Results of the search

The initial search in 2005 yielded 1647 records, of which we excluded 1631 after screening titles and abstracts. We assessed the full text of the remaining 16 records. WASH 2004 met our inclusion criteria. We excluded 12 studies, and the other 3 were ongoing (HELAS 2006; WATCH 2009; WARCEF 2000).

The updated search in 2010 yielded 1219 new records, of which we excluded 1203 after screening the titles and abstracts. We assessed 16 full‐text records and included two previously ongoing studies (HELAS 2006; WATCH 2009).

The updated search in 2013 yielded 1460 new records, of which we excluded 1452 after screening the titles and abstracts. Four records were abstracts of interest based on two retrospective analyses. However, they did not specify the proportion of participants with atrial fibrillation and did not provide information on adverse outcomes, so we excluded them. We reviewed four full‐text records, one of which met the inclusion criteria (WARCEF 2012).

The updated search in 2015 yielded 975 new records; we excluded 971 after screening the titles and abstracts and examined four records in detail. Pullicino 2013 was a journal publication with a subanalysis of the WARCEF 2012 study, looking at different types of stroke. The analysis did not add any new information relevant to this review. The three remaining records were abstracts of conference presentations of two small randomised studies (ASAP‐HF 2013; CACHE‐EHF 2014; CACHE‐EJHF 2014), which did not meet all inclusion criteria for this analysis.

In total, this review includes four studies (HELAS 2006; WARCEF 2012; WASH 2004; WATCH 2009). We present the reasons for excluding studies that closely missed the inclusion criteria in Characteristics of excluded studies.

In the two clinical trials registers, we checked all the retrieved entries but found no additional studies meeting review requirements.

We show the flow of studies, combined for all searches, in Figure 1.

Figure 1

---

---

PRISMA flow chart.

Included studies

We included four RCTs that met the review criteria for assessing the effectiveness of antiplatelet therapy (HELAS 2006; WARCEF 2012; WASH 2004; WATCH 2009; see Characteristics of included studies).

All studies recruited patients with heart failure with reduced LVEF only. Arterial fibrillation was not an absolute exclusion criterion in WASH 2004, but 92% of the study participants had no history of it at the time of recruitment. Other studies only included patients with sinus rhythm. The range of ages for the trials was 18 years and older with a mean age of 61 years across the trials. Most participants were men (61% to 93% across trials). We analysed all outcomes based on the intention‐to‐treat principle.

The studies were international (involving the USA, the UK, countries in Western and Eastern Europe, Canada and Argentina). The pharmaceutical industry provided some funding and support to the studies. The WASH trial was partly supported by the Bristol‐Myers Squibb, which manufactures warfarin in the USA. The Bristol‐Myers Squibb, Sanofi‐Synthelabo, and Dupont pharmaceutical companies partly supported the WATCH trial through unrestricted grants and provision of the study drugs. HELAS 2006 did not specify sources of funding, but Greece‐based centres led and coordinated the study, and Greece‐based companies manufactured all study agents.

WASH 2004 was a pilot study in 279 participants recruited from 17 centres in the UK and 3 centres in the USA. This review analysed only the 180 patients randomised to warfarin (target international normalised ratio (INR) 2.0 to 3.0) versus aspirin (300 mg once daily). The remaining patients were randomised to no antithrombotic therapy and thus were outside the scope of the analysis. Investigators recorded baseline data on all participants by telephone at the data centre before randomisation using prospectively constructed blocks of random numbers.

Participants had to have a clinical diagnosis of heart failure requiring treatment with diuretics and evidence of left ventricular systolic dysfunction on echocardiography within six months before entry (WASH 2004). Echocardiographic criteria of left ventricular dysfunction included an increased left ventricular end‐diastolic internal dimension (≥ 56 mm or ≥ 30 mm/m² body surface area) combined with a fractional shortening of at least 28% or LVEF of at least 35%. Investigators were encouraged to use warfarin in participants who developed arterial fibrillation and to reduce the dose of aspirin or use any appropriate antidyspeptic medication if dyspepsia occurred. This trial was limited by open‐label treatment with warfarin, small numbers and relatively short follow‐up (27 ± 1 months). Also, from the study manuscript is not clear whether any patients had incomplete outcome data.

HELAS 2006 recruited 197 participants in centres in Greece, Cyprus, Yugoslavia, Romania, Bulgaria, Poland and Georgia. We excluded the participants with dilated cardiomyopathy who were randomised to warfarin versus placebo, but we included participants with underlying ischaemic heart disease (n = 115) who were randomised to warfarin versus aspirin. The trial was double‐blinded and included participants with the New York Heart Association (NYHA) class II to IV and LVEF of more than 35% assessed by echocardiography, or by radionuclide or angiocardiographic ventriculography. The University coordination centre conducted computerised randomisation by fax or telephone.Investigators were blinded to therapy, and warfarin doses were recommended according to sham results where appropriate. However, the study manuscript does not provide details on the sham procedures, and it is unclear how robust they were. The participants were regularly followed up with a recording of any clinical outcomes. As with WASH 2004, HELAS 2006 is limited by the small number of participants and short follow‐up period (mean follow‐up time of 18.5 months for those taking aspirin and 18.7 months for those taking warfarin). It is not clear whether there was an independent assessment of study outcomes.

WATCH 2009 included 1587 participants from 142 centres in the USA, Canada and the UK. Patients were eligible if they had NYHA class II to IV heart failure for more than three months, LVEF of less than 35%, were in sinus rhythm and on a diuretic and ACE inhibitor (unless not tolerated). Patients were randomised into three arms: aspirin 162 mg once daily, clopidogrel 75 mg once daily in a double‐blind, double‐dummy manner, or open label warfarin (target INR 2.5 to 3.0). The Cooperative Studies Program Coordinating Center (CSPCC) generated randomisation sequences for each site and allocated treatment. The main outcome measures were adjudicated by a blinded committee. The mean duration of follow‐up was 1.9 years, for a total of 3073 patient‐years of follow‐up. Due to slow enrolment, the study failed to recruit the pre‐specified target of 4500 participants. The study was limited by open‐label management of warfarin. Also, similar to WASH 2004, , it is not clear whether any participants in WATCH 2009 had incomplete outcome data.

WARCEF 2012 was a large RCT conducted at 168 centres in 11 countries. The trial recruited 2305 participants with LVEF of 35% or less and sinus rhythm who were randomised to warfarin (target INR of 2.0 to 3.5) or aspirin (325 mg once daily). Investigators reported the primary outcomes of the first event in a composite endpoint of ischaemic stroke, intracerebral haemorrhage or death from any cause during a mean follow‐up period of of 3.5 ± 1.8 years. Two independent core echocardiography laboratories, which were unaware of the study treatment, assessed the accuracy of LVEF measurement. An independent endpoint adjudication committee blinded to the treatment assignments adjudicated all primary and secondary outcomes and cases of major haemorrhage. The WARCEF trial was a well‐designed, adequately powered study with low risk of the assessed bias types.

Excluded studies

We excluded 40 references and present our reasons for excluding 9 studies that most closely missed the inclusion criteria in Characteristics of excluded studies.

Four RCTs of ACE inhibitors assessed the use of antithrombotic therapy in participants with heart failure and/or left ventricular systolic dysfunction (Al‐Khadra 1998a; SAVE 1997; SOLVD 1997; V‐HeFT II 1992). These analyses were done outside the randomisation scheme. V‐HeFT II 1992 included participants with symptomatic heart failure and radiological, echocardiographic or radionuclide evidence of left ventricular systolic dysfunction. The SOLVD trials included participants with left ventricular systolic dysfunction, defined as a LVEF of 35% or less, who were either symptomatic and enrolled into the treatment trial (SOLVD‐treatment 1991), or asymptomatic and enrolled into the prevention trial (SOLVD‐prevent 1992). SAVE 1997 included participants post‐myocardial infarction with an LVEF of 40% or more and no overt heart failure (i.e. asymptomatic participants). PROMISE 1993 presented limited information from a retrospective analysis of one trial in abstract form only.

In one recent retrospective, observational, non‐randomised study, investigators used aspirin in 32% of 417 participants (warfarin in 28%) with an average follow‐up of five years (EPICAL 2002). Treatment with antithrombotic agents (either aspirin, warfarin or both) was associated with better survival at five years (40.4% versus 31% with no antithrombotic treatment, P = 0.01). However, we excluded this study because 24% of the participants had concomitant arterial fibrillation, investigators reported only survival rates, and they did not report event rates separately for the sinus rhythm group or those treated was aspirin, warfarin or no therapy.

Two almost identical abstracts presented results of the randomised open‐label clopidogrel versus aspirin in chronic heart failure (CACHE) study that compared 75 mg daily doses of the two drugs in 87 participants with heart failure in sinus rhythm and an N‐terminal‐pro‐brain natriuretic peptide (NT‐proBNP) of 400 ng/L (mostly NYHA II) (CACHE‐EHF 2014; CACHE‐EJHF 2014). By six months, treatment with aspirin was associated with worse renal function and higher diastolic blood pressure compared with clopidogrel, but investigators did not report any major adverse outcomes. The study did not have a comparator other than another antiplatelet agent (i.e. placebo or oral anticoagulant), so we excluded it from the analysis.

No adequately powered prospective studies have specifically examined the role of the antiplatelet agent dipyridamole in heart failure participants. There was a small randomised pilot study of dipyridamole plus aspirin (aggrenox) versus placebo in 18 participants with congestive heart failure (NYHA class II to IV) and LVEF of less than 40%, presented as a conference abstract in 2013 (ASAP‐HF 2013). After six months of therapy, there were no reported changes in haemodynamic characteristics in either group. BNP levels were reduced in the treatment group, but not in the placebo group (the report did not include a between‐group comparison). More participants in the treatment group showed improvement in six‐minute walking distance than in the placebo group. We excluded the study because no clinical events were recorded (although investigators mentioned the absence of bleeding events in both groups). There were two active treatments in the intervention arms, and investigators provided no information on whether the participants were in arterial fibrillation or sinus rhythm.

The ongoing COMMANDER HF trial (COMMANDER HF trial) compares rivaroxaban versusvs. placebo in participantpatients with heart failure of ischaemics aetiology with sinus rhythm.

Allocation

All studies used central randomisation with low risk of bias in relation to random sequence generation and allocation concealment.

Blinding

For blinding of participants and personnel, we assessed two studies as being at high risk (WASH 2004; WATCH 2009), one study as being at low risk (WARCEF 2012), and one study as being at unclear risk of blinding of participants and personnel (HELAS 2006).

We considered HELAS 2006 to be at unclear risk of detection bias in relation to the blinding of outcome assessors, while the other three studies were at low risk (WARCEF 2012; WASH 2004; WATCH 2009).

Incomplete outcome data

Two studies were also at unclear risk of incomplete outcome bias (WASH 2004; WATCH 2009), while two studies were at low risk of bias in this domain (HELAS 2006; WARCEF 2012).

Selective reporting

We deemed all four included studies to be at low risk of selective reporting bias.

All‐cause mortality

Four trials involving 3663 participants reported all‐cause mortality for warfarin versus aspirin (HELAS 2006; WARCEF 2012; WASH 2004; WATCH 2009). Participants in both groups were at a similar risk of all‐cause death (219 events per 1000 for both agents, Mantel‐Haenszel RR 1.00, 95% CI 0.89 to 1.13; I² = 0%; Analysis 1.1). The quality of the evidence was moderate.

WATCH 2009 (N = 1064) reported on all‐cause mortality for warfarin versus clopidogrel, but there was no evidence that either treatment was superior to the other (RR 0.93, 95% CI 0.72 to 1.21; participants = 1064; low quality of evidence; Analysis 2.1).

Non‐fatal cardiovascular events

Four trials with 3663 participants also reported non‐fatal cardiovascular events (such as non‐fatal stroke, myocardial infarction, pulmonary embolism, peripheral arterial embolism) for warfarin versus aspirin (HELAS 2006; WARCEF 2012; WASH 2004; WATCH 2009). Treatment with warfarin was associated with a reduction in non‐fatal cardiovascular events (66 per 1000 for warfarin, 83 per 1000 for aspirin, Mantel‐Haenszel RR 0.79, 95% CI 0.63 to 1.00; I² = 0%; Analysis 1.2). The quality of the evidence was moderate.

WATCH 2009) (N = 1064) reported non‐fatal cardiovascular events for warfarin versus clopidogrel. Patients in both groups were at similar risk (46 per 1000 for warfarin, 54 per 1000 for aspirin, Mantel‐Haenszel RR 0.85, 95% CI 0.50 to 1.45; Analysis 2.2). The quality of the evidence was low.

Major bleeding events

Four trials (N = 3663) reported rates of major bleeding events (HELAS 2006; WARCEF 2012; WASH 2004; WATCH 2009). In the pooled analysis warfarin use was associated with a twofold increase in the risk of major bleeding compared to aspirin (56 per 1000 for warfarin, 28 per 1000 for aspirin, Mantel‐Haenszel RR 2.00, 95% CI 1.44 to 2.78; I² = 4%; Analysis 1.3). The quality of the evidence was moderate.

WATCH 2009 (N = 1064) reported major bleeding events for warfarin versus clopidogrel. The rate of major bleeding events was 2.5 times higher in the warfarin group (49 per 1000 for warfarin, 20 per 1000 for aspirin, Mantel‐Haenszel RR 2.47, 95% CI 1.24 to 4.91; Analysis 2.3). The quality of the evidence was low.