirin or captured aspirin use from administrative databases, the CORRONA registry records all aspirin use, including OTC aspirin, prospectively. We found that aspirin independently predicted nonselective NSAID use rather than coxib use. This prescribing pattern may represent consideration of the CLASS trial, which found that aspirin use reduced the gastroprotective effect of supratherapeutic doses of celecoxib. Alternatively, this preference for nonselective NSAID use among aspirin users may represent reluctance to prescribe coxibs to patients at increased cardiovascular risk. However, this rationale for choosing nonselective NSAIDs among aspirin users was not found for patients with a history of hypertension, diabetes, myocardial infarction, or stroke. In fact, patients with a history of diabetes were more likely to receive a coxib. Thus, concern for a reduced gastroprotective effect of coxibs with concomitant aspirin use is a more plausible explanation for our findings.
In secondary analyses of celecoxib versus rofecoxib prescribing, we found no evidence of differential patterns of prescribing between these 2 agents. Although we hypothesized that patients with cardiovascular risk factors or established cardiovascular disease may be more likely to receive celecoxib than rofecoxib, our data did not support this hypothesis. In fact, the odds ratios for receiving celecoxib for patients with a history of myocardial infarction and stroke were both <1, although the confidence intervals were wide and the results not statistically significant. Nevertheless, our data suggest that rheumatologists do not differentiate between celecoxib and rofecoxib for patients at increased cardiovascular risk. Similarly, patients with gastrointestinal risk factors were no more likely to be prescribed rofecoxib than celecoxib. Whether or not recent pharmacoepidemiologic studies suggesting a differential cardiovascular risk associated with rofecoxib versus celecoxib will change future prescribing patterns is unknown (10, 17).
The cardiovascular safety of coxibs remains controversial. Although it has been suggested that the coprescription of aspirin with coxibs would reduce the likelihood of adverse cardiovascular events, this strategy remains unproven in randomized trials (18). Furthermore, the cardiovascular safety of concomitant aspirin therapy with nonselective NSAID use has also been challenged. In healthy volunteers, ibuprofen, but not diclofenac or rofecoxib, impaired aspirin's effect on thromboxane inhibition and platelet aggregation (19). Further evidence emerged from a subanalysis of the Physician's Health Study indicating that nonselective NSAIDs reduce the cardioprotective effect of aspirin in primary prevention of myocardial infarction (20). Moreover, a retrospective study from the United Kingdom concluded that concomitant aspirin and ibuprofen use was associated with an increased risk of cardiac-related and all-cause mortality (21). Additional randomized trials of the cardiovascular effects of aspirin and coxib coprescription are needed and currently are underway.
Many of the strengths of this study are attributable to the quality of the clinical and prescription data. In the CORRONA registry, all data is prospectively collected with comorbidity and prescription data recorded by the treating rheumatologist at the time of the visit. Diagnoses are recorded directly by the treating physician during the patient encounter. In contrast, the accuracy of administrative codes for arthritis-related diagnoses and other cardiovascular diseases has been shown to be highly variable (22–24). For the diagnosis of OA using Medicare physician claims, Katz et al found the positive predictive value of OA at any anatomic site to be 0.83 (95% CI 0.76–0.90), with particularly poor predictive value for OA of the hip (22). Furthermore, administrative datasets do not capture OTC aspirin or NSAID use. Studies have found that OTC aspirin represents 26–39% of total aspirin use (25, 26). Earlier studies of coxib prescribing patterns relied upon administrative data sources for which OTC aspirin was not captured (11–13). The CORRONA registry explicitly asks patients at each visit if they are currently taking aspirin.
This study has several potential limitations. Because our study utilized the CORRONA registry, which is limited to rheumatology practices, our results may not be generalizable to other physician specialties. In addition, the CORRONA registry groups all nonselective NSAID agents together for the purposes of data collection. Thus, differentiation among COX inhibitor agents was only possible for coxibs. It is also possible that residual confounding persisted despite our adjustment for covariates.
In summary, we found that aspirin use, but not cardiovascular comorbidities, predicted preferential prescribing of nonselective NSAIDs rather than coxibs. In light of recent evidence that ibuprofen and perhaps other nonselective NSAIDs may impair the cardioprotective effects of aspirin, additional studies of the safety of both coxib and nonselective NSAIDS among aspirin users are needed to guide therapeutic decisions.
While this manuscript was in press, rofecoxib was withdrawn worldwide due to results of an ongoing colorectal polyp trial demonstrating an increased risk of cardiovascular events associated with rofecoxib (27).