CLINICAL STUDIES
`
`Nonsteroidal Anti-Inflammatory Drug-Associated
`Gastropathy: Incidence and Risk Factor Models
`JAMES F. FRIES, M.D., CATHERINE A. WILLIAMS, M.A., DANIEL A. BLOCH,Ph.D.,
`BEAT A. MICHEL, M.D., Stanford, California
`
`PURPOSE: The most prevalent serious drug tox-
`icity in the United States is increasingly recog-
`nized as gastrointestinal (GI) pathology associ-
`ated with the use of nonsteroidal
`anti-inflammatory drugs (NSAIDs). The inci-
`dence of serious GI events (hospitalization or
`death) associated with NSAID use was therefore
`prospectively analyzed in patients with rheuma-
`toid arthritis (RA) and patients with
`osteoarthritis.
`PATIENTS, METHODS, AND RESULTS; The study
`consisted of 2,747 patients with RA and 1,091 pa-
`tients with ostecarthritis. The yearly hospital-
`ization incidence during NSAID treatment was
`1.58% in RA patients and was similar in all five
`populations studied. The hazard ratio of patients
`taking NSAIDs to those not taking NSAIDs was
`5.2. The incidence in osteoarthritis may be less.
`The risk of GI-related death in RA patients was
`0.19% per year with NSAIDs, Multivariate anal-
`yses assessing risk factors for serious GI events
`were performed in the 1,694 (98 with an event)
`RA patients taking NSAIDs at the predictive
`visit. The main risk factors were higher age, use
`of prednisone, previous NSAID GIside effects,
`prior GI hospitalization, level of disability, and
`NSAID dose. A rule is presented that allows esti-
`mation of the risk for the individual patient with
`RA.
`CONCLUSION: Knowledgeof the risk factors for
`NSAID-associated gastropathy andtheir inter-
`relationships provides a tool for identification of
`the patient at high risk and for initiation of ap-
`propriate therapeutic action.
`
`From the Division of Immunology and Rheumatology, Departmentof
`Medicine, Stanford University Schoo! of Medicine, Stanford, California.
`This work was supported by a grant from the National Institutes of
`Health (AR21393) to ARAMIS (the Arthritis, Rheumatism, and Aging
`Medical
`Information System) and in part by a grant from Searle
`Laboratories.
`Requests for reprints should be addressed to James F. Fries, M.D.,
`1000 Welch Road,Suite 203, Palo Alto, California 94304.
`Manuscript submitted November 30, 1990, and accepted in revised
`
`form March18, 1991.
`
`astrointestinal (GI) pathology associated with
`the use of nonsteroidal anti-inflammatory
`drugs (NSAIDs) is increasingly recognized as the
`most prevalent serious drug toxicity in the United
`States, resulting in an estimated 2,600 deaths and
`24,000 hospitalizations annually in patients with
`rheumatoid arthritis (RA) alone [1,2]. The predom-
`inant syndromeconsists of antral prepyloric ulcers,
`which may eventuate in GI hemorrhage or perfora-
`tion, although events in the duodenum,small bow-
`el, and the large bowel are also seen. Ulcerations
`visible on endoscopy havea point prevalence of 10%
`to 25%, and severe erosions are seen in additional
`patients [3-5]. The risk of GI hospitalization has
`been estimated at 1% to 1.5% per year in persons
`taking NSAIDs[1], and the risk of death is approxi-
`mately 0.13% per year in individuals treated with
`NSAIDs [1,6,7]. The importance of the syndrome
`has been emphasized by gastroenterologists [3,8,9],
`rheumatologists [2,5,10], and the Food and Drug
`Administration (FDA) [11].
`Importantinformation required for estimation of
`the magnitudeofthe problem and for development
`of strategies for resolution, however, has been lack-
`ing. For example, the prevalence of complications in
`conditions other than RA, such as osteoarthritis,
`has not been established. Generalizability of the
`observationsto different practice sites has not been
`presented. Quantitation of likely risk factors such
`as prior bleeding has not been reported, and the
`frequency of deaths has not been confirmed by pro-
`spective study. Most importantly, while individual
`risk factors have been suggested by a number of
`investigators [1,12,13], no multivariate risk factor
`modelthat permits estimation ofrisk in the individ-
`ual patient has been presented.
`This report addresses these issues in two steps:
`(1) with descriptive analyses of 2,747 patients with
`RA followed prospectively for an average of 4 years
`at five ARAMIS (Arthritis, Rheumatism, and Aging
`Medical Information System) data bank centers
`[14,15] and 1,091 patients with osteoarthritis, and
`(2) with risk factor analyses based on the 1,694 of
`these RA patients taking NSAIDsat the predictive
`visit.
`
`September 1991 The American Journal of Medicine Volume91
`
`213
`
`Page 1 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`Page 1 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`

`

`NSAID-ASSOCIATED GASTROPATHY/ FRIES ET AL
`
`TABLE|
`
`
`
`Rheumatoid Arthritis Gastrointestinal (GI) Hospitalization by Center
`
`
`Number of patients
`Person-years of observation
`Person-years taking NSAIDs
`Gl hospitalizations
`Numberof patients
`Rate per person-year (%)
`Number taking NSAIDs
`Rate per year while taking NSAIDs (%)
`Numberof years of observationafter Ist
`hospitalization
`Numberof additionalG! hospitalizations
`while taking NSAIDs
`Rate forat leastonemore Gl erie
`tion per yearwhile taking NSAIDs (%)
`Numberof patients with upper Gl hospi-
`talizations
`Rate of upperGI eins peryear
`while taking NSAIDs (%)
`Numberof patients with lower G! hospi-
`talizations
`Rate of lower GI hospitalizations per year
`while taking NSAIDs (%)
`
`PATIENTS AND METHODS
`
`Two thousand seven hundred and forty-seven
`patients with RA consecutively enrolled and fol-
`lowed at five ARAMIS centers,for a total of 9,525
`years of observation, were available for study (Ta-
`ble I). The Santa Clara County population of 302
`patients represents a community population re-
`cruited by advertisement. The other populations
`were formed by consecutive patient accrual at the
`site. The 679 Saskatoon patients are believed to
`make up the great majority of patients in Northern
`Saskatchewan province, the 307 Phoenix patients
`were drawn from a rheumatology private practice,
`as were the 1,080 Wichita patients, and the 379
`Stanford patients were enrolled from a tertiary care
`referral center. Data are collected in two modes.
`First, all routine clinical data including diagnosis,
`symptoms,signs, demographics,past history, labo-
`ratory tests, and treatment are entered for each
`patient encounter and hospitalization. Second, pa-
`tients complete the Health Assessment Question-
`naire (HAQ)[1,16-18] at 6-month intervals, provid-
`ing validated self-report of disability, discomfort,
`drug toxicity, and economic impact. All hospitaliza-
`tions and deaths are audited by abstraction of dis-
`charge summaries and NSAID usage at time of
`event confirmed. Deaths are reviewed by death cer-
`tificate discharge summary andrecent clinical
`notes. The system, procedures, andvalidation tech-
`niques have been previously described [2,14~18].
`The dependent variable in this study was “GI
`event,” consisting of an eventsufficiently serious as
`
`to result in hospitalization or death. Events are
`counted only when they are the primary basis for
`hospitalization or death, notif they occurred during
`a hospitalization or as part of a terminal illness
`sequence. Patients are considered to be taking
`NSAIDs if they report on the HAQ immediately
`prior to the event that they are taking any oneof the
`following drugs: aspirin, naproxen, ibuprofen,pir-
`oxicam,
`indomethacin, sulindac, meclofenamate,
`tolmetin, fenoprofen, ketoprofen, nonacetylated sa-
`licylates, salsalate, diflunisal, or diclofenac.
`Forrisk factor delineation, items considered were
`those available at the HAQ prior to the event—the
`predictive visit. Patients had had from oneto 13 6-
`month periods of observation. Obviously, patients
`analyzed for risk factors for NSAID gastropathy
`were required to be taking an NSAID atthe predic-
`tive visit. Among the 130 patients with GI events, 32
`patients were not included in the analysis because
`they were nottaking NSAIDs at the predictive visit,
`or did not have HAQ data within 9 months prior to
`the event. Therefore, 98 patients with GI events
`were available for comparison with patients with-
`out GI events (controls). Similarly, of the 2,617
`(2,747 minus 130) patients eligible as controls, 1,021
`were excluded because they did not meet one or
`moreof the abovecriteria. By definition, the non-
`event group could not have a GI event; however,
`control patients’ records were randomly truncated
`to match the numberof 6-month periods in which
`events occurred in patients with GI events. For ex-
`ample, 5% of the GI events occurred at the seventh
`
`214
`
`September 1991 The American Journal of Medicine Volume 91
`
`Page 2 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`Page 2 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`

`

`NSAID-ASSOCIATED GASTROPATHY/ FRIES ET AL
`
`
`TABLEI
`Variables Associated with Gastrointestinal (Gl) Hospitalization or Death at Predictive Visit
`
`Gl Events (n = 98)
`Number
`Mean
`Defined
`(SE)
`
`No Gi Events (n = 1,596)
`Number
`Mean
`Defined
`{SE)
`
`p Value
`
`98
`95
`98
`98
`7§
`92
`98
`93
`93
`74
`93
`
`65.47 (1.05)
`11.94 (0.25)
`18.75 (1.20)
`1.69(0.08)
`0.10 (0.04)
`0.20 (0.08)
`0.53 (0.04)
`1.03 (0.06)
`0.45 (0.06)
`0.28 (0.05)
`0.53(0.07)
`
`1,596
`1,525
`1,583
`1,596
`1,287
`1,529
`1,558
`1,582
`1,583
`1,314
`1,583
`
`58.65 (0.33)
`12.50 (0.07)
`16.86 (0.28)
`1.38 (0.02)
`0.17 (0.01)
`0.44 (0.12)
`0.59 (0,01)
`0.91 (0.01)
`0.56 (0.01)
`0.09 (0.01)
`0.22 (0.01)
`
`<0.001
`0.06
`O11
`<0.001
`0.08
`0.09
`0.15
`<0.05
`0.07
`<0.001
`<0.001
`
`Continuous variables
`Age (years)
`Education level (years)
`Disease duration {years}
`Disability index (0~3)
`Smoking (packs/day)
`Alcohol (drinks/day)
`Specialty care
`NSAID dose (see text)
`Numberof DMARDs
`Numberof Hz-antagonists
`Numberof antacids or Hp-antagonists
`Categorical variables (% positive)
`0.97
`76.7
`1,596
`76.5
`98
`Female sex
`0.06
`94.5
`1,545
`98.9
`95
`White race
`0.14
`17.2
`1,287
`10.7
`75
`Smoker
`0.21
`18.2
`1,529
`13.0
`92
`Alcohol
`<0.001
`18.6
`1,583
`32.3
`93
`NSAIDGl side effect ever
`<0.001
`31.0
`1,583
`51.6
`93
`rednisone
`<0.01
`10.1
`1,312
`20.3
`74
`Antacids
`<0.001
`91
`1,314
`28.4
`74
`Ho-antagonists
`Antacids or Hp-antagonists <0.001 93 40.9 1,583 19.2
`
`
`
`
`
`
`6-month observation period and,therefore, predic-
`tion was made using information at the sixth obser-
`vation period;similarly, a random 5% of the control
`subjects were analyzed at the sixth observation pe-
`riod as predictive of an “event” at the seventh ob-
`servation period. This method was applied to each
`of the 13 possible 6-month observation periods to
`adjust for differences that would otherwise have
`been present in the amount of data available for
`prediction in cases and controls. The mean time
`interval between the predictive data at the HAQ
`prior to the event and the HAQ associated with the
`event was 6.1 months.
`Univariate analyses were performed using a t-test
`for 11 continuous variables and by using a chi-square
`test on 2 X 2 tables for nine binary variables. The p
`Variable
`values are those computed for each individual com-
`parison and may be adjusted for multiple compari-
`sons by the Bonferroni adjustment. All variables sta- a
`tistically significant as predictors at the 0.05 level
`Smoker
`(two-tailed) and several additional variables consid-
`oecomplsht
`ered as potentially important were further analyzed
`Prednisone
`by multivariate analyses. Stepwise multiple logistic
`antagonist
`regression and recursive partitioning (classification
`Antacid or Ho-antagonist
`and regression tree) analyses were employed [19].
`NSAID dose > 1.0 (see text)
`RESULTS
`> sOseer
`Incidence
`eHyears
`> 70 years
`Ofthe 2,747 RA patients available for study, 130
`> 75 years
`Disability index (0-3)
`had GIevents sufficiently serious to result in hospi-
`>L
`>2
`talization or death. When hospitalizations were
`tabulated in these 130 patients, 116 patients had
`
`128 hospitalizations (10 patients had two hospital-
`izations and one patient had three) and 17 had GI-
`related deaths. Among the 17 GI deaths, three pa-
`tients had had prior GI hospitalizations. Data
`presented in Table I describe the 116 hospitaliza-
`tion events (the last event among patients with
`multiple hospitalizations). Data subsequently pre-
`sented in Tables II through VI result from analysis
`ofthe last GI event (death or hospitalization). Four-
`teen patients had previous GI events (three pa-
`
`TABLE tll
`OddsRatios for Selected Variables with Respect to Gastrointestinal (GI)
`Hospitalization or Death
`
`1.23-2.91
`
`95%Confidence
`
`pads
`10
`0.6
`of
`24
`23
`29
`14
`ie
`a2.0
`2.2
`
`89
`
`ee
`
`September 1991 The American Journal of Medicine Volume91
`
`215
`
`Page 3 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`Page 3 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`

`

`NSAID-ASSOCIATED GASTROPATHY/ FRIES ET AL
`
`
`TABLE \V
`Risk Factors for Gastrointestinal(Gl) Complications Related to Hospitalization or Death: Stepwise Logistic Regression
`
`for Odds Ratios 1,03-1.07
`
`Step
`Number
`
`Variable
`
`Age (years)
`Prednisone
`Previous NSAIDGl side effect
`NSAID
`dose
`Disability index (0-3)
`
`;
`Coefficient (SE)
`
`0.047(0.01)
`
`Improvement
`p Value
`<0,0001
`
`95% Confidence Interval
`
`1.15-1.76
`1.18-1.86
`0.95-1.90
`0.92-1.59
`
`TABLE V
`
`Characteristics of Classification Tree Subgroups with Respect to Gastrointestinal (G1) Hospitalization or Death
`
`GI Event Rate
`Subgroup
`NumberofPatients
`PercentofPatients
`YearTakingNSAIDs
`Number
`Characteristics
`GI/NoG! Event
`with GIEvent
`(%
`
`nsone
`1
`Age 76 years or older, no pred-
`12/79
`13.2
`4.2
`
`
`2
`
`Age 48 years or older, taking pred-
`nisone, disease duration > 3.7
`years,disability index > 1.7
`3
`Age 48 to 63 years, previous
`136
`16.3
`38
`NSAID side effects, disability
`index > 1.3
`
`4
`Gl event group: 10 second events
`3.1
`outof 98 (10.2%)
`
`33/199
`
`14.2
`
`3.9
`
`being taken, the overall frequency was 1.58% per
`tients with Gl-related deaths who had previous GI
`year. When the analysis was limited to upper GI
`hospitalizations, and 11 patients with multiple GI
`hospitalizations).
`hospitalizations, the rate during treatment with
`NSAIDs was 1.4% per year.
`Of the 116 last hospitalizations, 107 occurred
`Table I breaks down these hospitalizations by
`while patients were taking NSAIDs. Ninety-five of
`data bank center to evaluate the generalizability of
`these were noted as upper GI problems on discharge
`summaries, and 82 of these were gastric in location.
`the observations. The percentage of GI hospitaliza-
`Twelve were localized to the lower GI tract. The
`tions per year during NSAID treatment ranged
`from 1.2% in Santa Clara County to 1.8% in Saska-
`overall rate of GI hospitalizations per year of obser-
`toon and in Phoenix. Upper GI hospitalizations
`vation was 1.2%. During periods when NSAIDs were
`
`
`
`
`
`
`
`
`
`
`5
`
`Age 48 years orolder, taking pred-
`nisone,disease duration > 3.7
`years,disability index 0.3 to
`l7
`
`
`13148
`
`8.1
`
`2.2
`
`
`
`
`
`
`
`
`
`
`
`6
`7
`
`Age 63 to 76 years, no prednisone
`Age < 47 years
`
`27/369
`3/342
`
`68
`0.9
`
`1.9
`03
`
`8
`Age 47 to 63 years, no pred-
`3340
`0.9
`nisone, never NSAIDside ef-
`fects
`
`
`03
`
`9
`Age 47 to 63years,previous
`(37
`0
`0
`NSAID side effects, disability
`
`index < 1.3, no prednisone
`
`10
`Age 47 years or older,taking pred-
`(24
`0
`0
`nisone, disease duration < 3.7
`years
`
`
`ll
`
`(22
`
`0
`
`0
`
`
`
`
`
`Age 47 years or older, taking pred-
`nisone, disease duration > 3.7
`years, disability index <0.3
`
`
`216
`
`September 1991 The American Journal of Medicine Volume 91
`
`Page 4 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`Page 4 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`

`

`TABLE VI
`Description of Selected Variables Within Subgroupsof Tree (Figure 1)*
`
`
`
`NSAID-ASSOCIATED GASTROPATHY/ FRIES ET AL
`
`All
`
`1,694
`
`1
`
`91
`
`2
`
`232
`
`3
`
`43
`
`Disability index (0-3)
`Previous NSAID-related
`Glside effects (%)
`NSAIDdose (see text)
`
`
`Numberof patients
`
`Variables
`66
`57
`56
`39
`69
`63
`65
`57
`65
`79
`59
`Age (years)
`
`
`
`
`
`
`
`
`(03) (0.6)=(1.8)(0.3) (0.6) (0.7) (Ll (0.6) (0.2} {0.3) (0.2)
`
`
`Disease duration (years)
`17
`18
`21
`16
`19
`19
`20
`12
`17
`15
`2
`
`(0.3)
`(1.2)
`(0.7)
`(15)
`(L2
`(0.9)
`(0.7)
`(0.4)
`(0.5)
`(0.1)
`(0.2)
`
`14
`17
`23
`1.9
`17
`12
`1.5
`il
`1.2
`0.7
`15
`(0.02)
`(0.1)
`(0.02)
`(0.07)
`(0.1)
`(0.03)
`(0.04)
`(0.04)
`(0.04)
`(01)
`(0.2)
`20
`13
`20
`100
`32
`14
`20
`24
`0
`100
`8
`
`09
`0.8
`10
`1.0
`10
`0.9
`0.9
`09
`1.0
`1.0
`13
`
`(0.01)
`(0.1)
`(0.03)
`(0.1)
`(0.1)
`(0.04)
`(0.03)
`(0.03)
`(0.03)
`(0.1)
`(0.2)
`30
`0
`100
`0
`52
`100
`0
`29
`0
`0
`100
`Use of prednisone
`
`20
`20
`35
`26
`41
`28
`21
`15
`11
`24
`21
`Use of antacids or
`He-antagonists (%)
`
`Timetaking NSAIDs
`4l
`38
`43
`51
`40
`43
`43
`38
`41
`45
`16
`(months)
`(0.5)
`(2.2)
`(1.3)
`(3.3)
`(2.1)
`(1.6)
`(LL)
`(1.1)
`(1.2)
`(3.2)
`(1.2)
`
`“Valuesare means (SE).
`
`Subgroup Number (G! Event Group)
`4
`5
`6
`7
`
`98
`
`161
`
`396
`
`345
`
`8
`
`343
`
`9
`
`37
`
`10
`
`24
`
`ranged from 1.2% per year with NSAID treatment
`in Santa Clara County patients to 1.6% in Saska-
`toon. Results at all centers were similar, without
`statistically significant differences.
`Table I also lists patients taking NSAIDs who
`were hospitalized more than once. One subject was
`hospitalized three times. After the first hospitaliza-
`tion, the 10 additional hospitalizations occurred in
`only 201 years of observation, for a rate of 5% for at
`least one more GI hospitalization per year during
`treatment with NSAIDs. This rate varied from 0%
`in Phoenix to 8.3% in Santa Clara County. This
`overall rate of rehospitalization is approximately
`four times the rate of first hospitalization.
`There were 17 GI-related deaths occurring dur-
`ing these 9,525 years of observation. Thirteen of
`these deaths occurred in patients whowere reliably
`known to be taking NSAIDsat the time of death;
`each of the others might have been. This gives an
`overall GI death rate of 0.18% per year and a GI
`death rate during known treatment with NSAIDsof
`0.19% per year.
`
`Risk Factor Analyses
`Table II lists continuous variables analyzed uni-
`variately for their association with serious GI events
`(hospitalization or death) for patients having taken
`NSAIDsatthe predictive visit. The numberofsub-
`jects reflects the inclusion criteria previously de-
`scribed. The most significant differences were ob-
`tained for age, HAQ disability index, NSAID dose,
`and use of antacids or Hy-antagonists. The differ-
`ence in HAQdisability scores of 0.31 is clinically
`significant, representing the equivalentof 4 years of
`disease progression in RA. NSAID dose was calcu-
`
`lated by setting a value of 1.00 for the manufactur-
`er’s highest recommendeddose on the packagein-
`sert and normalizing the dose of each patientto this
`standard. Thus, the value 1.03 means that patients
`with events, on average, were taking 103% of the
`manufacturer’s highest recommended dose.
`“DMARDs”refers to prior use of“disease-modify-
`ing antirheumatic drugs.” The variable “specialty
`care” refers to the proportion of all doctor visits
`that were made to rheumatologists.
`The second part of Table II presents univariate
`comparisons of potentially predictive categorical
`variables dichotomized as present or absent (pre-
`sented as percent positive). The statistically most
`significant differences were seen for having previ-
`ously reported GI symptomsattributed to use of an
`NSAID (nausea, heartburn,loss of appetite, vomit-
`ing, or upper abdominal pain), use of prednisone
`(average dosage 7 mg/day) in the previous 6
`months, and use of antacids or Hy-antagonists with-
`in 9 months of the event. Female sex was not predic-
`tive. Race was not predictive, nor was cigarette
`smoking or alcohol use, although these analyses
`were limited in power because of small numbers of
`nonwhites, smokers, and heavy drinkers. Eight of
`the 10 statistically significant (p <0.05) differences
`remain significant after adjustment for the 20 mul-
`tiple comparisons of Table II.
`Odds ratios for selected variables with respect to
`GIevents along with 95% confidence intervals are
`shown in Table III. For example, the odds ratio for
`females was obtained as follows: the proportion of
`females with GI events was 75 of 98 = 76.5%; thus,
`the odds are 76.5/(100 — 76.5) = 3.26. Similarly, the
`proportion of females without GI events was 1,224
`
`September 1991 The American Journal of Medicine Volume 91
`
`217
`
`Page 5 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`Page 5 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`

`

`NSAID-ASSOCIATED GASTROPATHY/ FRIES ET AL
`
`
` Figure 1. Classification tree for Gl
`
`
`hospitalization or death (GI event).
`The value within the circles and
`boxes is the percentage of patients
`with GI events per year during
`treatment with NSAIDs. Numbers
`below the boxes are subgroup iden-
`tification numbers of final sub-
`groups. NSAID dose = 1.0 corre-
`sponds to the manufacturer's high-
`est recommended dose.
`
`of 1,596 = 76.7%; thus, the odds are 76.7/(100 —
`76.7) = 3.29. Therefore, the odds ratio for females
`with GI events to females without GI events is
`3.26/3.29 = 1.0. For the values considered, age
`showed the highest odds ratio when dichotomized
`at age 45 (only two events occurred below that age).
`Oddsratios are significant at p <0.05 if 95% confi-
`dence limits do not include the value of 1.
`Table IV summarizes the results of stepwise logistic
`regression analysis using most of the variables of Table
`II. Three variables had improvement p values less than
`0.05: age, use of prednisone, and previous NSAID GI
`side effects. NSAID dose and disability index were the
`next two variables stepped into the model. The odds
`ratio is the estimated multiplicative effect of a 1-unit
`increase in that variable on the odds of having a GI
`event, holding all other covariates constant.
`Figure 1 summarizes theresults of the classifica-
`tion tree analysis (recursive partitioning). Each node
`of the tree (represented by round and square boxes)
`contains the risk per year during NSAID treatment
`for patients in that node, in percent. With this meth-
`od, each node of the tree is recursively partitioned
`into two subgroups, one containing cases with vari-
`able values less than some cutoff point and the other
`containing cases with values greater than this cutoff.
`
`The “best” variable along with its splitting value
`(cutoff point) is selected with the use of a “goodness-
`of-split” criterion. The program (CART)cross-vali-
`dates the results to present a stable tree structure
`(19]. The same variable may appear twice in succes-
`sion in the tree at different cutpoints; this occurred
`with both age and disability.
`Thetree splitfirst on age, yielding a very-low-risk
`group (0.3% events per year during NSAID thera-
`py) below age 47. Thetreesplit next on prednisone
`use, and then on age and disease duration. Disabili-
`ty index and previous NSAID GIside effects ap-
`peared at lower levels of the tree. The tree displayed
`in Figure 1 contains 10 classifying subgroups (rep-
`resented by square boxes). Eleven groups are num-
`bered and ranked in order of risk, from a high of
`4.2% per year while taking NSAIDs in individuals
`over the age of 75 and not taking prednisone at the
`predictive visit to near zero: for example, in the
`groups with young individuals and with individuals
`between 48 and 63 years of age not taking predni-
`sone at the predictive visit and without previously
`having had an NSAID GIside effect.
`Table V provides descriptive data for Figure 1,
`and Table VI describes the average values of the
`major risk variables in each groupof Figure 1; the
`
`218
`
`September 1991 The American Journal of Medicine Volume 91
`
`Page 6 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`Page 6 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`

`

`very different patient characteristics are ap-
`parent.
`
`COMMENTS
`Results reported here confirm prospectively the
`highrelative risk of hospitalization or death in RA
`patients taking NSAIDs. Therate of GI hospitaliza-
`tions per year in patients taking NSAIDs is 1.58%,
`compared with a hospitalization rate of 0.3% (nine
`of 2,784) in patients not treated with NSAIDs,for a
`hazardratio of5.2. As previously reported [1], use of
`NSAIDs in RA does not appear to represent a sever-
`ity marker for RA, and mean HAQdisability scores,
`pain scores, age, disease duration, and use of pred-
`nisone are closely similar in patients taking
`NSAIDsandin those not taking NSAIDs. Without
`a randomized controlled trial of NSAID versus no
`NSAID use,one cannotabsolutely attribute causal-
`ity to the use of NSAIDs, but this use does not
`appear likely to be a marker of disease severity.
`Ourresults were consistent across all five popula-
`tions studied. It appears likely from preliminary
`data that the rate of hospitalizations in patients
`with osteoarthritis may be half or less of that ob-
`served in patients with RA. Previous GI hospital-
`ization is identified as a major risk factor.
`We performed a preliminary evaluation of pa-
`tients with osteoarthritis, with a smaller number of
`years of observation available. Patients with osteo-
`arthritis were drawn from the Wichita population
`(590 patients, 893 years of observation) and from a
`national pool of osteoarthritis patients who had
`taken the Arthritis Self-Management Course
`[20,21] offered by the Arthritis Foundation (501 pa-
`tients, 251 patient-years of observation). Only three
`GIhospitalizations occurred in these patients with
`osteoarthritis. All of these cases were gastric in loca-
`tion andall were in patients taking NSAIDs. Thus,
`the rate of upper GI hospitalizations per year dur-
`ing NSAID treatmentwas 1.4% in RA and only 0.4%
`in osteoarthritis. There were no GI-related deaths
`during this period of observation. This suggests the
`possibility of a higher hospitalization rate for RA
`patients, but the numbersof patients with osteoar-
`thritis were too small to achieve statistical
`significance.
`Other studies have addressed the question of in-
`cidence of serious GI events, and generally yield
`estimates consistent with those here. Thus, Carson
`et al [22,23] found hospitalization rates of 0.5% per
`year for bleeding ulcer in NSAID users forall indi-
`cations, and FDA estimates of “serious” problems
`are from 2% to 4% peryear [24]. Other estimates are
`somewhat lower (25-27], but underascertainment
`appears to be present in some of these studies. Our
`
`NSAID-ASSOCIATED GASTROPATHY/ FRIES ET AL
`
`data by study protocol do not allow for overascer-
`tainment. Indeed, even the data presented here are
`subject to some underascertainment, since if our
`patients did not report a hospitalization, we did not
`routinely seek discharge summaries, and addition-
`ally we failed to obtain discharge summaries about
`5% of the time. In general, the literature data are
`consistent with the highest incidences in RA pa-
`tients, who havethe highest levels of NSAID intake,
`and lower incidences for more occasional NSAID
`users [1,12,22-27].
`Two recent major publications further reinforce
`the results reported here. In a nested case-control
`study of Tennessee Medicaid GI hospitalizations,
`Griffin et al [28] found a relative risk of 4.1 for
`NSAID users versus nonusers regardless of disease
`indication,rising to a relative risk of 8.0 for current
`users at high NSAID dosage. Excess risk was found
`to be 17.4 events per 1,000 patient-years, as com-
`pared with 15.8 per 1,000 patient-years as reported
`here. Dosage was found important, and norole for
`confounding by current smoking or current heavy
`alcohol use could be identified. Soll et a! [29] pro-
`vide a careful review of published data on NSAID-
`associated ulcers, with a full discussion of relative
`risk, mechanisms, and approaches to prevention.
`Previous studies have suggested that one half to
`two thirds of GI deaths in RA patients taking
`NSAIDs are “excess,” while the remainder repre-
`sent “background” events [6,7]. Thus, these data
`are similar to the 0.18% per year excess GI death
`rate previously estimated [1].
`The individual clinical variables appearing to be
`predictive of serious GI events included age, dis-
`ability, NSAID dose, previous GI hospitalization,
`prior GI complaints with NSAIDs, and useof pred-
`nisone, antacids, or Ho-antagonists. Inadequate
`data were available for analysis of prior sucralfate
`use. Female sex was not predictive, in contrast to
`the preponderance of female admissions reported
`in a series of GI hospitalizations[9]. These findings
`are not contradictory, since the prevalence of
`NSAIDuse is muchhigherin females than in males
`[13]. A variable termed “specialty care” was not
`predictive, suggesting that hospitalization and
`death rates are not related to the experience or ex-
`pertise of the particular provider.
`The stepwise logistic regression model identified
`five variables that appear predictive (improvement
`p value <0.20), and theseare, in the main,intuitive.
`The univariate finding that prior use of GI protec-
`tive agents (antacids or H»-antagonists) was a
`strong predictive factor (p <0.001) was somewhat
`surprising. Most NSAID GIulcers are “silent” [3,8],
`but a substantial number of patients had prior
`
`September 1991 The American Journal of Medicine Volume91
`
`219
`
`Page 7 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`Page 7 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`

`

`NSAID-ASSOCIATED GASTROPATHY/ FRIES ET AL
`
`GI-EVENT SCORE TABLE FOR RA
`
`paceyens)dfeee
`HISTORY OF PREVIOUSNSAIDGI SIDE EFFECT|so=|
`Figure 2. GI event score table for
`RA. Derived from logistic regres-
`x10=
`DISABILITY INDEX (0-3) OR (ARA CLASS - 1)
`sion results, a simplified scheme is
`x15=
`NSAID DOSE(fraction of maximum recommended)
`presented here:
`the sum of (1) pa-
`tient age X 2, (2) 50 points if there
`CURRENT PREDNISONE USE
`is a prior history of NSAID dyspep-
`sia, (3) 0 points for American Rheu-
`matism Association (ARA) function-
`al class 1
`(normal), 10 for class 2
`(adequate), 20 for class 3 (limited),
`or 30 for class 4 (unable), (4) 15
`times the fraction patient NSAID
`dose /manufacturer's highest rec-
`ommended dose, and (5) 40 points
`for current prednisone use, When
`100 is subtracted from this sum
`and the result divided by 40, the
`risk of GI hospitalization or death
`over the next 12 months, in per-
`cent, is obtained,
`
`TOTALSCORE
`
`=
`
`GI-EVENT RISK PER YEAR ON NSAIDS IN RA
`
`
`
`
`RISK = (SCORE- 100) /40
`
`symptomatology that might have led to the use of
`“protective” agents. These agents are not believed
`to be effective in prevention of NSAID ulcers
`[4,5,10,30-33], and this is supported by the experi-
`ence reported here. There are two possible mecha-
`nismsby which these agents might prove harmful in
`
`practice: first, by treating dyspeptic symptoms and
`removing premonitory warning signs, and second,
`by reassuring the physician that preventive mea-
`sures had been taken. However, “protective” agents
`are probablygiven to patients already considered to
`be at higher risk, thus expressing the higher risk
`
`© Actual Values
`@ Values from
`RISK = SCORE - 100
`40
`
`Figure 3. Risk for GI hospitalization
`or death per year during treatment
`with NSAIDs. Comparison of risk
`scores for individual patients (Fig-
`ure 2) with actual measured risk
`(GI events/years at risk) for pa-
`tients grouped by 20-unit score
`categories, demonstrating a strong
`correlation between results pre-
`dicted by the simplified scoring rule
`and actual experience.
`
`220
`
`September 1991 The American Journal of Medicine Volume 91
`
`Page 8 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`Page 8 of 10
`
`Patent Owner Ex. 2003
`Mylan v. Pozen
`IPR2017-01995
`
`

`

`NSAID-ASSOCIATED GASTROPATHY/ FRIES ET AL
`
`mended dose, and contributes relatively little for
`most patients. Other factorsare readily ascertained
`by history. Figure 2 presents asimplescoring table.
`In these data sets, the estimated risk for serious
`GI events over the next year is well approximated
`by subtracting 100 from the score and dividing by
`40. Figure 3 displays the relationship between risk
`as estimated by the scoring rule and actual inci-
`dence of events per year during NSAID therapyin
`subjects grouped into 20-unit score categories.It is
`apparent that the scoring rule accurately models
`actual experience of events per year during NSAID
`therapy in these RA patients.
`These major risk factors are modifiable in part,
`since prednisone use and NSAID dose may be
`changed.It is likely, but not yet proven, that sub-
`stantial differences between NSAIDs, not analyzed
`here, may allow choice of less hazardous agents for
`some patients. Prophylaxis with exogenous prosta-
`glandins greatly reduces the rate of endoscopic ul-
`cerationsin similar patients, and current evidence
`suggests consideration of use of these agents in
`high-risk groups [1,8,13,25,28,29]. Increased under-
`standing of the role of clinical and demographic
`factors in determining the risk of NSAID gastro-
`pathy will provide an important means to reduce
`the magnitudeof this serious national problem.
`
`ACKNOWLEDGMENT
`We thank Hong Shi for statistical support and Lisa Cendejas for typing ot the
`manuscript We also thank Drs. Fred Wolfe, Don Mitchell, John Sibley, and
`Sanford Roth for permission to study their ARAMIS populations.
`
`le