The incidence of and mortality from upper gastrointestinal

Transcription

1 GASTROENTEROLOGY 2005;129: Risk of Serious Upper Gastrointestinal Toxicity With Over-the- Counter Nonaspirin Nonsteroidal Anti-inflammatory Drugs JAMES D. LEWIS,*,, STEPHEN E. KIMMEL,*,, A. RUSSELL LOCALIO,*, DAVID C. METZ, JOHN T. FARRAR,*,, LISA NESSEL,* COLLEEN BRENSINGER,*, KAREN MCGIBNEY,* and BRIAN L. STROM*,,, *Center for Clinical Epidemiology and Biostatistics, Department of Medicine, Department of Biostatistics and Epidemiology, and Department of Pharmacology, University of Pennsylvania, Philadelphia, Pennsylvania Background & Aims: Use of prescription nonaspirin nonsteroidal anti-inflammatory drugs (NANSAIDs) increases the risk of serious upper gastrointestinal toxicity. Less is known about over-the-counter (OTC) NANSAIDs, which are typically used at lower doses and for shorter durations. This study assessed the risk of toxicity with OTC NANSAIDs. Methods: A total of 359 case subjects hospitalized for upper gastrointestinal bleeding, perforation, or benign gastric outlet obstruction were recruited from 28 hospitals. A total of 1889 control subjects were recruited by random digit dialing from the same region. Data on medication use were collected via structured telephone interview. Results: Use of OTC NANSAIDs on >4 days during the most recent week had an adjusted odds ratio (OR) of 1.83 (95% confidence interval [CI], ). Use of high-dose OTC NANSAIDs during the index week had an adjusted OR of 5.21 (95% CI, ). In contrast, use of OTC NANSAIDs <4 times during the index week (adjusted OR, 0.67; 95% CI, ) and use of very low doses of prescription or OTC NANSAIDs during the index week (adjusted OR, 0.74; 95% CI, ) were not significantly associated with an increased risk of serious gastrointestinal toxicity. We did not observe a significant difference between the risk of toxicity with OTC naproxen versus OTC ibuprofen (adjusted OR, 0.84; 95% CI, ). Conclusions: Use of OTC NANSAIDs at recommended doses has a relatively good safety profile compared with prescription NANSAIDs. However, use of high-dose OTC NANSAIDs (comparable to a prescription dose) is associated with serious gastrointestinal toxicity. The incidence of and mortality from upper gastrointestinal bleeding have declined in recent decades. 1 Peptic ulcer disease is the most common cause of upper gastrointestinal bleeding. Most peptic ulcers are caused by consumption of aspirin and nonaspirin nonsteroidal anti-inflammatory drugs (NANSAIDs) and/or Helicobacter pylori infection. 2 5 In the United States, NANSAIDs are available by prescription or in lower-strength over-the-counter (OTC) preparations. Prior research related to gastrointestinal toxicity of NANSAIDs primarily focused on prescription use and consistently documented a higher risk of gastrointestinal bleeding, gastrointestinal surgery, and gastrointestinal-related mortality. 6 8 However, the risk of gastrointestinal complications varies with duration of use and NANSAID dosage and among different prescription NANSAIDs The extent to which lower-dose OTC NANSAIDs are associated with upper gastrointestinal toxicity is unknown. This study specifically addressed the risk of serious upper gastrointestinal toxicity from OTC NANSAIDs. As a secondary aim, we compared the risk of gastrointestinal toxicity between the 2 most commonly used OTC NANSAIDs (ibuprofen and naproxen). Materials and Methods Study Subjects We conducted a case-control study within the Delaware Valley Case Control Network. All hospitals in Philadelphia and the 8 surrounding counties were asked to participate. Twenty-eight hospitals approved the study and referred patients for inclusion. We obtained institutional review board approval in 17 additional hospitals but were unable to establish a recruitment system. To be eligible for inclusion as a case or control subject, participants had to reside in the 9-county region, be between 22 and 80 years old, have a telephone, and be able to complete a 30-minute interview. Case subjects had to be interviewed within 12 weeks of hospital admission (median, 38 days). In rare cases, interviews were completed by proxies if the patient had died or was too ill to complete the interview. Abbreviations used in this paper: CI, confidence interval; ICD-9, International Classification of Diseases, Ninth Revision; NANSAID, nonaspirin nonsteroidal anti-inflammatory drug; OR, odds ratio; OTC, overthe-counter by the American Gastroenterological Association /05/$30.00 doi: /j.gastro

2 1866 LEWIS ET AL GASTROENTEROLOGY Vol. 129, No. 6 Selection of Cases Cases included all patients admitted to the hospital for serious gastrointestinal toxicity, including acute upper gastrointestinal bleeding, perforation, or newly diagnosed gastric outlet obstruction with nonmalignant etiology. We applied the same definitions of these conditions as was described by Bombardier et al. 11 Case subjects were asked for written consent to allow the research team to review their medical records. Records were used to confirm that the patient met our inclusion criteria and establish the final diagnosis of gastrointestinal toxicity etiology. International Classification of Diseases, Ninth Revision (ICD-9) discharge diagnosis codes were requested from participating hospitals. We selected ICD-9 discharge diagnoses a priori that we believed would be strongly predictive of meeting our inclusion criteria (456.0, , 530.2, 530.4, 530.7, , xx, 535.0, , , , , 578.0, 578.1, and 578.9). We tested the positive predictive value of these codes by examining patients for whom we had ICD-9 codes and had reviewed the hospital record. A code was considered strongly predictive if the positive predictive value was 75% (530.7, , 531.xx, 532.xx, 534.xx, 535.3, , and 578.0). We excluded patients for whom chart review was unavailable and review of ICD-9 discharge diagnosis codes did not identify one of our prespecified codes with a positive predictive value of at least 75%. Patients without chart review or ICD-9 discharge diagnosis codes were included in primary analyses (n 20). Secondary analyses excluding these patients gave similar results and are not reported. Selection of Controls DataStat, Inc (Ann Arbor, MI) recruited controls from the same source population using random digit dialing. 12,13 If more than 1 potentially eligible control was available in a household, 1 subject was randomly selected for inclusion. Names and contact information for subjects agreeing to participate were referred to the University of Pennsylvania for the interview. Data Collection Data from cases and controls were collected using structured interviews administered by telephone. Interviews were completed using a structured standardized interview that was administered by experienced, trained interviewers following strict guidelines developed for this study. Case subjects were interviewed after discharge from the hospital so as to allow them to have access to their medications. For cases, the hospitalization date was the index date; for controls, the interview date was the index date. All exposures were measured backward from the index date. Data on exposure to NAN- SAIDs were obtained using a 3-stage process In stage I, patients were asked about medication use for symptoms commonly treated with NANSAIDs (eg, headache) in the preceding 8 weeks. In stage II, patients were shown a booklet (mailed before the interview) containing pictures of prescription and OTC NANSAIDs 18 available in the United States. While reviewing the pictures, patients were read the corresponding NANSAID names. In stage III, patients were asked about additional medication used in the preceding 8 weeks. Subjects were sent calendars as a memory aid. Pictures were also provided of commonly used aspirin- and acetaminophen-containing products, H 2 -receptor antagonists, proton pump inhibitors, and antidepressants. Definitions of NANSAID Use Our protocol specified that the primary definition of exposure to NANSAIDs, excluding selective cyclooxygenase-2 inhibitors, include only the week before the index date (the index week). Frequent use was defined as medication use for 4 days during the index week. A second definition of NANSAID exposure considered the 8-week study period, classifying subjects as always users if they reported use at least once in each of the preceding 8 weeks, sometimes users if they reported use in 1 7 of the 8 weeks, and nonusers if they reported no use. Thus, we classified NANSAID use as follows: always used prescription, always used OTC, sometimes used prescription, sometimes used OTC, and never used NANSAIDs. This ordering was hierarchical so that respondents reporting 2 use categories were assigned to the single higher category. A third NANSAID use classification, defined a priori, was based on the standardized daily dose during the index week. This was calculated for each NANSAID by dividing the average daily NANSAID dose consumed in the prior week (ie, total milligrams of the NANSAID consumed during the week divided by 7 days) by the minimum daily prescription dose recommended for rheumatoid arthritis. 10,19 If no dose information was provided for rheumatoid arthritis on the labeling/ package insert, we used the minimum recommended prescription dose for arthritis, if provided, or analgesia. One standardized unit equals 1200 mg of ibuprofen, 550 mg of naproxen sodium, and 200 mg of ketoprofen. The standardized daily dose for each NANSAID consumed by a subject during the index week was then summed to provide an overall standardized daily dose for all NANSAID consumption. Statistical Methods The association between NANSAID use and serious gastrointestinal toxicity was estimated using odds ratios (ORs) and 95% confidence intervals (CIs). 2 (or Fisher exact test when there were few observations) and Wilcoxon rank sum tests were used for unadjusted comparisons of categorical and continuous variables, respectively. Unconditional logistic regression was used for multivariable analysis. Checks for collinearity were performed before developing the regression models. 20 The starting model forced in all potential confounders is listed in Table 1. We then used a backward selection process to remove variables from the model that resulted in less than a 10% change in the OR, sequentially removing the variable that resulted in the smallest change in OR. 21 The exposure category of frequent OTC NANSAID use during the

3 December 2005 NONASPIRIN NSAIDS AND UPPER GI TOXICITY 1867 Table 1. Characteristics of the Study Population Category Cases (n 359) Controls (n 1889) P value Median age, y (interquartile range) 62.0 ( ) 54.0 ( ).001 Median body mass index, kg/m 2 (interquartile range) 26.2 ( ) 26.1 ( ).83 Median alcohol use over the past 2 months, no. of drinks (interquartile range) 0.0 ( ) 4.0 ( ).001 Male sex 203 (56.5) 782 (41.4).001 Race White 195 (55.6) 1512 (81.7).001 Black 126 (35.9) 286 (15.5) Other 30 (8.5) 53 (2.9) Socioeconomic status a Low 176 (49.4) 322 (17.1).001 Middle 120 (33.7) 907 (48.2) High 60 (16.9) 652 (34.7) Tobacco use b 98 (27.5) 390 (20.8).005 History of peptic ulcer disease 163 (45.5) 386 (20.5).001 History of gastroesophageal reflux disease 203 (56.9) 808 (42.8).001 History of cirrhosis 38 (10.7) 8 (0.4).001 Heart disease 81 (22.7) 139 (7.4).001 Hypertension 206 (57.7) 619 (32.8).001 Asthma 46 (12.9) 196 (10.4).16 Emphysema 22 (6.2) 31 (1.6).001 Chronic bronchitis 17 (4.8) 89 (4.7).97 Renal failure requiring dialysis 21 (5.9) 9 (0.5).001 Diabetes mellitus 93 (26.1) 189 (10.0).001 Irritable bowel syndrome 26 (7.3) 166 (8.8).35 Inflammatory bowel disease 11 (3.1) 39 (2.1).24 Osteoarthritis or rheumatoid arthritis 88 (24.6) 304 (16.1).001 Other arthritis 41 (11.5) 178 (9.4).23 Gout 32 (9.0) 57 (3.0).001 History of depression 81 (22.7) 274 (14.5).001 History of cancer 53 (14.8) 164 (8.7).003 Use during index week Aspirin 129 (35.9) 566 (30.0).02 Cyclooxygenase-2 inhibitors index week 33 (9.2) 121 (6.4).06 Acetaminophen use in index week 125 (34.8) 488 (25.8).001 Proton pump inhibitors 56 (15.6) 208 (11.0).01 H 2 - receptor antagonists Nonuser 322 (89.9) 1762 (93.4).02 OTC user 14 (3.9) 63 (3.3) Prescription user 22 (6.1) 61 (3.2) Nitrates 22 (6.1) 37 (2.0).001 Anticoagulants 27 (7.5) 24 (1.3).001 Antiplatelets 24 (6.7) 20 (1.1).001 Tricyclic antidepressants 11 (3.1) 34 (1.8).11 Serotonin reuptake inhibitors 61 (17.1) 197 (10.4).001 Bisphosphonates 15 (4.2) 104 (5.5).31 Corticosteroids 23 (6.4) 31 (1.6).001 NOTE. Values are expressed as number (%) unless otherwised indicated. Percentages are based on number of subjects without missing data for that variable. a High socioeconomic status, 17 or more years of education or household income of greater than $90,000; middle socioeconomic status, years of education or income between $30,000 and $90,000; low socioeconomic status, 13 years of education and household income less than $30,000. b Use during the index week of cigarettes, cigars, pipes, or chewing tobacco. index week was used to define the 10% change. Models including all 36 potential confounding variables also resulted in very similar point estimates and statistical significance for the primary exposure variables. As a result, only data from the more parsimonious models are presented. The final logistic regression model included the following covariates: age, sex, socioeconomic status, history of peptic ulcer disease, and use of anticoagulants. Secondary analyses comparing OTC NANSAIDs were based on fewer subjects. To avoid overfitting the multivariable model, we examined whether inclusion of each potential confounder altered the unadjusted OR for naproxen versus ibu-

4 1868 LEWIS ET AL GASTROENTEROLOGY Vol. 129, No. 6 profen by 10%. 22 Any variable meeting this criterion was included in the multivariable model. This strategy resulted in a final multivariable model including sex, race, use of acetaminophen, and cirrhosis as confounder variables. Effect modification was assessed by considering likelihood ratio and Wald tests of the statistical significance of interaction terms in this regression model and by inspecting estimated ORs for the clinical importance of their differences (at different levels of the effect modifiers). We only examined for an interaction among medications that were hypothesized a priori to have antiplatelet effects or had been shown to affect the risk of complications with aspirin or NSAIDs in prior studies These included aspirin, nitrates, serotonin reuptake inhibitors, antiplatelet medications, and corticosteroids. Because NANSAID use has historically been strongly related to peptic ulcer disease, subgroup analyses were performed limiting the case subjects to those with a diagnosis of peptic ulcer disease on the basis of chart review or ICD-9 discharge diagnosis codes if chart review was not available. Specific criteria for peptic ulcers were not applied; rather, we accepted the diagnosis as reported on the upper endoscopy or barium study. Additional subgroup analyses were conducted that minimized the potential impact of recall bias on our results. We have previously documented that recall of NANSAID use appears to be best among frequent prescription NANSAID users, that is, patients who likely take NANSAIDs every week. 29 Extrapolating from those data, subjects who report either use in each of the preceding 8 weeks or no use ever during the 8-week study period should have the best recall. Therefore, we performed a subgroup analysis limited to these subjects, while excluding subjects who reported any prescription NANSAID use and subjects who reported use of OTC NANSAIDs in 1 7 weeks of the 8-week study period. Within this subpopulation, we then examined the association between frequent and infrequent use of OTC NANSAIDs within the most recent week and the risk of upper gastrointestinal toxicity. To assess the potential impact of bias from nonparticipation, all controls who refused to complete the full interview were later recontacted and asked to answer questions regarding their use of analgesics in the previous week. Case subjects who were recruited in the final few months of the study and who refused the full interview were also asked to answer the same limited number of questions. These results were used in a sensitivity analysis to project the potential magnitude of bias from nonparticipation. We projected a sample size of between 382 and 544 eligible case subjects to have 90% power to detect an OR of 1.7 for OTC NANSAIDs based on the expected rates of NANSAID use. Our final case sample size was slightly smaller due to higher-than-anticipated rates of ineligibility discovered at chart review and ICD-9 code review. Ultimately, we had 359 cases, approximately a 5:1 ratio of controls to cases, and a 6.2% rate of frequent OTC NANSAID use by control subjects during the index week. This yielded 90% power to detect an OR of Results The study included 359 eligible cases and 1889 controls. From 7342 homes with eligible controls, DataStat, Inc recruited 2730 subjects (37%), of whom 1889 (69%) ultimately completed interviews. Of the 956 potentially eligible cases, physicians denied permission to contact 140 patients, 200 refused to participate, and 616 completed interviews. Of these, we excluded 177 (29%) after reviewing medical records and 80 (13%) based on ICD-9 discharge diagnosis codes. Of those excluded by chart review, 92 (52%) were hospitalized for lower gastrointestinal bleeding; 7 (4%) had upper gastrointestinal bleeding develop during the hospitalization; 24 (14%) were admitted with nausea, vomiting, or abdominal pain but did not meet our inclusion criteria; 8 (5%) were hospitalized for exacerbation of inflammatory bowel disease; and the rest had miscellaneous other reasons for hospitalization that did not meet our inclusion criteria (eg, bleeding from nasopharynx). Compared with controls, cases were slightly older, were more likely to be men, were more likely to be black, were more likely to use tobacco, were more likely to have a history of peptic ulcer disease and cirrhosis, and used aspirin more frequently (Table 1). As expected, peptic ulcer disease and erosive disease were the most common sources of serious upper gastrointestinal toxicity (Table 2). Formulation, dose, and frequency of use of NAN- SAIDs differed among cases and controls (Table 3). Cases were more likely frequent users of both prescription and OTC NANSAIDs, whereas controls reported higher rates of infrequent OTC NANSAID use. Among frequent OTC NANSAID users, the median daily dose was appreciably higher in cases than controls (0.8 standardized units daily compared with 0.4 standardized units daily; P.001). Concomitant OTC NANSAID and aspirin use was reported by 3.9% of cases and 5.6% of controls (P.19). A total of 1.4% of cases and 0.7% of controls used both a prescription and an OTC NANSAID during the index week (P.21). Only 1.0% of cases and 0.8% of controls used both OTC ibuprofen and OTC naproxen during the index week (P.75). Excluding people taking both OTC and prescription NANSAIDs or both OTC naproxen and OTC ibuprofen, consumption beyond the recommended maximum dose of OTC NAN- SAIDs during the index week was reported by 2.8% of cases and 0.7% of controls (P.005).

5 December 2005 NONASPIRIN NSAIDS AND UPPER GI TOXICITY 1869 Table 2. Final Diagnosis of Case Subjects Source of major gastrointestinal toxicity n (%) Chart review diagnosis (n 282) Peptic ulcer disease 140 (49.6) Gastric ulcer 87 (30.9) Duodenal ulcer 60 (21.3) Portal hypertension 29 (10.3) Esophageal varices 25 (8.9) Gastric varices 6 (2.1) Portal gastropathy 11 (3.9) Erosive gastritis/duodenitis 93 (33.0) Mallory Weiss tear 15 (5.3) Cancer 2 (0.7) Esophageal 2 (0.7) Vascular source of bleeding 4 (1.4) No source identified 4 (1.4) ICD-9 discharge diagnosis (n 57) Peptic ulcer disease 38 (66.7) Gastric ulcer 21 (36.8) Duodenal ulcer 17 (29.8) Erosive disease 9 (15.8) Mallory Weiss tear 7 (12.3) Vascular source of bleeding 3 (5.3) Miscellaneous other 3 (5.3) Gastrointestinal bleeding not otherwise specified 27 (47.4) NOTE. Numbers do not total 100% because some subjects had more than 1 diagnosis. No verification data were available for 20 case subjects. Multivariable analyses confirmed that frequent prescription NANSAID use was associated with increased risk of major gastrointestinal toxicity (adjusted OR, 2.57; 95% CI, ) (Table 4). Frequent OTC NANSAID use was also associated with increased risk of gastrointestinal toxicity, although the association was weaker (adjusted OR, 1.83; 95% CI, ). We Table 4. Association Between NANSAID Use and Upper Gastrointestinal Toxicity Category Upper gastrointestinal toxicity adjusted a OR (95% CI) Peptic ulcer disease adjusted a OR (95% CI) Index week use n 2234 n 2068 None Ref Ref Infrequent OTC b 0.67 ( ) 0.59 ( ) Infrequent Rx b 1.31 ( ) 2.00 ( ) Frequent OTC c 1.83 ( ) 2.33 ( ) Frequent Rx c 2.57 ( ) 3.68 ( ) Index week dose n 2207 n 2044 None Ref Ref ( ) 0.68 ( ) 0.33 to ( ) 2.13 ( ) ( ) 4.76 ( ) Use over 8 weeks n 2233 n 2067 None Ref Ref Some OTC d 0.67 ( ) 0.88 ( ) Some Rx d 1.12 ( ) 1.62 ( ) Always OTC e 1.50 ( ) 1.93 ( ) Always Rx e 2.51 ( ) 4.26 ( ) OTC, over the counter; Rx, prescription. a Adjusted for age, sex, socioeconomic status, history of peptic ulcer disease, and use of anticoagulants. Other variables included in Table 1 were tested as confounders, but inclusion in the multivariable model did not affect the results. b Infrequent use refers to use during 1 3 days of the index week. c Frequent use refers to use during 4 7 days of the index week. d Some use refers to use during a minimum of 1 week and a maximum of 7 weeks during the 8-week study period. e Always use refers to use of the medication for a minimum of 1 day in each of the 8 weeks. did not observe increased risk of major gastrointestinal toxicity with infrequent use of OTC NANSAIDs. When we examined any NANSAID use, there was a clear dose-response relationship, with the highest OR for Table 3. Comparison of the Pattern of NANSAID Use Among Cases and Controls Cases (n 359) Controls (n 1889) n(%) Median (IQR) n (%) Median (IQR) Index week use a Any NANSAID 89 (24.9) 0.5 ( ) 515 (27.3) 0.1 ( ) Any Rx NANSAID 34 (9.5) 1.1 ( ) 87 (4.6) 0.5 ( ) Frequent Rx NANSAID 27 (7.5) 1.5 ( ) 51 (2.7) 1.0 ( ) Infrequent Rx NANSAID 7 (2.0) 0.2 ( ) 32 (1.7) 0.1 ( ) Any OTC NANSAID 60 (16.8) 0.3 ( ) 437 (23.1) 0.1 ( ) Frequent OTC NANSAID 32 (9.0) 0.8 ( ) 117 (6.2) 0.4 ( ) Infrequent OTC NANSAID 27 (7.6) 0.1 ( ) 318 (16.9) 0.1 ( ) 8-Week use b 8-weeks of use of Rx NANSAID 24 (6.7) 56.0 ( ) 44 (2.3) 56.0 ( ) 8-weeks of use of OTC NANSAID 36 (10.0) 33.0 ( ) 162 (8.6) 24.0 ( ) 1 7 weeks of use of Rx NANSAID 30 (8.4) 7.0 ( ) 136 (7.2) 4.0 ( ) 1 7 weeks of use of OTC NANSAID 64 (17.8) 3.0 ( ) 667 (35.3) 3.0 ( ) None 218 (60.7) 942 (49.9) IQR, interquartile range; OTC, over the counter; Rx, prescription. a Median refers to dose. b Median refers to days.

6 1870 LEWIS ET AL GASTROENTEROLOGY Vol. 129, No. 6 average daily doses of 1.0 standard unit (adjusted OR, 3.53; 95% CI, ; test for trend, P.001). In secondary analyses looking specifically at OTC and prescription NANSAID doses, elevated ORs were observed for both high-dose groups (adjusted OR, 5.21 [95% CI, ] for OTC high dose; adjusted OR, 2.49 [95% CI, ] for prescription high dose). A similar dose response was observed for aspirin in the model with all potential confounders (test for trend, P.006). Compared with no use, aspirin at doses of 300 mg/day was not associated with an increased risk of upper gastrointestinal toxicity ( 100 mg/day: adjusted OR, 0.79 [95% CI, ]; mg/day: adjusted OR, 1.11 [95% CI, ]), while people taking 300 mg/day had an increased risk (adjusted OR, 1.89; 95% CI, ). We next examined NANSAID use in always users and serious gastrointestinal toxicity irrespective of dose. Always use of prescription NANSAIDs was associated with increased risk of serious gastrointestinal toxicity (adjusted OR, 2.51; 95% CI, ). The relative odds of always use of OTC NANSAIDs were also elevated (adjusted OR, 1.50; 95% CI, ) (Table 4). There was no apparent association between sometimes use of OTC or prescription NSAIDs and gastrointestinal toxicity. However, among OTC NANSAID users, each additional week of use during the 8-week period was associated with increased risk of serious gastrointestinal toxicity (adjusted OR, 1.21; 95% CI, ). Of note, the use analyses over 8 weeks had a slightly lower OR for always using OTC NANSAIDs in our model with 5 confounders than in a model including 20 confounder variables (adjusted OR, 1.68; 95% CI, ). We did not observe statistically significant interactions among NANSAID use categorized by dose and use of aspirin, nitrates, serotonin reuptake inhibitors, antiplatelet medications, or corticosteroids; however, the highest relative odds were observed for concomitant use of aspirin and high-dose NANSAIDs (adjusted OR, 7.69 [95% CI, ] among aspirin users; adjusted OR, 2.63 [95% CI, ] among aspirin nonusers; P value for interaction,.15). Secondary Analyses When we repeated analyses limiting the case group to patients with a chart review or ICD-9 discharge diagnosis code of peptic ulcer disease, the strength of all associations was stronger (Table 4). Index week use of NANSAIDs at an average daily dose of standard units and continuous use of OTC NSAIDs, both of which were nearly statistically significant in the primary analyses, all reached statistical significance in these analyses. Likewise, the magnitude of the association between aspirin and peptic ulcer disease was stronger than that with upper gastrointestinal toxicity and again demonstrated evidence of a dose response (test for trend, P.001), with a significant increased risk observed only in those taking 300 mg/day ( 100 mg/day: adjusted OR, 1.04 [95% CI, ]; mg/day: adjusted OR, 1.28 [95% CI, ]; 300 mg/day: adjusted OR, 3.46 [95% CI, ]). In our subgroup analyses designed to minimize the impact of recall bias (see Subjects and Methods), compared with subjects reporting no NANSAID use in the previous 8 weeks, those who reported frequent OTC NANSAID use during the index week and at least 1 day of use in each of the 8 weeks had higher rates of any upper gastrointestinal toxicity (adjusted OR, 1.97; 95% CI, ) and peptic ulcer disease (adjusted OR, 2.76; 95% CI, ). In contrast, patients who reported infrequent use during the index week and at least 1 day of use in each of the 8 weeks had no increased risk of upper gastrointestinal toxicity (adjusted OR, 1.00; 95% CI, ) or peptic ulcer disease (adjusted OR, 0.95; 95% CI, ). The reported rate of NANSAID use among control refusers (23%) was slightly lower than among participating controls (27%) (P.19). Similarly, the rate (16%) among the 38 case subjects who refused a full interview but responded to some questions about use during the index week was somewhat below the comparable rate (23%) for the participating case subjects (P.28). Assuming in a sensitivity analysis that all refusers had participated, the observed OR (1.83) for the association of frequent OTC NANSAID use and upper gastrointestinal toxicity would have decreased slightly to We did not observe differences in risk of gastrointestinal toxicity between frequent OTC naproxen and OTC ibuprofen use. There was little difference between the results adjusted for the 5 variables (age, sex, socioeconomic status, history of peptic ulcer disease, and use of anticoagulants) included in our primary models (OR, 1.00; 95% CI, ) and the results from our model adjusted for confounder variables identified from bivariate analyses (sex, race, use of acetaminophen, and cirrhosis) (OR, 0.84; 95% CI, ). Discussion Previous research established a strong relationship between prescription NANSAID use and risk of upper gastrointestinal toxicity. 6,7,9 Much less research has been

7 December 2005 NONASPIRIN NSAIDS AND UPPER GI TOXICITY 1871 performed on OTC NANSAID use. In this study, we documented a relatively good safety profile of OTC NANSAIDs used for short duration and at recommended doses. However, the risk of serious gastrointestinal toxicity is increased among OTC NANSAID users who take high doses. Of course, this is logical because the amount consumed, not the purchase method, influences gastrointestinal toxicity. Few studies have focused on severe gastrointestinal toxicity with OTC NANSAIDs. A pilot study by the American College of Gastroenterology suggested that OTC NANSAID use was associated with an approximately 3-fold increased risk of upper gastrointestinal bleeding 30 ; however, the study lacked detailed dose information collected directly from patients. In the United States, ibuprofen and naproxen are the 2 most commonly used OTC NANSAIDs. Several metaanalyses have examined the dose response of prescription NANSAID-related gastrointestinal toxicity. 6 9 Henry et al pooled results from 5 studies using variable definitions of low dose. 7 Both ibuprofen and naproxen had higher risks of gastrointestinal toxicity with high-dose therapy. In the low-dose groups, the pooled ORs were 1.6 (95% CI, ) and 3.7 (95% CI, ) for ibuprofen and naproxen, respectively. In a more recent meta-analysis of primary patient level data from 3 case-control studies, 8 use of ibuprofen at doses 1200 mg daily (1 standardized daily dose) was not associated with increased risk of upper gastrointestinal toxicity (pooled OR, 1.1). Although use of naproxen at 500 mg daily was associated with increased risk of upper gastrointestinal bleeding (OR, 4.8), the estimate was based on only a few exposed cases. In another study, Strom et al 31 observed that the risk of hospital admission for upper gastrointestinal bleeding within 14 days of initiating prescription therapy with OTC dose naproxen was approximately twice that of persons initiating therapy with OTC dose ibuprofen. However, this study used pharmacy records and consequently lacked detailed history on use patterns. In our study, the risk of gastrointestinal toxicity was not significantly different among users of OTC naproxen and OTC ibuprofen. Although not designed as an equivalence study, the results suggest little substantial differences in risk of toxicity between these OTC NANSAIDs. Use of other OTC NANSAIDs (ketoprofen) was too infrequent to permit meaningful comparisons. Aspirin was widely used in our population. Among control subjects, 30% reported use of aspirin and 27% reported use of NANSAIDs during the index week. However, the pattern and indication for use of aspirin are likely very different from NANSAIDs, with most aspirin use being daily and at very low doses for the purpose of prevention of cardiovascular disease. Because of the difference in pattern of use, we did not directly compare frequent aspirin use with frequent NANSAID use. However, our results are consistent with prior research demonstrating a strong dose response in risk of upper gastrointestinal toxicity and peptic ulcer disease with aspirin. 32,33 Because the dose of aspirin used for arthritis and other analgesic purposes is 300 mg/day, use for these indications would be expected to be associated with significant toxicity, as is seen with use of high-dose NANSAIDs. Because the various NANSAIDs have different potencies, a common metric is required. We elected to use the same metric that has been successfully used by Henry et al 10 and Griffin et al. 19 Importantly, as long as a common metric is used, there is little reason to believe that the dose-response results would have differed were a different metric used. Demonstration that there is a clear dose response in risk is the most important information from this analysis, because the available OTC NANSAIDs are generally used in lower doses than prescription NAN- SAIDs. This information can then be used in conjunction with our data on frequency of use and typical dose consumed to understand the risk-benefit profile of OTC NANSAIDs. We lacked data on the presence or absence of H pylori infection in this population. However, people who purchase OTC NANSAIDs almost never know whether they are infected with H pylori. As such, our results should be applicable to that population. Measurement of OTC medication use, particularly medications used on an as-needed basis, is a challenge. Methods commonly used to measure prescription medications, such as refill patterns in dispensing data, direct observational therapy, pill counts, electronic monitoring devices, and medication possession ratios, are impractical for OTC medications. As a result, in the absence of any one perfect method, we selected what we believed to be the best approach currently available to study this question. While misclassification resulting from self-report is possible, such misclassification would likely bias our results toward the null hypothesis. That we observed significant associations and a strong dose response suggests that misclassification from self-report was unlikely to have affected the conclusions of the study regarding higher doses and frequent use of OTC NANSAIDs. Although such misclassification could conceivably have masked an association with low doses and infrequent use of OTC NANSAIDs, the magnitude of the bias would have to be very large, given that our ORs for low doses and infrequent use of OTC NANSAIDs were less than

8 1872 LEWIS ET AL GASTROENTEROLOGY Vol. 129, No. 6 unity. Thus, misclassification bias is unlikely to have affected our conclusions. We tried to minimize recall bias in several ways. We used a previously validated 3-stage method for collecting our primary exposure data and used calendars and picture books as memory aids Controls were asked to recall medication use in the previous week, a relatively short interval, whereas cases were interviewed a median of 38 days after the start of their hospitalization. Searching for evidence of recall bias, we examined the proportion of cases reporting NANSAID use during the index week stratified by the time from the hospitalization date to the interview date. There was little variation in reported rates of index-week NANSAID use across these groups (30% vs 35% for those interviewed 1 6 weeks vs 7 12 weeks following the index date, respectively). In addition, reported rates of any OTC NANSAID use were nearly identical to data obtained at hospitalization in a prior prospective study of patients admitted for upper gastrointestinal bleeding. 34 Further, our stringent definitions of exposure (use every week versus no use and use on 4 days during the index week) should have minimized recall bias. 29 Furthermore, when we limited the analyses to patients reporting weekly OTC NANSAID use versus those reporting no NANSAID use in the previous 8 weeks, the risk of hospitalization with upper gastrointestinal toxicity was still strongly related to frequency of OTC NANSAID use during the index week. Although it is never possible to state with certainty that differential recall did not bias the results, the finding of similar results in this subgroup analysis argues strongly against recall bias as an explanation for our findings. A potential limitation of observational studies relates to nonrespondents. We found similar to slightly lower rates of NANSAID use among refusers who completed the abbreviated questionnaire. In a prior study using similar methodology in the Delaware Valley Case Control Network, Kimmel et al documented nearly identical rates of NANSAID use during the previous week among refusers and those who ultimately completed the questionnaire. 35 In addition, our sensitivity analysis focusing on frequent OTC NANSAID use suggested that relatively little bias may have resulted from nonparticipation. Likewise, if anything, inclusion of nonparticipants would have lowered the OR for infrequent OTC NAN- SAID use and therefore cannot explain the lack of effect on gastrointestinal toxicity seen with infrequent OTC NSAID use. Thus, it seems unlikely that an association between nonrespondents and use of OTC NANSAIDs had a major impact on our results. We collected data on a wide range of potential confounders and, as expected, the characteristics of the case subjects differed from those of the controls. Our models including all of these variables provided results similar to our final models. However, there could still be additional variables that were not adjusted for. For example, observational studies risk bias from confounding by indication. To explore this possibility, subjects who consumed NANSAIDs and acetaminophen were asked if they were taking these medications for abdominal pain. Nearly identical results were observed for both classes of medications, with more cases than controls reporting use of the medications for abdominal pain (data not shown). That these results were nearly identical for NANSAIDs and acetaminophen (which was not associated with gastrointestinal toxicity in the fully adjusted model) suggests that confounding by indication was unlikely to explain our results. While it is never possible to rule out entirely that unmeasured confounding may have affected the results of a study, several additional observations are reassuring that other unmeasured confounders are unlikely to have resulted in biased estimates. First, we observed the anticipated association between prescription NANSAIDs and gastrointestinal toxicity, which is reassuring that unmeasured confounding should not be a major problem. Second, one can consider the potential impact of an unmeasured confounder on the observed data and the clinical conclusions. For this study, one might ask whether unmeasured confounding could have obscured a true association between infrequent OTC NANSAID use and gastrointestinal toxicity. Because our adjusted point estimates of the association between infrequent OTC NANSAID use and gastrointestinal toxicity were less than unity, such an unmeasured confounder would need to have an enormous confounding effect to have obscured a true positive association. As such, we believe that unmeasured confounding is unlikely to have led to incorrect conclusions from our study. Our results are likely generalizable to patients hospitalized with upper gastrointestinal toxicity within the United States. Our case definition, selected a priori, did not permit us to study the effects of OTC NANSAIDs on less serious conditions such as dyspepsia or other conditions among hospitalized patients that may be related to NANSAID use. In our population, OTC NANSAIDs are overwhelmingly taken as directed. Our data show that OTC NAN- SAIDs, when used at recommended doses and for short duration, exhibit a relatively good safety profile. Our secondary analyses suggest similar rates of gastrointestinal toxicity among frequent users of OTC naproxen and OTC ibuprofen, although statistical power was sufficient only to rule out a 3-fold difference between the two. The

9 December 2005 NONASPIRIN NSAIDS AND UPPER GI TOXICITY 1873 small minority of people who use OTC NANSAIDs in doses comparable to prescription NANSAIDs (ie, greater than the maximum recommended OTC doses) or on a weekly basis (for at least 8 weeks) and are frequent users have the expected increase in risk of gastrointestinal toxicity. References 1. Lewis JD, Bilker WB, Brensinger C, Farrar JT, Strom BL. Hospitalization and mortality rates from peptic ulcer disease and GI bleeding in the 1990s: relationship to sales of nonsteroidal anti-inflammatory drugs and acid suppression medications. Am J Gastroenterol 2002;97: Hawkey CJ. Risk of ulcer bleeding in patients infected with Helicobacter pylori taking non-steroidal anti-inflammatory drugs. Gut 2000;46: Walsh JH, Peterson WL. The treatment of Helicobacter pylori infection in the management of peptic ulcer disease. N Engl J Med 1995;333: Singh G, Triadafilopoulos G. Epidemiology of NSAID induced gastrointestinal complications. J Rheumatol 1999;26: Wolfe MM, Lichtenstein DR, Singh G. Gastrointestinal toxicity of nonsteroidal antiinflammatory drugs. N Engl J Med 1999;340: Gabriel SE, Jaakkimainen L, Bombardier C. Risk for serious gastrointestinal complications related to use of nonsteroidal antiinflammatory drugs. A meta-analysis. Ann Intern Med 1991;115: Henry D, Lim LL, Garcia Rodriguez LA, Perez Gutthann S, Carson JL, Griffin M, Savage R, Logan R, Moride Y, Hawkey C, Hill S, Fries JT. Variability in risk of gastrointestinal complications with individual non-steroidal anti-inflammatory drugs: results of a collaborative meta-analysis. BMJ 1996;312: Lewis SC, Langman MJ, Laporte JR, Matthews JN, Rawlins MD, Wiholm BE. Dose-response relationships between individual nonaspirin nonsteroidal anti-inflammatory drugs (NANSAIDs) and serious upper gastrointestinal bleeding: a meta-analysis based on individual patient data. Br J Clin Pharmacol 2002;54: Langman MJ, Weil J, Wainwright P, Lawson DH, Rawlins MD, Logan RF, Murphy M, Vessey MP, Colin-Jones DG. Risks of bleeding peptic ulcer associated with individual non-steroidal antiinflammatory drugs. Lancet 1994;343: Henry D, Dobson A, Turner C. Variability in the risk of major gastrointestinal complications from nonaspirin nonsteroidal antiinflammatory drugs. Gastroenterology 1993;105: Bombardier C, Laine L, Reicin A, Shapiro D, Burgos-Vargas R, Davis B, Day R, Ferraz MB, Hawkey CJ, Hochberg MC, Kvien TK, Schnitzer TJ, VIGOR Study Group. Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. VIGOR Study Group. N Engl J Med 2000;343: Hartge P, Brinton LA, Rosenthal JF, Cahill JI, Hoover RN, Waksberg J. Random digit dialing in selecting a population-based control group. Am J Epidemiol 1984;120: Harlow BL, Hartge P. Telephone household screening and interviewing. Am J Epidemiol 1983;117: Metlay JP, Hardy C, Strom BL. Agreement between patient selfreport and a Veterans Affairs national pharmacy database for identifying recent exposures to antibiotics. Pharmacoepidemiol Drug Saf 2003;12: Klungel OH, de Boer A, Paes AH, Herings RM, Seidell JC, Bakker A. Influence of question structure on the recall of self-reported drug use. J Clin Epidemiol 2000;53: Kimmel SE, Lewis JD, Jaskowiak J, Kishel L, Hennessy S. Enhancement of medication recall using medication pictures and lists in telephone interviews. Pharmacoepidemiol Drug Saf 2003;12: Strom BL, Schinnar R. An interview strategy was critical for obtaining valid information on the use of hormone replacement therapy. J Clin Epidemiol 2004;57: The PDR electronic library. Montvale, NJ: Thomson Medical Economics, Griffin MR, Piper JM, Daugherty JR, Snowden M, Ray WA. Nonsteroidal anti-inflammatory drug use and increased risk for peptic ulcer disease in elderly persons. Ann Intern Med 1991;114: Belsey DA, Kuh E, Welsch RE. Regression diagnostics: identifying influential data and sources of collinearity. New York: Wiley, Mickey RM, Greenland S. The impact of confounder selection criteria on effect estimation. Am J Epidemiol 1989;129: Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol 1996;49: de Abajo FJ, Rodriguez LA, Montero D. Association between selective serotonin reuptake inhibitors and upper gastrointestinal bleeding: population based case-control study. BMJ 1999; 319: Dalton SO, Johansen C, Mellemkjaer L, Norgard B, Sorensen HT, Olsen JH. Use of selective serotonin reuptake inhibitors and risk of upper gastrointestinal tract bleeding: a population-based cohort study. Arch Intern Med 2003;163: Chan FK, Ching JY, Hung LC, Wong VW, Leung VK, Kung NN, Hui AJ, Wu JC, Leung WK, Lee VW, Lee KK, Lee YT, Lau JY, To KF, Chan HL, Chung SC, Sung JJ. Clopidogrel versus aspirin and esomeprazole to prevent recurrent ulcer bleeding. N Engl J Med 2005;352: Ng FH, Wong SY, Chang CM, Chen WH, Kng C, Lanas AI, Wong BC. High incidence of clopidogrel-associated gastrointestinal bleeding in patients with previous peptic ulcer disease. Aliment Pharmacol Ther 2003;18: Conn HO, Blitzer BL. Nonassociation of adrenocorticosteroid therapy and peptic ulcer. N Engl J Med 1976;294: Lanas A, Bajador E, Serrano P, Fuentes J, Carreno S, Guardia J, Sanz M, Montoro M, Sainz R. Nitrovasodilators, low-dose aspirin, other nonsteroidal antiinflammatory drugs, and the risk of upper gastrointestinal bleeding. N Engl J Med 2000;343: Lewis JD, Strom BL, Kimmel SE, Farrar JT, Metz DC, Brensinger C, Nessel L, Localio AR. Predictors of recall of over-the-counter and prescription non-steroidal anti-inflammatory drug exposure. Pharmacoepidemiol Drug Saf 2005 (in press); online at www. interscience.wiley.com; DOI: /pds Blot WJ, McLaughlin JK. Over the counter non-steroidal antiinflammatory drugs and risk of gastrointestinal bleeding. J Epidemiol Biostat 2000;5: Strom BL, Schinnar R, Bilker WB, Feldman H, Farrar JT, Carson JL. Gastrointestinal tract bleeding associated with naproxen sodium vs ibuprofen. Arch Intern Med 1997;157: Weil J, Colin-Jones D, Langman M, Lawson D, Logan R, Murphy M, Rawlins M, Vessey M, Wainwright P. Prophylactic aspirin and risk of peptic ulcer bleeding. BMJ 1995;310: Kelly JP, Kaufman DW, Jurgelon JM, Sheehan J, Koff RS, Shapiro S. Risk of aspirin-associated major upper-gastrointestinal bleeding with enteric-coated or buffered product. Lancet 1996;348: Wilcox CM, Shalek KA, Cotsonis G. Striking prevalence of overthe-counter nonsteroidal anti-inflammatory drug use in patients with upper gastrointestinal hemorrhage. Arch Intern Med 1994; 154: Kimmel SE, Berlin JA, Reilly M, Jaskowiak J, Kishel L, Chittams J, Strom BL. The effects of nonselective non-aspirin non-steroidal

10 1874 LEWIS ET AL GASTROENTEROLOGY Vol. 129, No. 6 anti-inflammatory medications on the risk of nonfatal myocardial infarction and their interaction with aspirin. J Am Coll Cardiol 2004;43: Received April 11, Accepted August 17, Address requests for reprints to: James D. Lewis, MD, MSCE, 720 Blockley Hall, 423 Guardian Drive, Philadelphia, Pennsylvania jlewis@cceb.med.upenn.edu; fax: (215) Supported by a grant from Bayer Healthcare. The authors were responsible for the design, implementation, data collection, data analysis, interpretation, and reporting of the data. Although the sponsor was allowed to comment on a near-final version of the manuscript, the sponsors did not have editorial authority and the authors made all final decisions. Dr Lewis has served as a consultant to Bayer, Wyeth, Pfizer, Merck, and GlaxoSmithKline. He has received other research support from Wyeth, Pfizer, and GlaxoSmithKline. He has received honorarium for speaking from AstraZeneca and GlaxoSmithKline. Dr Kimmel has served as a consultant to several companies that make NSAIDs, including Bayer, none directly related to non-aspirin NSAIDs. He has also received research funding from National Institutes of Health, Merck, and Pfizer for work on NSAIDs. Dr Metz has served as a consultant to AstraZeneca, TAP, Wyeth, Eisai, and Santarus. He has received research support from AstraZeneca, TAP, Wyeth, Novartis, Merck, and Pfizer. He has received honorarium for speaking from AstraZeneca, TAP, Wyeth, Merck, and Pfizer. His wife owns Johnson and Johnson stock and together they own stock in Pfizer. Dr Farrar has received research support from Cephalon and Pfizer. He has served as a consultant to Abbott, Adolor, Allergan, Celgene, Eli Lilly, OrthoMcNeil, and UCB Pharma. Dr Strom has testified on behalf of Bayer concerning issues related to rhabdomyolysis. He has also received research funding from and served as a consultant to most manufacturers of non-narcotic analgesics, including NSAIDs and COXIBs. The following hospitals identified case subjects for inclusion in the study (local principal investigators are listed in parentheses): Abington Memorial Hospital (Harvey Guttmann, MD), Albert Einstein Medical Center (Robert S. Wimmer, MD), Bryn Mawr Hospital (Matthew Astroff, MD), Central Montgomery Medical Center (Joanne Shupe, RN), Chester County Hospital (Laurence Weinberg, MD), Chestnut Hill Hospital (Joyann Krozer, MD), Christiana Hospital (Marciana Filippone, MD), Cooper University Hospital (Adam Elfant, MD), Delaware County Memorial Hospital (Mark Jacobs, MD), Doylestown Hospital (Philip Coff, MD), Elkins Park Hospital (Mark Tanker, DO), Graduate Hospital (Susan Gordon, MD), Holy Redeemer Hospital and Medical Center (Harvey Guttmann, MD), Hospital of the University of Pennsylvania (James D. Lewis, MD), Jeanes Hospital (Mark Codella, MD), Lankenau Hospital (Matthew Astroff, MD), Mercy Fitzgerald Hospital (Norman Cohen, MD), Paoli Memorial Hospital (Matthew Astroff, MD), Pennsylvania Hospital (Robin Rothstein, MD), Phoenixville Hospital (Kellen Kovalovich, MD), Pottstown Memorial Medical Center (Deborah Fleischer, MD), Presbyterian Medical Center (James D. Lewis, MD), Riddle Memorial Hospital (David Bobman, MD), St Mary Medical Center (David Popper, MD), Temple University Hospital (Robert Fisher, MD), Thomas Jefferson University Hospital (Sidney Cohen, MD), Underwood Memorial Hospital (Howard Kroop, MD), and Warminster General Hospital (Philip Coff, MD).