Nonsteroidal anti-inflammatory drugs (NSAIDs) are one

Nonaspirin Nonsteroidal Anti-Inflammatory Drugs and Risk of Hemorrhagic Stroke A Systematic Review and Meta-Analysis of Observational Studies Patompong Ungprasert, MD; Eric L. Matteson, MD; Charat Thongprayoon, MD Background and Purpose The association between and use of nonsteroidal anti-inflammatory drugs (NSAIDs) is not well established. We conducted a systematic review and meta-analysis of observation studies to further characterize this possible association. Methods Case control and cohort studies that reported odds ratio, relative risk, hazard ratio, or standardized incidence ratio comparing risk of among NSAIDs users versus nonusers were systematically searched. Point estimates from each study were extracted. Pooled risk ratios (RR) and 95% confidence intervals (CI) for all NSAIDs and individual NSAIDs were calculated using random-effect, generic inverse variance method. Results Ten studies were identified and included in our data analysis. As a single group, NSAIDs use was associated with a small but insignificant risk of with the pooled RR of 1.09 (95% CI, 0.98 1.22). Individual NSAIDs analysis revealed a significantly increased risk among diclofenac and meloxicam users (RR 1.27; 95% CI, 1.02 1.59 and RR 1.27; 95% CI, 1.08 1.50, respectively). The risk estimate for rofecoxib users was higher, but statistically nonsignificant (RR 1.35; 95% CI, 0.88 2.06). Conclusions Overall, the use of NSAIDs is not associated with an increased risk of, although this risk was modestly significantly elevated in diclofenac and meloxicam users. (Stroke. 2016;47:356-364. DOI: 10.1161/STROKEAHA.115.011678.) Key Words: cyclooxygenase meta-analysis nonsteroidal anti-inflammatory drugs stroke Nonsteroidal anti-inflammatory drugs (NSAIDs) are one of the most commonly used medications in the United States and worldwide. 1 Inhibition of cyclooxygenase (COX) enzyme, which is divided into 2 isoforms (COX-1 and COX- 2), is the fundamental mechanism of action of NSAIDs. COX-1 is a house-keeping enzyme that is constitutively expressed under normal physiological condition while the expression of COX-2 is generally inducible under inflammatory state. 2 Traditional NSAIDs have been used in clinical practice as analgesic for mild to moderate pain and as antiinflammatory agent for decades. However, the utility of these nonselective NSAIDs is limited by their adverse effects associated with inhibition of COX-1 enzyme, particularly gastrointestinal ulcer and bleeding. 3,4 Specific COX-2 inhibitors have higher specificity for the COX-2 enzyme and were marketed as a safer alternative to conventional NSAIDs after several randomized controlled trials demonstrated a superior gastrointestinal safety profile compared with traditional NSAIDs. 5,6 Nonetheless, rofecoxib, a specific COX-2 inhibitor, was later withdrawn from the market as a randomized-controlled trial published in 2004 demonstrated an increased incidence of acute coronary syndrome among its users. 7 The increased risk was confirmed in subsequent clinical trials and observational studies. 8,9 Since then, attention has turned to the cardiovascular adverse effect of other NSAIDs. In fact, several observational studies have suggested an increased incidence of cardiovascular event in some traditional NSAIDs, such as diclofenac. 10 Use of NSAIDs may also be associated with an increased risk of stroke, one of the major subtypes of cardiovascular disease. In fact, elevated risk of stroke was observed in the original study that led to the withdrawal of rofecoxib, even though not reaching statistical significance. 7 A recent meta-analysis of 6 observational studies reported an increased risk of ischemic stroke among rofecoxib and diclofenac users. 11 However, the data on the other major stroke subtype,, is limited, and it is still unclear whether NSAID users are at increased risk for. This study aimed to Received September 30, 2015; final revision received November 6, 2015; accepted November 10, 2015. From the Division of Rheumatology, Department of Internal Medicine, Mayo Clinic, Rochester, MN (P.U., E.L.M.); Department of Medicine, Faculty of Medicine Siriraj Hospital, Mahidol University, Bangkok, Thailand (P.U.); Division of Epidemiology, Department of Health Sciences Research, Mayo Clinic, Rochester, MN (E.L.M.); and Division of Nephrology and Hypertension, Department of Internal Medicine, Mayo Clinic, Rochester, MN (C.T.). The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/strokeaha. 115.011678/-/DC1. Correspondence to Patompong Ungprasert, MD, Division of Rheumatology, Mayo Clinic, 200 1st St SW, Rochester, MN 55905. E-mail Ungprasert. Patompong@mayo.edu 2015 American Heart Association, Inc. Stroke is available at http://stroke.ahajournals.org DOI: 10.1161/STROKEAHA.115.011678 356

Ungprasert et al NSAIDs and Hemorrhagic Stroke 357 comprehensively review all available epidemiological studies to characterize the relationship between and use of nonaspirin NSAIDs. Methods Search Strategy Two investigators (Drs Ungprasert and Thongprayoon) independently searched published English language studies indexed in MEDLINE and EMBASE databases from inception to August 2015. The search terms were compiled from terms for stroke and NSAIDs, including names of individual drugs in conjunction with the terms for nonrandomized studies developed by Furlan et al. 12 The detailed search strategy is provided in Supplemental material I in the online-only Data Supplement. A manual search of references of included studies and selected review articles was also performed. Inclusion Criteria We used the following criteria to determine study eligibility: (1) the study had to be case control or cohort study; (2) relative risk (RR), odds ratio, hazard ratio, or standardized incidence ratio with 95% confidence intervals (CI) of incident for nonaspirin traditional NSAIDs or specific COX-2 inhibitors was provided; and (3) NSAIDs nonusers were used as controls for cohort studies, whereas participants without were used as controls for case control studies. Study eligibility was independently determined by Drs Ungprasert and Thongprayoon. Quality assessment of the included studies was also independently performed by both investigators using the Newcastle-Ottawa quality assessment scale which evaluated each study in 3 domains, including (1) selection of the participants; (2) comparability between the groups; and (3) ascertainment of the exposure of interest for case control studies and the outcome of interest for cohort studies. 13 Differences in assessment was resolved by consensus among the investigators. Data Extraction A standardized data collection form was used to extract the following information: title of the study, first author s last name, country where the study was conducted, study design, year of recruitment and follow up, year of publication, study size, study population, names of studied NSAIDs, diagnosis and identification of hemorrhagic stroke, definition and verification of NSAIDs exposure, average duration of follow up (for cohort study), potential confounders that were adjusted for, and adjusted pooled effect estimates with 95% CI. Any data discrepancy was resolved by referring back to the original study. Statistical Analysis Data analysis was performed using Review Manager 5.3 software from the Cochrane Collaboration (London, United Kingdom). Adjusted point estimates were extracted from individual studies for both overall NSAIDs and individual NSAIDs. These point estimates were combined together to calculate the pooled estimates using the generic inverse variance method of DerSimonian and Laird, which assigned weight of each point estimate based on its standard error. 14 We performed 2 separate analyses, one for overall NSAIDs (as a single group) and the other one for individual NSAIDs. Pooled estimate for individual NSAIDs were calculated if there were at least 3 studies that provided a point estimate for those individual NSAIDs. We reported the pooled effect estimate of risk of using the combination of the data from case control and cohort studies to increase the precision and power of our estimates. Odds ratios of case control studies were used as an estimate of RR for this Figure 1. Study identification and literature review process.

358 Stroke February 2016 Table 1. Characteristics of Included Case Control Studies Thrift et al 17 Bak et al 18 Johnsen et al 19 Choi et al 20 Chang et al 21 Mangoni et al 22 Lapi et al 23 Country Australia Denmark (Funen County) Denmark (North Jutland County) Year of publication Cases Controls South Korea Taiwan Australia Italy 1999 2003 2003 2008 2010 2010 2015 All patients aged 18 y who were admitted at the participating hospitals with the primary diagnosis of hemorrhagic stroke between 1990 and 1995 were consecutively recruited. The diagnosis needed to be confirmed by imaging or postmortem examination. Patients with history of previous stroke were excluded (ie, only cases of first time were included). Sex, age, and residential area matched subject without history of stroke. Controls were recruited by home visit by the research nurses. NSAIDs Nonselective assessed in NSAIDs the study Definition of NSAIDs exposure Verification of NSAIDs exposure Use of any NSAIDs within 14 days before index date Structured interview by trained research nurses All patients 20 y who were first admitted with diagnosis of hemorrhagic stroke between January 1, 1994, and December 31, 1999. Cases were identified from Funen County discharge database. Medical records were reviewed by 3 neurologists for diagnostic accuracy. Funen county residents without history of stroke were randomly chosen as controls. Controls were identified from Funen Residence History Registry. Aceclofenac, azapropazone, diclofenac, etodolac, fenbufen, flurbiprofen, ibuprofen, indomethacin, ketoprofen, tiaprofenic acid, meloxicam, nabumetone, naproxen, piroxicam, sulindac, tenoxicam, tolfenamic acid, tolmetin, phenylbutazone, dexibuprofen, proquazone Supply of the most recent prescription lasted within 30 days before the index date Prescription information from Funen County Pharmacoepidemiologic Database All patients 18 y who were first admitted with diagnosis of between January 1, 1991, and December 31, 1999. Cases were identified from North Jutland County discharge database. Samples of medical records were reviewed by the investigators. Sex, age, and countymatched subjects without history of stroke who were randomly selected from the Danish Civil Registration system. Nonselective NSAIDs Supply of the most recent prescription lasted within 30 days before the index date Prescription information from North Jutland County Pharmacoepidemiologic Database All patients aged 30 y who were admitted at the 33 participating hospitals across south Korea with the primary diagnosis of hemorrhagic stroke between 2002 and 2004 were consecutively recruited. The diagnosis needed to be confirmed by imaging. Patients with history of previous stroke were excluded (ie, only cases of first time hemorrhagic stroke were included). All patients 20 y who were first admitted with diagnosis of between January 1, 2005, and December 31, 2006. Cases were identified from the Taiwan National Health Insurance database which covered 99% of the 23 million population in Taiwan. Medical records were reviewed by 3 neurologists for diagnostic accuracy. Patients with history of previous stroke were excluded. Sex and agematched subjects This study was a without history of stroke who were randomly selected from siblings, friends, or neighbors of cases. crossover study (ie, cases also serve as controls). For each patient, the author defined case period as 1 30 days before stroke and control period as 91 120 days before stroke. Aceclofenac, Diclofenac, diclofenac, ibuprofen, ibuprofen, ketoprofen, ketoprofen, meloxicam, meloxicam, nabumetone, naproxen, naproxen, piroxicam, piroxicam, acemetacin, mefenamic flufenamic acid, acid, celecoxib, mefenamic acid, sulindac, ketorolac, pranoprofen, indomethacin zaltoprofen, celecoxib, lornoxicam, nimelsulide, rofecoxib, talniflumate Use of any NSAIDs within 14 days before index date Structured interview by trained research nurses Supply of the most recent prescription lasted within 30 days of case and control period definition Prescription information from the All patients 20 y who were first diagnosed with between January 1, 2002, and June 30, 2006. Cases were identified from the Australia department of Veterans Affairs database which provided comprehensive medical care for 320 000 veterans and family members. Patients with history of previous stroke were excluded Sex, age, and statematched subjects without history of stroke who were randomly selected from the same database. Diclofenac, ibuprofen, indomethacin, ketoprofen, meloxicam, naproxen, piroxicam, tiaprofenic acid, mefenamic acid celecoxib, lumiracoxib, rofecoxib At least one prescription of NSAIDs in the last 2 y Prescription information from the All patients 40 y who were first admitted with diagnosis of hemorrhagic stroke between January 1, 2002, and December 31, 2011. Cases were identified from the Italian Health Search Longitudinal Patients Database which covered 1 million population across the country. Patients with history of previous stroke were excluded. Medical records were reviewed by the investigators. Sex, age, and cohort entry matched subjects without history of stroke who were randomly selected from the. Aceclofenac, diclofenac, ibuprofen, ketoprofen, meloxicam, nabumetone, naproxen, piroxicam, celecoxib, nimelsulide, rofecoxib, ketorolac, etoricoxib Supply of the most recent prescription lasted within 30 days of case and control period definition Prescription information from the Cases, n 331 4765 912 498 9456 1391 1546 (Continued)

Ungprasert et al NSAIDs and Hemorrhagic Stroke 359 Table 1. Continued Controls, n 331 40 000 9059 498 9456 27 747 9804 Percentage 39.6/39.6 48.8/51.6 46.9/46.9 42.0/42.0 42.0/42.0 35.9/35.9 61.1/63.4 of female in cases and controls Average age of cases and controls 63.4/63.4 NA 68.0/68.0 57.1/56.1 62.7/62.7 80.2/80.2 70.7/69.4 Confounders HTN, cholesterol Age, sex, and medications HTN, smoking, alcohol, adjusted for level, DM, history of DM, and medications CVD, BMI, exercise, alcohol, and smoking Newcastle- Ottawa scale Thrift et al 17 Bak et al 18 Johnsen et al 19 Choi et al 20 Chang et al 21 Mangoni et al 22 Lapi et al 23 HTN, salt intake, smoking family history of stroke Medications used between case and control period Age, sex, DM, HTN, dementia, History of CVD, renal disease, rheumatoid arthritis, and medications HTN, DM, history of CVD, renal disease, COPD, asthma, BMI, smoking, alcohol, and medications Selection 4 stars Selection 4 stars Selection 3 stars Selection 4 stars Selection 3 stars Selection 3 stars Selection 4 stars Comparability 2 stars Comparability 1 star Comparability 2 star Comparability 1 star Comparability 2 stars Comparability 1 star Comparability 1 star Exposure 3 stars Exposure 3 stars Exposure 3 stars Exposure 3 stars Exposure 3 stars Exposure 3 stars Exposure 3 stars BMI indicates body mass index; COPD, chronic pulmonary obstructive disease; CVD, cardiovascular disease; DM, diabetes mellitus; HTN, hypertension; and NSAIDs, nonsteroidal anti-inflammatory drugs. calculation. We expected a high between-study variance because of the different study design and population and, thus, used a randomeffect model rather than a fixed-effect model. The statistical heterogeneity of this study was assessed by Cochran s Q test. This test is complemented with the I 2 statistic, which quantifies the proportion of the total variation across studies that is because of heterogeneity rather than chance. A value of I 2 of 0% to 25% indicates insignificant heterogeneity, 26% to 50% indicates low heterogeneity, 51% to 75% indicates moderate heterogeneity, and 76% to 100% indicate high heterogeneity. 15 Publication bias was assessed by visualization of funnel plot and Egger s regression test. 16 Results Our search strategy yielded 9910 potentially relevant studies (3002 studies from Medline and 6908 studies from EMBASE). After exclusion of 2894 duplicates, 7016 studies underwent abstract and title review. 6896 articles were excluded at this stage because they were case reports, editorials, review articles, correspondences, animal studies, or clinical trials, leaving 120 articles for full-length article review. Sixty-five studies were excluded because they were interventional studies, whereas 27 studies were excluded because they were descriptive studies without a control group. Sixteen articles reported the risk of overall stroke or ischemic stroke without data on and were, thus, excluded. Twelve studies met our eligibility criteria. 17 28 However, 4 studies used the s (Chuang et al, 27 Chang et al, 21 Caughey et al, 28 and Mangoni et al 22 ). To avoid potential patient duplication, we included only the study with more comprehensive data (Chang et al 21 and Mangoni et al 22 ). Therefore, 10 studies (7 case control studies 17-23 and 3 cohort studies 24-26 ) with 1 489 120 participants were included in the final analysis. Figure 1 outlines our search methodology and literature review process. The detailed characteristics and Newcastle-Ottawa quality assessment scale of case control and cohort studies are described in Table 1 and Table 2, respectively. Baseline characteristics such as average age of participants, female to male ratio, and the studied NSAIDs were similar across studies. Most of the studies also defined the exposure to NSAIDs as use of any NSAIDs until the index date or within 30 days before index date. When NSAIDs were evaluated as a single group, we found a small but insignificantly increased risk of hemorrhagic stroke among the users with the pooled RR of 1.09 (95% CI, 0.98 1.22). I 2 of the overall analysis was 28%, which indicated a low statistical heterogeneity. Subgroup analysis according to study design revealed similar findings (RR 1.08; 95% CI, 0.99 1.18 for case control study and RR 1.13; 95% CI 0.72 1.76 for cohort study). However, it should be noted that the heterogeneity of the cohort study subgroup was high with an I 2 of 79%. Figure 2 demonstrates the forest plot of overall analysis. Meta-analysis of individual NSAIDs was performed for those with data available from at least 3 studies, which included diclofenac, ibuprofen, indomethacin, meloxicam, naproxen, piroxicam, celecoxib, and rofecoxib. A small increased risk was observe among the current users of ibuprofen, meloxicam, indomethacin, naproxen, and diclofenac, even though statistical significance was observed only for diclofenac (RR 1.27; 95% CI, 1.02 1.59) and meloxicam (RR 1.27; 95% CI, 1.08 1.50). The highest risk estimate was observed in rofecoxib users; however, the result did not reach statistical significance (RR 1.35; 95% CI, 0.88 2.06). The statistical heterogeneity across all individual NSAIDs was low with an I 2 of 47%. Figures 3 and 4 demonstrate the forest plots of individual NSAIDs. Sensitivity Analysis To confirm the robustness of our results, we performed several sensitivity analyses. First, we excluded the study by Mangoni et al 22 because it was the only study which defined NSAIDs exposure as use of any NSAIDs within the last 2 years (in contrast to 30 days in most studies). Exclusion of this study slightly increased the pooled RR to 1.12 but did not alter the statistical significance (RR 0.97 1.30). However, exclusion of this study increased the pooled RR of diclofenac and

360 Stroke February 2016 Table 2. Characteristics of Included Cohort Studies Haag et al 24 Roumie et al 25 Fosbol et al 26 Country The Netherlands United States Denmark Study design Prospective cohort Retrospective cohort Retrospective cohort Year 2008 2008 2014 Participants NSAIDs assessed in the study Definition of NSAIDs exposure Verification of NSAIDs exposure Controls Diagnosis of Follow up Number of subjects in the cohort Average age of subjects in the cohort, years Percentage of female in the cohort Average duration of follow up, years Person-years of NSAIDs Confounder adjusted for Newcastle-Ottawa scale All residences of Ommoord, a suburb city of Rotterdam, the Netherlands, who were 55 y were invited to participate in this study. 78% of eligible residences agreed to participate. The first cross-sectional study started in June 1990. Cross-sectional survey was conducted periodically by home interviews, participants visits to the research center, and bimonthly medical record review. Subjects with history of previous stroke were excluded. Diclofenac, ibuprofen, ketoprofen, meloxicam, nabumetone, naproxen, piroxicam, celecoxib, rofecoxib, valdecoxib, etoricoxib, indomethacin, flurbiprofen, apazone, sulindac Supply of the most recent prescription lasted until the index date Verified with the pharmacy database of the study which comprehensively covered all the prescriptions dispensed to the participants Subjects in the cohort who did not use any NSAIDs during the study period The continuous follow-up of all participants was aimed at identifying all events of interest, including. It was part of the routine followup procedure that all available data on the events of interest, such as hospital discharge summary, imaging, and outpatient visit note, were analyzed. One research physician and a neurologist who were blinded to drug exposure evaluated and verified the diagnosis. Until occurrence of study end point, death, emigration out of system, or September 30, 2004 All enrollees of Tennessee Medicaid program age >50 y who were enrolled between January 1, 1999, and December 31, 2004. Subjects with history of previous stroke were excluded. Celecoxib, rofecoxib, valdecoxib, ibuprofen, naproxen, indomethacin, diclofenac Supply of the most recent prescription lasted until the index date Prescription information from the NSAIDs use was analyzed as a timedependent variable(ie, duration of nonexposure to NSAIDs served as control) Diagnostic code from the same database. A systematic sample of 250 charts was reviewed. Until occurrence of study end point, death, disenrollment, or December 31, 2004 All Danish residences aged 10 on January 1, 1997, who had no contact with the Danish hospital 5 y before the index date. Participants were identified from Danish national health registry. Subjects with history of previous stroke were excluded. Study period lasted from January 1, 1997, to December 31, 2005. Celecoxib, rofecoxib, ibuprofen, naproxen, diclofenac Supply of the most recent prescription lasted within 30 days before the index date Prescription information from the NSAIDs use was analyzed as a timedependent variable(ie, duration of nonexposure to NSAIDs served as control) Diagnostic code from the same database Until occurrence of study end point, death, or December 31, 2004 7983 336 906 1 028 437 70.2 NA 39.0 61.3 64.8 42.0 9.2 3.0 5.0 41 718 177 066 2 000 027 Age, sex, HTN, BMI, cholesterol level, smoking, DM, history of CVD, and medications Age, sex, medications, hospitalization, enrollment category, smoking, history of CVD, and rheumatic diseases Age and sex Selection: 4 stars Selection: 4 stars Selection: 4 stars Comparability: 1 star Comparability: 1 star Comparability: 1 star Outcome: 3 stars Outcome: 2 stars Outcome: 3 stars BMI indicates body mass index; CVD, cardiovascular disease; DM, diabetes mellitus; HTN, hypertension; and NSAIDs, nonsteroidal anti-inflammatory drugs.

Ungprasert et al NSAIDs and Hemorrhagic Stroke 361 Figure 2. Forest plot of all nonsteroidal anti-inflammatory drugs (NSAIDs). meloxicam to 1.46 (95% CI, 1.30 4.64) and 1.48 (95% CI, 1.12 1.96), respectively. Second, we excluded the studies by Thrift et al 17 and Choi et al 20 because they were the only 2 studies that used structured interview to identify NSAIDs exposure. This approach was at a higher risk of recall bias compared with other studies that used pharmacy database. Nonetheless, the result of this sensitivity analysis was not significantly different from the complete analysis with pooled RR of 1.10 (95% CI, 0.97 1.25). Third, we excluded the study by Fosbol et al 26 because participants in their cohort were relatively younger compared with other studies (39.0 years compared with 56.1 80.2 years). Pooled RR of all NSAIDs slightly decreased to 1.06 (95% CI, 0.95 1.19) after the exclusion of this study, whereas individual NSAIDs analysis was not significantly affected by this sensitivity analysis. Evaluation for Publication Bias Comprehensive meta-analysis version 2.0 software (Biostat, Englewood, NJ) was used to perform the analysis for publication bias. The funnel plot used to evaluate publication bias for overall NSAIDs studies is provided in Supplemental material II in the online-only Data Supplement. Visualization of the funnel plot did not reveal any asymmetry and, therefore, did not provide evidence for publication bias. There was also no evidence of publication bias detected by Egger s regression test (P=0.51). Discussion To the best of our knowledge, this is the first systematic review of observational studies that summarizes the risk of hemorrhagic among nonaspirin NSAIDs users. We were also able to assess the risk for several individual NSAIDs. As a single group, NSAIDs use was not associated with a significantly elevated risk of. However, an increased risk was observed in several individual NSAIDs, and a significantly elevated risk was observed among diclofenac and meloxicam users. The highest risk estimate was found in rofecoxib users, even though the pooled RR did not achieve statistical significance which was probably because of the smaller number of rofecoxib users because this drug was withdrawn from worldwide market in 2004. Hemodynamic changes related to NSAID use are widely believed to be one of the main pathogenetic mechanisms for their adverse cardiovascular events. 29,30 Endothelium-derived prostaglandins, such as prostaglandin I 2, have a vasodilatory effect on the peripheral vasculature. Therefore, inhibition of COX enzymes by NSAIDs could result in vasoconstriction and increased peripheral arterial resistance. 31,32 Moreover, renal prostaglandins, particularly prostaglandins E 2 and prostaglandin I 2, have inhibitory effects on sodium reabsorption at the loop of Henle and on water reabsorption at the collecting tubules. These prostaglandins are locally derived from COX-2 enzyme in the renal cortex and juxtaglomerular cells. 31 33 Thus, inhibition of COX-2 enzyme could also lead to intravascular volume expansion. The combination of vasoconstriction and salt/fluid retention could eventually lead to hypertension, the prime risk factor for intracerebral hemorrhage. This might also explain the significant risk observed with diclofenac and meloxicam, the 2 traditional NSAIDs with the highest COX-2 selectivity. 34 The inhibition of platelet COX-1 activity by NSAIDs is known to cause platelet dysfunction and bleeding tendency. In fact, use of aspirin is a well-established risk factor for hemorrhagic stroke. However, in contrast to aspirin that irreversibly binds to platelet COX-1 enzyme, nonaspirin NSAIDs bind reversibly to the enzyme. The lack of significant association between nonaspirin NSAIDs and observed in this study might further highlight the pharmacological difference between these 2 medications. Data on the risk of from randomized controlled trials are limited because most of the studies

362 Stroke February 2016 Figure 3. Forest plot of individual nonsteroidal anti-inflammatory drugs (NSAIDs). A, Ibuprofen. B, Meloxicam. C, Piroxicam. D, Indomethacin. reported stroke as a composite event and do not separately report the incidence of ischemic and hemorrhagic subtype. Only few randomized controlled trials of rofecoxib have reported the incidence of, 7,35 37 and the data from those studies were difficult to interpret as the outcome of interest was uncommon. For example, a 4-year, randomized, placebo-controlled study of 1551 participants found one event of in the rofecoxib arm and 2 events in the control arm. 37 Even though the primary studies included in this metaanalysis were of high quality as evident by high Newcastle- Ottawa quality assessment scores, we acknowledge that there were some limitations, and thus, the results should be interpreted with caution. First, most of the included studies were conducted using coding-based medical registry, raising concern of coding inaccuracy and incompleteness. Most of the included studies also relied on prescription information from their prescription database which did not guarantee consumption of the medications and would not capture over-the-counter NSAIDs use. The exception is the study by Thrift et al 17 and Choi et al 20 that used structured interview to identify NSAIDs exposure. However, this method has the limitation that cases (participants with the event of interest) might recall exposure better than controls (ie, recall bias). 38 Second, most of the included studies (except for the study by Fosbol et al 26 ) were conducted in older populations with mean age ranging from 56 to 80 years. Therefore, the results might not be generalizable to general population, especially to younger patients with a different baseline cardiovascular risk. Third, even though the statistical heterogeneity of the overall analysis was low, the statistical heterogeneity of the cohort study subgroup was high. Fourth, confounders were not uniformly adjusted across the included studies. Therefore, we cannot exclude the possibility that potential confounders, such as hypertension, alcohol use, chronic kidney disease, liver disease, and polymorphisms for CYP2C9, involved in NSAID metabolism may have affected this association. Data on concurrent use of other medications that can alter the risk of intracerebral hemorrhage were not available for this meta-analysis. For example, a recent study reported that the

Ungprasert et al NSAIDs and Hemorrhagic Stroke 363 Figure 4. Forest plot of individual nonsteroidal anti-inflammatory drugs (NSAIDs). A, Naproxen. B, Diclofenac. C, Rofecoxib. D, Celecoxib. combination of antidepressants and NSAIDs was associated with a significantly increased risk of compared with antidepressants alone. 39 Meta-analysis of observational studies such as this are inherently at risk of several types of bias. 40 For example, participants who took NSAIDs might undergo more medical examinations and laboratory surveillance, resulting in detection bias. Physicians might also preferentially avoid prescribing NSAIDs to patients with increased baseline risk for cardiovascular disease, leading to selection bias of healthier participants to the exposed group. With regard to the risk of individual NSAIDs, some specific limitations are appreciated. First, as we performed several subgroup analyses (ie, several analyses for individual NSAIDs), it is possible that the statistical significance in a subgroup could occur by chance. Second, the number of studies and participants included in individual analysis was smaller compared with the overall analysis. Well done studies of a particular NSAID may receive more weight in this type of analysis. The significant results of individual NSAIDs, particularly meloxicam, were primarily driven by the study of Chang et al. 21 Conclusions As a single group, NSAIDs use was not associated with a significant risk of. However, a significantly increased risk was observed among diclofenac and meloxicam users. None. Disclosures Reference 1. Curhan GC, Bullock AJ, Hankinson SE, Willett WC, Speizer FE, Stampfer MJ. Frequency of use of acetaminophen, nonsteroidal antiinflammatory drugs, and aspirin in US women. Pharmacoepidemiol Drug Saf. 2002;11:687 693. doi: 10.1002/pds.732.

364 Stroke February 2016 2. Ungprasert P, Kittanamongolchai W, Price C, Ratanapo S, Leeaphorn N, Chongnarungsin D, et al. What is the safest non-steroidal ani-inflammatory drugs? Am Med J. 2012;3:115 123. 3. Vonkeman HE, van de Laar MA. Nonsteroidal anti-inflammatory drugs: adverse effects and their prevention. Semin Arthritis Rheum. 2010;39:294 312. doi: 10.1016/j.semarthrit.2008.08.001. 4. Harirforoosh S, Asghar W, Jamali F. Adverse effects of nonsteroidal antiinflammatory drugs: an update of gastrointestinal, cardiovascular and renal complications. J Pharm Pharm Sci. 2013;16:821 847. 5. Bensen WG, Zhao SZ, Burke TA, Zabinski RA, Makuch RW, Maurath CJ, et al. Upper gastrointestinal tolerability of celecoxib, a COX-2 specific inhibitor, compared to naproxen and placebo. J Rheumatol. 2000;27:1876 1883. 6. Mamdani M, Rochon PA, Juurlink DN, Kopp A, Anderson GM, Naglie G, et al. Observational study of upper gastrointestinal haemorrhage in elderly patients given selective cyclo-oxygenase-2 inhibitors or conventional non-steroidal anti-inflammatory drugs. BMJ. 2002;325:624. 7. Bresalier RS, Sandler RS, Quan H, Bolognese JA, Oxenius B, Horgan K, et al; Adenomatous Polyp Prevention on Vioxx (APPROVe) Trial Investigators. Cardiovascular events associated with rofecoxib in a colorectal adenoma chemoprevention trial. N Engl J Med. 2005;352:1092 1102. doi: 10.1056/NEJMoa050493. 8. Motsko SP, Rascati KL, Busti AJ, Wilson JP, Barner JC, Lawson KA, et al. Temporal relationship between use of NSAIDs, including selective COX-2 inhibitors, and cardiovascular risk. Drug Saf. 2006;29:621 632. 9. Gudbjornsson B, Thorsteinsson SB, Sigvaldason H, Einarsdottir R, Johannsson M, Zoega H, et al. Rofecoxib, but not celecoxib, increases the risk of thromboembolic cardiovascular events in young adults-a nationwide registry-based study. Eur J Clin Pharmacol. 2010;66:619 625. doi: 10.1007/s00228-010-0789-2. 10. McGettigan P, Henry D. Cardiovascular risk and inhibition of cyclooxygenase: a systematic review of the observational studies of selective and nonselective inhibitors of cyclooxygenase 2. JAMA. 2006;296:1633 1644. doi: 10.1001/jama.296.13.jrv60011. 11. Varas-Lorenzo C, Riera-Guardia N, Calingaert B, Castellsague J, Pariente A, Scotti L, et al. Stroke risk and NSAIDs: a systematic review of observational studies. Pharmacoepidemiol Drug Saf. 2011;20:1225 1236. doi: 10.1002/pds.2227. 12. Furlan AD, Irvin E, Bombardier C. Limited search strategies were effective in finding relevant nonrandomized studies. J Clin Epidemiol. 2006;59:1303 1311. doi: 10.1016/j.jclinepi.2006.03.004. 13. Stang A. Critical evaluation of the Newcastle-Ottawa scale for the assessment of the quality of nonrandomized studies in meta-analyses. Eur J Epidemiol. 2010;25:603 605. doi: 10.1007/s10654-010-9491-z. 14. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7:177 188. 15. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327:557 560. doi: 10.1136/ bmj.327.7414.557. 16. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315:629 634. 17. Thrift AG, McNeil JJ, Forbes A, Donnan GA. Risk of primary intracerebral haemorrhage associated with aspirin and non-steroidal anti-inflammatory drugs: case-control study. BMJ. 1999;318:759 764. 18. Bak S, Andersen M, Tsiropoulos I, García Rodríguez LA, Hallas J, Christensen K, et al. Risk of stroke associated with nonsteroidal anti-inflammatory drugs: a nested case-control study. Stroke. 2003;34:379 386. 19. Johnsen SP, Pedersen L, Friis S, Blot WJ, McLaughlin JK, Olsen JH, et al. Nonaspirin nonsteroidal anti-inflammatory drugs and risk of hospitalization for intracerebral hemorrhage: a population-based case-control study. Stroke. 2003;34:387 391. 20. Choi NK, Park BJ, Jeong SW, Yu KH, Yoon BW. Nonaspirin nonsteroidal anti-inflammatory drugs and risk: the Acute Brain Bleeding Analysis study. Stroke. 2008;39:845 849. doi: 10.1161/ STROKEAHA.107.497040. 21. 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: 10.1161/ STROKEAHA.110.585828. 22. Mangoni AA, Woodman RJ, Gilbert AL, Knights KM. Use of non-steroidal anti-inflammatory drugs and risk of ischemic and hemorrhagic stroke in the Australian veteran community. Pharmacoepidemiol Drug Saf. 2010;19:490 498. doi: 10.1002/pds.1945. 23. Lapi F, Piccinni C, Simonetti M, Levi M, Aprile PL, Cricelli I, et al. Non-steroidal anti-inflammatory drugs and risk of cerebrovascular events in patients with osteoarthritis: a nested case-control study [published online ahead of print August 14, 2015]. Int Emerg Med. http:// link.springer.com/article/10.1007%2fs11739-015-1288. Accessed September 30, 2015. 24. Haag MD, Bos MJ, Hofman A, Koudstaal PJ, Breteler MM, Stricker BH. Cyclooxygenase selectivity of nonsteroidal anti-inflammatory drugs and risk of stroke. Arch Intern Med. 2008;168:1219 1224. doi: 10.1001/ archinte.168.11.1219. 25. Roumie CL, Mitchel EF Jr, Kaltenbach L, Arbogast PG, Gideon P, Griffin MR. Nonaspirin NSAIDs, cyclooxygenase 2 inhibitors, and the risk for stroke. Stroke. 2008;39:2037 2045. doi: 10.1161/ STROKEAHA.107.508549. 26. Fosbøl EL, Olsen AM, Olesen JB, Andersson C, Kober L, Torp-Pedersen C, et al. Use of nonsteroidal anti-inflammatory drugs among healthy people and specific cerebrovascular safety. Int J Stroke. 2014;9:943 945. doi: 10.1111/j.1747-4949.2012.00863.x. 27. 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: 10.1161/STROKEAHA.114.007932. 28. Caughey GE, Roughead EE, Pratt N, Killer G, Gilbert AL. Stroke risk and NSAIDs: an Australian population-based study. Med J Aust. 2011;195:525 529. 29. Mamdani M, Juurlink DN, Lee DS, Rochon PA, Kopp A, Naglie G, et al. Cyclo-oxygenase-2 inhibitors versus non-selective non-steroidal antiinflammatory drugs and congestive heart failure outcomes in elderly patients: a population-based cohort study. Lancet. 2004;363:1751 1756. doi: 10.1016/S0140-6736(04)16299-5. 30. Kohli P, Steg PG, Cannon CP, Smith SC Jr, Eagle KA, Ohman EM, et al; REACH Registry Investigators. NSAID use and association with cardiovascular outcomes in outpatients with stable atherothrombotic disease. Am J Med. 2014;127:53 60.e1. doi: 10.1016/j.amjmed.2013.08.017. 31. Stichtenoth DO, Marhauer V, Tsikas D, Gutzki FM, Frölich JC. Effects of specific COX-2-inhibition on renin release and renal and systemic prostanoid synthesis in healthy volunteers. Kidney Int. 2005;68:2197 2207. doi: 10.1111/j.1523-1755.2005.00676.x. 32. White WB. Cardiovascular risk, hypertension, and NSAIDs. Curr Rheumatol Rep. 2007;9:36 43. 33. Aneja A, Farkouh ME. Adverse cardiovascular effects of NSAIDs: driven by blood pressure, or edema? Ther Adv Cardiovasc Dis. 2008;2:53 66. doi: 10.1177/1753944707088184. 34. Antman EM, DeMets D, Loscalzo J. Cyclooxygenase inhibition and cardiovascular risk. Circulation. 2005;112:759 770. doi: 10.1161/ CIRCULATIONAHA.105.568451. 35. 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. 36. Thal LJ, Ferris SH, Kirby L, Block GA, Lines CR, Yuen E, et al; Rofecoxib Protocol 078 study group. A randomized, double-blind, study of rofecoxib in patients with mild cognitive impairment. Neuropsychopharmacology. 2005;30:1204 1215. doi: 10.1038/sj.npp.1300690. 37. Kerr DJ, Dunn JA, Langman MJ, Smith JL, Midgley RSJ, Stanley A, et al; for the VICTOR Trial Group. Rofecoxib and cardiovascular adverse events in adjuvant treatment of colorectal cancer. N Engl J Med. 2007;357:360 369. doi: 10.1056/NEJMoa071841. 38. Coughlin SS. Recall bias in epidemiologic studies. J Clin Epidemiol. 1990;43:87 91. 39. Shin JY, Park MJ, Lee SH, Choi SH, Kim MH, Choi NK, et al. Risk of intracranial haemorrhage in antidepressant users with concurrent use of non-steroidal anti-inflammatory drugs: nationwide propensity score matched study. BMJ. 2015;351:h3517. 40. Signorello LB, McLaughlin JK, Lipworth L, Friis S, Sørensen HT, Blot WJ. Confounding by indication in epidemiologic studies of commonly used analgesics. Am J Ther. 2002;9:199 205.