Intended for healthcare professionals


Primary prevention of arterial thromboembolism in non-rheumatic atrial fibrillation in primary care: randomised controlled trial comparing two intensities of coumarin with aspirin

BMJ 1999; 319 doi: (Published 09 October 1999) Cite this as: BMJ 1999;319:958
  1. B S P Hellemons, general practitioner (Bep.Hellemons{at}HAG.Unimaas.NL)a,
  2. M Langenberg, general practitionera,
  3. J Lodder, neurologistb,
  4. F Vermeer, cardiologistc,
  5. H J A Schouten, clinical biostatistitiand,
  6. Th Lemmens, general practitionera,
  7. J W van Ree, professora,
  8. J A Knottnerus, professora
  1. a Department of General Practice, University of Maastricht, P O Box 616, 6200 MD Maastricht, Netherlands
  2. b Department of Neurology, University of Maastricht
  3. c Department of Cardiology, University of Maastricht
  4. d Department of Methodology and Statistics, University of Maastricht
  1. Correspondence to: B S P Hellemons-Boode
  • Accepted 15 September 1999


Objective: To investigate the effectiveness of aspirin and coumarin in preventing thromboembolism in patients with non-rheumatic atrial fibrillation in general practice.

Design: Randomised controlled trial.

Participants: 729 patients aged ≥60 years with atrial fibrillation, recruited in general practice, who had no established indication for coumarin. Mean age was 75 years and mean follow up 2.7 years.

Setting: Primary care in the Netherlands.

Interventions: Patients eligible for standard intensity coumarin (international normalised ratio 2.5-3.5) were randomly assigned to standard anticoagulation, very low intensity coumarin (international normalised ratio 1.1-1.6), or aspirin (150 mg/day) (stratum 1). Patients ineligible for standard anticoagulation were randomly assigned to low anticoagulation or aspirin (stratum 2).

Main outcome measures: Stroke, systemic embolism, major haemorrhage, and vascular death.

Results: 108 primary events occurred (annual event rate 5.5%), including 13 major haemorrhages (0.7% a year). The hazard ratio was 0.91 (0.61 to 1.36) for low anticoagulation versus aspirin and 0.78 (0.34 to 1.81) for standard anticoagulation versus aspirin. Non-vascular death was less common in the low anticoagulation group than in the aspirin group (0.41, 0.20 to 0.82). There was no significant difference between the treatment groups in bleeding incidence. High systolic and low diastolic blood pressure and age were independent prognostic factors.

Conclusion: In a general practice population (without established indications for coumarin) neither low nor standard intensity anticoagulation is better than aspirin in preventing primary outcome events. Aspirin may therefore be the first choice in patients with atrial fibrillation in general practice.

Key messages

  • Studies have shown that patients with non-rheumatic atrial fibrillation may benefit from anticoagulation

  • This study in a general practice population found no benefit of standard or low dose anticoagulation on risk of stroke, systemic embolism, major haemorrhage, or vascular death when compared with aspirin

  • Hypertension and age were prognostic factors for event occurrence

  • Aspirin is the treatment of choice for preventing thromboembolism in primary care patients at low risk


Coumarin and aspirin have been shown to reduce the risk of thromboembolic events by 68% and 36% respectively in patients with non-rheumatic atrial fibrillation.17 However, it is unclear whether these findings apply to patients with atrial fibrillation in primary care as the patients studied were selected by referral. Referred patients are generally at higher risk of thromboembolic events and the effect of antithrombotic treatment may be greater than that in primary care patients.812

We assessed the preventive effect of very low intensity and standard intensity anticoagulation among patients in general practice who had no clear indication for coumarin. Some evidence exists that low dose anticoagulation might be as effective as standard dose but with a lower risk of bleeding.1317 Since by 1990 placebo was not considered acceptable in trials of atrial fibrillation because of the proved effectiveness of coumarin and aspirin, we compared both anticoagulant doses with aspirin (150 mg/day).2

Participants and methods

The study was conducted from January 1990 to December 1996. The 284 participating general practitioners checked the pulse of all visiting patients aged ≥60 years. In addition, patients were identified from general practitioners' and pharmacists' databases. Patients were invited to have their pulse taken if they had not contacted the general practitioner spontaneously.18


Patients aged ≥60 years with electrocardiographically confirmed chronic atrial fibrillation or intermittent atrial fibrillation (electrocardiography within past two years) were eligible. Exclusion criteria were treatable causes of atrial fibrillation, previous stroke, rheumatic valvular disease, myocardial infarction or cardiovascular surgery in past year, cardiomyopathy (left ventricular ejection fraction <40%), chronic heart failure, cardiac aneurysm, history of systemic embolism, retinal infarction, coumarin use in the past three months, contraindications for aspirin or coumarin (haemoglobin concentration <7.0 mmol/l, ventricular or duodenal ulcer in the past three years, gastrointestinal or urogenital bleeding in the past year, aspirin intolerance, coagulation disorder, and severe hepatic or renal disease), pacemaker, and a life expectancy less than two years. Exclusion criteria for standard anticoagulation were age ≥78, retinopathy, ventricular or duodenal ulcer, history of gastrointestinal or genitourinary bleeding, and diastolic blood pressure >105 mm Hg or systolic pressure >185 mm Hg, or both.

Randomisation, data management, and ethical approval

Patients eligible for standard anticoagulation were randomly assigned (centrally, by telephone) to aspirin 150 mg/day, low anticoagulation (international normalised ratio 1.1-1.6), or standard anticoagulation (international normalised ratio 2.5-3.5; randomisation stratum 1). Adaptive biased urn randomisation guaranteed similar group sizes in each practice but kept treatment assignment unpredictable.19 Patients who were ineligible for standard anticoagulation were randomised between aspirin and low anticoagulation (randomisation stratum 2), giving five subgroups in the two strata. General practitioners followed up participants at four month intervals and checked compliance. Patients gave written informed consent. Patients were single blinded for the two intensities of anticoagulant, but end point ascertainments were blinded for treatment. Events were independently reviewed by two members of the (neurological, cardiological, vascular, ophthalmological, and internal medicine) event committees (or three, in case of disagreement; with deliberation to reach consensus). Either phenprocoumon or acenocoumarol (nicoumalone) was prescribed by the thrombosis centres according to normal prescription practice and monitored at intervals of 2-6 weeks. The medical ethics committee of Maastricht university hospital approved the study.

Outcome measures

The primary outcome events were defined as follows:

Stroke—Classified as not disabling, minor disabling, or major disabling and as ischaemic or haemorrhagic (based on computed tomography).

Systemic arterial embolism—Acute vascular occlusion resulting in recovery, permanent sequelae (major or minor), or death.

Major haemorrhage—Requiring hospital admission and blood transfusion or causing fall in haemoglobin concentration ≥2.0 mmol/l.

Vascular death—Within four weeks after stroke, systemic embolism, myocardial infarction, congestive heart failure, or major bleeding or sudden death (observed within one hour after onset of symptoms or patient found dead).

Secondary outcome events were non-fatal myocardial infarction (electrocardiographically or laboratory confirmed), retinal infarction, transient ischaemic attack, minor bleeding complication, or non-vascular death.

Sample size and statistical analysis

If the treatment results are equivalent, aspirin would be the treatment of choice since anticoagulation is more inconvenient for the patient Therefore, proof that aspirin is more effective than anticoagulation is not necessary and one sided testing is ethically appropriate as it requires experimental exposure of fewer patients.

Because of constraints on resources and time, we focused the sample size requirements on the comparison of low anticoagulation and aspirin, to be applied in both strata. With a power of 80%, a one sided significance of α=0.05 and an assumed four year cumulative primary event incidence of 24% on aspirin, 310 patients in both intervention groups were needed in order to detect a cumulative reduction in incidence to 16%.20 We aimed at balanced assignment of patients in both strata, so that 155 patients were required for each intervention in each stratum; 124 primary events were expected in both groups together. Standard anticoagulation was restricted to stratum 1. A reduction in cumulative incidence from 24% to 12% could be detected with 155 patients in each intervention group.

We analysed results using an intention to treat approach with a log rank test and Cox regression analysis. In the regression analysis we adjusted for baseline differences in prognostic factors and analysed potential effect modification20; we included age, sex, chronic or intermittent character of atrial fibrillation, recent onset of atrial fibrillation, no cardiovascular comorbidity (lone atrial fibrillation), current smoking, hypercholesterolaemia, history of hypertension (blood pressure ≥160/95 mm Hg), diabetes, raised body mass index, history of ischaemic heart disease, intermittent claudication, left ventricular ejection fraction on echocardiography ≤40%, and exclusion for standard anticoagulation (randomisation in stratum 2 being a covariable).21

Interim analyses were planned after 31, 62 and 93 primary events in the low anticoagulation and aspirin groups combined (both strata) with significance levels of 0.001, 0.008 and 0.017 as boundaries for the one sided P value from the stratified log rank test (Snapinn stopping rule).22 Accordingly, the trial would be stopped when there was at least a 90% probability that the final analysis would result in a significant difference if the trial was continued. Values of 0.71, 0.34 and 0.16 were used as successive boundaries to stop the trial with a P value above the boundary, when there is at least an 80% probability that the final analysis would result in a non-significant difference.


Baseline measurements

Of 1837 identified patients (fig 1), 966 were ineligible; reasons for exclusion were no atrial fibrillation on electrocardiography (309, 32%), current use of anticoagulants (290, 30%), sustained stroke or thromboembolism (193, 20%), aspirin intolerance (17, 2%), and valvular disease (161, 17%), and others (some met more than one criterion). Patients currently using anticoagulants had been previously referred. Nineteen patients who were thought to be eligible were later found to meet exclusion criteria (pacemaker (three), no atrial fibrillation on electrocardiography (three), hyperthyroidism (two), younger than 60 (seven), recent coumarin use (one), previous stroke (one), and too old for standard anticoagulation (two)).

Of the 871 eligible patients, 142 were not randomised; 23 (16%) refused, and in 119 patients reasons were not specified by the general practitioner. These 142 patients differed slightly from the included patients: fewer had new onset atrial fibrillation (16 (11%) v 237 (35%)), history of angina (9 (6%) v 81 (11%)), and history of diabetes (12 (8%) v 121(16%)) and there was a higher proportiona of men (72 (51%) v 327 (45%)).

A total of 335 patients met exclusion criteria for standard anticoagulation only (age ≥78 years (258), history of bleeding or ulcer,33 severe hypertension,32 and retinopathy12) and were randomised in stratum 2.

Table 1 gives the baseline characteristics. No patient was lost to follow up. There were 77 withdrawals during follow up for medical reasons: intolerance of study drug (five), dementia,13 cotreatment or contraindication problem,29 hospital admission,12 and non-specified medical reasons.18 Non-medical reasons were given in 92 patients. Before entering the trial, 204 (28%) patients were taking prophylactic aspirin, which was stopped before randomisation.

Table 1.

Baseline characteristics of study patients, according to assigned treatment

View this table:

Compliance with coumarin was checked by thrombosis services. For the 3983 measurements of international normalised ratio, the mean (SD) ratio was 3.1 (1.2) for standard anticoagulation and 1.4 (0.4) forlow anticoagulation. For standard anticoagulation 48% (328) of the measurements were within the target international normalised ratio (2.5-3.5; 28% were too low and 24 % too high) and for low anticoagulation 75% (2475) were within target (1.1-1.6; 13% too low and 12% too high). A quarter of the anticoagulation patients took phenprocoumon. Pill counts in the 319 patients taking aspirin showed non-compliance in three patients.

The first and second interim analyses gave inconclusive results.


In all, 157 major or fatal events occurred, including 108 primary events (analysed on the basis of “whichever came first”): 30 in stratum 1 and 78 in stratum 2 (table 2). Figure 2 shows the Kaplan-Meier curves. After 2.5-3 years of follow up the number of patients decreased. At four years 80 patients in stratum 1 were still at risk (28 low anticoagulation, 29 standard anticoagulation, 23 aspirin) and 34 in stratum 2 (17 each for low anticoagulation and aspirin).

Table 2.

Primary outcome events according to treatment groups

View this table:
Fig 2.
Fig 2.

Kaplan-Meier survival analysis for primary outcome events (non-fatal stroke, non-fatal systemic embolism, major haemorrhage, or vascular death) according to treatment group.

The annual event rate was 5.5%. Mean follow up was 2.7 years with a total of 1939 patient years. Compared with aspirin, the hazard ratio was 0.91 (95% confidence interval 0.61 to 1.4) for low anticoagulation (both strata) and 0.78 (0.34 to 1.8) for standard anticoagulation (stratum 1). The log rank test and the likelihood ratio test gave similar P values. For non-vascular death, the hazard ratio was 0.41 (0.20 to 0.82) comparing low anticoagulation with aspirin (table 3). High systolic and low diastolic blood pressure, and age were independent prognostic factors (table 4). Risk was lower in non-smokers with intermittent atrial fibrillation receiving low anticoagulation.

Table 3.

Number (percentage per year) of outcome events according to treatment

View this table:
Table 4.

Cox regression analysis determining the effect of prognostic factors and effect modifiers on primary outcome events in all patients

View this table:

In the per protocol analysis the risk of primary events was 0.77 (0.49 to 1.2) in the low anticoagulation group compared with aspirin; this fell to 0.66 (0.44 to 0.99) when non-vascular death was included. The hazards ratio for primary events with standard anticoagulation was 0.87 (0.34 to 2.2) and 0.62 (0.26 to 1.5) when non-vascular death was included.

Adverse effects

The annual bleeding rate was 3.9%, 1.2% for major or fatal bleeding (n=23) and 2.7% for minor bleeding (n=52). Seventeen of the major bleeds occurred in stratum 2, with 10 in the aspirin group. No significant difference in risk of bleeding was found between treatment groups (tables 2 and 5).

Table 5.

Major and minor bleedings in patients taking anticoagulants or aspirin. Values in parentheses are bleeding incidences per 100 patient years unless stated otherwise

View this table:


We found a low overall event rate in this elderly primary care population with non-rheumatic atrial fibrillation. The stroke rate was 1% for patients aged <78 years and 4% for the older patients. Since we could not prove that standard or low anticoagulation was more effective than aspirin, the prophylactic choice in primary care is aspirin if there is no clear indication for full dose anticoagulation.

Other trials have found that standard anticoagulation is better than aspirin Our results may differ because we included patients with less advanced disease without an established indication for oral anticoagulants. Patients in other trials had annual stroke rates of 3.5-8% with placebo and 2.5-6% with aspirin depending on age and risk factors, which suggests more severe disease. 23 24 Our result was not explained by lower age23 or the number of serious bleeds in the aspirin group.13 In retrospect, inclusion of a placebo group might have been helpful because of the low risk of events in our study population.

Time and resources forced us to stop follow up around the time that the third interim analysis would have taken place. The number of events in the aspirin and low anticoagulation patient groups slightly exceeded the 93 required for the third interim analysis. The P value (0.37) suggests that a significant difference would not be found with longer follow up, and the stopping rule advised stopping the trial.22 Our study therefore had sufficient power to compare low anticoagulation and aspirin.

Despite randomisation there were some baseline differences between the groups. They may be partly due to use of computer generated random lists for each practice, since some practices randomised just a few patients. However, the regression analysis adjusted for these covariables, and our results should not be biased.

Recommendations for primary care

Other trials of atrial fibrillation have generally studied younger patients (mean age 69 years v 75 in our study). The recommendation to give standard anticoagulation to patients aged >75 years is based on 245 patients taking warfarin,25 whereas in the general population about 70% of patients with atrial fibrillation are aged 65-85. 12 26 Of our 390 patients aged ≥75 years, 192 took coumarin (32 standard dose). Our data on low anticoagulation concur with the findings of other studies. 27 28

The rate of bleeding in patients taking aspirin was higher than in other trials.17 Differences in outcome between the aspirin prevention trials cannot be explained by differences in dose of aspirin. 1 2 29 30 New antiplatelet agents might be an option in future.31

The observed difference in numbers of non-vascular deaths between low anticoagulation and aspirin may be related to different drug effects in patients with cancer (30% of the non-vascular deaths), who might have a hypercoagulable state. 32 33 However, the finding could also be due to chance, as such an effect was not clear in the (younger) patients taking standard anticoagulation. The observed lower risk in non-smokers with intermittent atrial fibrillation taking low anticoagulation may also be due to chance.

The proportion of measurements of international normalised ratio that were out of range in patients taking coumarin varied from 58% to 20% in other atrial fibrillation trials compared with 25% for low dose and 52% for standard dose in our study.23 These data underline the difficulties of daily coumarin use. In addition, the effectiveness of anticoagulation in daily practice may be lower than under trial conditions because monitoring of international normalised ratio may be less accurate. 23 34


We thank the late They Lemmens, general practitioner, whose inspiring ideas initiated this study; the patients and general practitioners who participated in the study; and the specialists in cardiology, neurology, and internal medicine and staff from thrombosis services of hospitals in Maastricht, Heerlen, Kerkrade, Sittard, Roermond, ([A-Z]+), Diagnostisch Centrum Eindhoven, Helmond, Tilburg, Waalwijk, Breda, Rotterdam, Utrecht, Arnhem, and Venlo/Venray. We also thank M Olieslagers-van Haaren, A Pellaers-Peeters, and J Tatipata.

Contributors: All authors participated in the conduct of the trial and contributed to writing this paper. The external monitoring committee was J van Gijn, J Tijssen, F Huygen, A M P H van den Besselaar and H J A Schouten. Members of the event committees were F Vermeer, J Lodder, R Hupperts, K Hamulyak, M R Beintema, P J E H M Kitslaar, E vd Berg, and J Boiten. H J J Wellens, P Zwietering, and T Gorgels advised on study design; M de Leeuw conducted the final analysis; Th Krebber, A van Gennip, and J Metsemakers assisted in contacts with primary care practices, data set up, and data collection; P Meyer, Ch Rousseau, Y Bergmans, I Zijlstra, W Elzinga, S De Bie, I Bloemen, D Warndorff, and M Knapen assisted in patient contacts; and A van Thoor, A Hoes and P Trienekens advised on Rotterdam contacts.


  • Funding Prevention fund (grant 002817010), Zorg Onderzoek Nederland; Roche Nicholas BV, Bladel, Holland, donated aspirin.

  • Competing interests None declared.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
View Abstract