Antithrombotic therapies in patients with heart failure: hypothesis formulation from a research...

pharmacoepidemiology and drug safety 2010; 19: 911–920onlinelibrary.com) DOI: 10.1002/pds.1987
Published online 1 July 2010 in Wiley Online Library (wiley
ORIGINAL REPORT

Antithrombotic therapies in patients with heart failure: hypothesisformulation from a research databasey

Zhong Yuan, Rachel Weinstein, John Zhang, Mei Chengz, George Griffin, Robert Zolynas,Alexei N. Plotnikov, Mary S. Lee, Leonard Oppenheimer and Paul Burton*

Pharmaceutical Research and Development, Johnson and Johnson, Raritan, NJ, USA

SUMMARY

Purpose Heart failure is a significant public health problem. The present study is intended to explore in a research database whetherantithrombotic therapies (ATTs) affect cardiovascular outcomes in patients with incident heart failure (IHF).Methods Using the United Kingdom Health Improvement Network research database, several multivariable models (including logistic andCox’s regression models, as well as propensity score methods) were used to examine all-cause mortality and clinical outcomes among fivetreatment groups.Results The cohort included 24 554 patients with IHF (50.2% men), with a mean age (standard deviation [SD]) of 76.4 (11.0) years. Nearlythree-fourths of patients received at least one form of ATT. Patients receiving ATTs tended to be younger and more likely to be men, and hadmore cardiovascular comorbidities. During the 18-month follow-up period, the mortality rates were 11.1%, 14.6%, 17.8%, 19.5%, and 32.6%for warfarin combination therapy, warfarin alone, clopidogrel therapy, aspirin (ASA) alone, and no therapy, respectively, yielding odds ratios(95%confidence intervals [CI]) relative to no therapy of 0.28 (0.24, 0.33), 0.38 (0.34, 0.43), 0.46 (0.40, 0.52), and 0.49 (0.45, 0.53) for eachtherapy group, accordingly. The use of ATTs also appeared to be associated with a reduced risk for ischemic or thrombotic events.Conclusions These data contribute to the formulation of the hypothesis that use of ATTs in clinical practice decreases the risk of morbidityand mortality in patients with IHF, although findings require further confirmative studies. Copyright # 2010 John Wiley & Sons, Ltd.

key words — heart failure; antithrombotic therapy; survival; epidemiology

Received 23 November 2009; Revised 18 April 2010; Accepted 30 April 2010

INTRODUCTION

In the United States alone, heart failure (HF) affectsmore than 5 million people with 400 000–700 000 newcases diagnosed each year.1 HF remains a disease thatis difficult to manage, has a tremendous impact onquality of life, and is associated with a mortality rate ofover 30% in advanced stages of the disease.2

Through a multi-faceted pathophysiologic mechan-ism, HF might be associated with a hypercoagulable

* Correspondence to: P. Burton, Johnson and Johnson PharmaceuticalResearch and Development, 920 Route 202, Raritan, NJ 08879, USA.E-mail: [email protected]. Cheng is currently employed by Gilead Sciences Inc, Palo Alto, CA,USA.yThe data included in the manuscript were partially presented at the 24thInternational Conference on Pharmacoepidemiology and Drug Safety inAugust 2008, Copenhagen, Denmark.

Copyright # 2010 John Wiley & Sons, Ltd.

state, thus putting patients at an increased risk formyocardial infarction (MI), stroke, pulmonary embo-lism (PE), and other venous and arterial thromboem-bolic events. The current understanding of HF seemsto support the use of antiplatelet and anticoagulantagents in patients with HF who have pro-thromboticrisk factors, including coronary artery disease (CAD),atrial fibrillation (AF), a history of thromboembolicevents, and left ventricular mural thrombus.3 However,actual use of those drugs varies widely in clinicalpractice, and reliable estimates of the treatment effectsof those therapies for the management of HF aregenerally lacking.

Recent clinical trials have yielded inconclusiveevidence about the beneficial effect of antithrombotictherapy (ATT). The warfarin and antiplatelet therapy inchronic heart failure (WATCH) study was conducted toexamine the comparative efficacy of aspirin (ASA),

912 z. yuan ET AL.

clopidogrel, or warfarin for patients with HF. The studyshowed no significant difference among the treatmentgroups for the composite endpoint of death, acute MI(AMI), or stroke (20.5% for ASA, 21.8% for clopido-grel, and 19.8% for warfarin), although ASA wasassociated with a higher HF hospitalization rate ascompared with warfarin (22.2% vs. 16.1%, p¼ 0.01).4

Another study, warfarin/aspirin study in heart failure(WASH), compared antithrombotic strategies using adesign that randomized 279 participants (mean age 63years old) who were followed-up for an average of 27months. This study showed no significant difference inthe composite endpoint of death, nonfatal MI, andnonfatal stroke among the treatment groups (26% forno ATT, 32% for ASA, and 26% for warfarin).5

The goal of the study is to explore in a researchdatabase whether ATTs affect cardiovascular outcomesin patients with incident HF.

METHODS

Data source (the health improvement networkdatabase)

This study was based on an analysis of the healthimprovement network (THIN) database, which con-sists of primary care data from clinical practices in theUnited Kingdom (UK).6 Data are organized in files byindividual practice and provide longitudinal medicalinformation for each patient, including computerizedanonymous demographic, medical, and prescriptionrecords at the individual patient level. Medical eventsare recorded using Read Codes.7 Death records are alsocaptured in the file. At the time of this analysis, thedataset contained over 6 million UK patients (2.75million who were active) registered with 358 GeneralPractioners from 1985 to 2006, representing over37 million patient years of data.

Patients and study duration

Patients (�18 years of age) with a diagnosis of HF(codes available upon request) during 2000–2006 wereconsidered eligible for the study. HF was defined basedon Read Codes7 available in the study database that wereused by the primary care physician to documenta disease in clinical practice. The first diagnosis ofHF for each patient was defined as the index date. Inaddition, a patient had to meet the following criteria to bequalified for the analysis: (1) registered with the generalpractitioner for at least 12 months prior to the index date,and (2) HF had to be an incident event. Because aphysician’s decision to prescribe an ATT might be


affected by a patient’s predicted longevity, we excludedpatients who did not survive more than 28 daysfollowing the initial diagnosis of HF. All study patientswere to be followed up to 18 months from the index date,or to the last available follow-up date, or death.

Treatment exposure

Five mutually exclusive treatment groups were de-fined as: (1) warfarin combination therapy (warfarinplus clopidogrel and/or ASA); (2) warfarin alone; (3)clopidogrel, with or without ASA; (4) ASA alone; and(5) no therapy (none of the above). ATTs werecategorized primarily based on an a priori decision,taking into consideration the mechanism of action ofthe drugs (i.e., warfarin is a vitamin K antagonist, whileclopidogrel and aspirin are both antiplatelet agents)and preserving a reasonable sample size for each treat-ment group in order to achieve reliable estimates of thetreatment effects. For the analysis related to all-causemortality, patients must have been prescribed at least28 days of drug or received two prescriptions (if medi-cation days were missing) during the entire follow-upperiod to be considered on treatment. However, for theanalysis of thrombotic and bleeding events, the medi-cation had to be dispensed before an event for thepatient to be considered on treatment. The durations oftherapy for clopidogrel and ASA were generallyavailable in the dataset, but the duration for warfarintherapy was often missing, likely due to the fact thatdose titration was often needed for getting an optimalinternational normalized ratio (INR) value. Thus, weconsidered a patient to be on warfarin therapy, if he orshe received at least two prescriptions during thefollow-up period.

Main outcomes measures

Assessing all-cause mortality during the 18-monthfollow-up period was the primary aim of this study. Thecompleteness of death reporting in the study databasehas been examined by the THIN internal researchinvestigators.8 The findings suggest that the reportingof death varied over time for all practices. However,since the year 2000, the mortality rates reported peryear in the THIN database appeared to be closelycorrelated with age- and sex-specific national deathrates, thus suggesting overall acceptability of the mort-ality data. The occurrence of MI, stroke, deep venousthrombosis (DVT), PE, as well as bleeding events wasidentified using diagnostic Read Codes (available uponrequest) and evaluated in relation to ATTs.

Pharmacoepidemiology and Drug Safety, 2010; 19: 911–920DOI: 10.1002/pds

Figure 1. The inception of incident HF cohort

ANTITHROMBOTIC THERAPIES IN PATIENTS WITH HEART FAILURE 913

Statistical analysis

Descriptive statistics were provided for baselinecharacteristics. Mortality rates were assessed overtime, and the incidences of thrombotic and bleedingevents were calculated as number of events divided byperson-time at risk and the corresponding 95%confi-dence intervals (CIs) were estimated using the normalapproximation.

Unadjusted logistic regression models were first fitfor all baseline characteristics individually to assess theunadjusted association of each covariate with mort-ality, and were followed with a model adjusting formultiple confounders (named as conventional logisticmodel hereafter) and Charlson index scores. Thepotential confounding variables (including age, sex,comorbid illnesses, and prior medications of interest)were chosen based on the information available in thedatabase, as well as clinical judgment about commonlyknown associations with the treatment and outcomemeasures. In addition, we adjusted for the Charlsonindex a disease severity measurement that was develo-ped based on a wide spectrum of comorbid conditionswith a variety of pathologic conditions and is usedquite extensively in longitudinal studies.9,10 We testedmodel goodness-of-fit by means of the Hosmer–Lemeshow Chi-square test.11

A Cox proportional hazards model was used tocompare mortality rates among the treatment groups,while adjusting for baseline characteristics to corro-borate the overall findings. Patients were censoredaccording to the date of last available record inthe database or the end of the 18-month follow-upfrom the index date, if no death record was observed.The proportional hazards assumption was evaluatedby examining log (�log) plots. In addition, we alsoexplored the use of propensity scores to complementthe more conventional modeling approaches.12,13 Thepropensity score is a model-predicted probability ofreceiving treatment for any given patient based onspecific prognostic factors. We used logistic regres-sion to calculate the propensity score with the dep-endent variable being receipt of ATT, and the pre-dictor variables being the baseline characteristics aswell as the interaction of age and sex. Propensityscores were then categorized into deciles. Logisticregression models were used to compare mortalityrates between each of the active treatment groups andthe no therapy group, adjusting for propensity decilesusing a series of indicator variables, as well as ageand sex to adjust further for any potential residualconfounding,14 given that age was a strong predictor ofmortality.


As for treatment comparison on ischemic or throm-botic events (as well as the composite endpoint ofischemic or thrombotic events or all-cause death), amultivariable Cox regression model was used for theanalysis, while adjusting for all the potential con-founders identified at baseline.

Sensitivity analysis

To determine if the exclusion of patients who didnot survive more than 28 days affected the overallconclusion of the study, we conducted a sensitivityanalysis by including those patients. The treatmentdifferences were consistent with the findings from themain analysis, so these results are not presented.

The study proposal was reviewed and approved bythe THIN NHS Research Ethics Committee. All theanalyses were carried out using SAS, version 9.1 (SASInstitute Inc., Cary, North Carolina).

RESULTS

A total of 39 462 HF cases were identified during 2000through 2006. After excluding patients who met at leastone of the exclusion criteria (not registered with thegeneral practitioner at least one year before the indexdate (n¼ 2801), prevalent cases (n¼ 9792), and diedwithin 28 days of the index date (n¼ 2610), theinception cohort included 24 554 incident HF patients(Figure 1).

Table 1 presents the baseline characteristics. Patientsreceiving ATTs tended to be younger, more likely to bemen, had more cardiovascular comorbidities, and weremore likely to receive therapies (prior to the diagnosisof HF) that are commonly used to treat the HF con-dition. All these factors remained as independent


Table 1. Patient characteristics of incident heart failure by five mutually exclusive antithrombotic drug cohorts (2000–2006), THIN (0710 version)

Warfarinþ anyother� (N¼ 1685)

Warfarin alone(N¼ 3581)

Clopidogrelþ aspiriny

(N¼ 2039)Aspirin alone(N¼ 10 252)

None(N¼ 6997)

Total(N¼ 24 554)

Year of the cohort identified (n)2000 220 454 79 1613 1376 37422001 242 540 187 1795 1239 40032002 230 597 289 1791 1267 41742003 266 528 378 1586 1092 38502004 249 526 358 1304 736 31732005 239 462 389 1148 698 29362006 239 474 359 1015 589 2676

Age, mean (SD), years 73.3 (10.0) 73.6 (10.3) 74.7 (10.9) 77.1 (10.3) 77.9 (11.9) 76.4 (11.0)Median (range) 74.7 (26; 99) 75.5 (26; 98) 76.2 (32; 99) 78.4 (21; 104) 80.0 (18; 107) 77.9 (18; 107)Female (%) 35.4 42.9 45.0 48.9 59.6 49.8History (%)

Atrial fibrillation 37.4 54.7 12.6 14.1 10.2 20.4COPD 8.8 8.9 10.4 8.8 10.9 9.5Cancer 10.6 11.1 11.5 10.7 13.3 11.6Diabetesþ diabetes medication 18.3 15.4 24.6 18.1 11.8 16.5Hyperlipidemia 16.0 10.5 20.4 13.3 6.1 11.6Hypertension 50.0 45.9 54.1 49.1 40.6 46.7Liver disease 3.6 3.4 3.8 3.2 3.6 3.4Peripheral artery disease 7.2 5.8 10.2 7.4 4.5 6.6Prior CABG 10.8 5.8 9.9 6.6 1.4 5.6Prior myocardial infarction 26.5 16.1 39.4 23.5 6.8 19.2Prior PCI 2.7 0.8 5.7 1.6 0.4 1.6Prior stroke 11.8 10.8 14.4 11.4 6.1 10.1Renal disease 5.6 5.9 9.1 6.6 6.0 6.5

Prior medications (%)ACE inhibitor 53.2 50.7 51.7 48.6 36.4 46.0Angiotensin receptor blocker (ARB) 10.8 10.4 12.7 9.1 6.4 8.9b-blocker 38.1 32.9 44.3 31.8 18.1 29.5Digoxin 28.7 42.6 9.7 12.4 10.8 17.2Diuretic 78.2 77.5 72.4 76.2 78.0 76.7Spironolactone 10.0 9.7 8.5 6.2 6.3 7.2Statin 40.8 29.1 57.9 33.9 11.7 29.3

Medications during 18 months follow-up (%)ACE inhibitor 81.0 76.2 76.2 72.2 52.7 68.4Angiotensin receptor blocker 23.8 22.6 22.8 19.7 13.7 19.0b-blocker 57.6 47.1 60.5 41.8 21.5 39.7Digoxin 58.4 64.1 22.0 24.7 16.7 30.3Diuretic 96.0 94.2 91.0 92.5 82.9 90.2Spironolactone 39.7 35.6 33.1 26.1 17.4 26.6Statin 65.5 48.6 80.5 53.6 21.5 47.2

SD, standard deviation; COPD, chronic obstructive pulmonary disease; PCI, percutaneous coronary intervention; CABG, coronary artery bypass graft; ACE,angiotensin converting enzyme inhibitor.�Warfarin and another antiplatelet agent (either aspirin or clopidogrel or both). Does not include warfarin alone.yClopidogrel alone or clopidogrel plus aspirin.

914 z. yuan ET AL.

predictors for the use of ATTs in a multivariable model(data are not shown). In addition, it was notable thatpatients with AF were more likely to receive warfarintherapy (in combination or alone) and patients withprior MI were more likely to receive clopidogreltherapy.

Mortality

During the 18-month follow-up period, the mortalityrate was 11.1% for patients receiving warfarin combi-nation therapy, 14.6% for those receiving warfarin


alone, 17.8% for those receiving clopidogrel, and19.5% for patients receiving ASA alone, as comparedwith 32.6% for patients with no therapy (Table 2). Inaddition, the treatment effects on mortality were fur-ther explored separately among those patients with andwithout certain comorbid conditions (i.e., CAD andAF, which might be indicative of need for antithrom-botic prophylaxis). Regardless of the presence of thecomorbid conditions of interest, treatment benefitswere consistently observed. For example, among thosewithout CAD and AF, the mortality rate was 11.5%for patients receiving warfarin combination therapy,


Table 2. Unadjusted mortality by antithrombotic therapy during 6, 12, and 18 months of follow-up of patients with incident heart failure, 2000–2006, THINdatabase (10/07 version)

Mortality6-Month 1-Year 18-Month Person

yearsRate/100

person-years

Numberat risk

Numberdeaths

Percent(95%CI)

Numberdeaths

Percent(95%CI)

Numberdeaths

Percent(95%CI)

Personyears

Rate(95%CI)

Warfarinþ any other� 1685 30 1.8 (1.2, 2.5) 109 6.5 (5.4, 7.7) 187 11.1 (9.7, 12.7) 2352.1 8.0 (6.9, 9.2)Warfarin alone 3581 155 4.3 (3.7, 5.0) 358 10.0 (9.0, 11.0) 522 14.6 (13.4, 15.8) 4807.3 10.9 (9.9, 11.8)Clopidogrelþ aspiriny 2039 111 5.4 (4.5, 6.5) 228 11.2 (9.9, 12.6) 362 17.8 (16.1, 19.5) 2681.8 13.5 (12.1, 15.0)Aspirin alone 10 252 701 6.8 (6.4, 7.3) 1444 14.1 (13.4, 14.8) 2001 19.5 (18.6, 20.3) 13286.6 15.1 (14.4, 15.7)None 6997 1401 20.0 (19.1, 21.0) 1943 27.8 (26.7, 28.8) 2279 32.6 (31.5, 33.7) 7659.9 29.8 (28.5, 31.0)

�Warfarin and another antiplatelet agent (either aspirin or clopidogrel or both). Does not include warfarin alone.yClopidogrel alone or clopidogrel plus aspirin.


14.2% for those receiving warfarin alone, 16.7% forthose receiving clopidogrel, and 19.3% for patientsreceiving ASA alone, as compared to 29.2% for pati-ents with no therapy (Table 3). Based on univariate andmultivariable logistic regression models, the inter-actions for ATT use by AF versus no AF and for ATTuse by CAD versus no CAD were all statisticallysignificant (p< 0.001), which were probably driven bythe notable difference in mortality rates of 43.7% and43.6%, respectively, in AF and CAD patients with noATT treatment, as compare to a mortality rate of 29.2%in other patients with no ATT treatment.

Mortality rose sharply with advancing age, regard-less of the presence of ATTs, ranging from 10.3% forpatients less than 65 years of age to 35.0% for patientsolder than 84 years of age. In addition, men generallyhad higher mortality rates than women within differentage categories and treatment cohorts (Figure 2), altho-ugh the overall mortality rates for men and womenwere identical (21.8%) because, on average, womenwere older than men in our study cohort-(mean�standard deviation, SD) age for women 78.8� 10.3versus 74.0� 11.1 for men.

Univariate analyses revealed that age, most preexistingcomorbidities (including stroke, diabetes, peripheral

Table 3. Effect of antithrombotic therapies on mortality based on the comorbid

Mortality rate at 18-mCoronary artery disease

Treatment groups Patients at risk Percent (95%CI) Patients

Warfarinþ any other� 735 10.1 (8.0, 12.4) 63Warfarin alone 1113 18.1 (15.9, 20.4) 195Clopidogrelþ aspiriny 1102 18.1 (16.0, 20.5) 25Aspirin alone 4322 19.5 (18.4, 20.7) 144None 1180 43.7 (40.9, 46.6) 71

�Warfarin and another antiplatelet agent (either aspirin or clopidogrel or both). DyClopidogrel alone or clopidogrel plus aspirin.zOther¼ patients without coronary artery disease or atrial fibrillation.


vascular disease, chronic obstructive pulmonary dis-ease (COPD), renal disease, and cancer) and prior useof several medications (including diuretics, digoxin,and spironolactone) were associated with an increasedrisk for mortality, while other comorbidities (includinga history of percutaneous coronary intervention, coro-nary artery bypass surgery, hyperlipidemia, and AF)and prior use of other medications, including angio-tensin converting enzyme (ACE) inhibitors, angioten-sin receptor blockers (ARBs), beta-blockers, and statins,were associated with a decreased risk for mortality(Table 4). All those variables remained statistically( p< 0.05) or marginally ( p< 0.10) significant in themultivariable model (Table 4).

Adjusting for all the covariates identified, theconventional multivariable logistic regression modelyielded odds ratios (95%CI) relative to no therapy of0.28 (0.24, 0.33), 0.38 (0.34, 0.43), 0.46 (0.40, 0.52),and 0.49 (0.45, 0.53) for the warfarin combinationtherapy, warfarin alone, clopidogrel, and the ASAalone therapy, respectively. The other modeling met-hods (including a logistic model adjusting for Charlsonindex scores, a Cox proportional hazards model, andpropensity score models) generally resulted in similarconclusions (Table 5). However, it is worth noting that

ity of interest

onth follow-up by comorbidity of interestAtrial fibrillation Otherz

at risk Percent (95%CI) Patients at risk Percent (95%CI)

1 11.4 (9.1, 14.1) 617 11.5 (9.2, 14.2)9 13.5 (12.1, 15.1) 1137 14.2 (12.2, 16.3)6 20.7 (16.1, 26.0) 809 16.7 (14.2, 19.4)4 22.0 (19.9, 24.1) 5025 19.3 (18.2, 20.4)3 43.6 (40.0, 47.3) 5296 29.2 (28.0, 30.4)

oes not include warfarin alone.


Figure 2. Mortality (%) during 18-month follow-up period by age, gender, and treatment cohort. WarfþOther: warfarin and another antiplatelet agent (eitheraspirin or clopidogrel or both), which does not include Warfarin alone; ClopþASA: clopidogrel alone or clopidogrel plus aspirin

916 z. yuan ET AL.

the risk reductions associated with warfarin therapy(combination or alone) and clopidogrel therapy wereeven greater based on propensity score models ascompared with other modeling approaches (Table 5).Note that in all multivariable models, male sex is anindependent risk factor for mortality, with OR¼ 1.34(95%CI: 1.26–1.44) per the conventional logisticmodel.

Thrombotic and bleeding events

The overall rates (per 100 person-years) of stroke (2.3),AMI (3.5), DVT (0.7), PE (0.8), and bleeding events(2.2 for gastrointestinal [GI] bleeding and 0.9 for otherbleeding) were lower than the mortality rate for theentire cohort (Figure 3). However, data generally didnot show an increased risk for bleeding associated withATTs.

During the follow-up period, the event rates forstroke ranged from 1.2 to 2.2 per 100 person-years forpatients receiving ATTs as compared with 3.2 per 100person-years for no therapy; AMI rates ranged from 1.4to 4.5 per 100 person-years for patients who receivedATTs as compared with 6.0 per 100 person-years forpatients receiving no therapy; DVT rates ranged from


0.4 to 0.8 per 100 person-years for patients whoreceived ATTs as compared with 1.3 per 100 person-years for those receiving no therapy; while PE ratesranged from 0.4 to 0.6 per 100 person-years for patientswho received ATTs as compared with 1.7 per 100person-years for those receiving no therapy. When theendpoints for stroke, AMI, DVT, and PE were com-bined, the results based on Cox proportional hazardsmodel showed that ATTs were associated with asignificant risk reduction with hazard ratios (95%CI) of0.27 (0.21, 0.35), 0.34 (0.29, 0.40), 0.42 (0.35, 0.50),and 0.41 (0.37, 0.46) for the warfarin combinationtherapy, warfarin alone, clopidogrel, and ASA alonegroups, respectively, as compared with those receivingno therapy. It is also interesting to note that the the-rapeutic benefit associated with the ATTs was consi-stently demonstrated when a composite endpoint of theabove-defined ischemic or thrombotic events plusdeath was used for the multivariate analysis (results notshown).

DISCUSSION

This was an analysis of a population-based incident HFcohort, exploring the use of ATTs and their impact on


Table 4. Odds-ratios (ors) for mortality in incident heart failure, 2000–2006, THIN database (10/07 version)

Univariatelogistic model

N¼ 24 554

Multivariatelogistic model�

N¼ 24 554

Mutually exclusive therapy OR OR 95%CIWarfarinþ any othery 0.26 0.28 (0.24, 0.33)Warfarin alone 0.35 0.38 (0.34, 0.43)Clopidogrelþ aspirinz 0.45 0.46 (0.40, 0.52)Aspirin alone 0.50 0.49 (0.45, 0.53)None 1.0 1.0Age (1-year increment) 1.05 1.05 (1.04, 1.05)Male 1.01 ns 1.34 (1.26, 1.44)History

Prior myocardial infarction 0.97 ns 1.40 (1.28, 1.53)Prior stroke 1.33 1.38 (1.25, 1.53)Prior PCI 0.53 0.76 ns (0.55, 1.03)Prior CABG 0.56 0.77 (0.65, 0.91)Hypertension 0.85 0.88 (0.82, 0.94)Diabetes 1.12 1.43 (1.31, 1.56)Hyperlipidemia 0.65 0.89 ns (0.79, 1.01)Atrial fibrillation 0.90 0.90 (0.81, 0.99)Peripheral artery disease 1.49 1.56 (1.39, 1.76)COPD 1.32 1.32 (1.19, 1.47)Liver disease 1.10 ns 1.26 (1.06, 1.49)Renal disease 1.47 1.44 (1.27, 1.62)Cancer 1.62 1.42 (1.30, 1.56)

Prior medicationsACE inhibitor 0.75 0.84 (0.78, 0.90)Angiotensin receptor blocker 0.73 0.78 (0.69, 0.89)b-blocker 0.68 0.92 (0.85, 1.00)Diuretic 1.43 1.22 (1.12, 1.33)Digoxin 1.20 1.37 (1.24, 1.51)Statin 0.64 0.90 (0.82, 0.99)Spironolactone 1.32 1.43 (1.27, 1.61)

ns, not statistically significant at the 5% level; CI, confidence interval;COPD, chronic obstructive pulmonary disease; PCI, percutaneous coronaryintervention; CABG, coronary artery bypass graft; ACE inhibitor, angio-tensin converting enzyme inhibitor.�Final model with c¼ 0.703; Hosmer and Lemeshow goodness of fit test:p¼ 0.916.yWarfarin and another antiplatelet agent (either aspirin or clopidogrel orboth). Does not include warfarin alone.zClopidogrel alone or clopidogrel plus aspirin.


mortality and other clinical outcomes in routine clini-cal practice. To our knowledge, this is one of the largeststudies that has investigated the use of ATT in patientswith HF in the real-world clinical practice setting, andthe mean age of our study cohort (76.4� 11.0) wasgenerally similar to other observational studies ofincident HF published recently.15–19

Why is anticoagulation possibly linked withimproved outcome in HF patients?

Virchow first described in 1856 (Virchow RLK(1856). ‘Thrombose und Embolie. Gefassentzundungund septische Infektion’, Gesammelte Abhandlungenzur wissenschaftlichen Medicin. Frankfurt am Main:


Von Meidinger & Sohn, 219–732. Translation inMatzdorff AC, Bell WR (1998). Thrombosis andembolie (1846–1856). Canton, Massachusetts: ScienceHistory Publications. ISBN 0-88135-113-X) the triadthat led to an increased risk for thrombosis. Put simplythis includes disturbance in blood flow, damage to theblood vessel wall or a change in the constituents ofblood coagulation. There is evidence that patients withHF have abnormalities of all three of these systems andas such may be at increased risk for venous and arterialthrombosis that increase both mortality and morbid-ity.20 ATTs interfere with the coagulation cascade,thereby offering therapeutic benefits for reducingthrombotic events, and potentially associated morbid-ity and mortality, and are recommended for a subset ofspecific patients with HF.3 Our analyses showed thatATTs were associated with significant reductions inischemic and thrombotic events as well all-causemortality. Similar findings have been reported by otherinvestigators,21 which might support the notion thatthromboembolism as a cause for poor prognosis inpatients with HF might be underestimated. One reportshowed that about 31% of deceased patients originallythought to have a nonischemic cause of HF had signi-ficant CAD at autopsy.22 Recent clinical trial data,however, revealed no compelling evidence supportingthe beneficial effect of ATTs in reducing all-causemortality.4,5,23 In particular, the results from the WATCHtrial (N¼ 523, 524, and 540 for aspirin, clopidogrel, andwarfarin, respectively)23 have failed to demonstratesuperiority of warfarin versus aspirin or clopidogrelversus aspirin in the composite endpoint of death,nonfatal MI, or nonfatal stroke, although warfarin wasassociated with fewer nonfatal strokes as comparedwith aspirin or clopidogrel, and a reduced risk forhospitalization for worsening HF as compared withaspirin. Although the finding from this study does notrule out a potential benefit of ‘any’ therapy versus notherapy, it does raise another interesting dilemma abouthow to interpret the study results from clinical trialsand noninterventional studies.24 Note that the WATCHtrial was designed to determine the optimal antith-rombotic agent for HF patients with three treatmentarms: open-label warfarin and double-blind treatmentwith either aspirin or clopidogrel. In WATCH, thetreatment effect was measured by a composite endpointof death, nonfatal MI, or nonfatal stroke. The originalsample size of 1500 patients for each of the threegroups was intended to have 90% power to detect a20% between-group difference in annual event ratesadjusting for the three pair-wise comparisons. How-ever, the trial was stopped prematurely (N¼ 1587


Table 5. Results from various modeling approaches comparing the risk of mortality among mutually exclusive antithrombotic therapies within a cohort ofincident CHF, 2000–2006, THIN (0710)

Conventional multivari-ate logistic model�

Multivariate logisticmodel including the

Charlson index��

Cox proportionalhazards model��

Propensity scoremodelsy

N¼ 24 554 N¼ 24 554 N¼ 24 554 N varies by model

Mutually exclusive therapy OR 95%CI OR 95%CI HR 95%CI OR 95%CIWarfarinþ any otherz 0.28 (0.24, 0.33) 0.28 (0.24, 0.33) 0.28 (0.24, 0.33) 0.22 (0.18, 0.26)Warfarin alone 0.38 (0.34, 0.43) 0.38 (0.33, 0.42) 0.38 (0.34, 0.42) 0.31 (0.26, 0.35)Clopidogrelþ aspirinx 0.46 (0.40, 0.52) 0.47 (0.41, 0.53) 0.45 (0.40, 0.51) 0.33 (0.28, 0.39)Aspirin alone 0.49 (0.45, 0.53) 0.50 (0.46, 0.54) 0.49 (0.46, 0.52) 0.49 (0.46, 0.53)None 1.0 — 1.0 — 1.0 — — —

�The model adjusted for all the baseline characteristics.��The model adjusted for age, sex, Charlson index score and baseline medications, including ACE inhibitor, ARB, beta-blocker, diuretic, digoxin, statin, andspironolactone.��Model fitted using the same variables that were included in the conventional logistic regression model.yPropensity score models were fit using deciles of the propensity score, as well as age and sex.zWarfarin and another antiplatelet agent (either aspirin or clopidogrel or both). Does not include warfarin alone.xClopidogrel with or without aspirin.

918 z. yuan ET AL.

patients rather than 4500) due to slow enrollment. As aresult, the study is clearly underpowered to detect mod-est differences between treatments, in spite of treat-ment differences that seemed to be trending in adirection favoring warfarin therapy. In addition, clini-cal trials are typically conducted on a narrowly definedpatient population with limited comorbidities, the res-ults observed in such settings (particularly consideringthe sample size was significantly reduced in WATCH)might not be applicable to what would have been

Figure 3. Overall event rate (per 100 person-years) for the cohort with incidentversion). �Warfarin and another antiplatelet agent (either aspirin or clopidogrel or boaspirin. AMI, acute myocardial infarction; PE, pulmonary embolism; DVT, deep95%confidence intervals


observed in clinical practice. Furthermore, despite ourvarious modeling approaches generally reaching theconclusion that ATTs were associated with reducedrisks for mortality and ischemic/thrombotic events,there is always a challenge to inferring a causalinterpretation of the effect, given the nature of obser-vational studies.25 It is worth noting that anotherrandomized, double-blind clinical trial funded by theNational Institutes of Health is currently underway.26

The study is planned to enroll a total of 2860 patients

HF during 18-month follow-up period, 2000–2006, THIN database (10/07th). Does not include Warfarin alone. yClopidogrel alone or clopidogrel plusvenous thrombosis. The bars on the graph represent the upper bounds of


KEY POINTS

� Heart failure is a significant public healthproblem.

� The use of antithrombotic therapies (ATTs) inpatients with incident heart failure (IHF) varies inclinical practice.

� The data in this paper contribute to theformulation of the hypothesis that the use ofATTs in clinical practice affects morbidity andmortality in patients with IHF, although findingsrequire further confirmative studies.


with HF (cardiac ejection fraction �35%) from 70North American and 70 European sites, examining thelong-term treatment effect of warfarin versus aspirin onthe composite endpoint of death, or stroke (ischemic orhaemorrhagic).

Although the event rates for the main clinicaloutcome measures (i.e., mortality and thromboticevents) in our study were generally similar to those inreports from other studies, it was rather unexpected thatpatients receiving no ATTs had a higher bleeding ratethan their counterparts. As a result, we furtherinvestigated the use of other medications that mightpotentially lead to a higher bleeding risk during thefollow-up period, specifically nonsteroidal antiinflam-matory drugs (NSAIDs). However, we found thatprescriptions for the top 10 NSAIDs did not varysubstantially across the five groups investigated in thisstudy. As such, we postulate that the unexpected blee-ding results might be a reflection of two potentialissues: (1) there may have been underreporting ingeneral or specifically of bleeding events not requiringmedical intervention, and (2) patient channeling. It isconceivable that the treating physicians might haveselectively prescribed ATTs to those who are likely tobe at low risk for bleeding, although no information inthe database would be readily available to assess such abias.

Limitations of the study

This was a database study and as such may be subject tobias due to the completeness of data collection andreporting. The condition of HF was diagnosed perclinical practice, defined based on the Read Codes, andthe accuracy was not validated. Therefore, there is apotential for misclassification.

Additionally, some of the factors that might beassociated with the prognosis of HF, for example,ethnicity, socioeconomic status, smoking, alcohol use,body mass index, and the particular etiology of HF, aswell as the severity of the HF condition (e.g., the NYHAclass, and normal or reduced ejection fraction),19,27–30

are not available for analysis. Therefore, our multi-variable models have limitations. Thus, our findingsmight have been biased due to residual confounding.

In summary, these data based on a large, well char-acterized ‘real world’ cohort contribute to the formu-lation of the hypothesis that use of ATTs in clinicalpractice decreases the risk of morbidity and mortalityin patients with incident heart failure (IHF), althoughfindings require further confirmative studies.

Authors and Contributions:


Zhong Yuan and Paul Burton participated in theconception, design and conduct of the analyses, andcontributed to the write-up of the paper.

Rachel Weinstein and George Griffin participated inthe conduct of the analyses, and provided program-ming support of this article.

John Zhan, Mei Cheng, Robert Zolynas, AlexeiPlotnikov, Mary S Lee, and Leonard Oppenheimerparticipated in the conception, design, and conduct ofthe analyses of this article.

All authors were involved in data interpretation andthe review of the paper.

CONFLICT OF INTEREST

The authors (except for Dr. Cheng who has left thecompany) are employees of Johnson and Johnson Phar-maceutical Research and Development, LLC, which iscurrently developing an antithrombotic therapy for vari-ous indications, one of which could be heart failure.

DATA SHARING STATEMENT

The data that were included in this study will beavailable upon request, if such a request has beenapproved by the database owner: EPIC, part of theCegedim Strategic Data (CSD) Medical Researchgroup.

ETHICAL APPROVAL

The study proposal was reviewed and approved by theTHIN NHS Research Ethics Committee.

ACKNOWLEDGEMENTS

We thank Dr. Jesse Berlin for his review and thoughtfulcomments to the paper, and Tatiana Piotroff-Smith for hereditorial assistance.


920 z. yuan ET AL.

REFERENCES

1. Hunt Hunt SA, Abraham WT, Chin MH, et al. American College ofCardiology; American Heart Association Task Force on PracticeGuidelines; American College of Chest Physicians; InternationalSociety for Heart and Lung Transplantation; Heart Rhythm Society.ACC/AHA 2005 Guideline Update for the Diagnosis and Managementof Chronic Heart Failure in the Adult: a report of the American Collegeof Cardiology/American Heart Association Task Force on PracticeGuidelines (Writing Committee to Update the 2001 Guidelines forthe Evaluation and Management of Heart Failure): developed incollaboration with the American College of Chest Physicians andthe International Society for Heart and Lung Transplantation: endorsedby the Heart Rhythm Society. Circulation 2005; 112: e154–235.

2. Gomberg-Maitland M, Baran DA, Fuster V. Treatment of congestiveheart failure: guidelines for the primary care physician and the heartfailure specialist. Arch Intern Med 2001; 161: 342–352.

3. Farmakis D, Filippatos G, Lainscak M, Parissis JT, Anker SD, Kre-mastinos DT. Anticoagulants, antiplatelets, and statins in heart failure.Cardiol Clin 2008; 26: 49–58.

4. Cleland JG, Huan Loh P, Freemantle N, Clark AL, Coletta AP. Clinicaltrials update from the European Society of Cardiology: SENIORS,ACES, PROVE-IT, ACTION, and the HF-ACTION trial. Eur J HeartFail 2004; 6: 787–791.

5. Cleland JG, Findlay I, Jafri S, et al. The Warfarin/Aspirin Study inHeart failure (WASH): a randomized trial comparing antithromboticstrategies for patients with heart failure. Am Heart J 2004; 148(1): 157–164.

6. The Health Improvement Network (THIN). http://www.epic-uk.org/thin.htm (accessed 5 March 2008).

7. NHS Centre for Coding and Classification. The READ Codes, Version3. Stationary Office: London, 1996.

8. Maguire A, Blak B, Thompson M, Hall G. The importance of definingperiods of complete mortality reporting in automated patient data-bases. EPIC, CSD (presented in THIN research day, at the VillanovaConference Centre in Radnor, Pennsylvania, on June 17, 2008).

9. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method ofclassifying prognostic comorbidity in longitudinal studies: develop-ment and validation. J Chronic Dis 1987; 40: 373–383.

10. Subramanian U, Eckert G, Yeung A, Tierney WM. A single healthstatus question had important prognostic value among outpatients withchronic heart failure. J Clin Epidemiol 2007; 60: 803–811.

11. Hosmer DW, Taber S, Lemeshow S. The importance of assessing the fitof logistic regression models: a case study. Am J Public Health 1991;81: 1630–1635.

12. Rosenbaum PR, Rubin D. The central role of the propensity score inobservational studies for causal effects. Biometrika 1983; 70: 41–55.

13. D’Agostino RB, Jr., Propensity score methods for bias reduction in thecomparison of a treatment to a non-randomized control group. Stat Med1998; 17: 2265–2281.

14. Austin PC, Grootendorst P, Anderson GM. A comparison of the abilityof different propensity score models to balance measured variablesbetween treated and untreated subjects: a Monte Carlo study. Stat Med2007; 26: 734–753.


15. Cowie MR, Wood DA, Coats AJ, et al. Survival of patients with a newdiagnosis of heart failure: a population based study. Heart 2000; 83:505–510.

16. Cowie MR, Fox KF, Wood DA, et al. Hospitalization of patients withheart failure: a population-based study. Eur Heart J 2002; 23: 877–885.

17. Wang TJ, Larson MG, Levy D, et al. Temporal relations of atrialfibrillation and congestive heart failure and their joint influence onmortality: the Framingham Heart Study. Circulation 2003; 107: 2920–2925.

18. Varadarajan P, Pai RG. Prognosis of congestive heart failure in patientswith normal versus reduced ejection fractions: results from a cohort of2258 hospitalized patients. J Card Fail 2003; 9: 107–112.

19. van Jaarsveld CH, Ranchor AV, Kempen GI, et al. Gender-specific riskfactors for mortality associated with incident coronary heart disease – aprospective community-based study. Prev Med 2006; 43: 361–367.

20. Chong AY, Lip GY. Viewpoint: the prothrombotic state in heart failure:a maladaptive inflammatory response? Eur J Heart Fail 2007; 9: 124–128.

21. Echemann M, Alla F, Briancon S, et al. EPICAL Investigators. Epi-demiologie de l’Insuffisance Cardiaque Avancee en Lorraine. Antith-rombotic therapy is associated with better survival in patients withsevere heart failure and left ventricular systolic dysfunction (EPICALstudy). Eur J Heart Fail 2002; 4: 647–654.

22. Uretsky BF, Thygesen K, Armstrong PW, et al. Acute coronary findingsat autopsy in heart failure patients with sudden death: results from theassessment of treatment with lisinopril and survival (ATLAS) trial.Circulation 2000; 102: 611–616.

23. Massie BM, Collins JF, Ammon SE, et al. WATCH Trial Investigators.Randomized trial of warfarin, aspirin, and clopidogrel in patients withchronic heart failure: the Warfarin and Antiplatelet Therapy in ChronicHeart Failure (WATCH) trial. Circulation 2009; 119: 1616–1624.

24. Rawlins M. De Testimonio: on the evidence for decisions about the useof therapeutic interventions. Clin Med 2008; 8: 579–588.

25. Rubin DB. Estimating causal effects from large data sets using pro-pensity scores. Ann Intern Med 1997; 127: 757–763.

26. Pullicino P, Thompson JL, Barton B, Levin B, Graham S, Freuden-berger RS. WARCEF Investigators. Warfarin versus aspirin in patientswith reduced cardiac ejection fraction (WARCEF): rationale, objec-tives, and design. J Card Fail 2006; 12: 39–46.

27. Blackledge HM, Newton J, Squire IB. Prognosis for South Asian andwhite patients newly admitted to hospital with heart failure in theUnited Kingdom: historical cohort study. BMJ 2003; 327(7414): 526–531.

28. Yancy CW, Abraham WT, Albert NM, et al. Quality of care of andoutcomes for African Americans hospitalized with heart failure: find-ings from the OPTIMIZE-HF (Organized Program to Initiate Life-saving Treatment in Hospitalized Patients With Heart Failure) registry.J Am Coll Cardiol 2008; 51: 1675–1684.

29. Blair AS, Lloyd-Williams F, Mair FS. What do we know about socio-economic status and congestive heart failure? A review of the literature.J Fam Pract 2002; 51: 169.

30. Mahjoub H, Rusinaru D, Souliere V, Durier C, Peltier M, Tribouilloy C.Long-term survival in patients older than 80 years hospitalised for heartfailure. A 5-year prospective study. Eur J Heart Fail 2008; 10: 78–84.


Antithrombotic therapies in patients with heart failure: hypothesis formulation from a research...

Documents

Transcript of Antithrombotic therapies in patients with heart failure: hypothesis formulation from a research...