Traditional nonsteroidal anti-inflammatory drugs

GASTROENTEROLOGY 2008;135:1517 1525 Lower Gastrointestinal Events in a Double-Blind Trial of the Cyclo-Oxygenase-2 Selective Inhibitor Etoricoxib and the Traditional Nonsteroidal Anti-Inflammatory Drug Diclofenac LOREN LAINE,* SEAN P. CURTIS, MICHAEL LANGMAN, DENNIS M. JENSEN, BYRON CRYER, AMARJOT KAUR, CHRISTOPHER P. CANNON # *Division of Gastrointestinal & Liver Diseases, University of Southern California Keck School of Medicine, Los Angeles, California; Merck Research Laboratories, Rahway, New Jersey; University of Birmingham, Birmingham, United Kingdom; UCLA Geffen School of Medicine, CURE DDRC, and West Los Angeles VA Medical Center, Los Angeles, California; Division of Gastroenterology, Department of Medicine, University of Texas Southwestern Medical School, Dallas VA Medical Center, Dallas, Texas; # Thrombolysis in Myocardial Infarction (TIMI) Study Group, Cardiovascular Division, Brigham & Women s Hospital, Harvard Medical School, Boston, Massachusetts See Watanabe T et al on page 1279 in CGH. Background & Aims: Nonsteroidal anti-inflammatory drugs (NSAIDs) cause lower gastrointestinal (GI) clinical events such as bleeding. Cyclo-oxygenase (COX)-2 selective inhibitors decrease upper GI events, but no prospective trial has prespecified assessment of lower GI clinical events. Methods: Patients >50 years old with osteoarthritis or rheumatoid arthritis were randomly assigned to etoricoxib (60 or 90 mg qd) or diclofenac (150 mg qd). Lower GI clinical events, confirmed by a blinded adjudication committee, included perforation or obstruction requiring hospitalization or bleeding (gross or occult rectal bleeding without upper GI cause associated with hypotension, orthostatic changes in heart rate [>20 beats per minute] or blood pressure [>20 mmhg systolic or >10 mmhg diastolic], hemoglobin drop >2 g/dl, or transfusion; or observed active bleeding or stigmata of hemorrhage). Results: We enrolled 34,701 patients with mean duration of therapy of 18 months. Rates were 0.32 and 0.38 lower GI clinical events per 100 patient-years for etoricoxib and diclofenac (hazard ratio [HR] 0.84; 95% confidence interval [CI], 0.63 1.13). Bleeding was the most common event (rates of 0.19 and 0.23 per 100 patient-years, respectively). Multivariable analysis revealed significant risk factors to be prior lower GI event (HR 4.06; 95% CI, 2.93 5.62) and age >65 years (HR 1.98; 95% CI, 1.45 2.71). Conclusions: A statistically significant decrease in lower GI clinical events was not seen with the COX-2 selective inhibitor etoricoxib versus the traditional NSAID diclofenac. The risk of a lower GI clinical event with NSAID use seems to be constant over time, and the major risk factors are a prior lower GI event and older age. Traditional nonsteroidal anti-inflammatory drugs (NSAIDs) are a well-known cause of upper gastrointestinal (GI) clinical events such as bleeding ulcers. 1 Less evidence is available to evaluate the risk of NSAIDs for lower GI clinical events. A number of studies indicate that NSAIDs can cause damage to the small and large intestine based on assessment of mucosal integrity (eg, permeability) and on endoscopic or postmortem visualization of the intestinal mucosa. 2 In addition, a number of case-control studies report a significant association of traditional NSAID use with lower GI clinical events, such as bleeding and perforation. 2 Cyclo-oxygenase (COX)-2 selective inhibitors were developed to provide analgesic and anti-inflammatory activity comparable to traditional NSAIDs with a reduction in upper GI tract injury and antiplatelet activity. Large outcome studies have documented that COX-2 selective inhibitors significantly decrease upper GI clinical events such as bleeding. 3 6 In addition, a post hoc analysis of 1 outcome study also suggested a significant decrease in lower GI clinical events. 7 However, no study with a prespecified evaluation of lower GI tract clinical events has been published. The MEDAL Program is a randomized comparison of the COX-2 selective NSAID etoricoxib and the traditional NSAID diclofenac in 34,701 osteoarthritis and rheumatoid arthritis patients followed for a mean duration of 18 months. 8 10 The primary end point was cardiovascular thrombotic events, but lower GI clinical events were prespecified as the primary GI end point. We sought to determine whether the COX-2 selective inhibitor decreased lower GI clinical events as compared with the traditional NSAID in a study simulating real-world practice with the use of low-dose aspirin and proton pump Abbreviations used in this paper: CI, confidence interval; COX, cyclooxygenase; CT, computed tomography; GI, gastrointestinal; HR, hazard ratio; RR, relative risk. 2008 by the AGA Institute 0016-5085/08/$34.00 doi:10.1053/j.gastro.2008.07.067

1518 LAINE ET AL GASTROENTEROLOGY Vol. 135, No. 5 inhibitor (PPI) therapy. We also planned to determine risk factors for the development of lower GI clinical events, because this has not previously been assessed in a large, prospective outcomes trial. Methods Study Design and Patients The design of the MEDAL Program and the results for cardiovascular and upper GI outcomes have been presented in detail elsewhere. 8 10 This program was conducted from June 2002, to May 2006, at 1380 sites in 46 countries. The MEDAL Program was prospectively designed to pool data from 3 randomized, double-blind, clinical trials of similar pattern: The MEDAL study, the Etoricoxib vs Diclofenac Sodium Gastrointestinal Tolerability and Effectiveness (EDGE) study, and the EDGE II study. The ethics committee for each study site approved the trial at that site and all patients provided written informed consent before participation. Patients 50 years of age with osteoarthritis of the knee, hip, hand, or spine or with rheumatoid arthritis were eligible for enrollment if they required chronic therapy with an NSAID. Patients meeting entry criteria were randomly assigned using concealed allocation to treatment in equal proportions within each study site using a different computer-generated randomization schedule for each of the 3 component trials. The first 4333 osteoarthritis patients and all rheumatoid arthritis patients in the MEDAL study received etoricoxib 90 mg daily or diclofenac 75 mg twice daily. The subsequent osteoarthritis patients enrolled in the MEDAL study received etoricoxib 60 mg daily (the current recommended dose for osteoarthritis) or diclofenac 75 mg twice daily. In EDGE and EDGE II, patients received etoricoxib 90 mg daily, diclofenac 50 mg thrice daily (EDGE), or diclofenac 75 mg twice daily (EDGE II). A double-dummy design along with coded study medications provided blinding to treatment assignment in all 3 trials. Low-dose aspirin ( 100 mg/day) was strongly recommended for cardiovascular prophylaxis in patients with established cardiovascular, peripheral arterial, or cerebrovascular disease and was also encouraged for patients with diabetes. Use of medical cotherapy (PPIs or misoprostol) was also strongly recommended for patients at high risk of upper GI clinical events (age 65, prior history of GI ulcer or hemorrhage, or concurrent use of corticosteroid, anticoagulant, or antiplatelet therapy). If low-dose aspirin or a PPI or misoprostol was not given to a patient meeting these criteria, investigators were contacted and required to state their reasons for not providing the medication. For the MEDAL study, omeprazole (20 mg) and low-dose aspirin were provided free of charge; low-dose aspirin was provided free of charge in the EDGE and EDGE II trials. Patients returned for visits every 4 months and a scheduled telephone contact was made between visits. Patient compliance with study medication was assessed by pill count. An independent data and safety monitoring board monitored safety data from all 3 trials at regular intervals. Investigators were not gastroenterologists, and no algorithm or specific tests were mandated by the study protocol for patients presenting with potential lower GI clinical events. End Points and Data Analysis Potential lower GI clinical events were identified through active surveillance of reported adverse events, and were adjudicated by an independent blinded committee using predefined criteria (Table 1). All such potential lower GI clinical events were evaluated by the adjudication committee, which classified the event as confirmed (if the case met the predefined criteria) or as unconfirmed (if the case did not meet the predefined criteria) based on the available clinical data. Episodes of perforation, obstruction, and bleeding of the small or large intestine that were confirmed by the adjudication Table 1. Prespecified Criteria for Lower GI Clinical Events Criteria Perforation a Obstruction a Bleeding a Diverticulitis Ulcer Hospitalization for small or large intestinal perforation confirmed by endoscopy, radiography (intraperitoneal air, contrast extravasation, contiguous abscess), surgery, autopsy, or histologic examination of the intestine Hospitalization for small or large intestinal obstruction (including partial) confirmed by endoscopy, radiography, surgery, autopsy, or histologic examination of the intestine 1. Hematochezia or melena that is not due to upper GI event and has significant bleeding b or requires hospitalization; 2. Active bleeding, visible vessel or adherent clot by endoscopy, angiography, scintigraphy, or surgery; 3. Occult blood-positive stool temporally associated with significant bleeding b and negative upper endoscopy Hospitalization for diverticulitis with clinical presentation and course consistent with diagnosis and confirmed by response to antibiotics, radiography, surgery, autopsy, or histologic examination of the colon Hospitalization for small or large intestinal ulcer confirmed by endoscopy, contrast radiography, surgery, autopsy, or histologic examination of the intestine a Included in primary end point of lower GI clinical events. b Hypotension, orthostatic changes in heart rate ( 20 beats per minute) or blood pressure ( 20 mmhg systolic or 10 mmhg diastolic), hemoglobin drop 2 g/dl, or transfusion.

November 2008 LOWER GI EVENTS WITH ETORICOXIB VS DICLOFENAC 1519 committee comprised the prespecified primary end point of lower GI clinical events. Two other exploratory lower GI outcomes were also prespecified and adjudicated by the committee using predefined criteria: hospitalization for intestinal ulcer and hospitalization for diverticulitis (Table 1). The exploratory analysis of diverticulitis was included because observational studies have reported an association between NSAID use and complicated diverticular disease (eg, perforation, pericolic abscess) or symptomatic diverticular disease, although diverticulitis alone was not assessed separately in these studies. 2 The potential mechanism for NSAID induction of diverticulitis is not known, but theoretically could involve NSAIDs causing the microperforation that leads to diverticulitis. Sensitivity analyses combining hospitalizations for ulceration and for diverticulitis with the primary lower GI end point (perforation, obstruction, and bleeding) were planned. In addition, a sensitivity analysis including all investigatorreported lower GI clinical events (those confirmed and not confirmed by the adjudication committee) was also planned. Analyses were performed in the intention-totreat population of all patients randomly assigned to study therapy. The primary hypothesis was that etoricoxib would be associated with fewer lower GI clinical events than diclofenac. The MEDAL Program was event driven and continued until at least 635 confirmed thrombotic cardiovascular events had occurred; the number of lower GI events was not prespecified and was determined by the time required to accrue the necessary cardiovascular events. However, we prespecified that 125 confirmed lower GI events would be required to perform the formal analysis. This was based on the following sample size calculation: assuming a hazard ratio (HR) for etoricoxib versus diclofenac of 0.60, 121 events would be required to have 80% power to show this difference was significant at a 2-sided.05. Event rates were summarized by number per 100 patient-years and associated 95% confidence intervals (CIs). The 95% CI for the rates per 100 patient-years were calculated with the Poisson distribution assumption. Kaplan Meier time-to-event curves were generated and truncated when the number of patients remaining at risk was 500. The primary test for comparison of the number of patients with lower GI clinical events between etoricoxib and diclofenac was based on a Cox proportional hazards model using aspirin use at baseline as a stratification factor. Between-treatment HRs with 95% CIs were calculated from the Cox model, accompanied by the test to examine that the proportional hazards assumption was not violated. In a preliminary sensitivity analysis, the proportional hazards assumption was tested by including the factor treatment*log(time) in the model; nonsignificance (P.05) of this factor implies proportionality, that is, constancy of treatment effect over time. A list of baseline clinical variables was specified for assessment as potential risk factors for lower GI clinical events. A post hoc analysis using the Cox proportional hazard model was applied with inclusion of these covariates in the model to identify characteristics associated with the development of lower GI clinical events. In addition, prespecified subgroup analyses were performed to assess the influence of baseline variables on the treatment effect of etoricoxib versus diclofenac. The subgroup analyses were performed by adding terms for the subgroup factor and its interaction with treatment to the Cox model. P.05 was considered significant. All analyses were performed in the intention-to-treat population of all patients randomized and followed until 14 days after discontinuation of study medication. Subgroup analyses were also performed related to concomitant use of low-dose aspirin and/or PPI cotherapy defined as concomitant therapy during 75% of the study period (and for patients with an event, 75% in the period before the event); 75% was chosen because it was the prespecified definition of compliance for study drug in the MEDAL Program. In addition, a subgroup analysis of all patients taking low-dose aspirin (defined as concomitant low-dose aspirin during 10% of the study period) was performed. Results of a post hoc analysis generally have more limited importance than a predefined analysis because they occur after data are unblinded and increase the number of analyses performed, thereby increasing the possibility that significance will be demonstrated by chance. Furthermore, it is recognized that subgroup analysis has less power than an analysis based on the whole patient population, even if the subgroups have been prespecified, because the larger number of analyses increase the possibility that significance will be seen by chance and the smaller number of subjects in each subgroup increases the likelihood that true differences will not be shown to be significant. Results A total of 34,701 patients were enrolled in the MEDAL Program (MEDAL study, 23,504 patients; EDGE, 7111 patients; EDGE II, 4086 patients), including 24,913 (71.8%) with osteoarthritis and 9787 (28.2%) with rheumatoid arthritis (information on arthritis type missing for 1 patient). Baseline characteristics were similar in the 2 study groups (Table 2). The Kaplan Meier estimates for lower GI clinical events (primary end point of bleeding, perforation, and obstruction confirmed by adjudication committee) and lower GI bleeding are shown in Figure 1. The cumulative incidence rates with etoricoxib compared with diclofenac satisfied the proportional hazard assumption, indicating a constant HR over time. The number of patients with lower GI clinical events and the rates per 100 patientyears are shown in Table 3. The number of patients with lower GI clinical events was not significantly less with

1520 LAINE ET AL GASTROENTEROLOGY Vol. 135, No. 5 Table 2. Selected Baseline Characteristics Etoricoxib (n 17,412) Diclofenac (n 17,289) Age (yr) Mean (SD) 63.2 (8.5) 63.2 (8.5) 65 10,178 (58.5%) 10,127 (58.6%) 65 to 75 5201 (29.9%) 5261 (30.4%) 75 2033 (11.7%) 1901 (11.0%) Women 12,925 (74.2%) 12,823 (74.2%) Osteoarthritis 12,533 (72.0%) 12,380 (71.6%) Rheumatoid arthritis 4878 (28.0%) 4909 (28.4%) Smoker 2034 (11.7%) 2037 (11.8%) Prior lower GI clinical event 416 (2.4%) 403 (2.3%) Prior upper GI clinical event 1127 (6.5%) 1133 (6.6%) Established atherosclerotic 2014 (11.6%) 2010 (11.6%) cardiovascular disease Diabetes 1810 (10.4%) 1855 (10.7%) Low-dose aspirin use 6030 (34.6%) 5976 (34.6%) Proton pump inhibitor use 6742 (38.7%) 6664 (38.5%) Traditional NSAID use 14,209 (81.6%) 14,174 (82.0%) COX-2 selective NSAID use 4873 (28.0%) 4939 (28.6%) Systemic corticosteroid use 2685 (15.4%) 2705 (15.6%) Methotrexate use 2762 (15.9%) 2831 (16.1%) Other DMARD use 2246 (12.9%) 2208 (12.8%) DMARD, disease-modifying anti-rheumatic drug; GI, gastrointestinal. Table 5. A significant treatment-by-subgroup interaction was identified for only 1 of the multiple variables tested gender. The HR for females was 0.64 (0.45 0.92) and for males it was 1.55 (0.90 2.68). Because the daily dose of etoricoxib for osteoarthritis patients in the MEDAL Study was changed from 90 to 60 mg after the first 4333 patients, we also separately assessed lower GI events in the group given 60 mg compared with those concurrently randomized to diclofenac and in the group given 90 mg compared with those concurrently randomized to diclofenac. The rates per 100 patient-years for patients given etoricoxib 60 mg versus concurrent diclofenac were 0.24 and 0.30 (relative risk [RR], 0.79; 95% CI, 0.48 1.30). The rates for etoricoxib 90 mg versus concurrent diclofenac were 0.54 and 0.53 (RR, 1.04; 95% CI, 0.64 1.68), raising the possibility that the initial 4333 osteoarthritis patients may have been higher risks than those enrolled subsequently. Subgroup analyses assessing the influence of concomitant low-dose aspirin therapy and PPI use 75% of the study period are shown in Table 6. Misoprostol was used by only 25 (0.07%) patients in the overall study popula- etoricoxib than diclofenac (HR, 0.84; 95% CI, 0.63 1.13). The sensitivity analysis that included confirmed hospitalizations for diverticulitis or intestinal ulcers revealed a similar result (HR, 0.85; 95% CI, 0.64 1.12). Sixty-seven patients had investigator-reported potential events that were not confirmed by the adjudication committee as lower GI events per study criteria; 65 of these patients had reports of bleeding. Among the 50 patients with overt bleeding, 1 patient had a gastric ulcer as the cause and the other 49 had neither a hospitalization nor the clinically significant bleeding necessary to qualify as an event. Among the 15 potential bleeding episodes without overt bleeding, 13 had occult bleeding without hospitalization for the bleeding or clinically significant bleeding, and 2 had anemia without a positive fecal occult blood test. Two patients with unconfirmed events were submitted as obstructions, but imaging studies and clinical data did not confirm this diagnosis. The sensitivity analyses including all investigator-reported lower GI clinical events (those confirmed and not confirmed as lower GI events by the adjudication committee) showed 24% 25% relative risk reductions for etoricoxib versus diclofenac that were significant (Table 3). The post hoc multivariable analysis to identify significant independent predictors of lower GI clinical events (Table 4) revealed that a prior lower GI clinical event was the most important predictor of future events with approximately a 4-fold increase risk. Older age was the other significant predictor, with a 2-fold increase in risk. Subgroup analyses assessing the influence of baseline characteristics on the treatment effect of etoricoxib versus diclofenac for lower GI clinical events are shown in Figure 1. Time-to-event analyses in the intent-to-treat population. (A) Cumulative incidence of lower GI clinical events (bleeding, perforation, or obstruction confirmed by adjudication committee). (B) Cumulative incidence of lower GI bleeding events confirmed by the adjudication committee.

November 2008 LOWER GI EVENTS WITH ETORICOXIB VS DICLOFENAC 1521 Table 3. Patients With Lower GI Clinical Events Etoricoxib (n 17,412) Diclofenac (n 17,289) n (%) Rate a n (%) Rate a Hazard ratio (95% CI) Primary lower GI end point of bleeding, perforation, and obstruction Confirmed events 84 (0.48) 0.32 96 (0.56) 0.38 0.84 (0.63 1.13) All investigator-reported events 109 (0.63) 0.41 138 (0.80) 0.54 0.76 (0.59 0.98) Lower GI end point of bleeding, perforation, obstruction, diverticulitis, and ulcer Confirmed events 95 (0.55) 0.36 108 (0.62) 0.43 0.85 (0.64 1.12) All investigator-reported events 122 (0.70) 0.46 156 (0.90) 0.61 0.75 (0.59 0.95) a Events per 100 patient-years. tion, so its effects on lower GI events could not be assessed. For patients who experienced a lower or upper GI event to be considered regular aspirin or PPI users, they were required to have taken concomitant therapy for 75% of the period before the event. Therefore, the denominators in the aspirin and PPI subgroups for the lower GI clinical event analyses are slightly different than the values previously published for upper GI events. 10 Treatment-by-subgroup interactions were not significant for aspirin or PPI use, indicating that the treatment effect for etoricoxib versus diclofenac was consistent with and without low-dose aspirin and with and without PPI use. In addition, we assessed the influence of any concomitant aspirin use, defined as 10% of the study period. The rate of lower GI events per 100 patient-years in this population using low-dose aspirin was 0.49 in the etoricoxib group and 0.47 in the diclofenac group (RR, 1.04; 95% CI, 0.70 1.57); among the nonaspirin users by this definition the rates were 0.21 and 0.32 for etoricoxib versus diclofenac (RR, 0.66; 95% CI, 0.43 1.02). The absolute risk increases for aspirin versus no aspirin were 0.42 (95% CI, 0.17 0.67) in the etoricoxib group Table 4. Baseline Predictors of Lower GI Clinical Events a : Hazard Ratios (95% CIs) of the Effect of Individual Baseline Characteristics Baseline variable Hazard ratio (95% CI) Prior lower GI clinical event 4.06 (2.93 0.62) Age 65 1.98 (1.45 2.71) Low-dose aspirin use 1.39 (0.99 1.95) Smoker 1.46 (0.95 2.25) Systemic corticosteroid use 1.38 (0.86 2.20) Atherosclerotic cardiovascular disease 1.24 (0.83 1.85) Proton pump inhibitor use 1.12 (0.83 1.52) Diabetes 1.11 (0.71 1.73) Male gender 1.06 (0.77 1.47) Osteoarthritis (vs rheumatoid arthritis) 0.91 (0.57 1.47) Prior colonic neoplasm 0.91 (0.44 1.92) Prior upper GI clinical event 0.87 (0.51 1.50) Methotrexate use 0.83 (0.48 1.45) History of dyspepsia 0.74 (0.41 1.33) a Bleeding, perforation, and obstruction confirmed by adjudication committee. and 0.30 (95% CI, 0.04 0.56) in the diclofenac group. However, differences between these 2 nonrandomized, unmatched cohorts might also be due to confounding factors that influence the rate of lower GI events unrelated to aspirin use, and therefore such a comparison is problematic. The multivariable analysis, which adjusts for other factors, showed a trend to a small increase in risk with low-dose aspirin (RR, 1.39; 95% CI, 0.99 1.95). A breakdown of the component lower GI clinical events is shown in Table 7. Patients who had 1 type of clinical event (eg, bleeding and obstruction, perforation owing to diverticulitis) are listed in each category of event Table 5. Rates of Lower GI Clinical Events a and Treatment Effects Related to Selected Baseline Characteristics Rates per 100 patient-years Etoricoxib Diclofenac Hazard ratio (95% CI) Prior LGI event 2.05 2.56 0.82 (0.37 1.79) No prior LGI event 0.28 0.33 0.84 (0.61 1.15) Age (yr) 65 0.22 0.24 0.94 (0.59 1.49) 65 75 0.40 0.53 0.76 (0.47 1.21) 75 0.65 0.84 0.79 (0.41 1.53) Aspirin 0.49 0.51 0.96 (0.63 1.45) No aspirin 0.23 0.31 0.74 (0.49 1.12) Cigarette smoker 0.36 0.46 0.78 (0.36 1.73) Nonsmoker 0.31 0.37 0.85 (0.62 1.17) Corticosteroid 0.31 0.38 0.82 (0.60 1.14) No corticosteroid 0.35 0.38 0.93 (0.47 1.84) ASCVD 0.57 0.65 0.88 (0.45 1.72) No ASCVD 0.29 0.34 0.84 (0.60 1.16) Female b 0.26 0.40 0.64 (0.45 0.92) Male 0.50 0.32 1.55 (0.90 2.68) Prior UGI event 0.32 0.37 0.84 (0.64 1.17) No prior UGI event 0.35 0.56 0.63 (0.22 1.76) Rheumatoid arthritis 0.26 0.37 0.69 (0.39 1.20) Osteoarthritis 0.35 0.38 0.91 (0.64 1.29) ASCVD, atherosclerotic cardiovascular disease; LGI, lower gastrointestinal; PPI, proton pump inhibitor; UGI, upper gastrointestinal. a Bleeding, perforation, and obstruction confirmed by adjudication committee. b Significant treatment-by-subgroup interaction (P.008).

1522 LAINE ET AL GASTROENTEROLOGY Vol. 135, No. 5 Table 6. Lower GI Clinical Events a Related to Concomitant Use of Low-Dose Aspirin and PPIs 75% of Study Period Etoricoxib Diclofenac n/n (%) Rate b n/n (%) Rate b Hazard ratio (95% CI) Aspirin use (P.30 c ) Yes 44/5747 (0.77) 0.50 43/5673 (0.76) 0.51 0.99 (0.65 1.50) No 40/11,665 (0.34) 0.23 53/11,616 (0.46) 0.31 0.73 (0.48 1.09) PPI use (P.98 c ) Yes 44/6951 (0.63) 0.36 50/6910 (0.72) 0.43 0.85 (0.56 1.27) No 40/10,461 (0.38) 0.28 46/10,379 (0.44) 0.34 0.84 (0.55 1.28) CI, confidence interval; PPI, proton pump inhibitor. a Bleeding, perforation, and obstruction confirmed by adjudication committee. b Events per 100 patient-years. c Treatment-by-subgroup interaction. they developed. Bleeding was the most common lower GI event, followed by diverticulitis, perforation, and obstruction. The rates for each of the component events were slightly higher for diclofenac than etoricoxib, but in each case except ulcers, the 95% CIs of the HRs for etoricoxib versus diclofenac included 1 (Table 7). Lower GI bleeding in the etoricoxib group occurred in 28 (0.49%) of 5673 patients taking aspirin and 23 (0.20%) of 11,616 not taking aspirin; in the diclofenac group, it occurred in 29 (0.50%) of 5747 taking aspirin and 30 (0.26%) of 11,665 not taking aspirin. Among the 110 cases of bleeding, 16 were occult by the predefined criteria. All 16 had upper endoscopy and 15 had colonoscopy performed (with the 16th having had a negative colonoscopy within the past year). Three of the 16 underwent capsule endoscopy. Six of the 16 cases did not have a diagnosis for the cause of their bleeding determined. Among the 94 patients with overt lower GI bleeding, only 5 had no investigation recorded: 1 had bleeding immediately after a polypectomy, 1 had a prior colonoscopy for hematochezia that showed diverticula as the cause, 1 was on vacation in India and was asymptomatic upon return, 1 was not felt to require evaluation by the investigator, and no other follow-up information was available on the 5th patient. Diverticula were the most common cause of bleeding accounting for 28 cases with etoricoxib (14 with aspirin and 14 without aspirin) and 14 cases with diclofenac (9 with aspirin and 5 without aspirin). Among the 42 cases of bleeding due to diverticula, 34 had diagnosis by colonoscopy, 1 by flexible sigmoidoscopy, 2 by prior colonoscopy showing diverticula, 2 by barium enema, 2 by laparotomy, and 1 by computed tomography (CT) plus angiography. The cause of bleeding was attributed to NSAID ulcers or colitis/enteritis in 2 patients taking etoricoxib (1 aspirin, 1 without aspirin) and 12 patients taking diclofenac (6 aspirin, 6 no aspirin). Among these 14 cases, 6 had colonic ulcers, 2 had ileocecal valve ulcers, and 1 had ileal ulcers diagnosed by colonoscopy; 4 had NSAID colitis diagnosed by colonoscopy; and 1 had NSAID enteritis diagnosed by capsule endoscopy. The cause of bleeding was unknown in 10 patients in the etoricoxib group and 14 in the diclofenac group. Other less frequent causes of bleeding reported were colitis, ischemic bowel, inflammatory bowel disease, hemorrhoids, vascular ectasia, anal fissure, postpolypectomy or colonic resection bleeding, and aortoenteric fistula. Among the 96 cases with known causes of bleeding, 24 (5 of 41 in the etoricoxib group and 19 of 55 in the diclofenac group) had bleeding from lesions presumably caused de novo by NSAIDs (assuming NSAIDs cause ulcers, enteritis, colitis, and also cause inflammatory bowel disease or make subclinical disease clinically manifest). If we assume that cases labeled as ischemic enteritis or colitis are also caused de novo by NSAIDs, then 29 cases (5/41 with etoricoxib and 24/55 with diclofenac) were due to lesions caused de novo by NSAIDs. Table 7. Types of Confirmed Lower GI Clinical Events Etoricoxib (n 17,412) Diclofenac (n 17,289) n (%) Rate a (95% CI) n (%) Rate a (95% CI) Hazard Ratio (95% CI) Bleeding 51 (0.29) 0.19 (0.14 0.25) 59 (0.34) 0.23 (0.18 0.30) 0.83 (0.57 1.21) Perforation 21 (0.12) 0.08 (0.05 0.12) 24 (0.14) 0.09 (0.06 0.14) 0.84 (0.47 1.51) Obstruction 16 (0.09) 0.06 (0.03 0.10) 19 (0.11) 0.07 (0.05 0.12) 0.81 (0.42 1.57) Diverticulitis 29 (0.17) 0.11 (0.07 0.16) 32 (0.19) 0.13 (0.09 0.18) 0.87 (0.53 0.144) Ulcer 3 (0.02) 0.01 (0.00 0.03) 11 (0.06) 0.04 (0.02 0.08) 0.26 (0.07 0.94) Patients with 1 type of event are listed in each category of event they developed. a Events per 100 patient-years.

November 2008 LOWER GI EVENTS WITH ETORICOXIB VS DICLOFENAC 1523 Perforations in the etoricoxib group occurred in 12 (0.21%) taking aspirin and 9 (0.08%) not taking aspirin; for diclofenac, perforations occurred in 8 (0.14%) taking aspirin and 16 (0.14%) not taking aspirin. Surgery was performed in 43 of the 45 cases and the other 2 had diverticulitis diagnosed by CT and were treated conservatively. Diverticulitis was the cause of perforations in 36 (80%) of the 45 cases. Ulceration accounted for 3 cases: All had surgical management with the ulcerations ascribed to colitis in 2 cases and radiation enteritis (with small intestinal perforation) in 1 case. Other causes accounted for the remaining 6 perforation cases. Obstructions in the etoricoxib group occurred in 8 (0.14%) with aspirin and 8 (0.07%) without aspirin; the values were 8 (0.14%) with aspirin and 11 (0.09%) without aspirin in the diclofenac group. Surgery was performed in 25 of the 35 cases of obstruction; among the 10 treated conservatively, 8 had diagnosis by CT, 1 had diagnosis of small bowel obstruction by abdominal series alone, and 1 had diagnosis of colonic obstruction by barium enema. Obstruction was ascribed to adhesions in 21 (60%) of the 35 cases: 13 were based on surgical findings and 8 were based on imaging (CT scan in 7, abdominal series in 1) plus clinical data. Eight cases of obstruction were ascribed to diverticulitis (6 based on surgical findings, 1 based on CT, and 1 based on barium enema and colonoscopy), 4 small bowel obstructions were due to incarcerated hernias (all documented at surgery), and 2 cases were ascribed to other causes. Discussion NSAIDs increase the risk of lower GI clinical events such as bleeding or perforation, based on observational studies. 2 Although a great number of trials have examined management strategies to decrease the risk of NSAID-associated upper GI clinical events, a paucity of information is available to determine which therapies reduce lower GI events. Antisecretory therapy would not be expected to decrease the risk of lesions beyond the duodenum. No randomized controlled trials have directly assessed antisecretory therapy for lower GI events, but video capsule endoscopy studies report high rates of small intestinal mucosal breaks even when PPIs are given with traditional NSAIDs. 11,12 If inhibition of COX-1 mediated prostaglandin production in the lower GI tract plays a role in NSAIDinduced lower GI injury, then use of misoprostol or COX-2 selective inhibitors might be expected to have a therapeutic benefit. No randomized trials have assessed lower GI outcomes with misoprostol. A nonrandomized, controlled trial in 21 NSAID users with iron-deficiency anemia and small intestinal lesions identified a significant increase in hemoglobin with misoprostol but no increase in those not receiving misoprostol. 13 Somewhat more information is available regarding the effect of COX-2-selective inhibitors on the lower GI tract. Randomized trials assessing surrogate outcomes demonstrate that COX inhibitors have less effect on intestinal permeability and cause fewer mucosal breaks in the small intestine than traditional NSAIDs. 2 A post hoc analysis of lower GI clinical events in a previous GI outcome study revealed a significant decrease in events with rofecoxib versus naproxen (0.41% vs 0.89%). 7 In addition, reports of hematochezia in the first 6 months of another outcomes trial were significantly less frequent with celecoxib than with traditional NSAIDs, ibuprofen and diclofenac (0.4% vs 1.0%). 14 We prespecified assessment of lower GI clinical events, defined criteria for these events a priori, collected relevant clinical data prospectively, and had a blinded adjudication committee review all potential events. In this predefined comparison, a statistically significant decrease in the incidence of lower GI clinical events was not seen with the COX-2 selective inhibitor etoricoxib as compared with the traditional NSAID diclofenac. When all investigator-reported lower GI events were assessed, a slightly larger relative risk reduction of 24% was noted, with 95% CIs that did not cross unity. Because the diagnostic evaluation of lower GI clinical events is often more difficult and less likely to yield a definitive diagnosis, one might speculate that a small significant difference was missed owing to a number of cases not being able to be adjudicated as confirmed lower GI events based on insufficient or incomplete data. However, the proportion of all investigator-reported lower GI events that were confirmed by the adjudication committee was 77% with etoricoxib and 70% for diclofenac. These rates are slightly lower than the rates for confirmation of upper GI events in the MEDAL program of 88% and 87%. Our results suggest that a COX-2-selective inhibitor does not decrease lower GI clinical events as compared with a traditional NSAID or that the decrease is small and would require even a larger sample size to document significance. Even if a 16% relative risk reduction had been shown to be significant or the 24% relative risk reduction in investigator-reported events is correct, it is uncertain that this degree of benefit would be considered clinically meaningful. As mentioned, a post hoc analysis of a previous large outcome study did identify a significant 54% relative risk reduction in lower GI clinical events with the COX-2 selective inhibitor rofecoxib versus the traditional NSAID naproxen. 7 The difference in results could relate to the relative antiplatelet effects of the 2 comparator traditional NSAIDs, diclofenac and naproxen. Greater than 95% inhibition of COX-1 mediated thromboxane is required to impact platelet function. 15,16 Diclofenac s inhibition peaks at a mean of 87%, 17 which is not sufficient to meaningfully decrease platelet function in most patients. 18 If the antiplatelet effect of NSAIDs is a major factor in induction of lower GI bleeding, then one might expect little difference between NSAIDs that do not pro-

1524 LAINE ET AL GASTROENTEROLOGY Vol. 135, No. 5 duce significant antiplatelet activity, whether a COX-2 selective inhibitor such as etoricoxib or a traditional NSAID such as diclofenac. In contrast, naproxen is a more potent inhibitor of COX-1 mediated thromboxane with greater antiplatelet effect, 18 potentially explaining why a comparison between a COX-2-selective inhibitor and naproxen would show a significant difference in lower GI events. 7 In the previous comparison of rofecoxib and naproxen, the difference in lower GI events was largely driven by a difference in bleeding (0.3% vs 0.7%), whereas in the current comparison bleeding was nearly identical in the study groups (0.2% with etoricoxib and diclofenac). The results of our analysis apply to the specific drugs studied, and as mentioned, trials with other NSAIDs may produce different outcomes. The influence of aspirin intake also must be considered given the potential importance of the antiplatelet effect of medications in the development of lower GI bleeding. The previous study of lower GI outcomes mentioned 7 did not allow aspirin, whereas low-dose aspirin was allowed in our studies. The potent antiplatelet effect of aspirin may increase the rate of bleeding in both treatment groups, potentially diluting any small difference in bleeding between the 2 NSAIDs in the population using aspirin. The rates of events were nearly identical for etoricoxib versus diclofenac in the group taking aspirin during the study, with an HR of 0.99 (95% CI, 0.65 1.50), whereas the HR in those not using aspirin was 0.73 (95% CI, 0.48 1.09). However, the difference in these treatment effects was not statistically significant. The population-based incidence of hospitalization for lower GI bleeding in the United States is 20.5 per 100,000 person-years 17% of the combined rate of upper and lower GI bleeding hospitalizations (122.5 per 100,000 person-years). 19,20 Similar results are reported from Spain: The incidence of hospitalization for lower GI clinical events (primarily bleeding) was reported to be at least 13% 15% of the incidence of hospitalizations for all GI events, which was estimated at 121.9 events per 100,000 person-years. 21 Previous studies suggest that among traditional NSAID users, similar to the general population, upper GI clinical events are more common than lower GI clinical events, with lower GI events representing 13% 40% of all GI events. 7,22 25 In contrast, 50% of GI complications (bleeding, perforation, and obstruction) were due to lower GI events in the MEDAL Program: 0.38 lower and 0.32 upper GI events per 100 patient-years with diclofenac. 10 This seems to be largely due to a much lower rate of upper GI events in the MEDAL program than seen in previous studies with traditional NSAIDs ( 1 1.5 events per 100 patientyears 1 ). Potential explanations include frequent use of PPIs (not allowed in prior GI outcome studies), lower risk patients, and lower risk of upper GI mucosal injury and/or less antiplatelet activity with study medications. 10 Although most episodes of NSAID-associated upper GI bleeding are due to NSAID-induced lesions (ulcers), we found that a majority of lower GI bleeding cases in NSAID users were not due to lesions induced de novo by NSAIDs. Rather bleeding commonly occurred from preexisting conditions that became clinically manifest during NSAID use, such as diverticula, vascular ectasias, or hemorrhoids. Diverticula were the most important etiology of lower GI clinical events. Diverticulitis accounted for 80% of perforations and nearly 25% of obstructions, and diverticula were reported to be the source of nearly 40% of lower GI bleeding episodes. As seen in Figure 1, the risk of developing a lower GI clinical event appeared constant over time, at least up to 3 years of NSAID use. Thus, no drop-off in risk can be expected with continued NSAID use over 3 years, similar to the findings with upper GI clinical events in other prospective studies. 4,10,26 Our study had no placebo group and no previous placebo-controlled, prospective, randomized trial assessing lower GI events has been performed. We are therefore not able to document the presence or the magnitude of any increase in lower GI clinical events with NSAID therapy versus placebo or no therapy. A number of studies have assessed the risk factors for the development of upper GI complications, but few data are available on risk factors for lower GI clinical events owing to the lower frequency and the lack of prospective trials designed to assess lower GI complications. Given the large number of lower GI events in the MEDAL Program (n 180) and the prospective collection of relevant data, we were able to assess potential risk factors in a multivariable analysis. We found that a prior lower GI event and older age were the 2 independent risk factors for development of a lower GI clinical event with relative risk increases of 4-fold and 2-fold, respectively. Use of low-dose aspirin has been suggested to increase the risk of lower GI events, especially via its potent antiplatelet activity. The modest 35% relative risk increase with low-dose aspirin in our analysis did not quite achieve statistical significance the lower bound of the 95% CI (0.99) just crossed unity. When looking only at lower GI bleeding, the crude incidence of bleeding was 2- to 2.5-fold higher with aspirin versus no aspirin in the diclofenac and etoricoxib groups. In conclusion, lower GI clinical events (bleeding, perforation, and obstruction) accounted for just 50% of the GI complications in the MEDAL Program. A statistically significant decrease in lower GI clinical events was not seen with the COX-2 selective inhibitor etoricoxib versus the traditional NSAID diclofenac. The effect of traditional NSAIDs on lower GI bleeding may vary in relation to their antiplatelet activity. The risk of developing a lower GI clinical event with NSAID use remains constant over time, and the major risk factors are a prior lower GI event and older age.

November 2008 LOWER GI EVENTS WITH ETORICOXIB VS DICLOFENAC 1525 References 1. Laine L. Proton pump inhibitor co-therapy with nonsteroidal antiinflammatory drugs nice or necessary? Rev Gastroenterol Disord 2004;4(Suppl 4):S33 41. 2. Laine L, Smith R, Min K, et al. Systematic review: lower gastrointestinal adverse effects of nonsteroidal anti-inflammatory drugs. Aliment Pharmacol Ther 2006;24:751 767. 3. FDA Arthritis Advisory Committee meeting. February 7, 2001. Available at: http://www.fda.gov/ohrms/dockets/ac/01/briefing/ 3677b1.htm. Accessed June 15, 2008. 4. Bombardier C, Laine L, Reicin A, et al. Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. N Engl J Med 2000;343:1520 1528. 5. Schnitzer TJ, Burmester GR, Mysler E, et al. Comparison of lumiracoxib with naproxen and ibuprofen in the Therapeutic Arthritis Research and Gastrointestinal Event Trial (TARGET), reduction in ulcer complications: randomised controlled trial. Lancet 2004;364:665 674. 6. Singh G, Fort JG, Goldstein JL, et al. Celecoxib versus naproxen and diclofenac in osteoarthritis patients: SUCCESS-I Study. Am J Med 2006;119:255 266. 7. Laine L, Connors LG, Reicin A, et al. Serious lower gastrointestinal clinical events with non-selective NSAID or coxib use. Gastroenterology 2003;124:288 292. 8. Cannon CP, Curtis SP, Bolognese JA, et al. Clinical Trial Design and Patient Demographics of the Multinational Etoricoxib and Diclofenac Arthritis Long-term (MEDAL) Study Program: cardiovascular outcomes with etoricoxib vs. diclofenac in patients with osteoarthritis and rheumatoid arthritis. Am Heart J 2006;152:237 245. 9. Cannon CP, Curtis SP, FitzGerald GA, 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. 10. Laine L, Curtis SP, Cryer B, et al. Assessment of upper gastrointestinal safety of 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 2007;369:465 473. 11. Goldstein JL, Eisen GM, Lewis B, et al. Video capsule endoscopy to prospectively assess small bowel injury with celecoxib, naproxen plus omeprazole, and placebo. Clin Gastroenterol Hepatol 2005;3:133 141. 12. Goldstein JL, Eisen GM, Lewis B, et al. Small bowel mucosal injury is reduced in healthy subjects treated with celecoxib compared with ibuprofen plus omeprazole, as assessed by video capsule endoscopy. Aliment Pharmacol Ther 2007;25:1211 1222. 13. Morris AJ, Murray L, Sturrock RD, et al. Short report: the effect of misoprostol on the anaemia of NSAID enteropathy. Aliment Pharmacol Ther 1994;8:343 346. 14. Silverstein F, 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. JAMA 2000;284:1247 1255. 15. Reilly IAG, FitzGerald GA. Inhibition of thromboxane formation in vivo and ex vivo: implications for therapy with platelet inhibitory drugs. Blood 1987;69:180 186. 16. Grosser T, Fries S, FitzGerald GA. Biological basis for the cardiovascular consequences of COX-2 inhibition: therapeutic challenges and opportunities. J Clin Invest 2006;116:4 15. 17. Catella-Lawson F, Reilly MP, Kapoor SC, et al. Cyclooxygenase inhibitors and the antiplatelet effects of aspirin. N Engl J Med 2001;345:1809 1817. 18. Van Hecken A, Schwartz JI, Depré M, et al. Comparative inhibitory activity of rofecoxib, meloxicam, diclofenac, ibuprofen and naproxen on COX-2 versus COX-1 in healthy volunteers. J Clin Pharmacol 2000;40:1 12. 19. Longstreth GF. Epidemiology and outcome of patients hospitalized with acute lower gastrointestinal hemorrhage: a populationbased study. Am J Gastroenterol 1997;92:419 424. 20. Longstreth GF. Epidemiology of hospitalization for acute upper gastrointestinal hemorrhage: a population-based study. Am J Gastroenterol 1995;90:206 210. 21. Lanas A, Perez-Aisa MA, Feu F, et al. A nationwide study of mortality associated with hospital admission due to severe gastrointestinal events and those associated with nonsteroidal antiinflammatory drug use. Am J Gastroenterol 2005;100:1685 1693. 22. Wilcox CM, Alexander LN, Cotsonis GA, et al. Nonsteroidal antiinflammatory drugs are associated with both upper and lower gastrointestinal bleeding. Dig Dis Sci 1997;42:990 997. 23. Lanas A, Serrano P, Bajador E, et al. Evidence of aspirin use in both upper and lower gastrointestinal perforation. Gastroenterology 1997;112:683 689. 24. Lanas A, Sekar MC, Hirschowitz BI. Objective evidence of aspirin use in both ulcer and nonulcer upper and lower gastrointestinal bleeding. Gastroenterology 1992;103:862 869. 25. Singh G, Rosen RD. NSAID induced gastrointestinal complications: the ARAMIS perspective 1997. Arthritis, Rheumatism, and Aging Medical Information System. J Rheumatol Suppl 1998;51:8 16. 26. Kurata JH, Abbey DE. The effect of chronic aspirin use on duodenal and gastric ulcer hospitalizations. J Clin Gastroenterol 1990;12:260 266. Received March 12, 2008. Accepted July 24, 2008. Address requests for reprints to: Loren Laine, MD, Room 12-137, L.A. County U.S.C. Medical Center, 1200 N. State St, Los Angeles, CA 90033. E-mail: llaine@usc.edu; fax: 323-226-7573. Statistical analysis of the entire data sets pertaining to the primary safety end point of lower GI clinical events has been independently confirmed by Frontier Science Foundation (Madison, WI), under the supervision of C. Morton Hawkins, ScD and David DeMets, PhD. The corresponding author had full access to all of the data and takes full responsibility for the veracity of the data and analysis. The authors disclose the following: Supported by Merck & Co, Inc, Whitehouse Station, New Jersey. L. Laine: Research support: Merck, TAP, Novartis; Consultant: Merck, Novartis, AstraZeneca, Horizon, Santarus; DSMB: Pfizer, Bayer. S. Curtis, A. Kaur: Employees of Merck (own stock and/or stock options). M. Langman: Adjudication Committee and consultant: Merck; Symposium speaker: Bayer; Past consultant: Galpharm, Novartis, Procter & Gamble. D. Jensen: Adjudication Committee: Merck; Research support: AstraZeneca; Consultant: Wyeth, AstraZeneca. B. Cryer: Consultant: Merck, Pfizer, AstraZeneca, TAP. C. Cannon: Research support: Accumetrics, AstraZeneca, Merck, Schering Plough Consultant for or speaker at symposia sponsored by Alnylam, AstraZeneca, Biosite, Bristol Myers Squibb, Eisai, Glaxo SmithKline, Merck, Pfizer, Sanofi Aventis, Schering Plough, Tethys. The MEDAL Program was designed cooperatively by the sponsor (Merck Research Laboratories) and the Program Steering Committee, which is composed of experts in gastroenterology, cardiovascular medicine, rheumatology, pharmacology, statistical sciences, and epidemiology. The sponsor monitored the study, collected data, and performed statistical analysis. The corresponding author wrote the manuscript with input and review from the other authors but without any outside writing assistance. The decision to submit the report for publication was made by the corresponding author with agreement by the other authors. MEDAL Program trials registered at http://www.clinicaltrials. gov. Clinical trial registry numbers: NCT00092703; NCT00092742; NCT00250445.