Letters

Long term anticoagulation or antiplatelet treatment

Only warfarin has been shown to reduce stroke risk in patients with atrial fibrillation

Inclusion criteria determine results of review

Garbage in equals garbage out

Drug name was incorrect

Giving warfarin always depends on balancing risks

Patients at risk of stroke should be given warfarin

How do we decide between warfarin and aspirin?

Author's reply

Only warfarin has been shown to reduce stroke risk in patients with atrial fibrillation

EDITORThe conclusions of Taylor et al about the use of aspirin for atrial fibrillation are misleading and potentially dangerous for clinical practice.¹

Firstly, when considering anticoagulation or aspirin for the management of heart failure it is appropriate first to compare each against placebo. Overall, aspirin has no effect compared with placebo in preventing thromboembolic events or death among patients with atrial fibrillation, whereas warfarin exerts a significant reduction in both outcomes compared with placebo.²

Secondly, Taylor et al excluded a key study, stroke prevention in atrial fibrillation (SPAF) III, from their analysis.³ This study showed that full dose warfarin versus low dose warfarin combined with aspirin exerted a significantly greater reduction in stroke (1.7% v 5.6%; P=0.0007) with a trend to reduced total mortality (5.9% v 7.2%).³

Thirdly, the BMJ has previously published the mortality data from the aspirin (28 deaths) and placebo (30 deaths) arms of the AFASAK study, allowing mortality in the warfarin arm (13 deaths) to be calculated.⁴ Thus, the all cause mortality data from AFASAK is available contrary to the assertions of Taylor et al. Only warfarin has been shown to reduce the risk of stroke in atrial fibrillation, and only warfarin seems to reduce mortality in patients with atrial fibrillation. If the risk associated with atrial fibrillation is considered low enough to warrant treatment with aspirin then there is insufficient evidence to recommend any antithrombotic treatment at all.

John G F Cleland, foundation chair in cardiology. 
Gerry C Kaye, consultant cardiologist. 
School of Medicine, Academic Department of Cardiology, University of Hull, Hull HU16 5JQ

Competing interests: None declared.

1.	Taylor FC, Cohen H, Ebrahim S. Systematic review of long term anticoagulation or antiplatelet treatment in patients with non-rheumatic atrical fibrillation. BMJ 2001; 322: 321-326[Abstract/Free Full Text]. (10 February.)
2.	Cleland JGF, Cowburn PJ, Falk RH. Should all patients with atrial fibrillation receive warfarin? Evidence from randomised clinical trials. Eur Heart J 1996; 17: 674-681[Free Full Text].
3.	McBride R. Adjusted-dose warfarin versus low-intensity, fixed doses warfarin plus aspirin for high-risk patients with atrial fibrillation: stroke prevention in atrial fibrillation III randomised clinical trial. Lancet 1996; 348: 633-638[CrossRef][Medline].
4.	The Antiplatelet Trialists' Collaboration Collaborative overview of randomised trials of antiplatelet therapy1: Prevention of death, myocardial infarction, and stroke by prolonged antiplatelet therapy in various categories of patients. BMJ 1994;308:81-106.

Inclusion criteria determine results of review

EDITORWe agree with Taylor et al that questions remain unanswered over the relative benefits of anticoagulation and antiplatelet treatment for non-rheumatic atrial fibrillation.¹ Uncertainty remains over the optimum treatment of elderly patients, who were underrepresented in the anticoagulation trials, and in whom pathophysiological reasons and empirical evidence suggest higher risk of haemorrhage on warfarin.² There is also uncertainty over the generalisability of the trials to primary care.

Our community based Birmingham atrial fibrillation treatment of the aged (BAFTA) study, which is funded by the Medical Research Council, will randomise 1240 patients aged 75 or over to warfarin (target international normalised ratio 2.5) or aspirin (75 mg) and follow them up for an average of three years to address these issues.

The way the review has been conducted has, however, led to conclusions that overstate the case against warfarin. Two trials that included direct comparisons between warfarin with aspirin were excluded. The European atrial fibrillation study (EAFT) was excluded because it was difficult to interpret, although data for the 455 patients who were randomised to anticoagulation or aspirin are available on the Cochrane Library and in the Lancet paper that is cited.³ The stroke prevention in atrial fibrillation (SPAF) III study, which was not cited, will have been excluded because it evaluated combined use of anticoagulation with antiplatelet drugs.⁴ Both these studies found significant benefits of warfarin in adjusted dosage over aspirin (in combination with fixed low dose warfarin in SPAF III).

Excluding these trials will have affected the results. A systematic review published in 1999 that included the same trials as Taylor et al, but also included EAFT (but not SPAF III) reported a relative risk reduction of 36% (95% confidence interval 14-52%) for stroke (ischaemic or haemorrhagic) for patients on warfarin as compared to aspirin.⁵ Thus, inclusion criteria can have a substantial impact on the results of a systematic review. This creates a problem where the eligible studies are well known (as in this case) and the review is planned after the results are available, since the impact of different criteria can be predicted in advance.

With hindsight, it is difficult to say what the "correct" inclusion criteria should be, but where important studies are left out, these should be highlighted, since their results may influence how people choose to interpret the results of the review. Until more data are available from prospective randomised trials such as BAFTA, we would advocate caution in denying anticoagulation to patients with atrial fibrillation who are at high risk.

Jonathan Mant, senior lecturer. 
David Fitzmaurice, senior lecturer. 
Ellen Murray, research fellow. 
Department of Primary Care and General Practice, Division of Primary Care, Public and Occupational Health, University of Birmingham B15 2TT

Gregory Y H Lip, reader in medicine. 
University Department of Medicine, City Hospital, Birmingham B18 7QH, UK

F D Richard Hobbs, professor. 
Department of Primary Care and General Practice, Division of Primary Care, Public and Occupational Health, University of Birmingham B15 2TT

Competing interests: None declared.

Garbage in equals garbage out

EDITORThe systematic review by Taylor et al illustrates how the usefulness of a meta-analysis is limited by the quality of the studies included in the analysis.¹ They say that they assessed the quality of the reviewed trials based on the level of concealment of random allocation, degree of blinding used, and losses to follow up. This is not good enough. These criteria have more to do with the statistical validity of the trials than the equally important issue of their clinical validity. A critical aspect is whether the trials approximated clinical practice with regard to the characteristics of patients with non-rheumatic atrial fibrillation.

In this case, the decision to include the flawed PATAF study severely weakens any findings on meta-analysis.² The PATAF study has been extensively criticised on numerous groundsfor including a high proportion of low risk patients with lone atrial fibrillation, excluding patients with chronic heart failure, and arbitrarily excluding all patients aged 78 years or older from the standard anticoagulation arm of the study, and for a lack of statistical power.³ Furthermore, the PATAF study had a high dropout rate, ranging from 20% to 32% for the three treatment armsa fact that was not included by Taylor et al in their table 1. Clinicians should be wary of applying the implications drawn from the results of this imperfect meta-analysis to the care of their patients with atrial fibrillation.

Gregory Peterson, professor of pharmacy. 
G.Peterson{at}utas.edu.au

Shane Jackson, PhD student. 
School of Pharmacy, Faculty of Health Science, University of Tasmania, GPO Box 252-26, Hobart, Australia 7001

Competing interests: None declared.

Drug name was incorrect

EDITORWith reference to the article by Taylor et al, the drug used in the Studio Italiano Fibrillazione Atriale (SIFA) II trial is indobufen and not indoprofen.¹ Indobufen is a reversible cyclo-oxygenase inhibitor. Currently, a large trial is comparing indobufen with aspirin in primary and secondary prevention in patients with non-rheumatic atrial fibrillation.

Eduardo Stragliotto, head, cardiovascular therapeutics. 
Medical Department, Pharmacia, 20152 Milan, Italy eduardo.stragliotto{at}eu.pnu.com

Giving warfarin always depends on balancing risks

EDITORAs the originators of the AFASAK 1 and 2 trials we are happy that the field of anticoagulant preventive treatment in atrial fibrillation still attracts as much attention as shown by Taylor et al.¹ In the 12 years elapsed since AFASAK 1 was published we have spent most of our professional careers teaching and harassing the medical community that the margin between benefit and risk of anticoagulation treatment may be uncomfortably narrow² and that therefore the final, individual decision on whether to give warfarin or not always depends on balancing the expected reduction in stroke risk against the expected risk of bleeding.

This point of view has always been advocated in the Danish guidelines for anticoagulant treatment in cardiology. The aspirin dose in the AFASAK 1 study was 75 mg. This dose has been shown to be as effective as higher doses. The risk reduction with aspirin in the AFASAK 1 study of 18% (non-significant) is comparable to the effect obtained by aspirin in studies of secondary stroke prevention.

The atrial fibrillation investigator's studies included a true head to head analysis and reached very reliable conclusions. ^{3 4} Some critics have expressed doubt about the effects of aspirin in this setting at all.⁵

John Godtfredsen, chief of neurology. 
Herlev University Hospital, DK-2730 Herlev, Denmark

Gudrun Boysen, professor of neurology. 
Bispebjeg University Hospital, DK-2400 Copenhagen, Denmark

Palle Petersen, director. 
Division of Neurological Rehabilitation, Hvidovre University Hospital, DK-2650 Hvidovre, Denmark

Competing interests: None declared.

Patients at risk of stroke should be given warfarin

EDITORTaylor et al have produced a thorough analysis of head to head studies of the relative benefits and risks of anticoagulation and antiplatelet agents.¹ We believe, however, that their conclusions are influenced by their own hypotheses, potentially endangering patients who would benefit from warfarin at adjusted dosage.

The main reason to give anticoagulants to patients with atrial fibrillation is not to increase life expectancy; it is to prevent stroke. As several guidelines suggest, it should not be normal practice to treat all patients with atrial fibrillation with long term warfarin.^2-4 Patients at low risk are better served with aspirin. Despite the inclusion of a substantial proportion of such patients, they still show a significant benefit in favour of anticoagulation for stroke prevention. Taylor et al dismiss this as modest but then highlight a non-significant increase in major bleeding as an important harm. Reasons could be postulated for the exclusion of many of the trials, and not just for the one that weakens their argument. Taylor et al also raise the question of cost of anticoagulation services, while not mentioning the large hospital, community, and social costs of stroke, particularly as the large cortical infarcts associated with atrial fibrillation tend to be particularly severe and disabling.⁵

When we see patients with atrial fibrillation, we assess their risk of stroke and of bleeding, clinically and by the judicious use of echocardiography. We explore the potential for cardioversion, ablation or surgical treatment. If their risk of stroke is high, we would still advise them to take warfarin.

Andrew Evans, clinical lecturer. 
andy.evans{at}kcl.ac.uk

Lalit Kalra, professor of stroke medicine. 
Guy's, King's and St Thomas' School of Medicine, London SE5 9PJ

Competing interests: None declared.

How do we decide between warfarin and aspirin?

EDITORTaylor et al think that there is little to choose between the two treatment options for patients with atrial fibrillationantiplatelet treatment and anticoagulationexcept cost.¹ Different conclusions are drawn from analysis of a systematic review of a very similar data set recently published in the Cochrane Library.² The reviewers conclude that the evidence strongly supports the use of warfarin in atrial fibrillation for patients at average or greater risk of stroke, although there is a risk of haemorrhage. Although not definitively supported by the evidence, aspirin may prove to be useful for stroke prevention in subgroups with a low risk of stroke, with less risk of haemorrhage than with warfarin.

In these reviews it has not been possible to prove beyond reasonable doubt that aspirin is more efficacious than placebo or less efficacious than anticoagulant drugs. The disadvantages of using a 5% significance level to decide if we can be sure about results was highlighted earlier this year in the BMJ.³ Non-significant trends are open to subjective interpretation when results are handled dichotomously in this way. Moreover, aspirin is certainly more convenient than anticoagulation, but the cost argument employed by Taylor et al is flawed as the costs of caring for patients with stroke (or those with major bleeds) has not been considered.⁴

The directions of the differences found in trials randomising patients to warfarin or aspirin are the same as those found in the placebo controlled trials. If non-fatal strokes are compared with major bleeds the pooled odds ratios are almost reciprocal from the meta-analysis of the head to head trials. In practice therefore the trade off for individual patients depends on their assessed risk of having a stroke or a major bleed. In most trials included non-fatal strokes are roughly twice as common as bleeds, and therefore since both outcomes are rare the odds ratio behaves like a risk ratio. This means that in comparison with antiplatelet treatment, if 100 such patients are given anticoagulation for two years, roughly two non-fatal strokes will be prevented and one extra major bleed will occur.

In practice, the decision to prescribe anticoagulation or antiplatelet treatment therefore needs to be individually assessed and discussed with each patient. Some may well choose aspirin, but this needs to be on the basis of the risks that they face of having a stroke or bleeding, not on whether the pooled results of a meta-analysis reach 5% significance.

Christopher Cates, general practitioner. 
Bushey Health Centre, Bushey, Hertfordshire WD23 2NN chriscates{at}emailmsn.com

Competing interests: None declared.

1.	Taylor FC, Cohen H, Ebrahim S. Systematic review of long term anticoagulation or antiplatelet treatment in patients with non-rheumatic atrial fibrillation. BMJ 2001; 322: 321-326. (10 February.)
2.	Segal JB, McNamara RL, Miller MR, Powe NR, Goodman SN, Robinson KA, et al. Anticoagulants or antiplatelet therapy for non-rheumatic atrial fibrillation and flutter. In: Cochrane Collaboration,ed. Cochrane Library. Issue 1. Oxford: Update Software, 2001.
3.	Sterne JA, Smith GD. Sifting the evidencewhat's wrong with significance tests? BMJ 2001; 322: 226-231[Free Full Text]
4.	Cates CJ. Care is required with cost effectiveness approach. BMJ 2000; 321: 449[Free Full Text].

Author's reply

EDITORCleland and Kaye consider antiplatelet drugs ineffective in atrial fibrillation, citing a non-systematic review in support, although a recent Cochrane systematic review shows efficacy.¹ Direct head to head comparisons are the only way of making unbiased evaluations of efficacy, as direct comparisons are of limited value. Total mortality data for the warfarin arm of the AFASAK 1 trial can be derived from the published data on the aspirin and control arms, but the resulta significant reduction in total mortality in the warfarin armis inconsistent with the statement in the primary publication that an intention to treat analysis showed no difference in either vascular or total mortality.²

The BAFTA trial should be helpful, although it may be speculated that the choice of a low aspirin dose (75 mg, similar to that used in AFASAK 1²) may bias towards anticoagulation. The target sample size of 1240 participants is disappointingly small. About 5000 patients would be needed adequately to power a trial to detect a 25% advantage in fatal cardiovascular events of warfarin over aspirin. Peterson and Jackson confuse the internal validity of a trial with its generalisabilitywhether the patients randomised are representative of those seen in clinical practice. None of the trials included in our meta-analysis, including PATAF,³ were adequately powered.

PATAF losses to follow up were zero, as stated in our paper, which is the relevant trial quality indicator and not withdrawals from treatment. In British anticoagulation clinics, the majority of patients with atrial fibrillation have no past history of thromboembolism and are being treated with adjusted dose warfarin. Combined fixed low dose warfarin with aspirin, evaluated in SPAF-III, is seldom used and is not relevant to the question of adjusted dose anticoagulation versus antiplatelet drugs. The pooled relative risk of combined fatal and non-fatal outcomes in those trials studying predominantly patients without a history of transient ischaemic attacks or stroke was 0.74 (95% confidence interval 0.52 to 1.07, random effects, significant heterogeneity), although this falls to 0.85 (0.68 to 1.05, fixed effects, no heterogeneity) if the lower quality AFASAK 1 trial is excluded. Similar treatment effects that do not achieve significance are seen for non-fatal stroke with and without the inclusion of AFASAK 1 trial0.74 (0.50 to 1.10) and 0.85 (0.56 to 1.30) respectively. The wide confidence intervals suggest that the evidence is consistent with anticoagulation halving non-fatal strokes or increasing them by a third. Anticoagulants may be more effective than antiplatelet drugs in secondary prevention. The primary publication of the European atrial fibrillation trial (EAFT)⁴ did not present data allowing direct comparison of patients randomised to aspirin and warfarin, which resulted in its exclusion in our review. No details of specific non-fatal and fatal outcomes have been provided in an outdated Cochrane review of this trial.⁵ Both the EAFT and the Studio Italiano Fibrillazione Atriale (SIFA) trial⁶ were concerned with secondary prevention and for combined fatal and non-fatal outcomes, their pooled relative risk is 0.72 (0.56 to 0.93), although the findings of each trial are different.

All of us would wish to avoid a debilitating non-fatal stroke, but ascertainment of non-fatal strokes, particularly in non-blinded trials, may be difficult and potentially biased, hence our preference for fatal vascular events as the main outcome, as these are easier to count accurately and without bias. Godtfredsen et al believe that the margins of benefit and risk are narrow but ignore the options of improving our very imprecise estimates of these benefits and risks by organising bigger adequately powered trials,⁷ of asking about patients' treatment preferences, and considering the costs of treatment in their approach to decision making.

Fiona Taylor, systematic review training fellow. 
Bristol Heart Institute, University of Bristol, Bristol Royal Infirmary, Bristol BS2 8HW

Hannah Cohen, consultant in harmatology. 
Department of Haematology, University College London Hospitals, London WC1E 6DB

Shah Ebrahim, professor in epidemiology of ageing. 
Department of Social Medicine, University of Bristol, Bristol BS8 2HU

A longer version of this letter is published on bmj.com

Competing interests: None declared.