`
`Volume 115
`5 November 1991
`
`>)
`
`,
`
`Number 10
`
`. 1
`
`Annals ofInternal Medicine
`
`Biiched Twice Monthly by the American College of Physicians
`
`™ ARTICLES
`B Letters 830
`
`Increasing the Dietary Potassium Intake Reduces the Need
`753
`Siani, Strazzullo,
`Myonecrosis and Myofibrosis in
`for Antihypertensive Medication
`Giacco, and others
`Sickle Cell Anemia
`
`Muscle Infarction in Sickle Cell
`Treatment for Cerebral Toxoplasmosis Protects against
`760 Heald, Flepp, Chave,
`Anemia
`Pneumocystis carinii Pneumonia in Patients with AIDS
`Malinverni, and others
`
`Heterosexual Co-transmission of Hepatitis C Virus (HCV)
`764 Eyster, Alter, Aledort,
`and Human Immunodeficiency Virus (HIV)
`Quan, and others
`
`Reversible Renal Failure Associated with Angiotensin-
`769 Chapman, Gabow,
`Converting Enzyme Inhibitors in Polycystic Kidney Disease
`Schrier
`
`Screening for Alcohol Abuse Using CAGE Scores and
`774 Buchsbaum, Buchanan,
`Likelihood Ratios
`Centor, and others
`
`Thoracoscopic Tale Poudrage Pleurodesis for Chronic
`778 Aelony, King, Boutin
`Recurrent Pleural Effusions
`
`Dral Quinolone Treatment for
`Isteomyelitis
`
`QuinoloneProphylaxis during
`Yeutropenia
`
`Diagnosis of Pulmonary
`imbolism
`
`‘he Cardiac Conduction System
`o Unexplained Sudden Death
`
`f The Literature
`© BRIEF REPORTS
`
`of Medicine 335
`
`Combined Endoscopic Sphincterotomy and Laparoscopic 783=Aliperti,
`Cholecystectomy in Patients with Choledocholithiasis
`Edmundowicz, Soper,
`‘or complete contents, see pages
`-3 and I-5S
`and Cholecystolithiasis
`Ashley
`
`Recovery of Adrenal Function after Failure Resulting
`785
`Feuerstein, Streeten
`from Traumatic Bilateral Adrenal Hemorrhageseee
`@ REVIEWS
`
`Risk for Serious Gastrointestinal Complications Related
`787 Gabriel, Jaakkimainen,
`to Use of Nonsteroidal Anti-inflammatory Drugs:
`Bombardier
`A Meta-analysis
`
`Cocaine-induced Myocardial Infarction in Patients with
`797 Minor, Scott, Brown,
`Normal Coronary Arteries
`Winniford
`
`IMEAS 115(10)753-836 (1991)
`US ISSN 0003-4819
`
` NOV19199]
`
`™ MEDICINE AND PUBLIC ISSUES
`
`Cost Effectiveness of Colorectal Cancer Screening in
`807 Wagner, Herdman,
`the Elderly
`Wadhwa
`
`™@
`PERSPECTIVE
`
`How Should Results from Completed Studies Influence
`818 Laupacis, Connolly,
`Ongoing Clinical Trials?
`Gent, and others
`
`™ ON BEING A DOCTOR
`
`A Job Well Done?
`823 Curtis
`
`™ EDITORIALS
`
`Treatment of Rheumatoid Arthritis: Challenges to
`825
`Pincus, Wolfe
`Traditional Paradigms
`
`- Human Immunodeficiency Virus Infection in Women
`827
`Spence, Reboli
`
`Page 1 of 11
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Page 1 of 11
`
`

`

`REVIEWS
`
`
`
` This material may be protected by Copyrightlaw (Title 17 U.S. Code)
`
`Risk for Serious Gastrointestinal Complications Related
`to Use of Nonsteroidal Anti-inflammatory Drugs
`A Meta-analysis
`
`Sherine E. Gabriel, MD, MSc; Liisa Jaakkimainen, MSc; and Claire Bombardier, MD
`
`
`M Objective: To describe therelative risk for serious
`gasirointestinal complications due to nonaspirin non-
`steroidal anti-inflammatory drug (NSAID) exposure
`among NSAID users as well as in selected subgroups.
`Design: Overview and meta-analysis.
`H Data Identification: A literature search of English-
`language studies examining the association between
`NSAIDs and adverse gastrointestinal events for the pe-
`riod 1975 to 1990identified using MEDLINE and commu-
`nicating with three internationally recognized experts.
`i Data Analysis: A qualitative summary of study char-
`acteristics and a critical appraisal of study quality were
`done. The results of 16 primary studies were selected
`and combined statistically. Summary estimates were
`weighted by sample size and quality score.
`M Main Results: The overall odds ratio of the risk for
`adverse gastrointestinal events related to NSAID use,
`summarized from 16 studies (9 case-control and 7
`cohort) was 2.74 (95%Cl, 2.54 to 2.97). The summary
`odcis ratios were as follows: elderly patients, (aged
`= 60 years), 5.52 (Cl, 4.63 to 6.60); patients under 65
`years of age, 1.65 (Cl, 1.08 to 2.53); women, 2.32 (Cl,
`1.97 to 2.82); and men, 2.40 (Cl, 1.85 to 3.11). The
`summary oddsratio for NSAID users receiving concom-
`itant corticosteroids compared with NSAID users not
`receiving corticosteroids was 1.83 (Cl, 1.20 to 2.78). The
`summary odds ratio for the first gastrointestinal event
`was 2.39 (Cl, 2.16 to 2.65). The relative risk for a
`subsequent or unspecified gastrointestinal event was
`4.76(Cl, 4.05 to 5.59). The summary oddsratio for less
`than 1 month of NSAID exposure was8.00 (Cl, 6.37 to
`10.06); for more than 1 month but less than 3 months of
`exposure, the summary oddsratio was 3.31 (Cl, 2.27 to
`4.82); and for more than 3 months of exposure, the
`summary odds ratio was 1.92 (Cl, 1.19 to 3.13).
`™ Conclusions: Users of NSAIDs are at approximately
`three times greater relative risk for developing serious
`adverse gastrointestinal events than are nonusers. Ad-
`ditional risk factors include age greater than 60 years,
`Previous history of gastrointestinal events, and con-
`comitant corticosteroid use. Another possible risk fac-
`tor is the first 3 months of NSAID therapy. The risk for
`Serious gastrointestinal events appears to be equal
`among men and women. These data represent sum-
`marystatistics from 16 studies and cannot be consid-
`€red generalizable to all NSAID users.
`
`Annalsof Internal Medicine. 1991;115:787-796.
`
`From the Mayo Clinic and Mayo Foundation, Rochester, Min-
`Nesota; and Wellesley Hospital, Toronto, Ontario. For current
`author addresses, see end oftext.
`
`Nonsteroidal antiinflammatory drugs (NSAIDs) are
`the most widely used agents for the treatment of mus-
`culoskeletal and arthritic syndromes (1). Use of these
`agents has been increasingly associated with gastroin-
`testinal
`toxicity,
`including mild dyspepsia, as well as
`moreserious gastrointestinal reactions such as bleeding,
`perforation, and other events leading to hospitalization
`or death. Although researchers agree that an increased
`risk for gastrointestinal toxicity exists with NSAID use,
`the size of the reported risk has varied markedly, and
`thereis little agreement on the definition of ‘‘high risk”’
`groups (2-19).
`Wereviewedthe literature on NSAID-related adverse
`gastrointestinal events. First, we summarized study
`characteristics and appraised study quality. We then did
`a meta-analysis of all controlled trials that examined
`the risks
`for
`serious gastrointestinal events among
`NSAID users. Our primary objective was to estimate a
`summary oddsratio or relative risk for serious gastro-
`intestinal complications due to nonaspirin NSAID expo-
`sure.
`
`Methods
`
`A comprehensive search of the English-language literature
`from 1975 to 1990 was conducted using MEDLINE and search-
`ing the following terms: anti-inflammatory agents, non-sterol-
`dal; gastropathy, toxicity, adverse effects, or side effects; pep-
`tic ulcer or dyspepsia; gastric erosion, gastritis, gastric ulcer,
`gastric mucosa, endoscopy; and human. Wealso searched for
`specific NSAIDs by name.
`Five hundred twenty-six references were obtained. These
`were reviewed by one of the authors, and any citation that
`mentioned NSAID-related gastrointestinal events was selected
`(Figure 1). One hundred forty-two articles met this criterion
`and were entered into ‘‘Reference Manager’’ (20). Five addi-
`tional articles were identified by communication with three
`investigators (Marie Griffin, MD; Michael Langman, MD; and
`Richard Hunt, MD) from the United States, United Kingdom,
`and Canada, respectively. These 5 articles were added to the
`data set, for a total of 147 articles.
`From the 147 articles in the data set, 40 studies were se-
`lected that examined the association between NSAIDs and
`adverse gastrointestinal events. Specific inclusion and exclu-
`sion criteria were applied to these studies independently by
`two of the authors. All studies that contained a comparison
`group and provided an estimate of risk for serious gastrointes-
`tinal complications (defined as bleeding, perforation, or other
`adverse gastrointestinal events resulting in hospitalization or
`death) in NSAID users compared with nonusers, regardless of
`underlying disease, were included in the meta-analysis. A
`study was excluded if its primary objective was to assess
`effectiveness, if it involved the treatment of children (under 18
`years of age), if it described fewer than ten patients, if the only
`NSAIDstudied wassalicylate, or if the outcome examined was
`
`©1991 American College of Physicians
`
`787
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Page 2 of 11
`
`

`

`SELECTION OF STUDIES AND REVIEWS
`Searchstrategy
`Numberof citations
`Comprehensive medline
`526
`search and review of
`bibliographies of selected articles
`142
`Mention of NSAID-
`related GI adverse
`event(bycitation review)
`|. Additionalarticles identified
`by 3 internationai experts
`
`Reviews or commentaries
`(by citation review)
`7
`Reviews of humanstudies in
`English, which included data
`from 2 primary studies
`10
`Review articles selected
`for critical appraisal
`
`Studies examining the association
`between NSAIDs and adverse Gi events
`40
`
`Studies meeting specitic
`inclusion/exclusioncriteria
`16
`Primary studies selected
`for meta-analysis
`
`Figure 1. Selection of studies and reviews.
`
`the identification of ulcer rather than the presence of serious
`gastrointestinal complications. Disagreements between the two
`reviewers were resolved by consensus. Sixteen studies were
`selected (21-36) for meta-analysis (see Figure 1).
`
`Meta-analysis
`
`The following criteria were used to evaluate the quality of
`the studies included in the meta-analysis: blinding, definition of
`outcome, case selection, control selection, matching technique,
`definition of exposure, and control for confounders (Appendix
`A). The Methodssection of each study was photocopied, with
`care taken to exclude any mention of the authors’ names,
`study results, or journal title. Study quality was evaluated in a
`blinded fashion by two of the investigators. Quality scores
`were assigned to each criterion according to its relative impor-
`tance. A quality score of 0 indicated poor definitions and no
`attempt to avoid bias, and a score of 46 indicated the con-
`verse. The average score (between the two readers) among the
`first six categories constituted the baseline score for the study.
`For every 5 confoundersidentified in a primarystudy, 1 bonus
`point was awarded, to a maximum of5 points for studies that
`identified more than 25 confounders. Thus, the maximum qual-
`ity score attainable was 51. Agreement between the two read-
`ers regarding the quality score was evaluated using the kappa
`statistic (37).
`Data from all articles were abstracted in duplicate to avoid
`errors. The two observers met, discussed each item, and resolved
`all disagreements and errors. A final copy of the completed data
`collection forms was then created and entered into a database
`(ORACLE, Oracle Corporation, Belmont, California) (38).
`The results of the 16 primary studies were combinedstatis-
`tically using two different techniques. First, overall point esti-
`mates of the odds ratios and 95% confidence intervals (CIs)
`were calculated from the raw data of the 16 selected studics
`using the Mantel-Haenszelstatistic (39). The second technique
`involved combining the published odds ratios and CIs directly
`across studies to produce an overall estimate of the odds ratio
`and 95% CI (40). The latter will hereafter be referred to as the
`“direct”? method. The direct method was the primarystatisti-
`cal analysis technique used, and all results were calculated
`using this method unless otherwise stated.
`The purpose of this analysis was not to estimate a common
`parameter, but rather to compute an average or summary sta-
`tistic across the 16 selected studies. The CI for this statistic
`cannot,
`therefore, be generalized beyond the study samples.
`All summary estimates were weighted by sample size. The
`influence of the quality scores on the summary estimates was
`evaluated using logistic-regression analysis with quality score
`as a covariate.
`Overall odds ratios for all studies included in the meta-
`analysis as well as odds ratios for various subgroups were
`calculated. The overall odds ratios referred to the odds ratios
`
`combined from the main research questions of each of the
`studies. Summary odds ratios for various subgroups were cal-
`culated from those studies which provided data on these sub-
`groups. The method of Breslow and Day was used to test for
`homogeneity of the Mantel-Haenszel estimates (41). Tests of
`homogeneity were also performed for the direct method ac-
`cording to the method of Greenland (40).
`
`Results
`
`Weselected 16 studies (9 case-control and 7 cohort)
`that specifically examined the risks for clinically de-
`fined, NSAID-related, adverse gastrointestinal events
`(21-36). The reported relative risks varied from 1.0 (34)
`(indicating no increased risk for gastrointestinal events)
`to 13.7 (29) (indicating a risk for NSAID users 13.7
`times greater than that for nonusers). Two potentia!
`sources of variability were identified: differences in
`study characteristics and differences in study quality.
`
`Study Characteristics
`
`Study characteristics are shown in Appendix B. For
`both the case-control and cohort studies, serious gas-
`trointestinal events were defined among hospital-based
`cases. Among the case-control studies,
`the ascertain-
`ment of gastrointestinal outcome was not done in a
`uniform manner. Gastrointestinal events were assessed
`based on the results of endoscopy, roentgenography, or
`surgery (27-29, 33, 35) or on a clinical diagnosis of
`hematemesis or melena (26, 30-32). Some case-control
`studies used community controls (31, 33, 35); others
`compared cases with hospital controls (28-30, 32) or
`used both types of controls (26, 27). Most
`studies
`matched controls directly with cases (26-28, 30, 31, 33).
`Two case-control studies used a nested case-control
`
`Table 1. Study Quality Scores
`
`Total
`Bonus
`Baseline
`Study
`(reference)
`(range,
`(range,
`(range.
`0)-46)*
`0-5)T
`0-51)
`
`
`Griffin et al. (33)
`Levyet al. (32)
`McIntoshet al. (35)
`Somerville et al. (26)
`Bartle et al. (27)
`Henryet al. (30)
`Jick et al. (31)
`Carsonetal. (24)
`Guess et al. (25)
`Bloom (22)
`Beardet al. (23)4
`Beardonetal. (21)
`Armstrong and Blower(29)
`Collier and Pain (28)
`Jick et al. 34)¢
`Alexanderet al. (36)
`
`25.5
`24.5
`22.5
`22.5
`23.0
`20.5
`20.0
`15.5
`16.0
`14.5
`14.5
`13.5
`14.5
`13.5
`10.0
`9.50
`
`4.00
`5.00
`5.00
`4.00
`3.00
`2.00
`1.00
`5.00
`3.00
`4.00
`4.00
`2.00
`0.00
`1.00
`2.00
`1.00
`
`29.5
`29.5
`27.5
`26.5
`26.0
`22.5
`21.0
`20.5
`19.0
`18.5
`18.5
`15.5
`14.5
`14.5
`12.0
`10.5
`
`* Baseline scores were assigned based on an evaluation of the fol-
`lowing design items: explicit definitions of exposure, outcome, case and
`control status as well as the use ofblinding and matching.
`+ Bonus points were assigned based on the numberof confounders,
`which were accounted for in the analysis. See text for method of
`bonus-point assignment.
`£ Cohort studies.
`
`788
`
`15 November 1991 © Annals of Internal Medicine + Volume 115 + Number10
`
`Page3 of 11
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Page 3 of 11
`
`

`

`Individual Study Odds Ratios
`
`and
`study
`Figure 2. Individual
`Individual
`summary odds
`ratios.
`study odds ratios are arranged in
`order of
`increasing sample size
`(top to bottom).
`Individual study
`odds ratios were provided in the
`original studies (21-28, 30-32, 34,
`36) or calculated from data pro-
`vided in original studies (29, 33,
`35).
`(@, Individual study odds ra-
`tio; ®, summary odds ratio;
`the
`95% confidence intervals are indi-
`cated by the extended lines: * co-
`hort
`study;
`f case-control study;
`+t odds ratios summarized by ‘‘di-
`rect’? technique [40]; numbers in
`parentheses
`are
`the number of
`studies combined.)
`
`Reference No.
`32
`35
`25
`29
`36
`28
`27
`26
`33
`31
`30
`34
`23
`
`
`
`
`
`
`
`
`
`——e—— Gi surgery (3)
`
`——@— Gl death (4)
`
`
`Summary Odds Ratios *
`
`© Overall (16)
`Case control
`® Cohort
`
`Gl bleeding (9)
`
`1.0
`
`10
`
`OddsRatio
`
`design (31, 33). Determinations of NSAID exposure
`were made by an unblinded reviewofclinical notes
`(28-30), a structured questionnaire with interviewers
`who were blinded (26, 27, 32, 35), or an extraction of
`prescription data from pharmacy computerfiles (31, 33).
`In all cohort studies,
`the assessment of NSAID expo-
`sure was based on prescription files. Estimates of the
`duration of NSAID exposure varied from 30 days (24,
`25)
`to 90 days (22, 23, 31, 34). One cohort study (25)
`examined deaths from gastrointestinal causes, whereas
`the remainder looked at hospitalizations caused by gas-
`trointestinal complications. Samples examined in the
`cohort studies included the Group Health Cooperative
`in Puget Sound; the Pennsylvania Medicaid group; the
`residents of Saskatchewan, Canada; and the residents
`of the Tayside Region, Scotland. The Puget Sound
`Group Health Cooperative represents a younger, em-
`ployed population,
`the Medicaid group is elderly, and
`the Tayside and Saskatchewan groups represent resi-
`dents of geographically diverse districts.
`
`Study Quality
`
`Table 1 shows the study quality scores. Methodologic
`assessment of the 16 studies showed acceptable agree-
`Ment between two observers for the six study quality
`Categories evaluated (mean kappa, 0.70; minimum, 0.56;
`Maximum, 0.83). The mean kappa for the quality cate-
`gory of blinding was 0.67 (minimum, 0.0; maximum,
`1.0); for case selection, 0.75 (minimum, 0.66; maximum,
`0.90); for control selection, 0.68 (minimum, 0.4; maxi-
`
`mum, 1.0); for definition of exposure, 0.74 (minimum,
`0.59; maximum, 0.96);
`for matching technique, 0.83
`(minimum, 0.66; maximum, 1.0); and for definition of
`outcome, 0.56 (minimum, 0.0; maximum,
`1.0). Dis-
`agreements regarding control of confounders were re-
`examined and resolved by consensus. The six studies
`with the highest quality scores were case-control stud-
`ies (Table 1). These studies gave more explicit defini-
`tions of cases, controls, and exposure and used blinding
`more frequently. The study quality score was not found
`to be a significant covariate in the regression model
`(P > 0.2).
`
`Summary Odds Ratios
`
`Published odds ratios and summary odds ratios from
`the primary studies are shown in Figure 2. The overall
`odds ratio of the risk for adverse gastrointestinal events
`related to NSAID use (summarized from 16 case-con-
`trol and cohort studies) is 2.74 (CI, 2.54 to 2.97). The
`summary odds ratio (combined from 8 studies) for el-
`derly persons is 5.52 (CI, 4.63 to 6.60). In the cohort
`studies, the term ‘‘elderly’’ refers to persons 65 years of
`age or older. In the case-control studies, ‘‘elderly”’ re-
`fers to persons 60 years of age or older. The summary
`odds ratio for nonelderly persons, combined from 3
`studies,
`is 1.65 (CI, 1.08, 2.53). These data show a
`greater than threefold increase in relative risk for seri-
`ous gastrointestinal events among elderly NSAID users
`when compared with nonelderlyusers.
`Odds ratios were subdivided by gastrointestinal out-
`
`15 November 1991 * Annals of Internal Medicine * Volume 115 * Number10
`
`789
`
`Page 4 of 11
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Page 4 of 11
`
`

`

`Table 2. Comparison of Summary Odds Ratios and Confidence Intervals Obtained by Two Methods
`Category
`Numberof Studies
`Summary Odds Ratio
`95% Cl
`Combined
`
`Overall
`12*/167
`2.86*/2.74+
`2.62 to 3.12*; 2.54 to 2.97)
`Patient = 60 years of age
`6/8
`6.24/5.52
`5.21 to 7.48; 4.63 to 6.60
`Patient < 60 years of age
`2/3
`3.07/1.65
`1.62 to 5.82; 1.08 to 2.53
`Gastrointestinal bleeding
`7/9
`2.71/2.39
`2.26 to 3.24; 2.11 to 2.70
`Gastrointestinal surgery
`3/3
`7.04/7.75
`5.34 to 9.29; 5.83 to 10.31
`Gastrointestinal cause of death
`3/4
`4.22/4.79
`3.24 to 5.50; 3.64 to 6.22
`Unspecified adverse gastrointestinal event
`2/3
`2.68/1.79
`2.42 to 2.98; 1.70 to 1.90
`* Mantel-Haenszel technique for case-control studies only.
`+ Direct technique method of Greenland (reference 40).
`
`1.92 (CI, 1.19 to 3.13). The highest odds ratios were
`obtained from studies in which the duration of NSATD
`consumption was less than 1 month.
`Data were also subdivided by gastrointestinal eveit
`and age (Table 3). The relative risk for gastrointestinal
`surgery for nonelderly individuals, combined from three
`studies, was 0.44 (CI, 0.29 to 0.66), whereasthe risk for
`gastrointestinal surgery among elderly persons, com-
`bined from three studies, was 10.42 (CI, 7.40 to 14.66).
`These data suggest a tenfold increase in relative risk for
`gastrointestinal surgery among elderly users when com-
`pared with younger users.
`Estimates of the prevalence of serious gastrointestinal
`events among NSAID users were summarized from four
`cohort studies (7, 23, 25, 34). The summary, 1-year
`prevalence among NSAID users was | per 1000;
`the
`prevalence among elderly users (= 65 years of age) was
`3.2 per 1000; and the prevalence among younger users
`(< 65 years of age),was 0.39 per 1000.
`
`Sources of Heterogeneity
`
`Tests for homogeneity were statistically significant
`(P < 0.05) for all analyses,
`indicating that
`the differ-
`
`Summary Odds Ratios, Risk Factors
`= Overall (16)
`
`—)————_ 260 yr (8)
`
`——$=—=_ <60 yr (3)
`—*— Women (3)
`—+— Men (3)
`— First GI event(6)
`——e———_ Subsequent or unspecified GI event (10)
`—~e———_ Concomitant corticosteroids (3)
`
`Duration of NSAID Consumption
`—te—————_ >3 months (2)
`
`—<»——————_ >1, <3 months (2)
`
`<n6 ——_—_—X—X—3n—=—X—_——— OOO
`
`come. The odds ratio for gastrointestinal bleeding, com-
`bined from nine studies, was 2.39 (CI, 2.11 to 2.70).
`The odds ratio for gastrointestinal surgery, combined
`from three studies, was 7.75 (CI, 5.83 to 10.31). The
`summary odds ratio for gastrointestinal death, com-
`bined from four studies, was 4.79 (CI, 3.64 to 6.22).
`Thus,
`the relative risk for surgical or fatal outcomes
`among NSAID users is 2- or 3-fold higher than the
`relative risk for gastrointestinal bleeding.
`The summary odds ratio for women was 2.32 (Cl,
`1.91 to 2.82), whereas the summary oddsratio for men
`was 2.40 (CI, 1.85 to 3.11). The summary odds ratio for
`women compared with men was 1.15 (CI, 0.89 to 1.50).
`These data do not support gender as an independent
`risk factor. The risk for first compared with subsequent
`gastrointestinal event was also examined. The summary
`odds ratio for the first gastrointestinal event, combined
`from six studies, was 2.39 (CI, 2.16 to 2.65). The rela-
`tive risk for subsequent or unspecified gastrointestinal
`event, combined from the remaining 10 studies, was
`4.76 (CI, 4.05 to 5.59). These data suggest that patients
`with a history of gastrointestinal events may have an
`increased relative risk for further events. The use of
`concomitant corticosteroids was also examined. The
`summary odds ratio for NSAID users receiving con-
`comitant corticosteroids compared with NSAID users
`not receiving corticosteroids was 1.83 (CI, 1.20 to 2.78).
`This finding suggests an approximately twofold increase
`in the relative risk among NSAID users whoare receiv-
`ing corticosteroids compared with NSAID users not
`receiving corticosteroids.
`Summary odds ratios were also obtained using the
`Mantel-Haenszel statistic. A comparison of the results
`obtained by the two statistical techniques showed that
`the direct method enabled the use of data from more
`studies, resulting in narrower CIs. Summary odds ratios
`by both methods were similar in most categories (Table
`2).
`Summary odds ratios calculated according to individ-
`ual NSAID used and duration of NSAID exposure were
`as follows: piroxicam, 11.12 (CI, 6.19 to 20.23);
`indo-
`methacin, 4.69 (CI, 2.97 to 7.41); aspirin, 3.38 (CI, 2.26
`to 5.01); naproxen, 2.84 (CI, 1.68 to 4.82); and ibu-
`profen, 2.27 (CI, 1.85 to 2.80). There is substantial
`overlap in the Cls among NSAIDs. The duration of
`NSAID consumption may be related to the size of the
`odds ratio (Figure 3). The summary odds ratio for less
`than | month of NSAID exposure was 8.00 (CI, 6.37 to
`10.06); for longer than 1 month but less than 3 months,
`3.31 (CI, 2.27 to 4.82); and for longer than 3 months,
`
`
`
`6
`4
`Odds Ratio, Direct Method
`
`8
`
`10
`
`1.0
`
`2
`
`Figure 3. Summary odds ratios and risk factors. (®, Summary
`odds ratio; the 95% confidence intervals are indicated by the
`extended line; numbers in parentheses are the number of stud-
`ies combined.)
`
`790
`
`15 November 1991 * Annals of Internal Medicine + Volume 115 * Number 10
`
`Page 5 of 11
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Page 5 of 11
`
`

`

`the results
`determination of NSAID exposure bias
`toward a falsely large relative risk. The use of a struc-
`tured interview administered by an investigator who is
`blinded to the status (case patient or control) of the
`patient results in more valid estimates of relative risk
`(17, 26). Well-designed, nested, case-control studies
`minimize the selection bias, inherent in hospital-based
`case-control studies (33).
`Although there have been many studies examining
`the gastrointestinal
`risks of NSAID use,
`important
`methodologic limitations and differences in study char-
`acteristics contribute to the conflicting results. Retro-
`spective cohort studies probably underestimate the rel-
`ative risk, whereas some case-control studies probably
`overestimate it. The aggregation of the results from
`observational studies is controversial
`(46). The stron-
`gest studies are those that defined cases, controls, out-
`come, and exposure accurately and reproducibly (26,
`27, 32, 33, 35), as reflected by the quality-assessment
`scores in this meta-analysis (Table 1).
`We conducted a structured overview of all previous
`reviews of NSAID-related adverse gastrointestinal
`events. The quality of the 10 reviews selected (3, 6-12,
`18, 19) was assessed according to several criteria: the
`comprehensiveness of the literature search, the minimi-
`zation of bias in the selection of primary studies, the
`assessment of the quality of the primary studies,
`the
`appropriateness of the techniques used in data synthe-
`sis, and the validity of the conclusions made by the
`authors as supported by the data. Most of the published
`reviews on this topic cite only a portion of the available
`literature, do not provide a critical assessment of the
`quality of the studies cited, and fail
`to combine the
`results of these studies statistically. Only 1 of the 10
`reviews used a clearly defined, comprehensive search
`strategy (6). Inclusion criteria were stated for 2 of the
`10 reviews (6, 8). A quality assessment of the studies
`was done in only | review (6). Appropriate, explicitly
`stated methods of data synthesis were given in only 2
`reviews (6, 19).
`
`Table 3. Subgroup Odds Ratios Combined from Case
`Control and Cohort Studies Using the ‘‘Direct’’ Method*
`Variable
`Number
`Summary
`95%
`of Studies
`Odds
`cl
`
`Combined
`Ratio
`
`Gastrointestinal event by
`age*
`< 60 years
`Gastrointestinal
`bleeding
`Gastrointestinal
`surgery
`= 60 years
`Unspecified gastro-
`intestinal adverse
`event
`Gastrointestinal
`bleeding
`Gastrointestinal
`surgery
`3
`10.42
`7.40 to 14.66
`Gastrointestinal
`
`4 4.40cause of death 3.35 to 5.79
`
`I
`
`3
`
`3
`
`9
`
`
`
`1.03
`
`0.44
`
`1.78
`
`2.38
`
`
`
`0.60 to 1.76
`
`0.29 to 0.66
`
`1.69 to 1.87
`
`2.10 to 2.69
`
`ences among the results of individual studies are greater
`than can be expected on the basis of chance alone.
`We did two different
`types of analyses to identify
`sources of heterogeneity. Heterogeneity across studies
`is composed of intrastudy heterogeneity and inter-study
`heterogeneity. In an effort to describe intra-study het-
`erogeneity,
`tests of homogeneity were conducted for
`sever2! subgroups across studies. These subgroups were
`subdivided according to gastrointestinal outcome, age,
`age and gastrointestinal outcome, and use of individual
`NSAIDs. Each of these subgroups accounted for a por-
`tion of the variability, thus reducing the test statistic for
`homogeneity. There was, however, no subgroup identi-
`fied that accounted for most of the observed heteroge-
`neity.
`In an effort to describe interstudy heterogeneity,
`we did a multivariate regression analysis using the log
`of the study odds ratio as the dependent variable and
`study design, duration of NSAID use, gastrointestinal
`outcome, and average age as the independent variables.
`The regression was weighted using the individual study
`variances. The four independent variables accountedfor
`approximately half of the interstudy variability.
`
`Discussion
`
`Two research designs have been used to study the
`risk for gastrointestinal events related to NSAID ther-
`apy:
`retrospective cohort and case-control
`studies.
`Most of the cohort studies used secondary analysis of
`health insurance registries in which data were collected
`primarily for billing purposes. The computerized case
`definition for gastrointestinal events is subject to sub-
`stantial misclassification (42-44). Misclassification rates
`of up to 29% were noted in studies using retrospective
`chart review to confirm computerized diagnoses (23, 31,
`34), resulting in contamination of the case group by
`controls and of the control group by cases and thus
`reducing the relative-risk estimate. Similarly, the infor-
`mation on NSAID exposure obtained from these regis-
`tries may not have been of optimal quality. The dura-
`tion
`of NSAID exposure
`is often unknown and
`assumptions are made from prescription registries re-
`garding the average duration of NSAID use. Some stud-
`ies estimated an average prescription duration of 90
`days with full patient compliance (23, 31, 34). Such an
`assumption may overestimate the duration of NSAID
`exposure, biasing the results toward a falsely low rela-
`tive risk. The frequency of NSAID use in a study sam-
`ple determines the power of that study to detect. a
`Statistically significant relative risk (45). Nonsteroidal
`anti-inflammatory drug use among patients with prepaid
`health plans may be lower than that of the general
`population,
`further underestimating the relative risk.
`These factors contribute to the lower relative risks re-
`ported by the cohort studies when compared with the
`Case-control studies.
`In two case-control studies, different techniques were
`used to determine NSAID exposure among case pa-
`tients and controls (28, 29). Physicians hospitalizing pa-
`ents with gastrointestinal bleeding are more likely to
`Inquire about NSAID use than are physicians question-
`ig controls or their relatives. Such differences in the
`
`* Gastrointestinal events occurring in hospitalized patients.
`
`15 November 1991 * Annals of Internal Medicine « Volume 115 * Number 10
`
`791
`
`Page6 of 11
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Patent Owner Ex. 2001
`Mylan v. Pozen
`IPR2017-01995
`
`Page 6 of 11
`
`

`

`
`
`Meta-analysis is a systematic, quantitative, strategy
`of reviewing and summarizing data from the literature
`to address a specific research question. It differs from
`the traditional
`review article in that
`it uses explicit
`inclusion and exclusion criteria,
`incorporates a stan-
`dardized quality assessment, and provides a quantita-
`tive estimate of effect size. In this way, meta-analysis
`reduces the potential for error and bias implicit in the
`traditional review article (47).
`Meta-analyses have beencriticized for their emphasis
`on statistical techniques and their lack of attention to
`critical descriptions of methodologic and substantive is-
`sues discussed in the individual studies. The ‘‘best-
`evidence synthesis’” method combines the strengths of
`quantitative meta-analytic
`techniques with detailed,
`qualitative analysis of study characteristics typical of
`traditional review articles (48). We have examined crit-
`ically the study characteristics and quality and have
`provided a quantitative summary of the relative risks.
`Because meta-analysis is a retrospective form of re-
`search, it is limited by any biases inherent in the pri-
`mary studies. As with any reviewarticle, meta-analysis
`is subject to the preferential selection of studies dem-
`onstrating significant results (49). This publication bias
`is most problematic in studies of effectiveness in which
`it
`is assumed that studies showing no effect are less
`attractive to publishers and,
`therefore, remain unpub-
`lished. Such bias is less likely in studies of risk,
`in
`which the protective effect of an exposure on health
`status is of equal interest as the negative effect. Studies
`showing a protective effect of NSAIDs on the gastroin-
`testinal mucosa would be of great
`interest. Studies
`showing norisk for gastrointestinal complications asso-
`ciated with NSAIDs would also be of interest. Using
`the data from the 16 studies in this overview, we de-
`termined that it would require having missed approxi-
`mately 300 studies showing no gastrointestinal effect of
`NSAIDsto bring the summary odds ratio to unity. We
`believe such a scenario to be unlikely.
`An assumption underlying most meta-analyses is that
`of homogeneity, the belief that