REVIEW
`Annals of Internal Medicine
`Bleeding Risks With Aspirin Use for Primary Prevention in Adults:
`A Systematic Review for the U.S. Preventive Services Task Force
`
`Evelyn P. Whitlock, MD, MPH; Brittany U. Burda, MPH; Selvi B. Williams, MD, MPH; Janelle M. Guirguis-Blake, MD; and
`Corinne V. Evans, MPP
`
`Background: The balance between potential aspirin-related
`risks and benefits is critical in primary prevention.
`
`Purpose: To evaluate the risk for serious bleeding with regular
`aspirin use in cardiovascular disease (CVD) primary prevention.
`
`Data Sources: PubMed, MEDLINE, Cochrane Central Register
`of Controlled Trials (2010 through 6 January 2015), and relevant
`references from other reviews.
`
`Study Selection: Randomized, controlled trials; cohort studies;
`and meta-analyses comparing aspirin with placebo or no treat-
`ment to prevent CVD or cancer in adults.
`
`Data Extraction: One investigator abstracted data, another
`checked for accuracy, and 2 assessed study quality.
`
`Data Synthesis: In CVD primary prevention studies, very-low-
`dose aspirin use (≤100 mg daily or every other day) increased
`major gastrointestinal (GI) bleeding risk by 58% (odds ratio [OR],
`1.58 [95% CI, 1.29 to 1.95]) and hemorrhagic stroke risk by 27%
`(OR, 1.27 [CI, 0.96 to 1.68]). Projected excess bleeding events
`with aspirin depend on baseline assumptions. Estimated excess
`
`major bleeding events were 1.39 (CI, 0.70 to 2.28) for GI bleed-
`ing and 0.32 (CI, ⫺0.05 to 0.82) for hemorrhagic stroke per 1000
`person-years of aspirin exposure using baseline bleeding rates
`from a community-based observational sample. Such events
`could be greater among older persons, men, and those with
`CVD risk factors that also increase bleeding risk.
`
`Limitations: Power to detect effects on hemorrhagic stroke
`was limited. Harms other than serious bleeding were not
`examined.
`
`Conclusion: Consideration of the safety of primary prevention
`with aspirin requires an individualized assessment of aspirin's ef-
`fects on bleeding risks and expected benefits because absolute
`bleeding risk may vary considerably by patient.
`
`Primary Funding Source: Agency for Healthcare Research and
`Quality.
`
`Ann Intern Med. 2016;164:826-835. doi:10.7326/M15-2112 www.annals.org
`For author affiliations, see end of text.
`This article was published at www.annals.org on 12 April 2016.
`
`Although widely regarded as safe for patient-
`
`directed, over-the-counter use, aspirin is associ-
`ated with a range of harms. They vary in type and
`severity with the dosage and duration of use and un-
`derlying patient
`risk factors. By inhibiting cyclo-
`oxygenase-1 enzyme activity, low-dose aspirin leads to
`mucosal damage to the gastrointestinal
`(GI)
`tract
`and causes erosions, ulcers, and bleeding (1).
`Cyclooxygenase-mediated antiplatelet effects also in-
`crease non-GI bleeding events that range from trivial to
`serious,
`including intracranial bleeding events and
`hemorrhagic strokes (2). The advisability of using aspi-
`rin for the primary prevention of cardiovascular disease
`(CVD) events, with or without considering potentially
`beneficial effects on cancer, depends on accurately es-
`timating harms associated with a specific prevention
`regimen and the absolute and relative variability in
`harms for any individual or targeted subpopulation. We
`report serious bleeding-related harms from aspirin
`used for primary prevention. This review, along with 2
`companion reviews (3, 4) on CVD and cancer benefits,
`was used to inform updated U.S. Preventive Services
`Task Force (USPSTF) recommendations. These reviews
`
`See also:
`
`Related articles . . . . . . . . . . . . . . . 777, 804, 814, 836
`Editorial comment . . . . . . . . . . . . . . . . . . . . . . . . . 846
`
`all share a clinical focus on populations eligible for CVD
`primary prevention.
`
`METHODS
`Our full report describes our methods in detail (5).
`Data Sources and Searches
`We reviewed all included and excluded studies in 4
`relevant
`systematic reviews on aspirin-associated
`bleeding events (2, 6 – 8) and the 2 previous (9, 10) and
`updated USPSTF reviews (11, 12) to identify relevant
`literature. We supplemented this with newly identified
`studies
`found on PubMed, MEDLINE, and the
`Cochrane Central Registry of Controlled Trials from 1
`January 2010 to 6 January 2015.
`Study Selection
`Two investigators independently reviewed ab-
`stracts and full-text articles against prespecified criteria
`(5). We included trials and large longitudinal cohort
`studies conducted in adults with a mean age of 40
`years or older that evaluated regular oral aspirin use
`(≥75 mg at least every other day) for 1 year or longer
`for any indication compared with no treatment or pla-
`cebo. We required studies to report major GI or intra-
`cranial bleeding. Major GI bleeding included cases
`leading to death, those requiring hospitalization or
`transfusion, or those described by the trial investigator
`as serious. Intracranial bleeding included hemorrhagic
`stroke and intracerebral, subdural, and subarachnoid
`hemorrhage.
`
`826 Annals of Internal Medicine • Vol. 164 No. 12 • 21 June 2016
`
`www.annals.org
`
`Downloaded From: http://annals.org/ on 06/20/2016
`
`Page 1 of 12
`
`Patent Owner Ex. 2032
`CFAD v. Pozen
`IPR2015-01718
`
`

`
`Bleeding Risks With Aspirin Use
`
`Figure 1. Major GI bleeding in CVD primary prevention trials.
`
`REVIEW
`
`OR (95% CI)
`
`Events, n/N
`Aspirin
`No Aspirin
`
`2.02 (1.40–2.93)
`5.02 (0.87–29.05)
`1.73 (1.10–2.70)
`0.47 (0.09–2.57)
`2.73 (0.68–10.95)
`1.13 (0.43–2.92)
`1.37 (1.05–1.78)
`1.59 (1.32–1.91)
`
`77/9399
`4.5/1263
`49/11 037
`3/3429
`6/1268
`9/1675
`129/19 934
`277.5/48 005
`
`37/9391
`0.5/1278
`28/11 034
`3/1710
`2/1272
`8/1675
`94/19 942
`172.5/46 302
`
`Study, Year (Reference)
`
`Time Point, y
`
`Dose, mg/d
`
`Population
`
`75
`81 or 100
`162.5
`500
`75
`100
`50
`
`Men and women with hypertension
`Men and women with diabetes
`Male physicians
`Male physicians
`Men at high risk for IHD
`Men and women with ABI ≤0.95
`Female health professionals
`
`3.8
`4.4
`
`5 6 6
`
`.8
`8.2
`10.1
`
`HOT, 1998 (24)
`JPAD, 2008 (25)
`PHS, 1989 (26)
`BMD, 1988 (27)
`TPT, 1998 (29)
`AAA, 2010 (30)
`WHS, 2005 (32)
`Overall: I2 = 22.2%; P = 0.260
`
`index; BMD = British Doctor's Trial; CVD = cardiovascular disease;
`AAA = Aspirin for Asymptomatic Atherosclerosis; ABI = ankle brachial
`GI = gastrointestinal; HOT = Hypertension Optimal Treatment; IHD = ischemic heart disease; JPAD = Japanese Primary Prevention of Atheroscle-
`rosis With Aspirin for Diabetes; OR = odds ratio; PHS = Physicians' Health Study; TPT = Thrombosis Prevention Trial; WHS = Women's Health
`Study.
`
`5
`1
`0.1
`Aspirin No Aspirin
`
`Data Extraction and Quality Assessment
`One investigator abstracted data from the included
`studies; another checked data for accuracy. The same
`investigators assessed the quality of included studies
`using study design–specific criteria defined by the
`USPSTF (13) and supplemented with Newcastle–Ottawa
`Scale criteria for cohort studies (14). Good-quality stud-
`ies met most criteria and were downgraded to fair if not
`all criteria were met. Poor-quality studies (those with
`>40% attrition, >20% attrition between groups, other
`fatal flaws, cumulative effects of multiple minor flaws, or
`missing information significant enough to limit confi-
`dence in the validity of results) were excluded (5).
`
`Data Synthesis and Analysis
`Aspirin exposure was inferred from the intended
`dosages and treatment duration in trials, without ad-
`justment for actual adherence because of incomplete
`reporting. The average intended dose per day was cal-
`culated; 325 mg daily or less was defined as low-dose
`and 100 mg daily or less was defined as very-low-dose.
`Because harms were often rare, we explored whether
`broadening bleeding definitions (that is, any intracra-
`nial bleeding vs. hemorrhagic stroke alone) changed
`the results. The broader definition made little differ-
`ence, so we focused on hemorrhagic stroke (or intrace-
`rebral hemorrhage) results for consistency with an
`individual-participant data (IPD) meta-analysis (15) and
`our companion model (16). We used the Peto odds
`ratio (OR) for primary statistical analyses (17) because
`of rare events (that is, a control group event rate <1%)
`and repeated analyses using the Mantel–Haenszel OR;
`in both methods, we used a 0.5 continuity correction
`(18) with no major differences in results (Appendix Ta-
`ble 1, available at www.annals.org). We stratified results
`by population (primary prevention of CVD, secondary
`prevention of CVD, and colorectal cancer prevention)
`and conducted sensitivity analyses by dose, frequency,
`
`www.annals.org
`
`and duration of therapy. We also examined data by
`relevant a priori subgroups: age, sex, race/ethnicity, co-
`morbidities (diabetes, liver disease, ulcer disease, and
`previous GI bleeding), and concurrent medication use
`(selective serotonin reuptake inhibitors and nonaspirin
`nonsteroidal anti-inflammatory drugs [NSAIDs])
`(19 –
`21). Some subgroup analyses (for example, proton-
`pump inhibitor or statin use) were not specified a priori.
`Other aspirin-related harms (for example, age-related
`macular degeneration and ulcers) were addressed in
`our full report (5).
`We calculated absolute treatment effects for bleed-
`ing outcomes to represent the range of control group
`event rates from the CVD primary prevention trials
`about aspirin use. For each trial, we divided the num-
`ber of events for each outcome by the person-years at
`risk (approximated by multiplying the number of partic-
`ipants in the control group by the mean years of follow-
`up), assuming a constant risk over time. On the basis of
`the minimum, median, and maximum event rates (ex-
`cluding outliers and zeros) for each outcome, we calcu-
`lated a range of expected event rates after aspirin in-
`tervention using the pooled relative risks (RRs) from the
`included CVD primary prevention trials evaluating aspi-
`rin doses of 100 mg daily or less. Excess cases were
`calculated by subtracting the event rate per 1000
`person-years for aspirin users from event rates in the
`control groups for each risk level. We contrasted excess
`cases based on control group event rates from trials
`with results based on control group bleeding rates
`from the largest cohort study (22).
`
`Role of the Funding Source
`Agency for Healthcare Research and Quality staff
`provided oversight for the project. The USPSTF liaisons
`helped resolve review scope issues but were not in-
`volved in the conduct of the review.
`
`Annals of Internal Medicine • Vol. 164 No. 12 • 21 June 2016 827
`
`Downloaded From: http://annals.org/ on 06/20/2016
`
`Page 2 of 12
`
`Patent Owner Ex. 2032
`CFAD v. Pozen
`IPR2015-01718
`
`

`
`REVIEW
`
`Bleeding Risks With Aspirin Use
`
`Table 1. Sensitivity Analyses for Bleeding in CVD Primary Prevention Trials
`
`Study, Year (Reference)
`
`Dose
`
`Studies, k Participants, n Pooled OR (95% CI)
`
`Included Trials
`
`Major GI or extracranial bleeding
`Whitlock et al (main analysis), 2015 (5)*
`
`ATT Collaboration, 2009 (15)†
`
`Any
`
`≤100 mg
`
`Any
`
`De Berardis et al (cohort study), 2012 (22)*‡
`
`≤300 mg
`
`Hemorrhagic stroke
`Guirguis-Blake et al (meta-analysis), 2015 (11)
`
`Any
`
`≤100 mg
`
`ATT Collaboration (IPD meta-analysis), 2009 (15) Any
`
`7
`
`5
`
`6
`
`1
`
`9
`
`7
`
`6
`
`94 307
`
`67 097
`
`95 456
`
`372 850
`
`113 264
`
`86 054
`
`95 456
`
`1.59 (1.32–1.91);
`I2 = 22.2%
`1.58 (1.29–1.95);
`I2 = 28.6%
`1.54 (1.30–1.82)§;
`chi square = 3.1
`1.55 (1.46–1.65)兩兩
`
`1.33 (1.03–1.71);
`I2 = 0%
`1.27 (0.96–1.68);
`I2 = 0%
`1.32 (1.00–1.75)§;
`chi square = 4.7
`
`HOT, JPAD, PHS, BMD, TPT, AAA, WHS
`
`HOT, JPAD, TPT, AAA, WHS
`
`BMD, PHS, TPT, HOT, PPP, WHS
`
`NA
`
`PPP, HOT, JPAD, JPPP, PHS, BMD, TPT,
`AAA, WHS
`PPP, HOT, JPAD, JPPP, TPT, AAA, WHS
`
`BMD, PHS, TPT, HOT, PPP, WHS
`
`Intracranial hemorrhage, including
`hemorrhagic stroke
`Whitlock et al (main analysis), 2015 (5)
`
`Any
`
`10
`
`114 540
`
`1.34 (1.07–1.70);
`I2 = 0%
`1.30 (1.00–1.68);
`I2 = 0%
`1.54 (1.43–1.67)兩兩
`AAA = Aspirin for Asymptomatic Atherosclerosis; ATT = Antithrombotic Trialists; BMD = British Doctor's Trial; CVD = cardiovascular disease; GI =
`gastrointestinal; HOT = Hypertension Optimal Treatment; IPD = individual-participant data; JPAD = Japanese Primary Prevention of Atherosclerosis
`With Aspirin for Diabetes; JPPP = Japanese Primary Prevention Project; NA = not applicable; OR = odds ratio; PHS = Physicians' Health Study;
`POPADAD = Prevention of Progression of Arterial Disease and Diabetes; PPP = Primary Prevention Project; TPT = Thrombosis Prevention Trial;
`WHS = Women's Health Study.
`* Major GI bleeding.
`† IPD meta-analysis of GI or other major extracranial bleeding.
`‡ Hospitalizations for first major bleeding event.
`§ Year event rate ratio.
`兩兩 Incidence rate ratio.
`
`De Berardis (cohort study), 2012 (22)‡
`
`≤100 mg
`
`≤300 mg
`
`8
`
`1
`
`87 330
`
`372 850
`
`PPP, TPT, HOT, JPAD, PHS, JPPP, BMD,
`POPADAD, AAA, and WHS
`PPP, TPT, HOT, JPAD, JPPP, POPADAD,
`AAA, and WHS
`NA
`
`RESULTS
`Although we considered a larger set of trials that
`reported on harms associated with aspirin use (5), this
`review focuses on bleeding events from 10 of 11 CVD
`primary prevention trials in adults (mean age, 53.2 to
`70.1 years) that addressed 1 or more serious bleeding
`events due to aspirin use (23–32). Trial details are re-
`ported in our companion article (3). We also identified
`2 IPD meta-analyses (8, 15) of included trials that re-
`ported harms analyses complementing our trial-level
`results and 4 recent fair- or good-quality cohort studies
`(22, 33–35) of bleeding risks in persons with or without
`extended low-dose aspirin use; these studies were
`clearly or presumed for CVD primary prevention (Ap-
`pendix Table 2, available at www.annals.org). Most rel-
`evant cohort data came from a large good-quality Ital-
`ian study examining hospitalizations for all major
`bleeding events (intracranial and extracranial) after a
`median follow-up of 5.7 years in a population of
`372 850 community-dwelling adults (186 425 new us-
`ers of low-dose aspirin matched using propensity scor-
`ing with 186 425 never users; mean age, 69.4 years
`[range, 30 to 95 years]).
`Major GI Bleeding
`Seven CVD primary prevention trials of aspirin, 50
`to 500 mg daily or every other day, used over 3.8 to
`10.1 years (24 –27, 29, 30, 32), showed a 59% increased
`risk for major GI bleeding (OR, 1.59 [95% CI, 1.32 to
`
`828 Annals of Internal Medicine • Vol. 164 No. 12 • 21 June 2016
`
`1.91]; I2 = 22.2%) (Figure 1). Estimated bleeding risks
`remained similar when limited to trials of very-low-dose
`aspirin or when reported from an IPD meta-analysis ex-
`amining a slightly different outcome (extracranial
`bleeding) of 6 CVD primary prevention trials (Table 1)
`(15). In cohort data, the effect of aspirin on hospitaliza-
`tions for major GI bleeding events was similar (inci-
`dence rate ratio, 1.55 [CI, 1.46 to 1.65]) (22).
`
`Hemorrhagic Stroke
`Nine trials of aspirin, 50 to 500 mg daily or every
`other day, used for 3.6 to 10.1 years (23–27, 29 –32)
`showed an increased risk for hemorrhagic stroke by
`about one third (OR, 1.33 [CI, 1.03 to 1.71]; I2 = 0%),
`regardless of dose (Figure 2 and Table 1). The point
`estimate and its statistical significance varied slightly
`between pooled analyses depending on the studies in-
`cluded and whether the outcome included any cases of
`intracranial hemorrhage (3, 5, 15). The only study with a
`statistically significant increase in hemorrhagic stroke
`(OR, 1.84 [CI, 1.01 to 3.35]) was conducted in an older
`hypertensive Japanese population (31). Cohort data
`suggested that hospitalizations for intracranial bleed-
`ing events may contribute more prominently to
`bleeding-related hospitalizations in community settings
`(incidence rate ratio, 1.54 [CI, 1.43 to 1.64]) (22), repre-
`senting about one third of hospitalizations for all major
`bleeding events (22).
`
`www.annals.org
`
`Downloaded From: http://annals.org/ on 06/20/2016
`
`Page 3 of 12
`
`Patent Owner Ex. 2032
`CFAD v. Pozen
`IPR2015-01718
`
`

`
`Bleeding Risks With Aspirin Use
`
`Baseline Estimates of Major Bleeding Risks
`(Trial vs. Cohort)
`Mean major bleeding rates among control group
`participants from 6 CVD primary prevention trials were
`low (0.7 extracranial bleeding event and 0.3 hemor-
`rhagic stroke per 1000 person-years) based on an IPD
`meta-analysis (15) (Table 2). In contrast, hospitalization
`rates for GI bleeding among control participants in the
`cohort study (22) were much higher (2.4 per 1000
`person-years) than the highest GI bleeding rate sug-
`gested by the trials, with substantial variability by age
`(Table 2). The effect of baseline bleeding rate assump-
`tions on calculations of excess bleeding events is illus-
`trated in Table 3. Given a constant increase in the RR
`for bleeding associated with very-low-dose aspirin use,
`excess cases of major GI bleeding would vary consid-
`erably, depending on assumptions of the baseline rate
`(for example, 0.28 excess major GI bleeding event per
`1000 person-years based on median trial control group
`rates compared with 1.39 excess cases per 1000
`person-years based on cohort control group rates) (Ta-
`ble 3). For excess hemorrhagic strokes, variability is less
`extreme because baseline bleeding rates remain rela-
`tively rare whether estimated from trials or cohorts and
`some trials included participants with higher baseline
`bleeding risks.
`Baseline Estimates of Major Bleeding Risks, by
`Subgroup
`In both trial and cohort data, bleeding rates varied
`2- to 4-fold at baseline among subgroups defined by
`increasing age, male sex, and selected cardiovascular
`risk factors (5). The largest and most consistent statisti-
`cally significant differences in baseline bleeding risk oc-
`
`REVIEW
`
`curred with increasing age (increasing 1.5- to 2-fold in
`each subsequent decade after 50 years) and, to a lesser
`extent, male sex (Table 2). Multivariable analyses of
`both trial and cohort data suggested that age, sex, and
`other common factors independently modify baseline
`bleeding risks (Table 4). However, many trials restricted
`enrollment to participants without clear bleeding risk
`factors. After adjustment
`for bleeding risk factors—
`including aspirin use—a history of GI hospitalization was
`associated with the largest relative incidence rate of
`hospitalizations for major bleeding in cohort data
`(Table 4).
`Risk Factors for Increased Major Bleeding, by
`Site
`The RRs associated with participant characteristics
`differed somewhat between the 2 major bleeding sites.
`When analyses controlled for aspirin use,
`increasing
`age (per decade) had a greater effect on major GI or
`extracranial bleeding than on hemorrhagic stroke (Ta-
`ble 4). In addition to older age, male sex and diabetes
`mellitus increased the risk for serious bleeding, with
`possible variation in effect by site and due to imprecise
`magnitude.
`In an adjusted IPD meta-analysis of trial
`data (15), current smoking and mean blood pressure
`(BP) per 20 mm Hg were also independently associated
`with increased major extracranial bleeding events. For
`hemorrhagic stroke, only increasing age, current smok-
`ing, and elevated mean BP were clearly associated with
`increased risk, with elevated BP more strongly associ-
`ated with hemorrhagic stroke than GI bleeding risk. In-
`vestigators noted that coronary heart disease risk fac-
`tors associated with greater potential benefit
`from
`aspirin (that is, age, male sex, diabetes, current smok-
`
`OR (95% CI)
`
`Aspirin
`
`Events, n/N
`No Aspirin
`
`0.68 (0.12–3.95)
`
`2/2226
`
`3/2269
`
`0.93 (0.45–1.93)
`
`14/9399
`
`0.87 (0.29–2.58)
`
`6/1262
`
`15/9391
`
`7/1277
`
`1.88 (0.97–3.64)
`
`23/11 037
`
`12/11 034
`
`1.84 (1.01–3.35)
`
`28/7220
`
`15/7244
`
`1.08 (0.42–2.81)
`
`13/3429
`
`6/1710
`
`3.81 (0.40–36.66)
`
`2.5/1269
`
`0.5/1273
`
`1.25 (0.34–4.62)
`
`5/1675
`
`4/1675
`
`1.24 (0.83–1.87)
`
`51/19 934
`
`41/19 942
`
`1.33 (1.03–1.71)
`
`144.5/57 451
`
`103.5/55 815
`
`Figure 2. Hemorrhagic stroke in CVD primary prevention trials.
`
`Study, Year (Reference)
`
`Time Point, y
`
`Dose, mg/d
`
`Population
`
`100
`
`75
`
`81
`
`162.5
`
`100
`
`500
`
`75
`
`100
`
`50
`
`Men and women with ≥1 CVD risk
`factor
`
`Men and women with hypertension
`
`Men and women with diabetes
`
`Male physicians
`Men and women with ≥1 CVD risk
`factor
`
`Male physicians
`
`Men at high risk for IHD
`Men and women with ABI ≤0.95
`Female health professionals
`
`3.6
`
`3.8
`
`4.37
`
`5 5
`
`6 6
`
`.8
`
`8.2
`
`10.1
`
`PPP, 2001 (23)
`
`HOT, 1998 (24)
`
`JPAD, 2008 (25)
`
`PHS, 1989 (26)
`
`JPPP, 2014 (31)
`
`BMD,1988 (27)
`
`TPT, 1998 (29)
`
`AAA, 2010 (30)
`
`WHS, 2005 (32)
`Overall: I2 = 0.0%; P = 0.720
`
`AAA = Aspirin for Asymptomatic Atherosclerosis; ABI = ankle brachial index; BMD = British Doctor's Trial; CVD = cardiovascular disease; HOT =
`Hypertension Optimal Treatment; IHD = ischemic heart disease; JPAD = Japanese Primary Prevention of Atherosclerosis With Aspirin for Diabetes;
`JPPP = Japanese Primary Prevention Project; OR = odds ratio; PHS = Physicians' Health Study; PPP = Primary Prevention Project; TPT = Thrombosis
`Prevention Trial; WHS = Women's Health Study.
`
`www.annals.org
`
`Annals of Internal Medicine • Vol. 164 No. 12 • 21 June 2016 829
`
`5
`1
`0.1
`Aspirin No Aspirin
`
`Downloaded From: http://annals.org/ on 06/20/2016
`
`Page 4 of 12
`
`Patent Owner Ex. 2032
`CFAD v. Pozen
`IPR2015-01718
`
`

`
`REVIEW
`
`Bleeding Risks With Aspirin Use
`
`Table 2. Absolute Bleeding Rates Among Nonaspirin Control Groups, Overall and by Subpopulations*
`
`Baseline Characteristic Major GI or Extracranial Bleeding*,
`events per 1000 person-years
`
`Hemorrhagic Stroke†,
`events per 1000 person-years
`
`Hospitalization for Major Bleeding Event (95% CI)‡,
`events per 1000 person-years
`
`All control participants
`
`0.7
`
`0.3
`
`Age subgroups
`
`<65 y: 0.5
`≥65 y: 1.7
`
`Sex subgroups
`
`Men: 1.0
`Women: 0.5
`
`GI = gastrointestinal.
`* Resulting in hospitalization, transfusion, or death. Data from reference 15.
`† Data from reference 15.
`‡ Includes GI and intracranial bleeding. Data from reference 22.
`
`–
`
`–
`
`3.60 (3.48–3.72)
`Major extracranial bleeding (approximately): 2.40
`Major intracranial bleeding (approximately): 1.20
`<50 y: 0.61 (0.41–0.91)
`50–59 y: 1.40 (1.24–1.58)
`60–69 y: 2.58 (2.40–2.77)
`70–79 y: 4.61 (4.39–4.85)
`≥80 y: 6.93 (6.51–7.38)
`Men: 4.50 (4.30–4.70)
`Women: 2.86 (2.72–3.01)
`
`ing, and mean BP) were also associated with increased
`major bleeding risks for 1 or both outcomes, although
`somewhat more weakly (15). The influence of co-
`medications was assessed in the cohort study only (Ta-
`ble 4) (22); in adjusted analyses, NSAID use further in-
`creased the risk for bleeding (adjusted incidence rate
`ratio, 1.10 [CI, 1.05 to 1.16]), with a possible protective
`effect on bleeding risk from proton-pump inhibitor and
`statin use.
`Bleeding Events, by Aspirin Regimen
`We found very few within-trial direct comparisons
`of aspirin regimens for primary prevention, and
`between-trial
`comparisons were potentially
`con-
`founded by other between-study differences. Cohort
`studies were similarly uninformative because of restric-
`tions to a single low-dose regimen (35), lack of evalua-
`tion of dosage effects (22), or issues with exposure
`measurement (33, 34). In the 2 large U.S. cohorts (33,
`34), trend analyses strongly supported the effect of in-
`creasing the cumulative weekly aspirin dosage on
`lower or upper GI bleeding in both short- and long-
`term aspirin users, particularly women, and subarach-
`noid hemorrhages in men aged 55 years or older (36).
`Most bleeding cases (72.6%) involved daily, rather than
`less frequent, use of aspirin (33).
`Using available trial and cohort data, we found that
`the risk for bleeding associated with low-dose aspirin
`use probably persists throughout use but declines with
`discontinuation. In the Women's Health Study, the cu-
`mulative incidence of GI bleeding did not plateau in
`very-low-dose aspirin users compared with placebo re-
`cipients throughout 10 years of follow-up (37). In con-
`trast, a time point–stratified IPD meta-analysis sug-
`gested that the risk for major extracranial bleeding
`seen in early years decreased after 3 years (8). Because
`bleeding risks with placebo also declined with time,
`however, another mechanism for reduced bleeding
`events (such as unequal observation time) could have
`driven this observation (5, 38). Two cohort studies
`found that bleeding risk in regular aspirin users did not
`vary by duration of use (<5 years or ≥5 years) (33, 34).
`Weak evidence from the Women's Health Study sug-
`gested that excess GI bleeding risk rapidly attenuates
`after stopping aspirin (37).
`
`830 Annals of Internal Medicine • Vol. 164 No. 12 • 21 June 2016
`
`DISCUSSION
`in-
`We found relatively consistent estimates of
`creased risk for serious bleeding events with aspirin
`use in CVD primary prevention populations, whether
`based on trial or cohort data. For major GI bleeding,
`the best estimate with very-low-dose aspirin use in CVD
`primary prevention populations was an RR of 1.58 (CI,
`1.29 to 1.95; I2 = 28.6%). Although studies varied in the
`duration of aspirin use and data were sparse and some-
`what mixed on whether
`risk remains consistent
`throughout aspirin use, we believe that current empiri-
`cal data suggest a constant risk throughout use. In con-
`trast, due in part to rarer events and smaller effect
`size, the increased RR of hemorrhagic stroke was not
`statistically significant, with a best estimate of 1.27 (CI,
`0.96 to 1.68) for very-low-dose aspirin use in CVD pri-
`mary prevention. These are the estimates we provided
`for the companion model (16) based on a priori deci-
`sions to link harms estimates to the same population
`and aspirin dosages used for estimating benefits. For
`both types of bleeding, our pooled estimates were not
`statistically heterogeneous; their imprecision may re-
`flect inadequate power because of rare events and re-
`duced certainty of an average effect.
`Estimates of baseline bleeding risk are critical for
`accurately assessing the absolute risk for bleeding with
`aspirin use and determining net benefit. Control group
`trial participants had much lower average risks for
`bleeding than those from cohort studies (Table 2). This
`probably reflects the fact that, beyond the variability in
`risk represented by age and sex, participants at in-
`creased risk for bleeding had limited or no representa-
`tion in the CVD primary prevention trials (15). Our
`simulations illustrating a range of projected excess
`bleeding cases with very-low-dose aspirin use (Table 3)
`showed that assumptions about baseline bleeding rate
`are clearly important to avoid the underestimation of
`risk that could occur from applying trial-based aver-
`ages based on selective patient groups to a more un-
`selected general population.
`Nonetheless, the research basis for appropriately
`establishing community-based rates of serious bleed-
`ing remains insufficient, despite a long-standing inter-
`est in this issue. For example, we found little data be-
`
`www.annals.org
`
`Downloaded From: http://annals.org/ on 06/20/2016
`
`Page 5 of 12
`
`Patent Owner Ex. 2032
`CFAD v. Pozen
`IPR2015-01718
`
`

`
`Bleeding Risks With Aspirin Use
`
`yond 1 large cohort study to update a commonly cited
`baseline rate for major upper GI complications (that is,
`1 per 1000 person-years) that was previously derived
`from a systematic review of observational studies (39)
`and is not specific for bleeding. In subsequent work,
`the same researchers emphasized potential variability
`of harms from aspirin with differences in baseline GI
`risk. They clarified that their original estimate should be
`revised slightly upward (1 to 2 major upper GI compli-
`cations per 1000 person-years) but would still apply
`only to persons without significant risks (that is, men
`aged ≤60 years or women aged ≤70 years, all without
`history of GI pain, ulcers, and NSAID use) (40). This
`slightly increased range is consistent with another re-
`cent estimate of baseline risk for upper GI bleeding in
`aspirin nonusers with no CVD history (1.85 cases per
`1000 person-years) (41). Although we found the aver-
`age baseline GI bleeding rate to be slightly higher (ap-
`proximately 2.4 cases per 1000 person-years) when us-
`ing more recent cohort data, we believe these
`estimates are all reasonably similar (Table 2).
`From a clinical perspective, factors that either in-
`crease the risk for baseline bleeding or enhance aspi-
`rin's effect on bleeding can increase absolute rates of
`bleeding events with aspirin use. Although we found
`little evidence of effect modification for aspirin-related
`bleeding effects by medications or other factors that
`would be commonly present in candidates for CVD pri-
`mary prevention, baseline bleeding rates differed sub-
`stantially across expected patient risk factors. Older
`age and male sex consistently had an increased base-
`line bleeding risk, and some evidence indicated in-
`creased bleeding risk with CVD risk factors, such as di-
`abetes, current smoking, and elevated BP. Other
`researchers have determined that GI bleeding risk fac-
`
`REVIEW
`
`tors (that is, older age, male sex, history of GI ulcers or
`complications, and NSAID use) are relatively prevalent
`among aspirin users in the community (40), which sug-
`gests that substantial variability in cases of upper GI
`complications is to be expected among some users.
`These data are consistent with our findings, implying
`considerable potential variability in excess serious
`bleeding events with aspirin use because of risk factor
`differences among community-dwelling aspirin users.
`We found no adequately validated tools for assess-
`ing bleeding risks associated with aspirin use in primary
`prevention. A single risk prediction tool for upper GI
`complications has been published and is publicly avail-
`able (42). This tool has potential strengths but also de-
`ficiencies, including the incorporation of approaches to
`modifying the bleeding risk that are not empirically
`proven in a primary prevention population—for which
`caution clearly is warranted (43)—and insufficient exter-
`nal validation to confirm its readiness for clinical appli-
`cation (44). Therefore, selecting patients for aspirin
`prevention may be pragmatic through qualitative con-
`sideration of bleeding risk factors or candidate inclu-
`sion limited to those fitting trial selection criteria, which
`excludes those at increased risk (for example, previous
`ulcer or GI bleeding) and those with aspirin intolerance
`or contraindications (5).
`A stepwise practical approach, outlined by the Eu-
`ropean Society of Cardiology (45), is to select candi-
`dates for aspirin prevention on the basis of minimizing
`potential harms in those most likely to benefit. First, the
`risk for major CVD events is determined (with no further
`consideration of aspirin use in those below a 10-year
`risk threshold of 10%). Second, safety is assessed by
`eliminating candidates with a history of bleeding with-
`out reversible causes or with concurrent use of other
`
`Table 3. Absolute Events Caused or Prevented With Very-Low-Dose Aspirin Use for ≤10 y*
`
`Outcome
`
`Risk Level†
`
`Baseline Risk for Outcome,
`events per 1000 person-years
`
`Relative Risk (95% CI)
`
`Major GIB§ (k = 5)
`
`ICH, including HS (k = 8)
`
`HS (k = 7)
`
`Major bleeding event
`
`Low
`Median
`High
`Highest
`Low
`Median
`High
`Highest
`Low
`Median
`High
`Highest
`Cohort兩兩
`
`0.23
`0.49
`0.58
`1.04
`0.20
`0.47
`0.59
`1.25
`0.00
`0.37
`0.42
`1.26
`2.4 (GIB)
`1.2 (HS)
`3.6 (total)
`3.6
`
`1.58 (1.29 to 1.95)
`
`1.30 (1.00 to 1.68)
`
`1.27 (0.96 to 1.68)
`
`1.58 (1.29 to 1.95) (GIB)¶
`1.27 (0.96 to 1.68) (HS)¶
`
`1.55 (1.48 to 1.63)
`
`Events Caused per 1000
`Person-Years (95% CI)‡
`
`0.13 (0.07 to 0.22)
`0.28 (0.14 to 0.46)
`0.34 (0.17 to 0.55)
`0.60 (0.30 to 0.99)
`0.06 (0.00 to 0.14)
`0.14 (0.00 to 0.32)
`0.18 (0.00 to 0.40)
`0.38 (0.00 to 0.85)
`0.00 (0.00 to 0.00)
`0.10 (−0.01 to 0.25)
`0.11 (−0.02 to 0.29)
`0.34 (−0.05 to 0.86)
`1.39 (0.70 to 2.28) (GIB)
`0.32 (−0.05 to 0.82) (HS)
`1.71 (0.65 to 3.10) (total)
`1.98 (1.73 to 2.27)
`
`CVD = cardiovascular disease; GIB = gastrointestinal bleeding; HS = hemorrhagic stroke; ICH = intracranial hemorrhage.
`* Very-low-dose aspirin use was defined as ≤100 mg/d. Data are from 8 CVD primary prevention trials. Boldface values represent events clearly
`caused by aspirin use (i.e., 95% CI does not include both caused and prevented events).
`† Low (minimum), median, and high (maximum) control group rate for each outcome, excluding zeros and outliers from the set of CVD primary
`prevention trials. For major GIB and HS, "highest" indicates the maximum and "high" is the second highest.
`‡ Negative values indicate cases prevented.
`§ Data from companion systematic review on CVD primary prevention (11).
`兩兩 Baseline risk as reported by included cohort studies.
`¶ Using cohort baseline risk and trial relative risks to estimate events caused or prevented.
`
`www.annals.org
`
`Annals of Internal Medicine • Vol. 164 No. 12 • 21 June 2016 831
`
`Downloaded From: http://annals.org/ on 06/20/2016
`
`Page 6 of 12
`
`Patent Owner Ex. 2032
`CFAD v. Pozen
`IPR2015-01718
`
`

`
`REVIEW
`
`Bleeding Risks With Aspirin Use
`
`Table 4. Relative Rate Ratios for Bleeding Among Subpopulations From Trials and Cohort Studies
`
`Baseline Characteristic
`
`Adjusted Incidence Rate Ratio (95% CI)
`
`Major GI or Extracranial
`Bleeding*
`
`Hemorrhagic Stroke†
`
`Age (per decade)
`Male sex (vs. female se