E. Rahme 1, M. Bardou 2, K. Dasgupta 1, Y. Toubouti 1, J. Ghosn 1 and A. N. Barkun 1,3

Transcription

1 Rheumatology 2007;46: Advance Access publication 13 July 2006 doi: /rheumatology/kel223 Hospitalization for gastrointestinal bleeding associated with non-steroidal anti-inflammatory drugs among elderly patients using low-dose aspirin: a retrospective cohort study E. Rahme 1, M. Bardou 2, K. Dasgupta 1, Y. Toubouti 1, J. Ghosn 1 and A. N. Barkun 1,3 Objectives. Many elderly patients are prescribed both low-dose aspirin (ASA), for cardiovascular protection and non-steroidal anti-inflammatory drugs (NSAIDs) for pain control. Compared with non-selective NSAIDs (NS-NSAIDs), celecoxib has a superior gastrointestinal (GI) safety profile in general. It is unclear, however, whether this fact holds good among patients taking ASA. We compared GI hospitalization rates among elderly patients taking celecoxib, NS-NSAIDs, celecoxib and ASA or NS-NSAIDs and ASA. Methods. This was a retrospective cohort study using Quebec government databases. All patients 65 yrs of age or older who filled a prescription for celecoxib or an NS-NSAID between April 1999 and December 2002 were included. Cox regression models were used to compare the GI hospitalization rates between the four exposure categories adjusting for potential confounders. Results. A total of patients were included. Among celecoxib prescriptions, were filled by patients concurrently receiving ASA (celecoxib and ASA); of NS-NSAID prescriptions were filled by patients receiving ASA (NS-NSAID and ASA). Celecoxib without ASA was less likely than NS-NSAID without ASA to be associated with GI hospitalization [hazard ratio (HR) 0.41, 95% confidence interval (CI) ]; celecoxib and ASA was also less likely to be associated with GI hospitalization than NS-NSAID and ASA (HR 0.62, 95% CI ); GI hospitalization rates were similar for celecoxib and ASA and NS-NSAID without ASA (HR 1.01, 95% CI ). Conclusion. Among elderly patients receiving cardiovascular protection with ASA and pain control with anti-inflammatory drugs, celecoxib may be safer with regards to GI toxicity than NS-NSAIDs. KEY WORDS: Cox-2 inhibitors, NSAIDs, Aspirin, Gastrointestinal bleeding, Administrative databases, Elderly. Introduction Non-steroidal anti-inflammatory drugs (NSAIDs) are among the most frequently prescribed medications worldwide. Although these agents may increase the risk for gastrointestinal (GI) ulcers, hypertension and renal failure, they are commonly used among the elderly patients to control the pain associated with osteoarthritis (OA) and other chronic musculoskeletal conditions [1, 2]. Many elderly patients are treated with low-dose aspirin (ASA) for the prevention of cardiovascular events [3, 4]. Although ASA is an non-selective NSAID, it may provide an inadequate analgesic effect at the low doses ( mg/day) employed for cardioprotection. The cardiovascular effect of low-dose ASA is attributed to its inhibition of platelet cyclooxygenase (COX)-1 enzyme [5, 6]. ASA at low doses was found to be a relatively selective COX-1 inhibitor and higher levels of ASA are needed to inhibit COX-2 that is induced at sites of inflammation [7]. Consequently, many elderly patients who take ASA for cardioprotection also need another NSAID for pain control. A telephone survey of chronic NSAID users in north-eastern USA between 3 December 2001 and 19 January 2002 found that 50% of respondents aged 66 yrs or older used ASA also [8]. This practice is supported by a recent case control study that found no significant reduction in risk for myocardial infarction with any NS-NSAID other than ASA [9]. Indeed, it was recently reported that in the Alzheimer s Disease Anti-inflammatory Prevention Trial (ADAPT), there was an apparent increase in cardiovascular and cerebrovascular events among participants taking the NS-NSAID naproxen compared with those receiving placebo [10]. The association between NS-NSAID use, including lowdose ASA, and upper GI adverse events is well documented [1, 2, 11, 12]. Serious upper GI adverse events of gastric and duodenal ulcers, perforations and bleeding (PB) affect 1 2% of patients receiving these medications [13, 14]. PBs often require hospitalization and have been estimated to cause the death of one in every 1220 chronic NS-NSAID users [15]. Moreover, addition of a non-asa NS-NSAID to ASA therapy increases the risk of PB 2 4 times [11, 16, 17]. Clinical trials have demonstrated that the COX-2 inhibitor celecoxib is as effective as NS-NSAIDs for the treatment of arthritic pain but has a better GI safety profile (henceforth, the term NS-NSAID refers only to non-asa NS-NSAIDs) [18, 19]. Because advanced age is a risk factor for PB [20, 21], 1 Division of Clinical Epidemiology, McGill University Health Centre, Montreal General Hospital Site, Montreal, QC, Canada, 2 Clinical Pharmacology Unit, Faculte de Me decine de Dijon, Dijon, France and 3 Division of Gastroenterology, McGill University Health Centre, Montreal General Hospital Site, Montreal, QC, Canada. Submitted 24 November 2005; revised version accepted 23 May Correspondence to: Elham Rahme, Division of Clinical Epidemiology, Montreal General Hospital L10-408, 1650 Cedar Ave., Montreal, Quebec H3G 1A4, Canada. elham.rahme@mcgill.ca 265 ß The Author Published by Oxford University Press on behalf of the British Society for Rheumatology. All rights reserved. For Permissions, please journals.permissions@oxfordjournals.org

2 266 E. Rahme et al. clinical practice guidelines advocate the use of a COX-2 inhibitor instead of NS-NSAIDs among elderly patients, especially those concurrently receiving low-dose ASA [22 26]. To date, however, no studies have been specifically designed to evaluate the effect, on GI adverse events, of combining low-dose ASA with celecoxib vs NS-NSAIDs [26]. It has been recently reported that patients in the Adenoma Prevention with Celecoxib (APC) trial taking 400 and 800 mg of celecoxib daily had an 2.5-fold increase in risk of major fatal or non-fatal cardiovascular event compared with those taking placebo. In another long-term study, the Prevention of Spontaneous Adenomatopus Polyps (PreSAP) trial, there was no increased risk for patients taking 400 mg of celecoxib daily compared with those taking placebo [27]. Furthermore, current published evidence does not indicate that cardiovascular safety differs for celecoxib compared with most NS-NSAIDs or placebo [28 31]. Nevertheless, because celecoxib does not possess the antiplatelet activity of ASA, clinical practice guidelines recommend that arthritis patients who are also in need of cardiovascular risk reduction should concurrently take low-dose ASA [22 24]. The use of the NS-NSAID ibuprofen in combination with low-dose ASA has been shown to have a deleterious effect on the antiplatelet activity of ASA [32 34]. This further suggests that celecoxib may be preferable to at least some NS-NSAIDs when used in combination with ASA [35]. The GI adverse events with NS-NSAIDs remain an important concern with these drugs especially among the elderly patients. We assessed the GI risks of concurrent NS-NSAID and ASA treatment vs concurrent celecoxib and ASA treatment. Specifically, we compared GI hospitalization rates among elderly patients taking NS-NSAIDs only, celecoxib only, NS-NSAIDs in combination with ASA or celecoxib in combination with ASA using a large, province-wide, government-run administrative database. Methods Study design A retrospective cohort study was conducted based on prescriptions and hospital admissions in the province of Quebec, where all persons aged 65 yrs are covered under the medical plan administered by the Re gie de l Assurance Maladie du Que bec (RAMQ). The plan covers all out-patient and in-patient physician services. Adherence to the medical plan is free of charge for all Quebeckers. RAMQ also administer the provincial drug insurance plan. All patients 65 yrs are eligible for this plan but must pay an annual fee. Those who have access to a collective private drug plan may choose to stay with it. In 2002, 91% of individuals 65 yrs were registered with the RAMQ drug plan [36]. Celecoxib was added to the drug formulary in October Celecoxib, ASA and most NS-NSAIDs are reimbursed under the drug plan without restriction. The RAMQ database includes pharmaceutical, medical and demographic records. The hospital discharge summary database of Quebec (Med-Echo) provides information on all hospital admissions for the province, including primary and secondary discharge diagnoses, admission and discharge dates. The Institut de la Statistique du Que bec (ISQ) provides vital statistics including the date and cause of death, coded according to the ICD-9 codes (before 2000) or ICD-10 codes (since 2000). The RAMQ, Med-Echo and ISQ databases are linkable through a unique patient identifier. These databases have been used in previous pharmacoepidemiological studies [37 39]. RAMQ data were analysed for the period from April 1998 to December 2002, while Med-Echo data were analysed from January 1997 to December Study population The study cohort included RAMQ beneficiaries 66 yrs of age who filled a prescription for celecoxib or an NS-NSAID between April 1999 and December For each patient, the date of the first filled prescription for NS-NSAIDs or celecoxib during the study period was the study index date. All patients included had at least 365 days of physician and prescription claims data and 27 months of hospitalization data prior to the index date. Patients who filled prescriptions for both celecoxib and an NS-NSAID as well as for celecoxib/ns-nsaid in combination with either rofecoxib or meloxicam at the index date were excluded. Baseline characteristics Patient characteristics determined at the index date included demographic variables, and diagnoses of musculoskeletal disease, GI conditions and chronic diseases identified during the prior year. Chronic diseases included cancer, anaemia or diseases of the blood, diabetes, hypertension or cardiovascular, cerebrovascular or renal diseases. Disease diagnoses were inferred from ICD-9 codes and, in the case of diabetes and hypertension, also from use of disease-specific medications. The use of gastroprotective agents (GPAs), misoprostol, proton pump inhibitors (PPIs) and histamine-2 receptor antagonists (H2RAs), was assessed at the index date and during the year prior to that date. Use of medications known to increase the risk for GI bleeding (e.g. anticoagulant or corticosteroid therapy) was determined for the 90-day period preceding the index date. The follow-up period for subjects in this retrospective cohort study was from the index date until the first one of the following events: termination of NS-NSAID/celecoxib use, GI hospitalization, death or the end of the study period. Drug exposure The number of days of supply for celecoxib or NS-NSAID prescriptions was ascertained from the RAMQ database. For study purposes, exposure to celecoxib/ns-nsaid was defined to be 1.25(number of days of supply) because the GI effects of celecoxib and NS-NSAIDs may occur some time after the actual exposure period. Most prescriptions for NS-NSAID in Quebec were dispensed with a medication supply for 30 days [40, 41]. Extending the exposure period by a grace period of 25% of the number of days of supply implies that GI hospitalizations that occurred during therapy or up to a week immediately following that therapy would be attributed to NS-NSAID/celecoxib use. Exposure episodes were classified into four categories: NS-NSAID only; celecoxib only; NS-NSAID and ASA; celecoxib and ASA. In our main analysis, a prescription was classified as celecoxib and ASA if there was any overlap in the days in which ASA and celecoxib were supplied. The NS-NSAID and ASA category was defined in a similar manner. Exposure to celecoxib/ns-nsaid was terminated at a switch to rofecoxib or meloxicam. Patients who filled a celecoxib and an NS-NSAID prescription at the same time during follow-up were not excluded from the study, but their data for the period during which they were exposed to both drugs were excluded from analyses. Outcome measures and validation The primary comparison was GI hospitalization among the four exposure categories. A GI hospitalization was the one with a primary discharge diagnosis of gastric or duodenal PB. GI hospitalizations were identified from the Med-Echo hospital discharge summary data by the ICD-9 codes 531.x 534.x, 578.0, and Such codes have been shown to identify hospitalization for upper GI bleeding with a positive predictive value of 86% and have been used in other studies [42, 43]. GI hospitalizations recorded as in-hospital complications were excluded.

3 Association of GI bleeding with NSAIDs in elderly patients 267 To examine whether some secondary discharge diagnoses of PB would in fact be the cause of admission to hospital, we conducted an ad hoc analysis and reviewed charts of 30 patients from the Montreal General Hospital-McGill University Health Centre. All charts were requested from the hospital on the sole basis of an ICD-9 code for upper GI bleeding (as defined in our study) on the discharge summary. Eleven of these charts had the upper GI bleeding coded as primary diagnosis and the other 19 had upper GI bleeding coded as secondary diagnosis. Among the 11 patients who had GI bleeding coded as primary diagnosis, all (100%) had bleeding on admission, thus confirming that upper GI bleeding was the cause of hospitalization. All of the 11 patients had their bleeding originating in the stomach or duodenum. Among the 19 patients who had the upper GI bleeding, coded as secondary diagnosis, two (11%) had bleeding on admission, nine had bleeding that occurred during the hospitalization and the remaining eight had a history of GI bleeding but did not experience bleeding while in hospital. These data indicate a positive predictive value of 100% (11/11) and a sensitivity of 85% (11/13). A GI hospitalization that occurred within an exposure episode was attributed to the current drug therapy defining that exposure category. GI hospitalizations that occurred during overlapping days of suppy for both celecoxib and NS-NSAIDs were attributed to the drug dispensed last. GI death that occurred outside hospitals were identified from ISQ data. A GI death was defined as one that was caused by a GI PB as identified by the ICD-9 codes listed above or their corresponding ICD-10 codes. All assumptions regarding the definitions of exposure and attributable GI hospitalizations were tested in sensitivity analyses as described below. Statistical analyses Descriptive analyses (means S.D. and proportions) were used as appropriate to compare patient characteristics at the index date among the four exposure categories. The rates of GI hospitalizations per patient-year were compared among exposure categories using multivariable Cox regression models, with exposure considered as a time-dependent variable. At a particular point in time, the model considers a patient in the risk set corresponding to the category he/she is exposed to. Times when the patient was not exposed to either celecoxib or NS-NSAIDs were not included in the model to facilitate the analysis. Therefore, discontinuous time intervals were considered. Such time intervals remove the subjects from the risk sets but not from the study [44]. GI hospitalization rates were adjusted for baseline patient characteristics that could be potential confounders [45]. These included age, sex, diagnosis of OA or rheumatoid arthritis (RA), GPA use (prescription at the index date), use of corticosteroids or anti-coagulants (during 90 days preceding the index date) and use of clopidogrel (1 yr period prior to the index date). Models were adjusted for occurrence of GI events during the year prior to the index date. They were also adjusted for the following chronic conditions as identified during the year prior to the index date: diabetes, hypertension, heart failure, ischaemic heart disease, diseases of the blood, psychiatric disorders, cancer and renal failure. Backward selection of variables was performed. Exposure categories were forced in the model and an independent variable was kept in it if it was statistically significant at the 10% level of significance or if its exclusion caused >5% change in the parameter estimate of any one of the exposure categories. Colinearity between independent variables was examined. To examine whether the effect of concomitant use of PPIs with NS-NSAIDs or celecoxib on the prevention of GI hospitalizations differed between patients at high risk of GI events vs those with no GI risk factors, interaction terms were tested in the model. All statistical analyses were performed using SAS version 8.2 (SAS Institute Inc., Cary, NC, USA). Sensitivity analyses Definitions of exposure and outcome were varied in five sensitivity analyses. First, the analyses were repeated including only the patients who did not have any prescription for an NS-NSAID or celecoxib in the year before the index date, did not have a GI hospitalization in the preceding 27 months (principal or secondary diagnosis identified from the hospital discharge summary database) or did not have an ulcer diagnosis during the preceding year (identified from the physician claim database). As explained previously, the main analysis attributed a GI hospitalization that occurred during the overlapping days of supply for celecoxib and NS-NSAIDs to the drug dispensed last. In a sensitivity analysis, GI hospitalizations were instead attributed to the drug dispensed first. In a third sensitivity analysis, exposure episodes and patient follow-up were stopped at the date of a medication switch if it occurred. In a fourth sensitivity analysis, the definitions of NS-NSAID and ASA and celecoxib and ASA episodes were restricted to include only the cases in which ASA exposure overlapped 100% of the duration of the NS-NSAID or celecoxib exposure episode. The main analysis accounted for GI hospitalizations that occurred within the 125% of the days of supply for the medication. In a fifth sensitivity analysis, the 25% grace period was removed and only GI hospitalizations that occurred during the actual days of supply were counted. In a sixth sensitivity analysis, a 50% grace period was considered. Results Patient characteristics Eighty-one patients were excluded because they filled prescriptions for both celecoxib and NS-NSAID or celecoxib/ns-nsaid and rofecoxib or meloxicam at the index date. Of patients included in the study, received an ASA prescription with the days of supply overlapping those for their index prescription (celecoxib or NS-NSAID) (Table 1). Baseline characteristics of these patients are shown in Table 1. Compared with patients receiving NS-NSAIDs, a higher proportion of patients receiving celecoxib were women, were older and had OA during the year prior to the index date. Patients receiving celecoxib were also more likely to have had GI hospitalizations in the prior 27 months than those receiving NS-NSAIDs. These differences were also seen with patients in the celecoxib and ASA category compared with those in the NS-NSAID and ASA category. Despite the increased risk for PB with combined NS-NSAID and ASA therapy, 76% of patients in the NS-NSAID and ASA category did not fill a GPA prescription at the index date or did not have a previous GPA prescription overlapping that date. Drug exposure Patients included in the analyses filled celecoxib and NS-NSAID prescriptions during the study period. Naproxen accounted for 24% of all NS-NSAID prescriptions, the combination diclofenac/misoprostol 22%, diclofenac 18%, ibuprofen 10%, nabumetone 6%, indometacine 5%, etodolac 4%, ketoprofen 3%, the remaining 8% for 10 other NS-NSAIDs. In total, prescriptions were classified in the celecoxib and ASA group and in the NS-NSAID and ASA group (Table 2). Patients given ASA used it at a median dose of 325 mg/day [interquartile range ( mg/day)], and 23% of patients used ASA at dose of 80 mg/day. Similar overlap between antiinflammatory therapy exposure and ASA exposure for the celecoxib and ASA category compared with the NS-NSAID and ASA category was observed (91.8 vs 92.7% overlap, respectively).

4 268 E. Rahme et al. TABLE 1. Patient characteristics at baseline Drug exposure category NS-NSAID Celecoxib NS-NSAID and ASA Celecoxib and ASA Number of patients Demographics (%) Women Age yrs (%) and above Musculoskeletal disease in prior year (%) Osteoarthritis Rheumatoid arthritis Visited a rheumatologist Cardiovascular disease in prior year (%) Congestive heart failure Ischaemic heart disease Cerebrovascular disease Anti-hypertensive medication Anti-diabetic agents Visited a cardiologist Clopidogrel Chest pain Renal failure Other comorbid conditions a (%) Cancer Chronic obstructive pulmonary disease Anaemia or blood disease Psychiatric disorders GI events in prior year (%) GI hospitalizations in prior 27 months Diagnosis with PB in prior year Visited a gastroenterologist Upper GI diagnostic test Dyspepsia or heartburn in prior year GPA in prior year GPA overlapping the index date GPA filled at baseline PPI filled at baseline Prescribed ASA b in prior year (%) Prescribed acetaminophen in prior year (%) Opiates in prior year (%) Anti-coagulant use in the prior 90 days (%) Corticosteroid use in the prior 90 days (%) ASA, acetylsalicylic acid; GI, gastrointestinal; PB, perforation or bleeding; GPA, gastroprotective agent; NS-NSAID, non-selective non-steroidal antiinflammatory drug; PPI, proton pump inhibitors. a Disease conditions were inferred from physician diagnoses in the RAMQ database (ICD-9 codes) and medication use for diabetes and hypertension. Prior GI hospitalization was inferred from the Med-Echo database. b Not including those that overlap the index date. TABLE 2. Prescriptions and unadjusted GI hospitalization rates in follow-up Drug exposure NS-NSAID Celecoxib NS-NSAID and ASA Celecoxib and ASA Number of prescriptions Prescriptions with a medication switch (%) (3.4) (2.6) 6879 (3.5) 9561 (2.2) Total drug exposure duration (yrs) Average % overlap with ASA prescription Number of GI hospitalizations (rate per 1000 patient-yrs) 223 (5.1) 197 (2.5) 111 (8.9) 182 (6.2) Number of out-patient death due to GI bleeding (rate per 1000 patient-yrs) 5 (0.1) 5 (0.1) 0 7 (0.2) ASA, acetylsalicylic acid; NS-NSAID, non-selective non-steroidal anti-inflammatory drug. GI hospitalization Table 2 shows the total number of days and crude rates of GI hospitalization for each exposure category. Unadjusted GI hospitalization rates were higher in the NS-NSAID group than in the celecoxib group, and in the NS-NSAID and ASA group than in the celecoxib and ASA group. Patients who received ASA coprescriptions had higher GI hospitalizations than those who did not.

5 Association of GI bleeding with NSAIDs in elderly patients 269 TABLE 3. Cox regression model with time-dependent exposure to determine the association between drug exposure and GI hospitalization Variable Hazard ratio (95% CI) a Women 0.73 (0.63, 0.85) Age yrs yrs 1.52 (1.18, 1.95) yrs 1.80 (1.40, 2.32) yrs 2.94 (2.28, 3.81) 85 yrs and above 3.36 (2.53, 4.45) Anaemia or blood disease 1.78 (1.41, 2.25) Psychiatric disorders 1.69 (1.23, 2.31) Renal failure 1.62 (1.01, 2.60) Anti-coagulant use in prior 90 days 3.20 (2.43, 4.23) Corticosteroid use in prior 90 days 1.24 (0.96, 1.61) Prescribed acetaminophen in prior year 1.17 (1.00, 1.36) Prescribed NS-NSAIDs in prior year 0.77 (0.65, 0.92) Opiates in prior year 1.53 (1.14, 2.06) GI hospitalization in prior 27 months 3.65 (2.56, 5.19) Diagnosis with PB in prior year (out-patient) 1.67 (1.08, 2.58) PPI vs no GPA use at baseline 0.47 (0.32, 0.69) Other b GPA vs no GPA use at baseline 1.29 (0.97, 1.72) NS-NSAID Celecoxib 0.41 (0.33, 0.50) NS-NSAID with ASA 1.63 (1.29, 2.05) Celecoxib with ASA 1.01 (0.81, 1.25) ASA, acetylsalicylic acid; NS-NSAID, non-selective non-steroidal anti-inflammatory drug; GPA, gastroprotective agents; PPI, proton pump inhibitor; GI, gastrointestinal. a The model was adjusted for other patient characteristics at baseline. Only those significant at the 10% level were kept in the model. b Misoprostol or H2RA. The results of the time-dependent Cox regression model, comparing the adjusted GI hospitalization hazard rates between the four exposure categories are shown in Table 3. The effect of PPI at baseline did not differ according to whether or not the patient was at high risk of GI bleeding (prior GI bleeding and/or anti-coagulant use in the prior 90 days). Therefore, interaction terms were removed from the final model. After adjustment for baseline patient characteristics including GPA use at the index date and prior GI events, factors that were positively associated with GI hospitalization included age 70 yrs, prior GI hospitalizations or prior diagnoses with PB, male gender and use of anticoagulant in the prior 90 days. Use of acetaminophen or opiates in the prior year was also positively associated with GI hospitalization. In addition, patients with diagnoses of psychiatric disorders, anaemia or diseases of the blood or renal failure were found to be at an increased risk for GI hospitalization. Filling a prescription for a PPI at baseline decreased the adjusted hazard rate of GI hospitalizations but filling a prescription for other GPAs did not seem to have a significant effect. Compared with NS-NSAID use alone, only celecoxib had 59% reduction in the adjusted hazard rate of GI hospitalization. No statistically significant difference in adjusted GI hospitalization hazard rate was found for celecoxib and ASA compared with NS-NSAID use alone. However, the NS-NSAID and ASA combination had a 63% increase in hazard rate compared with NS-NSAID alone and the celecoxib and ASA combination conferred a significant 38% hazard rate reduction compared to NS-NSAID and ASA treatment. GI hospitalization and/or GI death Table 2 shows the crude rates of GI hospitalizations and those of death from GI bleeding occurring outside a hospital setting for each exposure category. As indicated in Table 2, the number of GI death that occurred outside a hospital was very small. The results of the time-dependent Cox regression model comparing the combined outcome GI hospitalization and/or GI death were not substantially different from those of the model analysing the single outcome GI hospitalization (data not shown). Sensitivity analyses The effects of varying definitions of exposure and outcome in the sixth sensitivity analysis are shown in Table 4. As for the base-case scenario, GI hospitalization hazard rates were adjusted for baseline patient characteristics (including GPA use and GI events) and chronic conditions. Results of all sensitivity analyses were similar to those in the base-case scenario. Discussion Our results confirm that use of the combination of celecoxib/ NS-NSAID with ASA is associated with a higher rate of serious adverse GI events than celecoxib/ns-nsaid used alone. However, among older patients using both ASA for cardioprotection and celecoxib/ns-nsaid for pain control, the combination of celecoxib and ASA is associated with a lower likelihood of adverse GI events than the combination of NS-NSAID and ASA. In the base-case analysis, celecoxib use was associated with fewer GI hospitalizations than was NS-NSAID use, whether comparing the celecoxib-only and NS-NSAID-only categories, or the celecoxib and ASA and NS-NSAID and ASA categories. The higher rate of GI hospitalization seen in NS-NSAID-only recipients than in celecoxib-only recipients was consistent with results of other retrospective database analyses in Ontario, Canada [42] and Denmark [46], which found significantly higher rates of GI bleeding among NS-NSAID users than among users of COX-2 inhibitors. Another database analysis examined the impact of the introduction of COX-2 inhibitors on the incidence of GI hospitalization among the elderly population of Ontario [47]. The rate of GI hospitalizations per seniors aged at least 66 yrs increased from 15.4 before the introduction of COX-2 inhibitors to 17.0 after. This result accompanied an increase of more than additional elderly individuals using NSAIDs annually. Importantly, one cannot infer the causation from an ecological study, so it is unclear whether the increase in GI hospitalization rate at the population level was attributable to COX-2 inhibitor use. Adding low-dose ASA to celecoxib increased the GI hospitalization rate such that it was no longer significantly different than for NS-NSAID-only recipients. The increased rate of GI hospitalization observed among celecoxib users who used lowdose ASA is consistent with the results of the Celecoxib Longterm Arthritis Safety Study (CLASS). In the CLASS study, celecoxib recipients who also used low-dose ASA had a significantly higher risk of ulcer complications than those who did not [18]. Recent prescribing guidelines judged as uncertain the appropriateness of coprescribing a PPI for gastroprotection among elderly celecoxib recipients who also used ASA [26]. However, in this study, elderly patients with concomitant celecoxib and ASA exposure had a higher rate of GI hospitalization than did patients who received celecoxib only. This supports the conclusion of another recent review [48] that additional gastroprotection is warranted for elderly chronic users of both low-dose ASA and a COX-2 inhibitor. Several animal and human studies showed that the intake of celecoxib together with 100 mg ASA blocked the synthesis of aspirin-triggered lipoxin which increased the gastric damage induced by ASA [49 54]. This harmful gastric effect was not observed when low-dose ASA was used together with 5-lipooxygenase inhibitors, a new group of anti-inflammatory drugs [54]. It should also be noted that most of the ASA prescriptions in this population were for 325 mg/day. Studies have shown that ASA at 75 mg/day was sufficient for cardioprotection and higher doses ASA may increase the risk of GI PB by a factor of 4 10 at analgesic doses [7, 55].

6 270 E. Rahme et al. TABLE 4. Sensitivity analyses a using Cox regression models to determine the association between drug exposure and GI hospitalization under different definitions of exposure and outcome Hazard ratio (95% CI) b Drug exposure Base-case NS-NSAID (n ¼ ) (n ¼ ) (n ¼ ) (n ¼ ) (n ¼ ) (n ¼ ) (n ¼ ) Celecoxib 0.41 (0.33, 0.50) (n ¼ ) 0.36 (0.27, 0.48) (n ¼ ) 0.43 (0.35, 0.52) (n ¼ ) 0.34 (0.26, 0.44) (n ¼ ) 0.42 (0.34, 0.52) (n ¼ ) 0.43 (0.34, 0.53) (n ¼ ) 0.33 (0.25, 0.43) (n ¼ ) NS-NSAID and ASA 1.63 (1.29, 2.05) (n ¼ ) 1.50 (1.00, 2.56) (n ¼ ) 1.62 (1.31, 2.02) (n ¼ ) 1.59 (1.23, 2.04) (n ¼ ) 1.79 (1.40, 2.29) (n ¼ ) 1.49 (1.16, 1.91) (n ¼ ) 1.91 (1.47, 2.48) (n ¼ ) Celecoxib and ASA 1.01 (0.81, 1.25) (n ¼ ) 0.83 (0.60, 1.13) (n ¼ ) 1.00 (0.81, 1.23) (n ¼ ) 0.77 (0.58, 1.02) (n ¼ ) 1.09 (0.87, 1.38) (n ¼ ) 1.00 (0.80, 1.26) (n ¼ ) 0.91 (0.70, 1.19) (n ¼ ) ASA, acetylsalicylic acid; GI, gastrointestinal; n, number of prescriptions; NS-NSAID, non-selective non-steroidal anti-inflammatory drug. a Sensitivity analysis scenario: 1. Patients with prior NS-NSAIDs and patients with GI hospitalizations in the prior 27 months were excluded. 2. GI hospitalizations in patients who switched medications were attributed to the first prescription. 3. Follow-up was terminated at the date of the first switch if it occurred. 4. Celecoxib and ASA and NS-NSAID and ASA categories were restricted to prescriptions with at least 100% overlap with ASA exposure. 5. Exposure was 100% of the days of supply (no grace period of 25%) 6. Exposure was 150% of the days of supply (grace period of 50%) b Cox regression models were adjusted for patient characteristics at baseline. The reasons for prescribing ASA at 325 vs 80 mg/day have not been identified. We speculate that cost may be a contributor factor to this choice since the acquisition price of ASA at 80 mg/day is about five times higher than that at 325 mg/day. A recently published study conducted in the US reported that one-third of US primary care physicians who participated in the study recommended 325 mg rather than 81 mg of ASA/day for cardioprotection [56]. Despite numerous publications and clinical practice guidelines that recommend the use of gastroprotective agents with NS-NSAIDs in the elderly patients [57], the utilization of these drugs remains suboptimal. In fact, only 24% of patients in the NS-NSAID and ASA category had a GPA prescription at the index date. These results are consistent with results from another study examining NSAIDs utilization in the UK, where it was reported that 26% of patients using NS-NSAIDs also used a GPA [58]. The adjusted rate of GI hospitalizations among patients who did not receive a PPI at baseline was approximately twice as high as among those who did, and was similar to that of patients who received another GPA (H2RA or misoprostol). The protective effect of PPI against severe complications (haemorrhage or perforation) has not been well quantified in the literature among NS-NSAID users and less so among celecoxib users. Our study was not designed to answer this important question, nor was it designed to study the effect of misoprostol or H2RA separately. However, in a separate study that we conducted to investigate these issues [59], we found that the risk of GI hospitalizations associated with NS-NSAID plus PPI was similar to that of celecoxib alone. In that study, the risk among those using celecoxib plus PPI was 30% lower than that of patients using NS-NSAIDs plus PPI. The use of a PPI with celecoxib was beneficial in high-risk patients, those aged 75 yrs and those using ASA. Analyses were conducted using a large, population-based medical database that has been well-validated [60]. Use of the RAMQ database provides the advantage of large sample size, generalizability and the broad inclusion of patients with multiple GI risk factors, who are typically excluded from clinical trials but whose representation is very much relevant to the clinical question at hand. We used Cox regression models with time-dependent exposure to compare the time with event between patients in the four exposure categories, NS-NSAID only; celecoxib only; NS-NSAID and ASA; celecoxib and ASA. Several assumptions concerning overlaps and switches between these categories had to be adopted in the main analyses. However we conducted extensive sensitivity analyses and found that our results were robust with respect to these assumptions. Nevertheless, this study has limitations. First, the drug exposure groups may be subject to selection bias [61], because differences may exist between patients in different groups in variables that were not available in the database. For example, although gastroduodenal ulcer within the 12 months prior to the index date and GI hospitalization within 27 months prior were adjusted for in the analyses, residual bias might still be present. However, in a sensitivity analysis we excluded patients with these prior GI conditions and the results did not change. Use of any medical claim database might underestimate the prevalence of medical conditions that frequently go untreated, such as alcohol and drug abuse. Therefore, some of the comorbidities included in the model, such as psychiatric disorders, anaemia or blood disease or renal failure, may have been underestimated. Second, celecoxib and NS-NSAIDs are not always taken as prescribed, and patients who filled prescriptions for these drugs may not have consumed them. Depending on the existence and direction of differential non-adherence by drug exposure category, the results could be biased in different ways. Recent database analyses have found that celecoxib recipients are less likely to discontinue the therapy than are NS-NSAID recipients [62, 63], and the rate of medication switching in this study was higher among NS-NSAID recipients than among celecoxib recipients (Table 2). Both findings suggest that rates of non-adherence may be higher among NS-NSAID users than among celecoxib users. If this was the case in our study cohort, more patients in the NS-NSAID group than in the celecoxib group did not consume their drugs, which would be predicted to reduce any observed differences in GI hospitalization rates between celecoxib and NS-NSAID categories. Third, the data on over-the-counter medication purchases are not available from RAMQ, so non-prescription use of NS-NSAIDs, ASA or GPAs could not be assessed. Over-thecounter use of NS-NSAIDs is relatively high among the general Quebec population; during the study period, according to Sante Que bec, the government health agency, 17.0, 2.2 and 1.1% of the elderly patients who consumed NS-NSAIDs, ASA or GPAs, respectively, acquired them over-the-counter. However, since

7 Association of GI bleeding with NSAIDs in elderly patients 271 patients were included in this study cohort only if they had a prescription for an NS-NSAID or a celecoxib, they are less likely to use over-the-counter NS-NSAIDs. As in other population-based observational studies of celecoxib and NS-NSAID use in Canada [42], patients had strong financial incentives to obtain NS-NSAIDs via prescription rather than by over-the-counter purchase, which is not reimbursed by the provincial drug plan. Furthermore, any over-the-counter use of NS-NSAIDs during a celecoxib prescription would be predicted to increase the GI risk in the celecoxib cohort and subsequently reduce any observed difference in GI hospitalization rate between the celecoxib and NS-NSAID categories. Finally, meloxicam use is not common in Quebec. Among all NS-NSAID prescriptions, 3.6% were for meloxicam and the number of GI hospitalizations that occurred during exposure to meloxicam was 13, precluding any robust analysis using meloxicam-only and meloxicam and ASA subgroups. Therefore, patients who used meloxicam during the study period were excluded. In summary, these results suggest that for those patients taking ASA for cardioprotection as well as anti-inflammatory therapy for pain control, there is a lower risk of GI complications if celecoxib is used instead of NS-NSAIDs. Prospective, randomized trials are required to more accurately measure the size of the effect observed in this observational, population-based study. Acknowledgements This study was largely supported by a grant from the Canadian Institutes of Health Research and partially supported by funding from the Arthritis Society. This article was prepared with the assistance of BioMedCom Consultants Inc., Montreal, Canada. Dr E.R. has served as a consultant for Merck & Co. Inc. and Pfizer Inc. and Boehringer Ingelheim received research funding from Pfizer Inc., the manufacturers of celecoxib. Dr M.B. has served as a consultant for AstraZeneca and has received research support from Sanofi-Aventis. Dr A.N.B. has served as a consultant and a member of the speakers bureau, and has received research funding from Merck & Co. Inc., AstraZeneca Canada, Altana Pharma Canada and Janssen-Ortho Canada Inc. K.D., Y.T. and J.G. have declared no conflicts of interest. References 1. Griffin MR. Epidemiology of nonsteroidal anti-inflammatory drugassociated gastrointestinal injury. Am J Med 1998;104:23S 9S. 2. MacDonald TM. Epidemiology and pharmacoeconomic implications of non-steroidal anti-inflammatory drug-associated gastrointestinal toxicity. Rheumatology 2000;39(Suppl 2):S Patrono C. Aspirin: new cardiovascular uses for an old drug. Am J Med 2001;110:62S 65S. 4. Mahe I, Leizorovicz A, Caulin C, Bergmann JF. Aspirin for the prevention of cardiovascular events in the elderly. Drugs Aging 2003;20: Patrono C, Rodriguez LAG, Landofi R, Baigent C. Low-dose aspirin for the prevention of atherothrombosis. N Engl J Med 2005;353: Patrignani P, Filabozzi P, Patrono C. Selective cumulative inhibition of platelet thromboxane production by low-dose aspirin in healthy subjects. Clin Invest 1982;69: Patrono C, Coller B, FitzGerald GA, Hirsch J, Roth G. Plateletactive drugs; the relationships among dose, effectiveness, and side effects: the Seventh ACCP Conference on Antithrombotic and Thrombolytic Therapy. Chest 2004;126:234S 64S. 8. Cox ER, Frisse M, Behm A, Fairman KA. Over-the-counter pain reliever and aspirin use within a sample of long-term cyclooxygenase 2 users. Arch Intern Med 2004;164: Garcia Rodriguez LA, Varas-Lorenzo C, Maguire A, Gonzalez-Perez A. Nonsteroidal anti inflammatory drugs, the risk of myocardial infarction in the general population. Circulation 2004;109: Use of non-steroidal anti-inflammatory drugs suspended in large Alzheimer s disease prevention trial. pr/dec2004/od-20.htm National Institutes of Health, US, Department of Health and Human Services. 11. Weil J, Colin-Jones D, Langman M et al. Prophylactic aspirin and risk of peptic ulcer bleeding. Br Med J 1995;310: Derry S, Loke YK. Risk of gastrointestinal haemorrhage with long term use of aspirin: meta-analysis. Br Med J 2000;321: MacDonald TM, Pettitt D, Lee FH, Schwartz JS. Channelling of patients taking NSAIDs or cyclooxygenase-2-specific inhibitors and its effect on interpretation of outcomes. Rheumatology 2003;42: Hernandez-Diaz S, Garcia Rodriguez LA. Association between nonsteroidal anti-inflammatory drugs and upper gastrointestinal tract bleeding/perforation: an overview of epidemiologic studies published in the 1990s. Arch Int Med 2000;160: Tramer MR, Moore RA, Reynolds DJ, McQuay HJ. Quantitative estimation of rare adverse events which follow a biological progression: a new model applied to chronic NSAID use. Pain 2000;85: Williams HJ, Ward JR, Egger MJ et al. Comparison of naproxen and acetaminophen in a two-year study of treatment of osteoarthritis of the knee. Arthritis Rheum 1993;36: Sorensen HT, Mellemkjaer L, Blot WJ et al. Risk of upper gastrointestinal bleeding associated with use of low-dose aspirin. Am J Gastroenterol 2000;95: Silverstein FE, Faich G, Goldstein JL et al. Gastrointestinal toxicity with celecoxib vs nonsteroidal anti-inflammatory drugs for osteoarthritis and rheumatoid arthritis: the CLASS study: a randomized controlled trial. Celecoxib Long-term Arthritis Safety Study. JAMA 2000;284: Deeks JJ, Smith LA, Bradley MD. Efficacy, tolerability, and upper gastrointestinal safety of celecoxib for treatment of osteoarthritis and rheumatoid arthritis: systematic review of randomised controlled trials. Br Med J 2002;325: Gabriel SE, Jaakkimainen L, Bombardier C. Risk for serious gastrointestinal complications related to use of nonsteroidal anti-inflammatory drugs. A meta-analysis. Ann Intern Med 1991;115: Laine L, Bombardier C, Hawkey CJ et al. Stratifying the risk of NSAID-related upper gastrointestinal clinical events: results of a double-blind outcomes study in patients with rheumatoid arthritis. Gastroenterol 2002;123: Guidelines for the management of rheumatoid arthritis: 2002 update. Arthritis Rheum 2002;46: American College of Rheumatology Committee on Osteoarthritis Guidelines. Arthritis Rheum 2000;43: Tannenbaum H, Peloso PM, Russell AS, Marlow B. An evidencebased approach to prescribing NSAIDs in the treatment of osteoarthritis and rheumatoid arthritis. Can J Clin Pharmacol 2000;7(Suppl A):4A 16A. 25. Hunt RH, Barkun AN, Baron D et al. Recommendations for the appropriate use of anti-inflammatory drugs in the era of the coxibs: defining the role of gastroprotective agents. Can J Gastroenterol 2002;16: Dubois RW, Melmed GY, Henning JM, Laine L. Guidelines for the appropriate use of non-steroidal anti-inflammatory drugs, cyclooxygenase-2-specific inhibitors and proton pump inhibitors in patients requiring chronic anti-inflammatory therapy. Aliment Pharmacol Ther 2004;19: Pfizer statement on new information regarding cardiovascular safety of Celebrex. mn_2004_1217.cfm Linton MF, Fazio S. Cyclooxygenase-2 and inflammation in atherosclerosis. Curr Opin Pharmacol 2004;4:

8 272 E. Rahme et al. 29. White WB, Faich G, Borer JS, Makuch RW. Cardiovascular thrombotic events in arthritis trials of the cyclooxygenase-2 inhibitor celecoxib. Am J Cardiol 2003;92: Solomon DH, Schneeweiss S, Glynn RJ et al. Relationship between selective cyclooxygenase-2 inhibitors and acute myocardial infarction in older adults. Circulation 2004;109: Ray WA, Stein CM, Daugherty JR, Hall K, Arbogast PG, Griffin MR. COX-2 selective non-steroidal anti-inflammatory drugs and risk of serious coronary heart disease. The Lancet 2002;360: Catella-Lawson IA, Reilly MP, Kapoor SC et al. Cyclooxygenase inhibitors and the antiplatelet effects of aspirin. N Engl J Med 2001;345: Kurth T, Glynn RJ, Walker AM et al. Inhibition of clinical benefits of aspirin on first myocardial infarction by nonsteroidal antiinflammatory drugs. Circulation 2003;108: MacDonald TM, Wei L. Effect of ibuprofen on cardioprotective effect of aspirin. Lancet 2003;361: Baigent C, Patrono C. Selective cyclooxygenase 2 inhibitors, aspirin, and cardiovascular disease: a reappraisal. Arthritis Rheum 2003;48: accessed March 21, De Wals P, Trochet C, Pinsonneault L. Prevalence of neural tube defects in the province of Quebec, Can J Public Health 1999;90: Rahme E, Marentette MA, Kong SX, LeLorier J. Use of NSAIDs, COX-2 inhibitors, and acetaminophen and associated co-prescriptions of gastroprotective agents in an elderly population. Arthritis Care Res 2002;47: Rahme E, Barkun AN, Toubouti Y, Bardou M. The cyclooxygenase- 2-selective inhibitors rofecoxib and celecoxib prevent colorectal neoplasia occurrence and recurrence. Gastroenterol 2003;125: Rahme E, Pettit D, LeLorier J. Determinants and sequelae associated with acetaminophen utilization versus traditional NSAIDs in an elderly population. Arthritis Rheum 2002;46: Rahme E, Joseph L, Kong SX, Watson DJ, LeLorier J. Gastrointestinal health care resource use and costs associated with nonsteroidal antiinflammatory drugs versus acetaminophen: retrospective cohort study of an elderly population. Arthritis Rheum 2000;43: Mamdani M, Rochon PA, Juurlink DN et al. Observational study of upper gastrointestinal haemorrhage in elderly patients given selective cyclo-oxygenase-2 inhibitors or conventional non-steroidal antiinflammatory drugs. Br Med J 2002;325: Raiford DS, Gutthann SP, Rodriguez LAG. Positive predictive value of ICD-9 codes in the identification of cases of complicated peptic ulcer disease in the Saskatchewan Hospital automated database. Epidemiol 1996;7: Fleming T, Harrington D. Counting process and Survival analysis. New York, Wiley, Barkun A, Bardou M, Marshall JK. Consensus recommendations for managing patients with nonvariceal upper gastrointestinal bleeding. Ann Intern Med 2003;139: Norgard B, Pedersen L, Johnsen SP et al. COX-2-selective inhibitors and the risk of upper gastrointestinal bleeding in high-risk patients with previous gastrointestinal diseases: a population-based casecontrol study. Aliment Pharmacol Ther 2004;19: Mamdani M, Juurlink DN, Kopp A, Naglie G, Austin PC, Laupacis A. Gastrointestinal bleeding after the introduction of COX 2 inhibitors: ecological study. Br Med J 2004;328: Kimmey MB, Lanas A. Review article: appropriate use of proton pump inhibitors with traditional nonsteroidal anti-inflammatory drugs and COX-2 selective inhibitors. Aliment Pharmacol Ther 2004;19(Suppl 1): Wallace JL, Zamuner SR, Mcknight W, Dicay M et al. Aspirin, but not NO-releasing aspirin NCX-4016 interacts with selective COX-2 inhibitors to aggravate gastric damage and inflammation. Am J Physiol Gastrointest Liver Physiol 2004;286: Fiorucci S, Distrutti E, de Lima OM et al. Relative contribution of acetylated cyclo-oxygenase (COX)-2 and 5-lipooxygenase (LOX) in regulating gastric mucosal integrity and adaptation to aspirin. FASEB Journal 2003;17: Hochberg MC, Altman RD, Brandt KD et al. Guidelines for the medical management of osteoarthritis. Part II. Osteoarthritis of the knee. American College of Rheumatology. Arthritis Rheum 1995;38: Nandurkar S, Talley NJ, Xia H, Mitchell H, Hazel S, Jones M. Dyspepsia in the community is linked to smoking and aspirin use but not to Helicobacter pylori infection. Arch Int Med 1998;158: Blum AL, Talley NJ, O Morain C et al. Lack of effect of treating Helicobacter pylori infection in patients with nonulcer dyspepsia. Omeprazole plus Clarithromycin and Amoxicillin Effect One Year after Treatment (OCAY) Study Group. N Engl J Med 1998;339: Fiorucci S, Santucci L, Wallace JL et al. Interaction of a selective cyclooxygenase-2 inhibitor with aspirin and NO-releasing aspirin in the human gastric mucosa. Proc Natl Acad Sci USA 2003;100: Lanas A, Bajador E, Serrano P, Fuentes J, Carreno S, Guardia J. Nitrovasodilators, low-dose aspirin, other nonsteroidal anti inflammatory drugs, and the risk of upper gastrointestinal bleeding. N Engl J Med 2000;343: Chey WD, Eswaren S, Howden CW, Inadomi JM, Fendrick AM, Scheiman JM. Primary care physician perceptions of non-steroidal anti-inflammatory drug and aspirin-associated toxicity: results of a national survey. Aliment Pharmacol Ther 2006;23: American College of Rheumatology. Update of ACR guidelines for osteoarthritis: role of the coxibs. Journal of Pain and Symptom Management 2002;32(Suppl 4):S Price-Forbes AN, Callaghan R, Allen ME, Rowe IF. A regional audit of the use of COX-2 selective non-steroidal anti-inflammatory drugs (NSAIDs) in rheumatology clinics in the West Midlands, in relation to NICE guidelines. Rheumatol 2005;44: Rahme E, Barkun A, Toubouti Y, Rochon S, LeLorier J. Do proton pump inhibitors infer additional gastrointestinal protection in patients given celcoxib. Pharmacoepidemiol Drug Saf 2005;14:S Tamblyn R, Lavoie G, Petrella L, Monette J. The use of prescription claims databases in pharmacoepidemiological research: the accuracy and comprehensiveness of the prescription claims database in Quebec. J Clin Epidemiol 1995;48: Savitz A. Interpreting epidemiologic evidence: Strategies for study design and analysis. Oxford University Press, New York, Rahme E, Toubouti Y, Hunsche E. Therapy switching and associated costs in elderly patients receiving COX-2 selective inhibitors or nonselective non-steroidal anti-inflammatory drugs in Quebec, Canada. Rheumatology (Oxford) 3 February 2006 [Epub ahead of print]. 63. Moride Y, Ducruet T, Rochon S, Lavoie F. Persistency of use of COX-2-specific inhibitors and non-specific non-steroidal antiinflammatory drugs (NSAIDs) in Quebec. Rheumatology 2003;42(Suppl 3):iii17 22.