Abstract
Aims
Previous studies have suggested that nonsteroidal anti‐inflammatory drugs (NSAIDs) may be associated with higher cardiovascular risks. However, few have been active comparison studies that directly assessed the potential differential cardiovascular risk between NSAID classes or across individual NSAIDs. We compared the risk of major cardiovascular events between cyclooxygenase 2 (COX‐2)‐selective and nonselective NSAIDs in patients with hypertension.
Methods
We conducted a cohort study of patients with hypertension who initiated COX‐2‐selective or nonselective NSAIDs in a population‐based Taiwanese database. The outcomes included hospitalization for the following major cardiovascular events: ischaemic stroke, acute myocardial infarction, congestive heart failure, transient ischaemic attack, unstable angina or coronary revascularization. We followed patients for up to 4 weeks, based on the as‐treated principle. We used inverse probability weighting to control for baseline and time‐varying covariates, and estimated the on‐treatment hazard ratios (HRs) and 95% conservative confidence interval (CIs).
Results
We identified 2749 eligible COX‐2‐selective NSAID users and 52 880 eligible nonselective NSAID users. The HR of major cardiovascular events comparing COX‐2‐selective with nonselective NSAIDs after adjusting for baseline and time‐varying covariates was 1.07 (95% CI 0.65, 1.74). We did not observe a differential risk when comparing celecoxib to diclofenac (HR 1.17; 95% CI 0.61, 2.25), ibuprofen (HR 1.36; 95% CI 0.58, 3.18) or naproxen (HR 0.75; 95% CI 0.23, 2.44). There was an increased risk with COX‐2‐selective NSAIDs, however, when comparing COX‐2‐selective NSAIDs with mefenamic acid (HR 2.11; 95% CI 1.09, 4.09).
Conclusions
Our results provide important information about the comparative cardiovascular safety of NSAIDs in patients with hypertension.
Keywords: cardiovascular safety, cohort study, COX‐2 selective, hypertension, nonsteroidal anti‐inflammatory drugs, pharmacoepidemiology
What is Already Known about this Subject
A number of studies have suggested that nonsteroidal anti‐inflammatory drugs (NSAIDs), including cyclooxygenase 2(COX‐2)‐selective and nonselective NSAIDs, are associated with an increased risk of cardiovascular adverse outcomes.
It is unclear if there is a differential cardiovascular risk between NSAID classes or across individual NSAIDs because there have been limited active comparison studies, especially in patients with hypertension.
What this Study Adds
In this population‐based cohort study, which included 55 629 patients with hypertension, we observed no difference in cardiovascular risk between patients who were on COX‐2‐selective and nonselective NSAID treatment.
There was no apparent difference in cardiovascular risk when comparing celecoxib with diclofenac, ibuprofen or naproxen, although a significantly increased risk was observed when comparing celecoxib with mefenamic acid.
Our results provide important information about the comparative safety of various NSAIDs in patients with hypertension in real‐world settings. The findings suggest that with a low‐to‐moderate daily dose and a short‐term treatment period, most commonly used NSAIDs, including celecoxib, diclofenac, ibuprofen and naproxen, have similar cardiovascular safety profiles.
Introduction
Nonsteroidal anti‐inflammatory drugs (NSAIDs) inhibit http://www.guidetopharmacology.org/GRAC/FamilyDisplayForward?familyId=269 enzymes (COX, including COX‐1 and COX‐2) and are classified as COX‐2‐selective and nonselective NSAIDs based on their selectivity for COX‐2 inhibition. Both types of NSAIDs are widely used for their anti‐inflammatory and pain‐relieving effects. The cardiovascular safety of NSAIDs, however, has been a subject of considerable debate for decades 1, 2. The Vioxx Gastrointestinal Outcome Research (VIGOR) trial raised a concern about a higher risk of myocardial infarction (MI) associated with rofecoxib in patients with rheumatoid arthritis, leading to the voluntary withdrawal of rofecoxib from the market in 2004 3, 4. Since then, multiple randomized controlled trials (RCTs) and observational studies reported that other COX‐2‐selective and nonselective NSAIDs may also be associated with adverse cardiovascular thrombotic events 5, 6, 7, 8, 9.
There is still conflicting evidence about the comparative cardiovascular safety of NSAIDs. Some research has suggested that the harmful cardiovascular effect observed with rofecoxib may not be a class effect for all COX‐2‐selective NSAIDs, and that the cardiovascular risk may also vary across individual nonselective NSAIDs, with naproxen having a better safety profile 10, 11. One early meta‐analysis of RCTs, mainly conducted in arthritis patients, suggested that COX‐2‐selective NSAIDs, as a class, have a higher risk of MI than nonselective NSAIDs, with a hazard ratio (HR) of 1.46 [95% confidence interval (CI) 1.02, 2.09] 12. However, two large‐scale, non‐inferior RCTs in arthritis patients 13, 14 and some active comparison observational studies in elderly patients or those hospitalized for coronary heart disease (CHD) 15, 16, 17, 18 found no difference in cardiovascular risk when comparing individual COX‐2‐selective NSAIDs (celecoxib and etoricoxib) with individual nonselective NSAIDs (diclofenac, ibuprofen and naproxen). A recent observational study in Canadian and European healthcare databases also suggested that, as compared to non‐use of NSAIDs, the risk of MI with celecoxib did not appear to be higher than that with diclofenac, ibuprofen and naproxen in the general population 19.
Hypertension is a major risk factor for cardiovascular diseases 20, and the destabilization of blood pressure is a potential effect modifier of NSAID‐related cardiovascular events 21, 22. Few studies, however, have used an active comparison approach to compare the safety spectrum among different NSAIDs in this population. Previous studies have found that patient characteristics and treatment duration may differ between COX‐2‐selective and nonselective NSAID users 15, 16, 17, 18. We aimed to examine the risk of major cardiovascular events in patients with hypertension taking COX‐2‐selective vs. nonselective NSAIDs. Specifically, we used inverse probability weighting to account for potential baseline and time‐dependent confounding and differential follow‐up, and modelled the cardiovascular risk associated with up to 4 weeks of continuous NSAID use.
Methods
Data source and source population
A single‐payer National Health Insurance programme began in Taiwan in 1995, achieving an enrolment rate of 99% by 2014. The National Health Research Institute constructed the National Health Insurance Research Database (NHIRD), which consisted of linked data from demographic and enrolment records, hospital admissions, outpatient visits and pharmacy dispensing claims from hospitals, outpatient clinics and community pharmacies. The source population for the study included all beneficiaries in the Longitudinal Health Insurance Database, which comprised a random sample of one million individuals from the NHIRD with longitudinally linked, annually updated data, with the exception of information about deaths. The study protocol was approved by the National Taiwan University Hospital Research Ethics Committee.
Study population and study drugs
We identified patients initiating treatment with oral COX‐selective or nonselective NSAIDs at outpatient visits between 1 January 2010 and 31 December 2012 using the World Health Organization's Anatomical Therapeutic Chemical (ATC) classification system codes (see Table S1 for a list of codes). We defined NSAID initiation as the first oral NSAID dispensed during the study period, with no dispensing for oral or intravenous forms of NSAIDs in the preceding 180 days. The index date was the date of the first dispensing of a study NSAID. To ensure that sufficient data were available to capture baseline characteristics, we excluded patients with fewer than 180 days of continuous enrolment before the index date and required patients to have at least one inpatient or outpatient visits during this time window. We further restricted the cohort to patients having at least one inpatient or two outpatient diagnoses of hypertension [International Classification of Diseases, 9th Revision, Clinical modification (ICD‐9‐CM) codes 401–405] in the 180 days before the index date 23. We excluded patients aged under 20 years on the index date, patients with missing age or ambiguous gender information, and patients initiating both a COX‐2‐selective and nonselective NSAID on the index date.
Follow‐up and outcomes
We defined the outcomes of interest – major cardiovascular events – as the first hospitalization for ischaemic stroke, acute MI, congestive heart failure, transient ischaemic attack, unstable angina and coronary revascularization (including coronary artery bypass grafting and percutaneous transluminal coronary angioplasty) during follow‐up. We defined each outcome using validated claims‐based algorithms (Table S2) with positive predictive values of 88% for ischaemic stroke 24, 80% for acute MI 25, 84% for congestive heart failure 26, 77% for transient ischaemic attack 27, 88% for unstable angina 28, 98% for coronary artery bypass grafting 29 and 95–96% for percutaneous transluminal coronary angioplasty 29. As congestive heart failure may be indirectly related to renal effects rather than the thromboembolic effects of NSAIDs, we also conducted another analysis that excluded congestive heart failure from the outcome definition.
To minimize the possibility of exposure misclassification during the follow‐up, we followed patients based on the as‐treated principle. Specifically, we followed patients for up to 28 days (i.e. 4 weeks), from the index date to the earliest of the following: outcome occurrence; index NSAID treatment discontinuation; NSAID treatment change; or last date of hospital discharge, outpatient visit or community pharmacy visit. NSAID treatment discontinuation occurred when more than 14 days elapsed between the end of one dispensing and the date of the next dispensing, if any. NSAID treatment change (switch or addition) occurred when an NSAID was dispensed that was different from the index study drug (i.e. a patient with an index COX‐2‐selective NSAID filling a nonselective NSAID, and vice versa). As our data source did not include information on deaths, we censored patients on the last date of their hospital discharge, outpatient visit or community pharmacy visit during follow‐up, to reduce potential outcome misclassification or biases associated with the inability to capture cardiovascular‐related deaths (i.e. misclassifying patients who have died due to cardiovascular causes as being alive and having no cardiovascular outcomes). Disenrollment from the National Health Insurance programme was uncommon.
Covariate assessment and adjustment
We considered two groups of covariates: baseline and time‐varying covariates. Baseline covariates included age on the index date, gender, calendar year of the index date, comorbidities, other medication use and resource utilization recorded within 180 days before the index date. We ascertained comorbidities based on inpatient and outpatient diagnosis files, and determined medication use based on the pharmacy dispensing claims of outpatient clinics and community pharmacies. We measured resource utilization based on the records of hospital admissions and outpatient visits.
Time‐varying covariates included cardiovascular comorbidities and cardiovascular medications, updated weekly during the follow‐up period. For cardiovascular comorbidities, provided that patients had one comorbidity at any time during follow‐up, the status would be carried forward through the remainder of their follow‐up period. For cardiovascular medications, whether patients used one medication in a week depended on the pharmacy claim records in that week. Therefore, patients might use one drug in one week and then discontinue it in the following week. Tables S3 and S4 provide more detailed information on covariates.
Statistical analysis
We applied two analytical approaches. In the first approach, we adjusted for confounding and potential selection bias arising from differential loss to follow‐up using measured baseline covariates only. Specifically, we used all measured baseline covariates to estimate the inverse probability of treatment weight (IPTW) and the inverse probability of censoring weight (IPCW) 30, 31. The IPTW was calculated as the inverse of the treatment probability, given all measured baseline covariates, and the IPCW was calculated as the inverse of the probability of remaining uncensored, given treatment status and all measured baseline covariates. To reduce the variability of the weights, we ‘stabilized’ them by replacing the numerator of the IPTW by the average treatment probability, and the numerator of the IPCW by the probability of remaining uncensored, given the treatment status and select baseline covariates. We fitted logistic regression models to estimate all the probabilities. Appendix S1 describes the computation steps in detail.
We then multiplied the stabilized IPTW and the stabilized IPCW, to obtain an overall weight for each patient. This allowed us to create a weighted population in which the receipt of NSAID treatment and loss to follow‐up were independent of the measured baseline covariates. We truncated the overall weight at the first and 99th percentile cut‐points, to minimize further the influence of extreme weights. With a person‐week data structure, we used the generalized estimating equations and the robust variance estimators to fit a weighted pooled logistic regression model, which approximated a weighted Cox model, and estimated HRs and conservative 95% CIs of major cardiovascular events. As we followed patients based on the as‐treated principle, the HRs would represent the on‐treatment effect while patients remained on their index COX‐2‐selective NSAID treatment vs. patients who remained on their index nonselective NSAID treatment during the follow‐up.
In the second approach, we repeated the same steps as above, except that now we estimated the stabilized IPTW and stabilized IPCW using both baseline and time‐varying covariates updated weekly, and fitted pooled logistic regression models to estimate all the probabilities 32, 33, 34, 35. The denominator of the stabilized IPTW was the treatment probability, given the treatment history, all measured baseline covariates and all measured time‐varying covariates; the denominator of the stabilized IPCW was the probability of remaining uncensored, given the treatment history, all measured baseline covariates and all measured time‐varying covariates. The numerator of the stabilized IPTW was the treatment probability, given the treatment history and select baseline covariates; the numerator of the stabilized IPCW was the probability of remaining uncensored, given the treatment history and select baseline covariates. Appendix S2 provides detailed computation steps.
Subgroup and sensitivity analyses
To examine if the HR would be different across individual nonselective NSAIDs, we conducted additional analyses to compare COX‐2‐selective NSAIDs (celecoxib and etoricoxib) separately with diclofenac, ibuprofen, naproxen and mefenamic acid, which are expected to be commonly used in clinical settings.
We also conducted several sensitivity analyses, to examine the robustness of our results. First, use of low‐dose aspirin is a potential marker for patients' underlying cardiovascular risk. Therefore, we additionally included the interaction term of NSAID use and low‐dose aspirin in the outcome model, to examine if there was any effect modification by patients' underlying cardiovascular risk. Secondly, we performed a ‘first exposure carried forward’ (or ‘intention‐to‐treat’) analysis by following patients regardless of NSAID treatment discontinuation or change and censoring them only at the first outcome occurrence, the last date of accessing medical resources or 28 days from the index date.
Thirdly, besides using inverse probability weighting to adjust for both baseline and time‐varying covariates, we estimated propensity scores (PSs) – the probabilities of initiating COX‐2‐selective NSAIDs – using all measured baseline covariates and conducted PS matching to minimize confounding effects by baseline covariates. We also performed inverse probability weighting after PS matching, to account for time‐varying covariates. Finally, we used high‐dimensional PSs (hd‐PSs) to identify and include an additional 50 empirically identified variables in the PS model 36, 37.
Nomenclature of targets and ligands
Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY 38, and are permanently archived in the Concise Guide to PHARMACOLOGY 2015/16 39.
Results
Baseline characteristics
We identified a total of 55 629 eligible patients (Figure 1). Among patients initiating COX‐2‐selective NSAIDs (N = 2749), celecoxib (65%) was prescribed more commonly than etoricoxib (35%). Among patients initiating nonselective NSAIDs (N = 52 880), diclofenac (34%), mefenamic acid (22%) and ibuprofen (15%) were most commonly prescribed; only 3% of patients initiated naproxen. The mean [standard deviation (SD)] age of the cohort was 64 (13) years and 51% were male. Compared with patients initiating nonselective NSAIDs, patients initiating COX‐2‐selective NSAIDs were more likely to be older and female, and appeared to have more comorbidities, more medication use and higher resource utilization (Table 1).
Figure 1.
Flow chart of the study cohort assembly. COX, cyclooxygenase; NSAIDs, non‐steroidal anti‐inflammatory drugs
Table 1.
Baseline characteristics by NSAID group: COX‐2‐selective NSAIDs vs. nonselective NSAIDs
| Total study cohort (N = 55 629) | Weighted cohorta | |||||
|---|---|---|---|---|---|---|
| COX‐2 selective NSAIDs (N = 2749) | Nonselective NSAIDs (N = 52 880) | Standardized difference | COX‐2‐selective NSAIDs (N = 2749) | Nonselective NSAIDs (N = 52 880) | Standardized difference | |
| Age at initiation, years, mean (SD) | 74.1 (9.7) | 63.0 (12.9) | 0.97 | 70.1 (8.3) | 63.6 (13.0) | 0.60 |
| Male, % | 39.8 | 51.3 | −0.23 | 45.2 | 50.7 | −0.11 |
| Initiation year, % | ||||||
| 2010 | 61.3 | 64.6 | −0.07 | 62.9 | 64.4 | −0.03 |
| 2011 | 25.0 | 24.5 | 0.01 | 24.7 | 24.6 | 0.00 |
| 2012 | 13.7 | 10.9 | 0.09 | 12.4 | 11.0 | 0.04 |
| Comorbidities, % | ||||||
| Ischaemic heart disease b | 22.1 | 16.3 | 0.15 | 19.2 | 16.6 | 0.07 |
| Myocardial infarction b | 1.4 | 0.8 | 0.06 | 1.1 | 0.8 | 0.03 |
| Cardiac dysrhythmia b | 9.5 | 6.0 | 0.13 | 8.0 | 6.1 | 0.07 |
| Congestive heart failure b | 7.8 | 5.2 | 0.10 | 7.2 | 5.3 | 0.08 |
| Cerebrovascular disease b | 19.1 | 10.2 | 0.25 | 15.7 | 10.7 | 0.15 |
| Ischaemic stroke b | 10.1 | 5.3 | 0.18 | 8.5 | 5.6 | 0.11 |
| Haemorrhagic stroke b | 1.9 | 1.0 | 0.07 | 1.5 | 1.2 | 0.04 |
| Peripheral vascular disease b | 2.0 | 1.0 | 0.08 | 1.5 | 1.2 | 0.04 |
| Disorders of lipid metabolism b | 32.2 | 33.1 | −0.02 | 33.3 | 33.2 | 0.01 |
| Diabetes mellitus b | 31.6 | 29.6 | 0.04 | 32.6 | 29.7 | 0.06 |
| Thyroid disease | 8.1 | 4.9 | 0.13 | 6.6 | 5.0 | 0.07 |
| COPD | 5.0 | 3.5 | 0.07 | 4.3 | 3.6 | 0.04 |
| Asthma | 7.6 | 8.0 | −0.01 | 8.4 | 7.9 | 0.02 |
| Chronic liver disease | 21.9 | 16.3 | 0.14 | 21.0 | 16.6 | 0.11 |
| Gastritis or peptic ulcer disease | 7.1 | 3.4 | 0.17 | 6.0 | 3.6 | 0.11 |
| Chronic kidney disease | 50.6 | 11.2 | 0.94 | 22.3 | 13.2 | 0.24 |
| Rheumatoid arthritis/osteoarthritis | 12.2 | 2.4 | 0.39 | 6.0 | 2.9 | 0.15 |
| Osteoporosis | 7.5 | 8.1 | −0.02 | 8.0 | 8.0 | −0.00 |
| Gout | 7.5 | 4.4 | 0.13 | 6.8 | 4.6 | 0.10 |
| Any cancer | 4.8 | 1.8 | 0.17 | 3.3 | 2.0 | 0.08 |
| Dementia | 22.1 | 16.3 | 0.15 | 19.2 | 16.6 | 0.07 |
| Medication use, % | ||||||
| ACEIs/ARBs c | 61.9 | 56.9 | 0.10 | 61.3 | 57.2 | 0.08 |
| β‐Blockers c | 41.0 | 42.0 | −0.02 | 39.8 | 42.0 | −0.04 |
| Calcium channel blockers c | 65.0 | 62.8 | 0.05 | 63.7 | 62.9 | 0.02 |
| Diuretics c | 26.2 | 20.3 | 0.14 | 23.7 | 20.6 | 0.07 |
| Other antihypertensive agents c | 10.4 | 8.8 | 0.06 | 9.7 | 8.8 | 0.03 |
| Nitrates c | 15.3 | 8.5 | 0.21 | 11.3 | 8.8 | 0.08 |
| Antiarrhythmic agents c | 3.9 | 1.9 | 0.12 | 3.3 | 2.0 | 0.08 |
| Digoxin c | 2.8 | 1.3 | 0.10 | 2.1 | 1.4 | 0.05 |
| Aspirin c | 35.7 | 27.8 | 0.17 | 33.8 | 28.2 | 0.12 |
| Clopidogrel c | 6.0 | 2.6 | 0.17 | 4.4 | 2.7 | 0.09 |
| Warfarin c | 1.9 | 1.0 | 0.07 | 1.7 | 1.0 | 0.06 |
| Statins/fibrates c | 28.5 | 27.6 | 0.02 | 29.3 | 27.6 | 0.04 |
| Insulin c | 5.3 | 3.4 | 0.09 | 4.6 | 3.5 | 0.06 |
| Oral antidiabetic agents c | 28.3 | 27.0 | 0.03 | 29.1 | 27.1 | 0.05 |
| Histamine‐2 antagonists/PPIs | 19.6 | 18.2 | 0.04 | 20.3 | 18.3 | 0.05 |
| Antiepileptic agents | 9.5 | 5.0 | 0.17 | 7.7 | 5.3 | 0.10 |
| Antidepressants | 10.5 | 6.1 | 0.16 | 7.8 | 6.3 | 0.06 |
| Anxiolytic agents | 37.0 | 29.1 | 0.17 | 34.7 | 29.5 | 0.11 |
| Hypnotic agents | 18.3 | 13.0 | 0.15 | 16.6 | 13.3 | 0.09 |
| Antipsychotic agents | 7.6 | 6.0 | 0.06 | 7.1 | 6.1 | 0.04 |
| Oestrogen | 1.2 | 1.4 | −0.02 | 1.1 | 1.4 | −0.02 |
| Uric acid‐lowering agents | 10.4 | 10.4 | −0.00 | 11.1 | 10.4 | 0.02 |
| Resource utilization, mean (SD) | ||||||
| No. of hospitalizations | 1.8 (2.7) | 1.2 (2.3) | 0.25 | 1.6 (2.3) | 1.2 (2.4) | 0.19 |
| No. of outpatient visits | 17.5 (11.1) | 14.2 (9.1) | 0.33 | 16.4 (8.8) | 14.3 (9.2) | 0.23 |
ACEI, angiotensin‐converting enzyme inhibitor; ARB, angiotensin II receptor blocker; COPD, chronic obstructive pulmonary disease; COX, cyclooxygenase; NSAID, non‐steroidal anti‐inflammatory drug; PPI, proton pump inhibitor; SD, standard deviation
Weighted by the inverse probability of treatment weight derived by baseline covariates
Cardiovascular comorbidities included ischaemic heart disease, myocardial infarction, cardiac dysrhythmia, congestive heart failure, cerebrovascular disease, ischaemic stroke, haemorrhagic stroke, peripheral vascular disease, disorders of lipid metabolism, and diabetes mellitus
Cardiovascular medications included ACEIs/ARBs, β‐blockers, calcium channel blockers, diuretics, other antihypertensive agents, nitrates, antiarrhythmic agents, digoxin, aspirin, clopidogrel, warfarin, statins/fibrates, insulin and oral antidiabetic agents
After applying the stabilized IPTW derived from baseline covariates, the balance in baseline covariates between COX‐2‐selective and nonselective NSAID groups improved, with most of the covariates showing an absolute standardized difference of less than 0.1, although there were still slight differences between treatment groups (Table 1). Tables S5–S8 show differences in baseline covariates when comparing COX‐2‐selective NSAIDs to individual nonselective NSAIDs. Table S9 presents the distributions of stabilized IPTW and IPCW weights.
Follow‐up and outcomes
The mean follow‐up duration was 24 days for COX‐2‐selective NSAIDs and 18 days for nonselective NSAIDs. The crude incidence rate was 122 per 1000 person‐years (based on 22 outcome events) in the COX‐2‐selective NSAID group and 76 per 1000 person‐years (based on 193 events) in the nonselective NSAID group (Table 2). There were 7919 patients (14% of the study cohort) who remained uncensored 4 weeks after the cohort entry. COX‐2‐selective NSAID initiators were more likely to be uncensored than nonselective NSAID initiators (proportion of remaining uncensored: 50% vs. 12%). The mean daily dose used for individual NSAIDs was: 210 mg for celecoxib, 60 mg for etoricoxib, 107 mg for diclofenac, 1084 mg for ibuprofen, 694 mg for naproxen and 1248 mg for mefenamic acid.
Table 2.
Follow‐up and incidence rates of major cardiovascular events, by NSAID group
| COX‐2‐selective NSAIDs | Nonselective NSAIDs | Celecoxib | Diclofenac | Ibuprofen | Naproxen | Mefenamic acid | |
|---|---|---|---|---|---|---|---|
| Total No. of patients | 2749 | 52 880 | 1779 | 17 882 | 7927 | 1497 | 11 830 |
| Total follow‐up person‐days | 65 782 | 932 905 | 42 521 | 318 051 | 130 405 | 26 280 | 195 681 |
| Mean follow‐up days (SD) | 23.93 (6.30) | 17.64 (5.84) | 23.90 (6.35) | 17.79 (5.64) | 16.45 (5.15) | 17.56 (6.24) | 16.54 (4.72) |
| No. of patients with events | 22 | 193 | 17 | 73 | 24 | 7 | 30 |
| Crude incidence rate per 1000 person‐years (95% CI) | 122.15 (80.43–185.52) | 75.56 (65.62–87.01) | 146.03 (90.78–234.90) | 83.83 (66.65–105.45) | 67.22 (45.06–100.29) | 97.29 (46.38–204.08) | 56.00 (39.15–80.09) |
CI, confidence interval; COX, cyclooxygenase; NSAID, non‐steroidal anti‐inflammatory drug; SD, standard deviation
Risk of major cardiovascular events comparing COX‐2‐selective vs. nonselective NSAIDs
The crude HR of major cardiovascular events comparing COX‐2‐selective vs. nonselective NSAIDs in the primary on‐treatment analysis was 1.99 (95% CI, 1.28, 3.09) (Table 3). The adjusted HR was 1.09 (95% CI 0.64, 1.86) in the analysis that adjusted only for baseline covariates. After adjusting for both baseline and time‐varying covariates, the adjusted HR was 1.07 (95% CI 0.65, 1.74). There were 137 events (13 for COX‐2‐selective NSAIDs and 124 for nonselective NSAIDs) when we excluded congestive heart failure from the outcome definition. The adjusted HR did not change materially (1.07; 95% CI 0.56, 2.04).
Table 3.
Hazard ratios of major cardiovascular events in patients who were on COX‐2‐selective NSAID treatment vs. patients who were on nonselective NSAID treatment during the follow‐up
| Comparison | Adjusted hazard ratio (95% CI) | Adjusted hazard ratio (95% CI) | |
|---|---|---|---|
| Crude hazard ratio (95% CI) | Weighted by stabilized IPTW and IPCW estimated from baseline covariates | Weighted by stabilized IPTW and IPCW estimated from baseline and time‐varying covariates | |
| COX‐2‐selective NSAIDs vs. | |||
| nonselective NSAIDs | 1.99 (1.28, 3.09) | 1.09 (0.64, 1.86) | 1.07 (0.65, 1.74) |
| Diclofenac | 1.80 (1.12, 2.88) | 0.89 (0.51, 1.57) | 0.91 (0.51, 1.63) |
| Ibuprofen | 2.29 (1.29, 4.05) | 0.82 (0.37, 1.81) | 1.22 (0.58, 2.57) |
| Naproxen | 1.49 (0.64, 3.44) | 0.49 (0.19, 1.30) | 0.67 (0.24, 1.91) |
| Mefenamic acid | 2.77 (1.61, 4.77) | 1.65 (0.86, 3.17) | 1.63 (0.91, 2.93) |
| Celecoxib vs. | |||
| Diclofenac | 2.16 (1.28, 3.65) | 1.20 (0.65, 2.22) | 1.17 (0.61, 2.25) |
| Ibuprofen | 2.77 (1.49, 5.13) | 1.06 (0.45, 2.46) | 1.36 (0.58, 3.18) |
| Naproxen | 1.80 (0.76, 4.31) | 0.50 (0.19, 1.35) | 0.75 (0.23, 2.44) |
| Mefenamic acid | 3.35 (1.85, 6.04) | 2.17 (1.04, 4.54) | 2.11 (1.09, 4.09) |
CI, confidence interval; COX, cyclooxygenase; IPCW, inverse probability of censoring weight; IPTW, inverse probability of treatment weight; NSAID, non‐steroidal anti‐inflammatory drug
Comparing COX‐2‐selective NSAIDs vs. individual nonselective NSAIDs yielded adjusted HRs of: 0.91 (95% CI 0.51, 1.63) vs. diclofenac; 1.22 (95% CI 0.58, 2.57) vs. ibuprofen; 0.67 (95% CI 0.24, 1.91) vs. naproxen; and 1.63 (95% CI 0.91, 2.93) vs. mefenamic acid (Table 3). We observed no differential risks of major cardiovascular events when comparing celecoxib with diclofenac (adjusted HR 1.17; 95% CI 0.61, 2.25), ibuprofen (adjusted HR 1.36; 95% CI 0.58, 3.18) or naproxen (adjusted HR 0.75; 95% CI 0.23, 2.44). Nevertheless, there was a significantly increased risk when comparing celecoxib with mefenamic acid (adjusted HR 2.11; 95% CI 1.09, 4.09). We could not compare etoricoxib with individual nonselective NSAIDs separately because of limited outcome events for etoricoxib. Use of low‐dose aspirin did not significantly modify the adjusted HRs of celecoxib vs. diclofenac (P‐value for interaction: 0.17), celecoxib vs. ibuprofen (P = 0.27), celecoxib vs. naproxen (P = 0.20) or celecoxib vs. mefenamic acid (P = 0.13).
Table S10 shows the findings for each comparison using the ‘first exposure carried forward‘ approach; these were similar to the results using the as‐treated approach. When we applied PS matching to control for baseline covariates, we observed numerically elevated HRs comparing celecoxib vs. diclofenac (1.44; 95% CI 0.68, 3.04), ibuprofen (1.43; 95% CI 0.57, 3.50) and mefenamic acid (2.12; 95% CI 0.88, 3.08). The HR comparing celecoxib vs. naproxen was 0.74 (95% CI 0.21, 2.65). After applying inverse probability weighting to handle time‐varying covariates further in the PS‐matched analysis, the adjusted HRs of celecoxib vs. diclofenac (1.20; 95% CI 0.51, 2.80) and celecoxib vs. ibuprofen (1.05; 95% CI 0.37, 2.98) attenuated toward the null, while celecoxib still showed a different, but not significantly so, risk profile vs. mefenamic acid (adjusted HR 1.83; 95% CI 0.70, 4.81). The HR comparing celecoxib vs. naproxen did not change materially (0.74; 95% CI 0.21, 2.68) (Table S11). In general, the results based on baseline PS matching followed by inverse probability weighting did not differ substantially compared with those from the main analysis of using inverse probability weighting for both baseline and time‐varying covariates, although the risk estimates tended to be more imprecise. The analysis with hd‐PS matching followed by inverse probability weighting yielded similar results, although risk estimates for some comparisons could not be calculated because we were unable to match enough patients initiating nonselective NSAIDs to each patient initiating COX‐2‐selective NSAIDs in the matching step (Table S12).
Discussion
In the present longitudinal, population‐based cohort study, we examined the cardiovascular safety of NSAIDs in patients with hypertension by using inverse probability weighting to account for baseline and time‐varying covariates alike. There was no strong evidence to suggest a higher cardiovascular risk with COX‐2‐selective NSAIDs (celecoxib and etoricoxib) or celecoxib alone compared with diclofenac, ibuprofen and naproxen. However, we observed an elevated risk when comparing celecoxib with mefenamic acid.
Biological plausibility of NSAID‐associated cardiovascular thrombotic effects
The mechanisms of NSAID‐associated cardiovascular thrombotic effects have not been comprehensively elucidated. It has been hypothesized that COX‐2‐selective NSAIDs inhibit the synthesis of http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=1915 but not http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=4482, which leads to an imbalance between these two eicosanoids and an increase in platelet aggregation, vasoconstriction and cardiovascular‐related adverse events 1, 2. The extent of COX‐2 inhibition can be quantified based on the ratio of the concentrations required to produce a 50% inhibition in the activity of the isozymes (i.e. COX‐1/COX‐2 IC50 value), with a higher value indicating greater COX‐2 selectivity. COX‐2 selectivity for rofecoxib (255) is higher than for etoricoxib (162), celecoxib (32), diclofenac (24), ibuprofen (0.6) and naproxen (0.5) 40, 41.
Comparison with other studies
It remains to be determined whether COX‐2‐selective NSAIDs have a different cardiovascular safety profile than nonselective NSAIDs. COX‐2‐selective NSAIDs as a class, including rofecoxib, lumiracoxib, valdecoxib, celecoxib and etoricoxib, have been linked to a higher risk of MI than nonselective NSAIDs in a meta‐analysis of RCTs that included primarily arthritis patients 12. However, two RCTs of arthritis patients found no significant differences in thrombotic cardiovascular events across individual NSAIDs 13, 14. In the 18‐month Multinational Etoricoxib and Diclofenac Arthritis Long‐term (MEDAL) programme 13, the HR for thrombotic cardiovascular events was 0.95 (95% CI 0.81, 1.11) for etoricoxib (60–90 mg daily) vs. diclofenac (150 mg daily). In the 30‐month Prospective Randomized Evaluation of Celecoxib Integrated Safety versus Ibuprofen or Naproxen (PRECISION) trial 14, the HR for thrombotic cardiovascular events was 0.82 (95% CI 0.69, 0.97) for celecoxib (around 200 mg daily) vs. ibuprofen (around 2000 mg daily) and 0.95 (95% CI 0.80, 1.13) for celecoxib (around 200 mg daily) vs. naproxen (around 850 mg daily). Similarly, cohort studies conducted in elderly patients or in patients hospitalized for serious CHD detected no difference in the risk of MI, stroke or recurrent serious CHD comparing celecoxib with diclofenac 15, ibuprofen 15, 16 or naproxen 17, 18.
Our findings of no significant difference in cardiovascular risk between COX‐2‐selective NSAIDs (celecoxib and etoricoxib) or celecoxib alone and several individual nonselective NSAIDs (ibuprofen, diclofenac and naproxen) in patients with hypertension are in line with previous non‐inferior RCTs 13, 14 and active comparison observational studies 15, 16, 17, 18. However, we observed that celecoxib had an apparent higher cardiovascular risk compared with mefenamic acid. There have been a number of studies evaluating the cardiovascular risk associated with mefenamic acid compared with no NSAID treatment, but the results have been mixed 42, 43, 44. The mean daily dose of mefenamic acid used in our study was lower than the normal daily dose (1248 mg vs. 1500 mg). Mefenamic acid is commonly used for dysmenorrhoea and dental pain, rather than arthritis in clinical settings, and patients may tend to take the drug irregularly. Mefenamic acid might thus be less likely to interfere with the COX system and present a lower cardiovascular risk than celecoxib. However, this was the only statistically significant result out of the many subgroup analyses. Therefore, the possibility that this was a chance finding could not be ruled out. Further studies are required to examine the underlying mechanisms and replicate our findings.
The dosage and treatment duration of NSAIDs might play a role in cardiovascular risk 6, 19, 45. Patients included in the MEDAL programme and in the PRECISION trial received moderate‐to‐high daily doses of NSAIDs with a longer treatment duration 13, 14. The mean daily dose was lower and the treatment duration shorter in our study compared with the MEDAL program and the PRECISION trial, but the use pattern reflected actual use in real‐world clinical settings. These differences may partially explain our neutral findings across these individual NSAIDs. The pattern of low‐dose, short‐term use of NSAIDs was also observed in three German, UK and French adult cohorts 46, 47, 48.
Strengths and weaknesses of the study
As hypertension is a prevalent condition and a major risk factor for cardiovascular diseases 20, our results provide important information about the comparative safety of alternative NSAID use, and suggest that most commonly used NSAIDs, including naproxen, have similar cardiovascular safety profiles in this vulnerable population. Our results also suggest that the relationship between NSAIDs and cardiovascular risk is complex, and that it might be more appropriate to describe NSAID‐associated cardiovascular safety by individual drug rather than as a whole drug class.
Consistent with previous observational studies 15, 16, 17, 18, our study found that patients receiving COX‐2‐selective NSAIDs were sicker and tended to have a longer treatment duration than those receiving nonselective NSAIDs. Unlike previous studies, which usually considered baseline confounding only, our study simultaneously accounted for baseline and time‐varying confounding and potential selection bias due to informative censoring using inverse probability weighting. The method accounts for time‐varying confounders more appropriately compared with conventional regression approaches 32, 33, 34, 35. The results were also qualitatively similar when we applied conventional PS or hd‐PS matching to adjust for baseline covariates, and inverse probability weighting to account further for time‐varying covariates.
There were several limitations to our study. First, pharmacy claims data provide accurate information about the prescriptions that are filled but do not necessarily reflect whether and when patients take the drugs. However, most NSAIDs (>90% of approved products) are available via a prescription and are reimbursed by the National Health Insurance programme in Taiwan. This allowed us to ascertain exposure status more comprehensively than other database studies, which often have incomplete capture of NSAIDs due to considerable over‐the‐counter use. Secondly, the number of outcome events was limited for individual NSAIDs in our study, and several analyses might have been underpowered. For example, there were only seven outcome events for naproxen, so the adjusted HRs comparing COX‐2‐selective NSAIDs vs. naproxen (0.67; 95% CI 0.24, 1.91) or celecoxib vs. naproxen (0.75; 95% CI 0.23, 2.44) were unstable, with wide confidence intervals. However, reassuringly, our findings were in line with previous active comparison RCTs and observational studies 13, 14, 15, 16, 17, 18. Furthermore, we only had five outcome events for etoricoxib and were unable to compare this agent with individual nonselective NSAIDs. Similarly, we were also not able to examine individual components of the composite cardiovascular outcome. Further research, using the whole Taiwanese population, using an active comparison study design, is likely to provide more precise findings.
Thirdly, our data source did not include information on deaths, which prevented us from evaluating the association between use of NSAIDs and overall or cardiovascular death. Finally, the prescribing habits or preferences of health professionals, different NSAID indications, different safety warnings, and contraindications might have led to an imbalance in covariates between patients receiving different NSAIDs. In the current study, we observed that patients who initiated COX‐2‐selective NSAIDs were older and had more comorbidities. We applied different analytical approaches, including inverse probability weighting, to account for potential baseline and time‐varying confounders. The balance in baseline covariates between treatment groups improved after inverse probability weighting. However, patients receiving COX‐2‐selective NSAIDs still tended to be slightly sicker than those receiving nonselective NSAIDs, which might have biased the results further away from the null. In addition, although we considered a large number of covariates in the analysis, we could not rule out the possibility of unmeasured confounding, such as education level, smoking status or body mass index.
Conclusion
In conclusion, the present population‐based cohort study found that, under low‐to‐moderate daily dose and a short‐term treatment period, there was no apparent difference in cardiovascular risk comparing celecoxib with diclofenac, ibuprofen and naproxen in patients with hypertension. A potential increased cardiovascular risk with celecoxib when compared with mefenamic acid warrants further investigation.
Competing Interests
There are no competing interests to declare.
This study was supported by the Ministry of Science and Technology, Taiwan grant MOST 103–2314‐B‐002 ‐037 and MOST 104–2314‐B‐002 ‐036 ‐MY2. The funding source had no role in the study design; the collection, analysis and interpretation of the data; the writing of the report; or in the decision to submit the article for publication.
Contributors
Y.‐H.D., C.‐H.C. and S.T. were responsible for the study concept and design; L.‐C.W. carried out data analysis; all authors contributed to the interpretation of the data; Y.‐H.D. drafted the manuscript; C.‐H.C., J.‐S.H. and S.T. carried out critical revision of the manuscript for important intellectual content; C.‐H.C. was responsible for acquisition of the funding and data; C.‐H.C. was the principal investigator. All of the authors have read, contributed to and approved the final manuscript, and agreed to transfer the copyright ownership in the event of acceptance. Y.‐H.D. and C.‐H.C. are the grantors.
Supporting information
Table S1 Anatomical Therapeutic Chemical (ATC) classification system codes used to identify initiators of cyclooxygenase 2 (COX‐2)‐selective and nonselective NSAIDs
Table S2 International Classification of Diseases, 9th Revision, Clinical Modification (ICD‐9‐CM) diagnosis or procedure codes, or Taiwan health insurance service claims codes used to identify outcomes of interest
Table S3 International Classification of Diseases, 9th Revision, Clinical Modification (ICD‐9‐CM) diagnosis codes used to identify baseline and time‐varying comorbidities
Table S4 Anatomical Therapeutic Chemical (ATC) classification system codes used to identify baseline and time‐varying medication use
Table S5 Baseline characteristics by nonsteroidal anti‐inflammatory drug (NSAID) group after weighting: cyclooxygenase 2 (COX‐2)‐selective NSAIDs vs. diclofenac
Table S6 Baseline characteristics by nonsteroidal anti‐inflammatory drug (NSAID) group after weighting: cyclooxygenase 2 (COX‐2)‐selective NSAIDs vs. ibuprofen
Table S7 Baseline characteristics by nonsteroidal anti‐inflammatory drug (NSAID) group after weighting: cyclooxygenase 2 (COX‐2)‐selective NSAIDs vs. naproxen
Table S8 Baseline characteristics by nonsteroidal anti‐inflammatory drug (NSAID) group after weighting: cyclooxygenase 2 (COX‐2)‐selective NSAIDs vs. mefenamic acid
Table S9 Distribution of the stabilized inverse probability of treatment weight (IPTW) and the stabilized inverse probability of censoring weight (IPCW) estimated from baseline covariates when comparing cyclooxygenase 2 (COX‐2)‐selective nonsteroidal anti‐inflammatory drugs (NSAIDs) with nonselective NSAIDs
Table S10 Hazard ratios of major cardiovascular events in patients who initiated cyclooxygenase 2 (COX‐2)‐selective nonsteroidal anti‐inflammatory drug (NSAID) treatment vs. patients who initiated nonselective NSAID treatment, using the ‘first exposure carried forward’ or ‘intention‐to‐treat’ approach
Table S11 Hazard ratios of major cardiovascular events in patients who were on cyclooxygenase 2 (COX‐2)‐selective nonsteroidal anti‐inflammatory drug (NSAID) treatment vs. patients who were on nonselective NSAID treatment during the follow‐up, using the propensity score matching approach followed by inverse probability weighting
Table S12 Hazard ratios of major cardiovascular events in patients who were on cyclooxygenase 2 (COX‐2)‐selective nonsteroidal anti‐inflammatory drug (NSAID) treatment vs. patients who were on nonselective NSAID treatment during the follow‐up, using the high‐dimensional propensity score matching approach, followed by inverse probability weighting
Appendix S1 Estimation of the stabilized inverse probability of treatment weight (IPTW) and the stabilized inverse probability of censoring weight (IPCW) based on baseline covariates only
Appendix S2 Estimation of the stabilized inverse probability of treatment weights (IPTW) and the stabilized inverse probability of censoring weight (IPCW) based on previous treatment histories, baseline covariates and time‐varying covariates simultaneously
Dong, Y.‐H. , Chang, C.‐H. , Wu, L.‐C. , Hwang, J.‐S. , and Toh, S. (2018) Comparative cardiovascular safety of nonsteroidal anti‐inflammatory drugs in patients with hypertension: a population‐based cohort study. Br J Clin Pharmacol, 84: 1045–1056. doi: 10.1111/bcp.13537.
References
- 1. Patrício JP, Barbosa JP, Ramos RM, Antunes NF, de Melo PC. Relative cardiovascular and gastrointestinal safety of non‐selective non‐steroidal anti‐inflammatory drugs versus cyclo‐oxygenase‐2 inhibitors: implications for clinical practice. Clin Drug Investig 2013; 33: 167–183. [DOI] [PubMed] [Google Scholar]
- 2. Bruno A, Tacconelli S, Patrignani P. Variability in the response to non‐steroidal anti‐inflammatory drugs: mechanisms and perspectives. Basic Clin Pharmacol Toxicol 2014; 114: 56–63. [DOI] [PubMed] [Google Scholar]
- 3. Bombardier C, Laine L, Reicin A, Shapiro D, Burgos‐Vargas R, Davis B, et al Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. N Engl J Med 2000; 343: 1520–1528. [DOI] [PubMed] [Google Scholar]
- 4. Karha J, Topol EJ. The sad story of Vioxx, and what we should learn from it. Cleve Clin J Med 2004; 71: 933–934. [DOI] [PubMed] [Google Scholar]
- 5. Kearney PM, Baigent C, Godwin J, Halls H, Emberson JR, Patrono C. Do selective cyclo‐oxygenase‐2 inhibitors and traditional non‐steroidal anti‐inflammatory drugs increase the risk of atherothrombosis? Meta‐analysis of randomised trials. BMJ 2006; 332: 1302–1308. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. McGettigan P, Henry D. Cardiovascular risk with non‐steroidal anti‐inflammatory drugs: systematic review of population‐based controlled observational studies. PLoS Med 2011; 8: e1001098. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Trelle S, Reichenbach S, Wandel S, Hildebrand P, Tschannen B, Villiger PM, et al Cardiovascular safety of non‐steroidal anti‐inflammatory drugs: network meta‐analysis. BMJ 2011; 342: c7086. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Varas‐Lorenzo C, Riera‐Guardia N, Calingaert B, Castellsague J, Salvo F, Nicotra F, et al Myocardial infarction and individual nonsteroidal anti‐inflammatory drugs meta‐analysis of observational studies. Pharmacoepidemiol Drug Saf 2013; 22: 559–570. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Bhala N, Emberson J, Merhi A, Abramson S, Arber N, Baron JA, et al Vascular and upper gastrointestinal effects of non‐steroidal anti‐inflammatory drugs: meta‐analyses of individual participant data from randomised trials. Lancet 2013; 382: 769–779. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Solomon DH, Glynn RJ, Rothman KJ, Schneeweiss S, Setoguchi H, Schneeweiss S, et al Subgroup analyses to determine cardiovascular risk associated with nonsteroidal antiinflammatory drugs and coxibs in specific patient groups. Arthritis Rheum 2008; 59: 1097–1104. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Fosbøl EL, Folke F, Jacobsen S, Rasmussen JN, Sørensen R, Schramm TK, et al Cause‐specific cardiovascular risk associated with nonsteroidal antiinflammatory drugs among healthy individuals. Circ Cardiovasc Qual Outcomes 2010; 3: 395–405. [DOI] [PubMed] [Google Scholar]
- 12. Chen LC, Ashcroft DM. Risk of myocardial infarction associated with selective COX‐2 inhibitors: meta‐analysis of randomised controlled trials. Pharmacoepidemiol Drug Saf 2007; 16: 762–772. [DOI] [PubMed] [Google Scholar]
- 13. Cannon CP, Curtis SP, FitzGerald GA, Krum H, Kaur A, Bolognese JA, et al 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. [DOI] [PubMed] [Google Scholar]
- 14. Nissen SE, Yeomans ND, Solomon DH, Lüscher TF, Libby P, Husni ME, et al Cardiovascular safety of celecoxib, naproxen, or ibuprofen for arthritis. N Engl J Med 2016; 375: 2519–2529. [DOI] [PubMed] [Google Scholar]
- 15. Rahme E, Watson DJ, Kong SX, Toubouti Y, LeLorier J. Association between nonnaproxen NSAIDs, COX‐2 inhibitors and hospitalization for acute myocardial infarction among the elderly: a retrospective cohort study. Pharmacoepidemiol Drug Saf 2007; 16: 493–503. [DOI] [PubMed] [Google Scholar]
- 16. Solomon DH, Avorn J, Stürmer 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. [DOI] [PubMed] [Google Scholar]
- 17. Abraham NS, El‐Serag HB, Hartman C, Richardson P, Deswal A. Cyclooxygenase‐2 selectivity of non‐steroidal anti‐inflammatory drugs and the risk of myocardial infarction and cerebrovascular accident. Aliment Pharmacol Ther 2007; 25: 913–924. [DOI] [PubMed] [Google Scholar]
- 18. Ray WA, Varas‐Lorenzo C, Chung CP, Castellsague J, Murray KT, Stein CM, et al Cardiovascular risks of nonsteroidal antiinflammatory drugs in patients after hospitalization for serious coronary heart disease. Circ Cardiovasc Qual Outcomes 2009; 2: 155–163. [DOI] [PubMed] [Google Scholar]
- 19. Bally M, Dendukuri N, Rich B, Nadeau L, Helin‐Salmivaara A, Garbe E, et al Risk of acute myocardial infarction with NSAIDs in real world use: Bayesian meta‐analysis of individual patient data. BMJ 2017; 357: j1909. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. D'Agostino RB Sr, Vasan RS, Pencina MJ, Wolf PA, Cobain M, Massaro JM, et al General cardiovascular risk profile for use in primary care: the Framingham Heart Study. Circulation 2008; 117: 743–753. [DOI] [PubMed] [Google Scholar]
- 21. Spalding WM, Reeves MJ, Whelton A. Thromboembolic cardiovascular risk among arthritis patients using cyclooxygenase‐2‐selective inhibitor or nonselective cyclooxygenase inhibitor nonsteroidal anti‐inflammatory drugs. Am J Ther 2007; 14: 3–12. [DOI] [PubMed] [Google Scholar]
- 22. Solomon DH, Glynn RJ, Rothman KJ, Schneeweiss S, Setoguchi S, Mogun H, et al Subgroup analyses to determine cardiovascular risk associated with nonsteroidal antiinflammatory drugs and coxibs in specific patient groups. Arthritis Rheum 2008; 59: 1097–1104. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Hsu CC, Wang H, Hsu YH, Chuang SY, Huang YW, Chang YK, et al Use of nonsteroidal anti‐inflammatory drugs and risk of chronic kidney disease in subjects with hypertension: nationwide longitudinal cohort study. Hypertension 2015; 66: 524–533. [DOI] [PubMed] [Google Scholar]
- 24. Hsieh CY, Chen CH, Li CY, Lai ML. Validating the diagnosis of acute ischemic stroke in a National Health Insurance claims database. J Formos Med Assoc 2015; 114: 254–259. [DOI] [PubMed] [Google Scholar]
- 25. Floyd JS, Blondon M, Moore KP, Boyko EJ, Smith NL. Validation of methods for assessing cardiovascular disease using electronic health data in a cohort of Veterans with diabetes. Pharmacoepidemiol Drug Saf 2016; 25: 467–471. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Birman‐Deych E, Waterman AD, Yan Y, Nilasena DS, Radford MJ, Gage BF. Accuracy of ICD‐9‐CM codes for identifying cardiovascular and stroke risk factors. Med Care 2005; 43: 480–485. [DOI] [PubMed] [Google Scholar]
- 27. Benesch C, Witter DM Jr, Wilder AL, Duncan PW, Samsa GP, Matchar DB. Inaccuracy of the International Classification of Diseases (ICD‐9‐CM) in identifying the diagnosis of ischemic cerebrovascular disease. Neurology 1997; 49: 660–664. [DOI] [PubMed] [Google Scholar]
- 28. Sundbøll J, Adelborg K, Munch T, Frøslev T, Sørensen HT, Bøtker HE, et al Positive predictive value of cardiovascular diagnoses in the Danish National Patient Registry: a validation study. BMJ Open 2016; 6: e012832. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Lee DS, Stitt A, Wang X, Yu JS, Gurevich Y, Kingsbury KJ, et al Administrative hospitalization database validation of cardiac procedure codes. Med Care 2013; 51: e22–e26. [DOI] [PubMed] [Google Scholar]
- 30. Cole SR, Hernán MA. Adjusted survival curves with inverse probability weights. Comput Methods Programs Biomed 2004; 75: 45–49. [DOI] [PubMed] [Google Scholar]
- 31. Hernán MA, Robins JM. Causal Inference. Boca Raton, FL: Chapman & Hall/CRC, 2018. [In Press]. [Google Scholar]
- 32. Toh S, Hernán MA. Causal inference from longitudinal studies with baseline randomization. Int J Biostat 2008; 4: 22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Hernán MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV‐positive men. Epidemiology 2000; 11: 561–570. [DOI] [PubMed] [Google Scholar]
- 34. Hernán MA, Alonso A, Logan R, Grodstein F, Michels KB, Willett WC, et al Observational studies analyzed like randomized experiments: an application to postmenopausal hormone therapy and coronary heart disease. Epidemiology 2008; 19: 766–779. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Yang S, Eaton CB, Lu J, Lapane KL. Application of marginal structural models in pharmacoepidemiologic studies: a systematic review. Pharmacoepidemiol Drug Saf 2014; 23: 560–571. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Rassen JA, Doherty M, Huang W, Schneeweiss S. Pharmacoepidemiology toolbox. Available at http://www.drugepi.org/dope-downloads/#Pharmacoepidemiology Toolbox (last accessed 28 February 2015).
- 37. Rassen JA, Glynn RJ, Brookhart MA, Schneeweiss S. Covariate selection in high‐dimensional propensity score analyses of treatment effects in small samples. Am J Epidemiol 2011; 173: 1404–1413. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Harding SD, Sharman JL, Faccenda E, Southan C, Pawson AJ, Ireland S, et al The IUPHAR/BPS Guide to PHARMACOLOGY in 2018: updates and expansion to encompass the new guide to IMMUNOPHARMACOLOGY. Nucl Acid Res 2018; 46: D1091–D1106. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Alexander SPH, Fabbro D, Kelly E, Marrion NV, Peters JA, Faccenda E, et al The Concise Guide to PHARMACOLOGY 2017/18: Enzymes. Br J Pharmacol 2017; 174: S272–S359. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Warner TD, Mitchell JA. Cyclooxygenases: new forms, new inhibitors, and lessons from the clinic. FASEB J 2004; 18: 790–804. [DOI] [PubMed] [Google Scholar]
- 41. Patrignani P, Tacconelli S, Capone ML. Risk management profile of etoricoxib: an example of personalized medicine. Ther Clin Risk Manag 2008; 4: 983–997. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. van Staa TP, Rietbrock S, Setakis E, Leufkens HG. Does the varied use of NSAIDs explain the differences in the risk of myocardial infarction? J Intern Med 2008; 264: 481–492. [DOI] [PubMed] [Google Scholar]
- 43. Chang CH, Shau WY, Kuo CW, Chen ST, Lai MS. Increased risk of stroke associated with nonsteroidal anti‐inflammatory drugs: a nationwide case‐crossover study. Stroke 2010; 41: 1884–1890. [DOI] [PubMed] [Google Scholar]
- 44. Chuang SY, Yu Y, Sheu WH, Tsai YT, Liu X, Hsiung CA, et al Association of short‐term use of nonsteroidal anti‐inflammatory drugs with stroke in patients with hypertension. Stroke 2015; 46: 996–1003. [DOI] [PubMed] [Google Scholar]
- 45. García Rodríguez LA, Tacconelli S, Patrignani P. Role of dose potency in the prediction of risk of myocardial infarction associated with nonsteroidal anti‐inflammatory drugs in the general population. J Am Coll Cardiol 2008; 52: 1628–1636. [DOI] [PubMed] [Google Scholar]
- 46. Sift R, van Staa TP, Abenhaim L, Ebner D. A study of the longitudinal utilization and switching‐patterns of non‐steroidal anti‐inflammatory drugs using a pharmacy based approach. Pharmacoepidemiol Drug Saf 1997; 6: 263–268. [DOI] [PubMed] [Google Scholar]
- 47. Langman M, Kahler KH, Kong SX, Zhang Q, Finch E, Bentkover JD, et al Drug switching patterns among patients taking non‐steroidal anti‐inflammatory drugs: a retrospective cohort study of a general practitioners database in the United Kingdom. Pharmacoepidemiol Drug Saf 2001; 10: 517–524. [DOI] [PubMed] [Google Scholar]
- 48. Duong M, Salvo F, Pariente A, Abouelfath A, Lassalle R, Droz C, et al Usage patterns of 'over‐the‐counter' vs. prescription‐strength nonsteroidal anti‐inflammatory drugs in France. Br J Clin Pharmacol 2014; 77: 887–895. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1 Anatomical Therapeutic Chemical (ATC) classification system codes used to identify initiators of cyclooxygenase 2 (COX‐2)‐selective and nonselective NSAIDs
Table S2 International Classification of Diseases, 9th Revision, Clinical Modification (ICD‐9‐CM) diagnosis or procedure codes, or Taiwan health insurance service claims codes used to identify outcomes of interest
Table S3 International Classification of Diseases, 9th Revision, Clinical Modification (ICD‐9‐CM) diagnosis codes used to identify baseline and time‐varying comorbidities
Table S4 Anatomical Therapeutic Chemical (ATC) classification system codes used to identify baseline and time‐varying medication use
Table S5 Baseline characteristics by nonsteroidal anti‐inflammatory drug (NSAID) group after weighting: cyclooxygenase 2 (COX‐2)‐selective NSAIDs vs. diclofenac
Table S6 Baseline characteristics by nonsteroidal anti‐inflammatory drug (NSAID) group after weighting: cyclooxygenase 2 (COX‐2)‐selective NSAIDs vs. ibuprofen
Table S7 Baseline characteristics by nonsteroidal anti‐inflammatory drug (NSAID) group after weighting: cyclooxygenase 2 (COX‐2)‐selective NSAIDs vs. naproxen
Table S8 Baseline characteristics by nonsteroidal anti‐inflammatory drug (NSAID) group after weighting: cyclooxygenase 2 (COX‐2)‐selective NSAIDs vs. mefenamic acid
Table S9 Distribution of the stabilized inverse probability of treatment weight (IPTW) and the stabilized inverse probability of censoring weight (IPCW) estimated from baseline covariates when comparing cyclooxygenase 2 (COX‐2)‐selective nonsteroidal anti‐inflammatory drugs (NSAIDs) with nonselective NSAIDs
Table S10 Hazard ratios of major cardiovascular events in patients who initiated cyclooxygenase 2 (COX‐2)‐selective nonsteroidal anti‐inflammatory drug (NSAID) treatment vs. patients who initiated nonselective NSAID treatment, using the ‘first exposure carried forward’ or ‘intention‐to‐treat’ approach
Table S11 Hazard ratios of major cardiovascular events in patients who were on cyclooxygenase 2 (COX‐2)‐selective nonsteroidal anti‐inflammatory drug (NSAID) treatment vs. patients who were on nonselective NSAID treatment during the follow‐up, using the propensity score matching approach followed by inverse probability weighting
Table S12 Hazard ratios of major cardiovascular events in patients who were on cyclooxygenase 2 (COX‐2)‐selective nonsteroidal anti‐inflammatory drug (NSAID) treatment vs. patients who were on nonselective NSAID treatment during the follow‐up, using the high‐dimensional propensity score matching approach, followed by inverse probability weighting
Appendix S1 Estimation of the stabilized inverse probability of treatment weight (IPTW) and the stabilized inverse probability of censoring weight (IPCW) based on baseline covariates only
Appendix S2 Estimation of the stabilized inverse probability of treatment weights (IPTW) and the stabilized inverse probability of censoring weight (IPCW) based on previous treatment histories, baseline covariates and time‐varying covariates simultaneously