Anticoagulantes para Derrame Isquêmico Agudo

- Não alterado Anticoagulants for acute ischaemic strokeary, Issue 3, 2008. Oxford: Update Software.

A substantive amendment to this syste

Anticoagulants for acute ischaemic stroke

Gubitz G, Sandercock P, Counsell C

---

This review should be cited as: Gubitz G, Sandercock P, Counsell C. Anticoagulants for acute ischaemic stroke (Cochrane Review). In: The Cochrane Libr

matic review was last made on 02 March 2004. Cochrane reviews are regularly checked and updated if necessary.

Abstract

Background

Most ischaemic strokes are caused by blood clots blocking an artery in the brain. Clot prevention with anticoagulant therapy could have a significant impact on patient survival, disability and stroke recurrence.

Objective

The objective of this review was to assess the effect of anticoagulant therapy versus control in the early treatment of patients with acute ischaemic stroke.

Search strategy

We searched the Cochrane Stroke Group trials register (last searched 30 October 2003). For previous updates of this review, we searched the register of the Antithrombotic Trialists' (ATT) Collaboration, consulted MedStrategy (1995), and contacted relevant drug companies.

Selection criteria

Randomised trials comparing early anticoagulant therapy (started within two weeks of stroke onset) with control in patients with acute presumed or confirmed ischaemic stroke.

Data collection and analysis

Two reviewers independently selected trials for inclusion, assessed trial quality and extracted the data.

Main results

Twenty-two trials involving 23,547 patients were included. The quality of the trials varied considerably. The anticoagulants tested were standard unfractionated heparin, low-molecular-weight heparins, heparinoids, oral anticoagulants, and thrombin inhibitors. Based on nine trials (22,570 patients) there was no evidence that anticoagulant therapy reduced the odds of death from all causes (odds ratio (OR) = 1.05, 95% confidence interval (CI) 0.98 to 1.12) at the end of follow-up. Similarly, based on six trials (21,966 patients), there was no evidence that anticoagulants reduced the odds of being dead or dependent at the end of follow-up (OR = 0.99; 95% CI 0.93 to 1.04). Although anticoagulant therapy was associated with about 9 fewer recurrent ischaemic strokes per 1000 patients treated (OR = 0.76; 95% CI 0.65 to 0.88), it was also associated with a similar sized 9 per 1000 increase in symptomatic intracranial haemorrhages (OR = 2.52; 95% CI 1.92 to 3.30). Similarly, anticoagulants avoided about 4 pulmonary emboli per 1000 (OR = 0.60, 95% CI 0.44 to 0.81), but this benefit was offset by an extra 9 major extracranial haemorrhages per 1000 (OR = 2.99; 95% CI 2.24 to 3.99). Sensitivity analyses did not identify a particular type of anticoagulant regimen or patient characteristic associated with net benefit.

Reviewers' conclusions

Immediate anticoagulant therapy in patients with acute ischaemic stroke is not associated with net short- or long-term benefit. The data from this review do not support the routine use of any type of anticoagulant in acute ischaemic stroke. People treated with anticoagulants had less chance of developing deep vein thrombosis (DVT) and pulmonary embolism (PE) following their stroke, but these sorts of blood clots are not very common, and may be prevented in other ways.

---

Synopsis

Anticoagulants (blood thinners) are of no net benefit for the early treatment of strokes caused by blood clots.

Millions of people around the world have strokes every year. Most strokes take place when a blood clot forms or gets stuck in a blood vessel leading to the brain. Without a proper blood supply, the brain quickly suffers damage, which can be permanent. The damage from a stroke can cause arm or leg weakness, or difficulties with language or vision. Strokes are sometimes fatal, but will more often leave the survivor unable to do the things that they used to do. Because strokes are common and cause such damage, researchers are trying to find ways to get rid of the blood clot soon after the stroke happens. One way to do this is with blood thinning drugs called 'anticoagulants'. If anticoagulants work, the bad effects of the stroke might be avoided. The main problem with anticoagulants is that they can cause bleeding, which can sometimes be very serious. This systematic review was designed to find out whether people treated with anticoagulants early after stroke got better or not, and whether they had problems with bleeding. There is a lot of information in this systematic review - about 24,000 people with stroke have been involved in research studies to answer this question. The systematic review did not provide any evidence that the early use of anticoagulants benefits people with strokes caused by blood clots. Anticoagulants did not reduce disability, and caused more bleeding. People treated with anticoagulants had less chance of developing blood clots in their legs and in their lungs following their stroke, but these sorts of blood clots are not very common, and may be prevented in other ways.

Background

There are approximately 4.6 million deaths from ischaemic stroke each year world-wide, with about 75% of these occurring in developing countries (WDR 1993). However, this may be an underestimate, as these data are based on extrapolation from total mortality rates, and death certificate data, which are only available for about one third of the world's population (Bonita 1995). Stroke incidence is considerably more difficult to ascertain globally, but more accurate data tend to come from developed countries. There are about 600,000 new strokes each year in the European Union (EUSI 2003) and over 700,000 new strokes each year in the USA (Broderick 1998). Approximately 85% of these strokes will be ischaemic (Bamford 1990). Stroke is a leading cause of adult disability; for those who survive, between 15 to 30% will be permanently disabled, and 20% will require institutional care at three months after onset (AHA 2003).

In order to reduce the mortality and disability caused by stroke, simple and effective early treatments are needed. In theory, the immediate use of anticoagulants might reduce the volume of infarcted cerebral tissue, thereby reducing neurological deficit, and the risk of death or disability. Immediate anticoagulant therapy might also reduce the risk of early recurrent thromboembolic stroke, deep vein thrombosis and pulmonary embolus. However, anticoagulants may also increase the risk of intracranial and extracranial haemorrhage which might offset any of these benefits.

Recent evidence provided by randomised controlled clinical trials investigating the use of anticoagulants for the treatment of acute ischaemic stroke has been reflected in the international development of several new clinical practice guidelines, as well as updates of several older guidelines. Readers of this review are encouraged to seek out their own relevant national guidelines. Many of these guideline statements now reference the previous (1999) version of this Cochrane review. As a possible result of the evidence provided by the systematic review, these new and updated guidelines have provided more definitive statements regarding the use of anticoagulants for acute ischaemic stroke. For example, the Scottish Intercollegiate Guideline Network stroke guidelines (SIGN 1997) state that "no benefit has been demonstrated for heparin in reducing mortality in patients with acute ischaemic stroke". The Royal College of Physicians (UK) National Electronic Library for Health - Stroke (NHS 2002) states that, after considering the early use of aspirin, and possibly thrombolytic treatment in special circumstances, "no other drug treatment (including the use of anticoagulants) aimed at treatment of the stroke should be given unless as a part of a randomised controlled trial". The European Union Stroke Initiative (EUSI 2003) has indicated that "there is no recommendation for general use of heparin, low-molecular heparin or heparinoids after ischaemic stroke", although there remains some controversy in specific clinical circumstances, where it is suggested that "full-dose heparin may be used when there are selected indications such as atrial fibrillation, other cardiac sources with high risk of re-embolism, arterial dissection or high-grade arterial stenosis". The recently updated American Stroke Association guidelines for the treatment of acute ischaemic stroke (ASA 2002) now include a number of specific recommendations against the general use of anticoagulants for the treatment of acute ischaemic stroke. A recent study of 36 American academic institutions (ISBP 1998), found that heparin use had decreased to about 56% of stroke patients, although the range was quite broad (from 10 to 93%).

This paper is a scheduled update of the 1999 systematic review of all the randomised trials of anticoagulants versus control in patients with acute presumed or confirmed ischaemic stroke. Our aim is to establish the balance of risk and benefit of immediate anticoagulation.

Objectives

We wished to determine the effectiveness and safety of immediate anticoagulation in patients with acute presumed or confirmed ischaemic stroke. Our hypotheses were that, compared with a policy of avoiding their use, immediate anticoagulation would be:
(1) associated with a reduced risk of being dead or dependent in activities of daily living a few months after stroke onset;
(2) associated with a reduced risk of early recurrent ischaemic stroke;
(3) associated with an increased risk of symptomatic intracranial and extracranial haemorrhage; and
(4) associated with a reduced risk of deep vein thrombosis and pulmonary embolism.

Criteria for considering studies for this review

Types of studies

All unconfounded, truly randomised trials in which early treatment with anticoagulants was compared with control in patients with acute presumed or confirmed ischaemic stroke were sought. Patients with ischaemic stroke due to cerebral venous thrombosis were not specifically included in these trials, and so are not represented in this review. Trials in which allocation to treatment or control group was not truly random, or where allocation was not adequately concealed (e.g. allocation by alternation, date of birth, hospital number, day of the week, or open random number list) were not included, since foreknowledge of treatment allocation could lead to biased treatment allocation, and thereby overestimate the treatment effect by up to 30% (Schulz 1995). If it was unclear whether the method of randomisation provided adequate concealment of allocation, the trial was included.

Types of participants

This systematic review was confined to the early treatment of acute ischaemic stroke, and therefore excluded trials: which randomised patients more than 14 days after stroke onset; in patients with transient ischaemic attacks only; which only included patients with intracerebral haemorrhage, confirmed by computed tomography (CT) or magnetic resonance (MRI) scanning before entry. Trials in which the pathological type of stroke was not confirmed by scanning before entry were included, as the majority of such strokes are ischaemic, at least in Caucasians (Bamford 1990).

Types of intervention

Anticoagulants are broadly defined as those which act on the coagulation cascade to exert an anticoagulant effect, excluding thrombolytic agents (e.g. streptokinase) and defibrinogenating agents (e.g. ancrod). The use of thrombolytic agents in acute ischaemic stroke is the subject of a separate Cochrane review (Wardlaw 2003), as is the use of fibrinogen depleting agents (Liu 2003).

The following anticoagulants are included in this review: subcutaneous and intravenous standard unfractionated heparin, low-molecular-weight heparins, subcutaneous and intravenous heparinoids, oral anticoagulants and specific thrombin inhibitors.

This review includes all trials which assessed the effects of anticoagulants on any clinical outcome, provided that treatment was started within 14 days of the initial stroke. Therefore, the included trials aimed to assess different clinical problems, including the treatment of the acute ischaemic stroke, the prevention of new arterial or intracranial venous thrombus manifesting as stroke progression, early recurrent ischaemic stroke, deep vein thrombosis or pulmonary embolism.

Types of outcome measures

For each trial, the number of patients originally randomly allocated to each treatment and control group was identified. In both groups, outcome information was sought regarding the number of patients:
(1) who were either dead, or dependent for help from other people for their activities of daily living, at least one month after their stroke. This minimum interval was used to allow time for recovery from the initial stroke. Comparable definitions of dependency were used in all of the trials assessed in this review;
(2) who died from any cause during the scheduled treatment period (generally shorter than the scheduled follow-up period);
(3) who died from any cause during the scheduled follow-up period (greater than one month after their stroke);
(4) with objective evidence of deep vein thrombosis detected by the systematic use of imaging techniques such as iodine 125 fibrinogen scanning (I-125 scan), ultrasound of the leg, plethysmography, or X-ray contrast venography in all patients during the scheduled treatment period and during scheduled follow-up. These methods therefore detected clinically silent deep vein thrombosis as well as confirming or refuting the diagnosis in patients with clinical features suggestive of deep vein thrombosis. The outcome was therefore 'symptomatic or asymptomatic deep vein thrombosis'. Screening of patients by clinical observation alone was not considered adequate;
(5) with at least one confirmed symptomatic pulmonary embolus diagnosed during life, or at autopsy (symptomatic or not) within the scheduled treatment period and during scheduled follow-up;
(6) with recurrent stroke during the treatment period and during follow-up which was either definitely ischaemic (haemorrhage excluded by CT scan or autopsy), or of unknown type (no CT scan or autopsy performed);
(7) with symptomatic intracranial (intra and extracerebral) haemorrhage, including symptomatic haemorrhagic transformation of the cerebral infarct, during the scheduled treatment period and during follow-up. The haemorrhage must have been confirmed by CT (or MRI) scanning after clinical deterioration, or by autopsy;
(8) with any recurrent stroke or symptomatic intracranial haemorrhage during the treatment period or during long-term follow-up (as previously defined);
(9) with any major extracranial haemorrhage during the scheduled treatment period. The definition of major haemorrhage was usually taken from the original article but if none was given it was defined as any fatal bleed, or bleeding severe enough to require transfusion or operation.

Search strategy for identification of studies

See: Cochrane Stroke Group search strategy

See: methods used in reviews.

This review has drawn on the search strategy developed for the Cochrane Stroke Group as a whole. Relevant trials were identified in Group's trials register. The register was last searched by the Review Group Co-ordinator for relevant trials on 30 October 2003.

Three additional methods were used in the 1997 and 1999 updates of this systematic review:

(1) The following anticoagulant manufacturers were contacted to identify any unpublished trials.

Alfa Wasserman (parnaparin and dermatan sulphate)

Kabi (dalteparin)

Knoll (reviparin)

Leo (tinzaparin)

Mediolanum (dermatan sulphate)

Mitsubishi Chemical (argatroban/MD-805)

Novo (tinzaparin)

Organon (danaparoid)

Rhone-Poulenc Rorer (enoxaparin)

Sandoz (Sandoz LMWH)

Sanofi Winthrop (nadroparin and CY 222)

(2) A comprehensive guide to pharmaceutical development in the field of stroke (MedStrategy 1995) was consulted in the 1999 update of this review, but has not been searched since that time, as the data contained within it are also known to be contained within the trials register of the Cochrane Stroke Group.

(3) A search of the trials register held by the Antithrombotic Trialists' (ATT) Collaboration in August 1998, but not since that time, as the data contained within it are also known to be contained within the trials register of the Cochrane Stroke Group.

Methods of the review

A pair of reviewers (PS & CC for the trials included in the first version of this review; PS & GG for updates since then) independently selected which trials to include in the review. Disagreements were resolved by discussion. The methodological quality of each trial was assessed by the same reviewers. A scoring system was not used to assess trial quality, but we simply recorded details of randomisation method, blinding, whether intention-to-treat analyses were possible from the published data (i.e. whether there were any exclusions from the trial after randomisation) and whether any patients were lost to follow-up. Data were independently extracted by two reviewers and cross-checked. Data on the number of patients with each outcome event, by allocated treatment group, irrespective of compliance, and whether or not the patient was subsequently deemed ineligible or otherwise excluded from treatment or follow-up, were sought to allow an 'intention-to-treat' analysis. Data on the use of CT or MRI scanning prior to randomisation, the delay from stroke onset to trial entry, the type of patients included, and the type of anticoagulant regimen used were also sought. If any of the above data were not available in the publications, further information was sought by correspondence with the trialists.

The results reported in the text are odds ratios (i.e. the ratio of the odds of an unfavourable outcome among treatment-allocated patients to the corresponding odds amongst controls) which were calculated using the Peto fixed-effect method (APT 1994). The significance of any differences between odds ratios (in relation to subgroup analyses) were calculated using a standard method (Altman 1996). For some outcomes (e.g. deep vein thrombosis and any intracranial haemorrhage), intention-to-treat analyses were not possible because all patients did not have the relevant investigation performed to detect the event. In these analyses, the number of patients in each group who had the appropriate investigation was used as the denominator for the main analyses. However, best and worst case scenarios were also analysed: the 'best case' scenario (with regards to treatment) assumed that none of the patients excluded from the analysis in the treatment group had an adverse outcome, whilst all those excluded from the control group did, and vice versa for the 'worst case' analysis. Where indicated, absolute outcome events (i.e. the number of events avoided for every 1000 patients treated) were calculated using risk reduction statistics. Heterogeneity between trial results was tested for using a standard chi-squared test.

In the analyses, trials of each type of anticoagulant (e.g. unfractionated heparin, low-molecular-weight heparins, heparinoids, oral anticoagulants, thrombin inhibitors) were grouped together to assess whether there were any significant differences between classes of anticoagulant agent. It should be noted that this was an INDIRECT rather than a direct randomised comparison.

We specified the following classification of anticoagulant dosing regimens.
(1) 'Low fixed dose' anticoagulant, i.e. a dose intended to be sufficient for the prevention of deep vein thrombosis and pulmonary embolism.
(2) 'Medium fixed dose' anticoagulant, i.e. a dose intended to have effects on the arterial circulation, but not enough to require monitoring.
(3) 'Adjusted dose' anticoagulant, i.e. a dose adjusted by blood testing or body weight to meet a specific target.

The following sensitivity analyses were pre-specified for the review.

Trials in which the method of randomisation ensured adequate concealment of treatment allocation.

The type of anticoagulant agent used.

Trials in which all patients had intracerebral haemorrhage excluded by CT or MR scanning prior to trial entry, since the risk from anticoagulation may be higher in patients with intracerebral haemorrhages.

Several additional post-hoc sensitivity analyses were also evaluated where possible.

The time from stroke onset to randomisation (i.e. less than 48 hours, more than 48 hours).

Trials in which treatment with antiplatelets were used in addition to anticoagulants (in unconfounded studies, i.e. anticoagulant plus antiplatelet versus antiplatelet alone).

Trials which evaluated different anticoagulant dosages; 'high dose' (adjusted iv plus fixed medium dose) versus 'low dose'.

Trials in which the stroke was of suspected cardioembolic origin versus non-cardioembolic.

With and without the IST, as this trial contains most of the data for the review (IST 1997).

We have performed many analyses, and so by chance alone, about one in 20 will reach statistical significance at the p <>

Description of studies

Twenty-two trials with a total of 23,547 patients were included in this review. Summary details of the trials are given in the 'Characteristics of included studies' section. Nineteen trials were excluded for a variety of reasons (see 'Characteristics of excluded studies'). Two trials are awaiting further assessment before inclusion in this review. One involves a comparison of the direct thrombin inhibitor argatroban with placebo in about 180 patients (ARGIS-1); the other compares the low molecular weight heparin dalteparin with placebo in 143 patients (Dan 2000).

The age of patients in the included studies ranged from 28 to 92 years. A significant proportion of patients were over 70 years old. For example, 61% of patients enrolled in the IST (IST 1997) were aged 70 or older. Most trials included slightly more males than females.

Most trials excluded patients thought to be at high risk of bleeding (e.g. clotting disorders, hepatic or renal failure). In addition, ten trials excluded patients with significant degrees of hypertension (generally diastolic pressures greater than 120 mmHg or systolic pressures greater than 180 mmHg), and nine trials excluded comatose patients.

Thirteen trials enrolled patients within 48 hours of stroke onset, three within 72 hours, three within seven days and three within 14 days. In the latter two trials, patients could also be entered after two weeks but in one of these the authors provided data on only those entered within 14 days (Pambianco 1995) and in the other all except four of those with a presumed embolic stroke were entered within one week, and so all those with embolic stroke were included in this review (NAT-COOP 1962).

The scheduled period of anticoagulant treatment in the included trials was one to two weeks in 17 trials and one month in four. The anticoagulants used were:

standard unfractionated subcutaneous heparin (six trials);

standard unfractionated intravenous heparin (two trials);

low-molecular-weight heparins (seven trials - two dalteparin, two nadroparin, one tinzaparin, one fraxiparine and one CY 222);

subcutaneous heparinoid (two trials - one danaparoid and one mesoglycan);

intravenous heparinoid (one danaparoid trial);

oral anticoagulants (two trials); and

thrombin inhibitors (two MD805 trials).

In the trials using oral anticoagulants (Marshall 1960; NAT-COOP 1962), heparin was given intravenously for the first few days to provide immediate anticoagulation. Three trials (FISS 1995; FISS-bis 1998; IST 1997) randomised between two doses of anticoagulant as well as control; for the main analyses in this review we combined the two anticoagulant groups for these trials.

Fourteen trials routinely performed a CT head scan in all patients to rule out haemorrhage before randomisation (Cazzato 1989; CESG 1983; Duke 1986; Elias 1990; FISS 1995; FISS-bis 1998; Kwiecinski 1995; Pambianco 1995; Prins 1989; Sandset 1990; Tazaki 1986; Tazaki 1992; TOAST 1998; Turpie 1987). Three trials performed CT in most patients: (Duke 1983; IST 1997; Vissinger 1995). Eighty-one per cent of patients in the Duke trial were scanned. In the IST, 67% were scanned before randomisation, and 29% after randomisation, so that overall, 96% of patients were scanned. In the Vissinger trial, 66% of patients were scanned, and the remainder had cerebral scintigraphy to exclude haemorrhage. Three trials performed almost no CT scans (McCarthy 1977; McCarthy 1986; Pince 1981), and two trials were undertaken before CT scanning was introduced (Marshall 1960; NAT-COOP 1962). It is therefore likely that some patients with intracerebral haemorrhage were inadvertently included in the main analyses of this review. This may have biased the results against anticoagulation if the risks of anticoagulation are greater in those with intracerebral haemorrhage, although such a bias is unlikely given the relatively small numbers of patients with intracerebral haemorrhage involved in these trials, and because the IST (IST 1997) provided well over 80% of the overall data. Nevertheless, a sensitivity analysis was performed based on only those trials where all patients had CT scanning prior to randomisation.

Three trials included only patients with presumed cardioembolic stroke (CESG 1983; NAT-COOP 1962; TOAST 1998), and one trial evaluated subsets of patients with atrial fibrillation (IST 1997).

The duration of follow-up in the trials was generally short, although this was mainly a characteristic of the smaller trials, which contributed less to the overall analysis. Four trials in which the primary outcome of interest was deep vein thrombosis did not follow the patients beyond 14 days (Elias 1990; McCarthy 1977; Pince 1981; Prins 1989), and only nine trials followed patients for longer than one month (Duke 1986; FISS 1995; FISS-bis 1998; IST 1997; Kwiecinski 1995; Marshall 1960; McCarthy 1986; TOAST 1998; Turpie 1987). This lack of long-term follow-up is a weakness of many of the smaller studies, as a significant proportion of deaths after one month could have been due to stroke related thromboembolic events and might therefore have been prevented by early anticoagulation. Similarly, disability is best assessed when most of the recovery has taken place (i.e. between 3 to 6 months), rather than in the first week or so.

Relatively few trials assessed the clinically most important outcome of long-term functional status. Treatments that prevent death from stroke may lead to survival in a disabled state - an outcome considered by many to be worse than death. The composite outcome of 'dead or dependent at follow-up' is therefore the most important outcome in acute stroke trials. Eight trials assessed this combined outcome, but only six trials reported their results. These trials (Cazzato 1989, FISS 1995, FISS-bis 1998, IST 1997, Kwiecinski 1995; TOAST 1998) contain data from well over 90% of the patients included in this review, and evaluated the outcomes of death and dependency adequately. Other important outcomes, including recurrent stroke or intracranial haemorrhage, were assessed, but once again, only by the more recent trials which included large numbers of patients. Quality of life assessments were not undertaken in any of these trials.

Methodological quality

Randomisation
There was marked variation in the quality of the trials. In 13 trials, the method of randomisation adequately prevented foreknowledge of treatment allocation. One trial (IST 1997) used a central telephone randomisation service. One trial (TOAST 1998) used permuted blocks to generate a randomisation list controlled by the hospital pharmacy. Eight trials utilised numbered or coded containers administered sequentially to enrolled participants (FISS 1995; FISS-bis 1998; Prins 1989; Sandset 1990; Tazaki 1986; Tazaki 1992; Turpie 1987; Vissinger 1995). Three trials used random number tables controlled by an independent party (Cazzato 1989; Duke 1983; Duke 1986). The 2:1 treatment to control allocation ratio in the trials by Turpie et al (Turpie 1987), Tazaki (Tazaki 1986), FISS (FISS 1995) and in FISS-bis (FISS-bis 1998) was deliberate.

The method of randomisation was unclear in nine trials. Seven trials stated that sealed envelopes were used but in five of these it was unclear whether the envelopes were opaque and sequentially numbered (Elias 1990; McCarthy 1977; McCarthy 1986; NAT-COOP 1962; Pince 1981). In one trial, the envelopes were not numbered (Pambianco 1995). The exact method of randomisation was unknown in two trials (Kwiecinski 1995; Marshall 1960).

Blinding
Adequate blinding may be important to reduce bias in the detection of deep vein thrombosis, pulmonary embolism, symptomatic intracranial haemorrhage, recurrent stroke and functional outcome. Eleven trials were double-blind, i.e. treatment allocation was concealed from patients, physicians and outcome assessors (Duke 1983; Duke 1986; FISS 1995; FISS-bis 1998; Prins 1989; Sandset 1990; Tazaki 1986; Tazaki 1992; TOAST 1998; Turpie 1987; Vissinger 1995), and in two other trials (McCarthy 1977; McCarthy 1986), the assessment of deep vein thrombosis was made by radiologists blinded to treatment allocation. One study (Cazzato 1989) had a blinded outcome assessor only. The IST (IST 1997) was not designed as a blinded study. However, an analysis of 207 patients from the UK enrolled in the IST pilot study showed that at the six month follow-up, the majority of patients could not remember whether or not they had been treated, and so these patients were effectively 'blinded' (Lindley 1993). In the IST, follow-up data were collected by self-completed questionnaire mailed to the patient six months after randomisation, or by telephone interview by a person blinded to treatment allocation. The remainder of the trials did not appear to use any form of blinded assessment.

Losses to follow-up
In total, only 213 patients (0.9% overall) were reported to be excluded from analysis after randomisation or lost to follow-up, with the vast majority of patients enrolled in studies in which an intention-to-treat analysis was performed. However, a number of patients in the smaller trials which did not report an intention-to-treat analysis may have been omitted from the analysis.

None of the trials reported significant imbalances in important baseline prognostic variables, although the small size of many suggests that they might only be ruling out gross differences.

In trials with follow-up, differences in the long-term use of antiplatelet treatment between the anticoagulant and control groups after hospital discharge may have biased the results, as antiplatelet treatment has been shown to reduce the risk of further vascular events by about 25% (ATC 2002). Aspirin was given to all survivors in the FISS trial (FISS 1995). Long-term treatment with aspirin was encouraged, but optional in several other trials, including the IST (IST 1997), FISS-bis (FISS-bis 1998) and TOAST (TOAST 1998) trials.

Results

Death or dependence at final follow-up more than one month after randomisation
Six trials including randomised data from 21,966 patients (93.3% of patients included in the overall review) evaluated death and long-term disability. The degree of dependence was determined by noting whether the patients required help from other people for their activities of daily living at the time of final follow-up.

Treatment with immediate anticoagulants was not associated with a significant reduction in the odds of being dead or dependent at final follow-up (odds ratio (OR) = 0.99; 95% confidence interval (CI) 0.93 to 1.04). There was significant heterogeneity of treatment effect in the LMWH subgroup, which was chiefly attributable to one trial of 312 oriental patients (FISS 1995). There was no obvious explanation other than the play of chance, or possibly ethnic differences in the study population, although the latter would seem to be unlikely. Interestingly, a subsequent trial (FISS-bis 1998), designed to replicate the results of the FISS trial, did not show any clear benefit overall, suggesting that the FISS result was due to the play of chance. We examined whether the estimate of treatment effect differed using a fixed-effect and a random-effects model; the result was similar with both.

A number of pre-specified sensitivity analyses were performed to identify whether patients with particular characteristics or particular anticoagulant regimens were associated with a net benefit. These sensitivity analyses showed:

(1) that all of the trials evaluating death and dependence at final follow-up had adequate concealment of the randomisation process;

(2) no statistically significant difference in the effect of treatment on death or dependence at final follow-up between the different types of anticoagulant used. The majority of the patient data (89.8%) came from a single trial evaluating unfractionated heparin (IST 1997), which demonstrated no net benefit (OR = 1.00; 95% CI 0.94 to 1.06). However, data from trials of low molecular weight heparin (FISS 1995; FISS-bis 1998; Kwiecinski 1995) (OR = 0.85; 95% CI 0.66 to 1.08), and heparinoid (TOAST 1998) (OR = 0.92; 95% CI 0.72 to 1.19), also showed no clear evidence of benefit;

(3) no statistically significant difference in the effect of treatment on death or dependence at final follow-up between trials in which patients had a CT head scan before randomisation (OR = 1.01; 95% CI 0.95 to 1.08), or after randomisation or not at all (OR = 1.01; 95% CI 0.91 to 1.12). For this analysis, the results of the IST (IST 1997) were subdivided by whether CT scanning was first performed before randomisation or not.

Other post-hoc sensitivity analyses showed:

(1) no statistically significant difference in the effect of treatment on death or dependence at final follow-up between trials in which anticoagulants were added to aspirin during the treatment period (OR = 1.00; 95% CI 0.92 to 1.09), or if anticoagulants alone were given during the treatment period (OR = 0.98; 95% CI 0.91 to 1.06);

(2) that all trials evaluating death or dependence at final follow-up enrolled patients within 48 hours of stroke onset. Within the IST (IST 1997), there was no evidence that effect of treatment increased or decreased with increasing delay to randomisation up to 48 hours;

(3) no statistically significant difference in death or dependence at final follow-up between trials in which low fixed-dose anticoagulants were compared with control (OR = 1.00; 95% CI 0.93 to 1.09), or if medium fixed-dose anticoagulants or adjusted-dose anticoagulants were compared with control (OR = 0.98; 95% CI 0.91 to 1.05) during the treatment period;

(4) no apparent difference in the effect of treatment on death or dependence at final follow-up if data from the IST were included (OR = 0.99; 95% CI 0.94 to 1.05) or excluded (OR = 0.89; 95% CI 0.74 to 1.06);

(5) no statistically significant difference in the effect of treatment on death or dependence at final follow-up if a cardiac source of embolism was suspected (OR = 1.00; 95% CI 0.85 to 1.18) or not suspected (OR = 1.00; 95% CI 0.94 to 1.06) to be the cause of the stroke. Only two trials (IST 1997, TOAST 1998) reported results separately for patients randomised with suspected cardioembolic stroke.

Death from all causes during the scheduled treatment period
Data from nineteen trials including randomised data from 22,356 patients (94.9% of patients included in the review) were available for this outcome. Anticoagulants were not associated with a significant reduction in death at the end of the treatment period (OR = 0.99; 95% CI 0.90 to 1.09).

The pre-specified sensitivity analyses showed:

(1) no statistically significant difference in the effect of treatment on death during the treatment period between trials that used well concealed methods of randomisation (OR = 0.99; 95% CI 0.90 to 1.08) and those that did not (OR = 1.14; 95% CI 0.62 to 2.11);

(2) no statistically significant heterogeneity in the effect of treatment on death during the treatment period between the different types of anticoagulant used. The majority of the patient data (93%) come from a single trial evaluating unfractionated heparin (IST 1997);

(3) no statistically significant heterogeneity in the effect of treatment on death during the treatment period between trials in which patients had a CT head scan before randomisation (OR = 0.99; 95% CI 0.98 to 1.11), or after randomisation or not at all (OR = 0.98; 95% CI 0.83 to 1.16).

Other post-hoc sensitivity analyses showed:

(1) a significant excess of deaths during the treatment period in those trials in which patients could be entered after 48 hours (OR = 1.94; 95% CI 1.15 to 3.29) compared with patients randomised within 48 hours of stroke onset (OR = 0.97; 95% CI 0.88 to 1.06);

(2) no significant heterogeneity in the effect of treatment on deaths during the treatment period between high dose anticoagulant regimens designed to prevent neurological deterioration (OR = 1.02; 95% CI 0.89 to 1.16) and low dose regimens recommended for deep vein thrombosis prophylaxis (OR = 0.96; 95% CI 0.84 to 1.09);

(3) no significant heterogeneity in the effect of treatment on death during the treatment period if the data from the IST (IST 1997) were included (OR = 0.99; 95% CI 0.90 to 1.09) or excluded (OR = 1.39; 95% CI 0.97 to 1.98).

Death from all causes at follow-up greater than one month after randomisation
Data were available for nine trials including 22,570 patients (95.9% of patients included in the overall review). Anticoagulants were not associated with any significant reduction in the odds of death at final follow-up of greater than one month (OR = 1.05, 95% CI 0.98 to 1.12).

The pre-specified sensitivity analyses showed:

(1) an effect of trial quality. There was a marginally significant trend favouring anticoagulants with respect to death at final follow-up in two small trials (386 patients) using poorly concealed methods of randomisation (OR = 0.63; 95% CI 0.40 to 1.00) compared with six trials (22, 094 patients) using well concealed methods of randomisation (OR = 1.06; 95% CI 0.99 to 1.13);

(2) no statistically significant heterogeneity in the effect of treatment on death at final follow-up between the different types of anticoagulant used, with the confidence intervals for each of the different drug treatments including the possibility of either harm or benefit. The majority of the patient data (90.2%) come from a single trial evaluating unfractionated heparin (IST 1997);

(3) no statistically significant heterogeneity in the effect of treatment on death at final follow-up between trials in which patients had a CT head scan before randomisation (OR = 1.06; 95% CI 0.84 to 1.34), or after randomisation or not at all (OR = 1.04; 95% CI 0.98 to 1.12). Several smaller trials had incomplete data for this outcome but this is unlikely to have any effect on the overall result of the analysis because of the very small numbers of patients involved.

Other post-hoc sensitivity analyses showed:

(1) no statistically significant heterogeneity in the effect of treatment on death at final follow-up between trials in which all patients were randomised within 48 hours (OR = 1.05; 95% CI 0.98 to 1.12), and those in which some patients were randomised after 48 hours (OR 0.80; 95% CI 0.31 to 2.05);

(2) a significant increase in the odds of death at final follow-up compared with control in the trials which used a high dose anticoagulant (OR = 1.10; 95% CI 1.00 to 1.20). There was no significant excess of deaths in the subset of low dose regimens (OR = 1.01; 95% CI 0.92 to 1.11). However, these two estimates of treatment effect are not significantly different from each other;

(3) no significant heterogeneity in the effect of treatment on death from all causes at final follow-up if data from the IST (IST 1997) were included (OR = 1.05; 95% CI 0.98 to 1.12) or excluded (OR = 0.95; 95% CI 0.77 to 1.18).

Interestingly, the trial with the longest follow-up (Duke 1986) showed a qualitatively different result at one year, with more deaths in the anticoagulant group. There are several possible explanations for this. It may be a chance finding, as the number of randomised patients was quite small. Randomisation may have failed to prevent an important imbalance between the prognosis of the patients in the two groups (although no such imbalance was reported) or there may have been a difference in the long term treatment of patients in the two groups, with more patients in the control group receiving antiplatelets to prevent recurrent vascular disease. The use of antiplatelets was not reported.

Deep vein thrombosis during the treatment period
Ten trials including randomised data from 916 patients (only 3.9% of patients included in the overall review) sought to systematically determine the effect of anticoagulants on the occurrence of 'symptomatic or asymptomatic deep vein thrombosis' at the end of the treatment period, as detected by:

I-125 fibrinogen scanning (Duke 1983; Elias 1990; McCarthy 1977; McCarthy 1986; Pince 1981; Prins 1989; Turpie 1987),

B mode and Doppler ultrasound (Pambianco 1995) or

X-ray contrast venography (Sandset 1990; Vissinger 1995).

Despite the small numbers of patients studied, anticoagulation was associated with a highly significant reduction in the odds of deep vein thrombosis (OR = 0.21; 95% CI 0.15 to 0.29), although the majority of deep vein thromboses detected were subclinical and asymptomatic. This result was equivalent to the prevention of deep vein thrombosis in 281 patients for every 1000 patients treated (95% CI 230 to 332 deep vein thromboses prevented).

Fifteen patients (10 in the treatment group, 5 in the control) did not have an adequate assessment of deep vein thrombosis and therefore were excluded from this analysis, but even if these patients were included under a 'worst case' scenario the results did not change significantly (OR 0.23).

There was significant heterogeneity between the trials results (Chi square 31.61 (df = 9) p < href="http://cochrane.bvsalud.org/cochrane/show.php?db=reviews&mfn=21&id=&lang=pt&dblang=&lib=COC#CD000024-bbs2-0014">Pambianco 1995; Sandset 1990; Vissinger 1995), and two trials which did (Elias 1990; McCarthy 1986). The three negative trials were the only ones which did not use I-125 fibrinogen scanning. One used ultrasound assessment (Pambianco 1995); the other two used venography (Sandset 1990; Vissinger 1995). In addition, in one of these trials, patients were randomised up to 14 days after their initial stroke (Pambianco 1995) whereas the other trials all randomised patients within 7 days. The two most positive trials had very small numbers of patients, with the resultant possibility that the results may have been due to chance. In addition, the McCarthy trial (the most positive), was poorly concealed, introducing another potential source of bias (McCarthy 1986). A funnel plot analysis (Figure 01) did not suggest any asymmetry in the numbers of positive or negative trials, but did show asymmetry for the DVT outcome, suggesting that some trials may have measured DVT but not reported it, which may also have added to the heterogeneity..

Sensitivity analyses showed that there was no significant heterogeneity in the reduction in deep vein thrombosis from the above result if the analysis was restricted to trials where the concealment of allocation was secure (OR = 0.45; 95%CI 0.26 to 0.78) or to trials in which radiographic assessment was blinded (OR = 0.21; 95% CI 0.15 to 0.29). One of the trials excluded from this review (Dahan 1986) did provide data on the numbers of deep vein thromboses in the stroke patients by allocated treatment group (1/19 heparin, 3/27 placebo), but inclusion of these results did not significantly alter the analysis. No trials systematically sought deep vein thrombosis after the end of the treatment period.

Symptomatic pulmonary embolism during the treatment period
Fourteen trials, including data from 22,544 patients (95.7% of patients included in the overall review), assessed reported fatal and non-fatal symptomatic pulmonary embolism, but no trial had systematically sought asymptomatic pulmonary embolism by performing ventilation - perfusion scans in all patients at the end of the treatment period.

Anticoagulation was associated with a significant reduction in the odds of pulmonary embolism (OR = 0.60; 95% CI 0.44 to 0.81), which translated into about 4 pulmonary emboli avoided for every 1000 patients treated with anticoagulants (95% CI 1 to 6 pulmonary emboli avoided), although this is likely to be an underestimate of the reduction in risk due to incomplete ascertainment.

In the trials described, the frequency of pulmonary embolism during the treatment period was variable, but quite low (1% in IST versus 7% in ELIAS/ PRINS). Although not reported systematically, and thereby potentially under-reported, the rate of pulmonary embolism in the IST (IST 1997) in patients not receiving heparin was only 0.8%. This observation is supported by data from prospective hospital based studies which reported pulmonary embolism as a complication of acute stroke. One such study (Davenport 1996) reported pulmonary embolus as a complication in 1% of patients; a review of the literature in the same paper quotes a range of frequencies from 1% to 3%. The IST contributed 80% of the patient data to this analysis, which suggests that clinically important pulmonary embolism is an uncommon complication of acute ischaemic stroke, at least within the first two weeks.

Patients may continue to be at risk of pulmonary embolism after the early treatment period. This was suggested by data from three trials (FISS 1995; TOAST 1998; Turpie 1987) which continued to seek events systematically during the follow-up period. Eight further pulmonary emboli were recorded, with six in the control group. The potential use of antiplatelet or anticoagulant agents after the trial period may have influenced the results of several trials (FISS 1995; TOAST 1998; Turpie 1987). One trial (McCarthy 1986), with an 80% autopsy rate, did show a significant reduction in the risk of symptomatic and asymptomatic pulmonary embolism detected at autopsy in the anticoagulation group (7/24 versus 33/47, OR = 0.19; 95% CI 0.07 to 0.52).

Recurrent ischaemic stroke or recurrent stroke of unknown pathological type TYPE during the treatment period
Eleven trials including 21,605 patients (91.8% of patients included in the overall review) systematically sought to record early recurrent strokes which were definitely ischaemic (CT scan excluded haemorrhage) or probably ischaemic, i.e. in which the cerebral pathology was unknown because a CT scan had not been performed.

Anticoagulation was associated with a statistically significant reduction in recurrent ischaemic strokes (OR = 0.76; 95% CI 0.65 to 0.88), which translated into 9 additional ischaemic strokes avoided per 1000 patients treated (95% CI 4 to 13 strokes avoided). The majority of the data (95%) were obtained from one trial (IST 1997).

Symptomatic intracranial haemorrhage during the treatment period
Fifteen trials including randomised data from 22,794 patients (96.8% of patients included in the overall review) reported data on symptomatic (fatal and non-fatal) intracranial haemorrhage confirmed by CT scanning or autopsy.

Immediate anticoagulation significantly increased symptomatic intracranial haemorrhages by more than twofold (OR = 2.52; 95% CI 1.92 to 3.30), an absolute excess of 9 intracranial haemorrhages per 1000 patients treated (95% CI 6 to 11 per 1000). The majority of data (76%) were contributed by one trial (IST 1997).

There was no significant heterogeneity in the number of excess haemorrhages with different types of heparin. However, in the IST (IST 1997), intracranial haemorrhage significantly increased with increasing heparin dose. Patients allocated to avoid heparin, low-dose, and medium-dose heparin had rates of intracranial haemorrhage of 0.3%, 0.7% and 1.8% respectively. A review of the direct randomised comparisons of different doses of the same anticoagulant agent is under way (Gubitz, in preparation).

There is the possibility of some bias within these data, as there may have been a lower threshold for re-scanning patients who had deteriorated clinically if they were known to be receiving anticoagulants (for example, in the IST which was not blinded). In addition, even in blinded trials, a physician is likely to be unblinded if bruising is observed at heparin injection sites. An unbiased assessment of the effect of anticoagulants on the occurrence of intracranial haemorrhage would come from systematic studies, in which all patients undergo a CT scan before the beginning of treatment to exclude haemorrhage, and all survivors have a repeat CT scan at the end of the scheduled treatment period, regardless of their clinical status. In such an unbiased assessment, all patients who died during the study would also have to undergo an autopsy. Unfortunately, it is rarely possible to achieve repeat CT scans in all survivors, or autopsies in all deaths. Four trials in this review made a systematic attempt to detect both symptomatic and asymptomatic intracranial haemorrhage in this way (CESG 1983; FISS 1995; Prins 1989; Sandset 1990). All of the confirmed intracranial haemorrhages were intracerebral. In one trial, the use of systematic CT scanning was introduced during the trial and so not all patients were eligible for this analysis (FISS 1995). However, the numbers of patients and events in this analysis was small (symptomatic plus asymptomatic haemorrhages occurring on 20/266 (7.5%) allocated anticoagulant versus control 27/264 (10.2%), so the estimate of risk of 'symptomatic plus asymptomatic' haemorrhage is imprecise (OR = 0.76; 95% CI 0.38 to 1.52). In these trials, 25 patients (5% overall, 15 treated versus 10 control) did not have a repeat CT scan or autopsy. Including these 25 patients in hypothetical best and worst case analyses changes the odds ratio significantly (OR = 0.44 and 1.44 respectively), which suggests that the results are compatible with either substantial reductions or increases in the risk of 'symptomatic plus asymptomatic' intracranial haemorrhages with treatment.

Any recurrent stroke or symptomatic intracranial haemorrhage during the treatment period and during long term follow-up
Immediate anticoagulation reduces the odds of ischaemic stroke but also increases the odds of symptomatic intracranial haemorrhage. An outcome which combines these two (without double counting - i.e. patient allowed only one of ischaemic stroke or intracranial haemorrhage, with the first event being the one which is included) is useful for assessing the net short-term effects of anticoagulants.

Eleven trials including randomised data from 22,605 patients (96.0% of patients included in the overall review) evaluated the occurrence of 'any recurrent stroke or symptomatic intracranial haemorrhage' during the treatment period. Anticoagulation was not associated with a net reduction in the odds of this outcome (OR = 0.97; 95% CI 0.85 to 1.11). The majority of the data (93.6%) were obtained from one trial (IST 1997).

An analysis of only fatal recurrent strokes or intracranial haemorrhages during the treatment period similarly demonstrated no net benefit from anticoagulants (OR = 0.98; 95% CI 0.75 to 1.27).

An analysis of the recurrent strokes or intracranial haemorrhages during the follow-up period included data from three small studies (FISS 1995; Marshall 1960; Turpie 1987). There were far too few events for a reliable analysis.

Major extracranial haemorrhage during the treatment period
Sixteen trials included randomised data from 22,049 patients (93.6% of patients included in the overall review) reported data on major extracranial haemorrhage (defined as bleeding serious enough to cause death or require hospitalisation or transfusion). Anticoagulation was associated with a significant three-fold increase in major extracranial haemorrhage (OR = 2.99; 95% CI 2.24 to 3.99), or about 9 additional major extracranial haemorrhages per 1000 patients treated (95% CI 7 to 12 additional haemorrhages).

Other sensitivity analyses
(1) Comparison of different anticoagulant agents
Sensitivity analyses did not show any heterogeneity in the effects of the different types of anticoagulant on any of the outcomes assessed. However, such indirect comparisons are not the most reliable method of assessing differences (APT 1994) and the analyses do not exclude the possibility that important differences do exist. Indirect comparisons may be confounded by differences in the types of patients entered into the different trials. The numbers of patients randomised to treatment with most agents other than subcutaneous unfractionated heparin were small and so random error may have also obscured real differences between anticoagulant regimens. For example, one trial (FISS 1995) randomising patients between nadroparin and placebo showed a significant trend toward improved outcome with anticoagulants. However, the subsequent publication of the FISS-bis trial (FISS-bis 1998) suggested that the FISS results were most likely to have been due to the play of chance.

A systematic review of the small number of randomised controlled trials that directly compared one type of anticoagulant with another in acute ischaemic stroke can be found elsewhere in The Cochrane Library (Counsell 2003).

(2) Comparison of different doses of anticoagulants
In a previous version of this review, a sensitivity analysis of data from 1034 patients in 10 smaller trials demonstrated no significant difference in the risk of symptomatic intracranial haemorrhage or major extracranial haemorrhage with anticoagulants between those trials which used higher anticoagulant doses as generally recommended for treatment of arterial thrombosis (OR = 0.98; 95% CI 0.24 to 4.02), and those that used lower doses as recommended for prevention of venous thrombosis (OR = 1.54; 95% CI 0.46 to 3.33). This analysis was based on limited data. Data from the direct randomised comparison of 'medium' versus 'low' dose subcutaneous unfractionated heparin in the IST (IST 1997) indicated a dose-related increase in major intracranial haemorrhage in patients treated with anticoagulant, with an increase in the absolute risk of bleeding from 0.3% to 0.7% to 1.8% for control, low dose, and medium dose respectively. A similar dose-related increase was noted for major extracranial haemorrhage, with an increase in the absolute risk of bleeding from 0.3% to 0.4% to 1.4% for control, low dose, and medium dose respectively.

A systematic review of randomised controlled trials that evaluates different doses of anticoagulants in acute ischaemic stroke is currently being developed for the Cochrane Stroke Group, but has not yet been published (Gubitz, in preparation).

(3) Inclusion of additional data
An earlier version of this review (Sandercock 1993) included the Czechanowski trial (Czechanowski 1981) which was excluded from the present review because it was found to be confounded by the concomitant use of dihydroergotamine, a vasoactive agent. Inclusion or exclusion of this trial makes no material difference to the results or conclusions of this review.

(4) Publication bias
To determine whether or not we might have missed an important number of small negative trials (these are the trials most likely to remain unpublished) we undertook a funnel plot analysis (Egger 1997). The analyses with the most number of trials included (and hence the greatest statistical power) were the effects of treatment on death during the treatment period, death from all causes at final follow-up and on deep vein thrombosis. For these outcomes, a plot of the sample size for each trial versus the odds ratio for that trial showed an approximate 'funnel distribution' with 'tails' in both the positive and negative treatment effect directions (except in the outcome of deep vein thrombosis) indicating that we were unlikely to have missed a substantial number of negative trials.

Discussion

The evidence provided in this updated systematic review has not changed any of the major conclusions of the previous review, published in 1999, and can be summarised as follows.

Net effect of immediate anticoagulants in acute ischaemic stroke
Acute stroke treatments should aim to prevent disability as well as death, lest patients survive their acute stroke only to remain severely disabled. The present randomised evidence indicates that routine immediate anticoagulation does not provide any significant net short or long-term reduction in death or disability. Although immediate anticoagulation leads to fewer recurrent ischaemic strokes, avoiding 9 ischaemic strokes per 1000 patients treated, this benefit is entirely offset by a similar-sized increase in the number of intracranial haemorrhages. The net result is no short- or long-term benefit.

Hazards of immediate anticoagulants in acute ischaemic stroke
In order to be useful, a medical therapy must be safe. The present randomised evidence demonstrates a clinically and statistically significant risk of major intra- and extracranial haemorrhage with the immediate use of anticoagulants in patients with acute ischaemic stroke, with an additional 9 symptomatic intracranial and 9 extracranial haemorrhages for every 1000 patients treated.

Prevention of deep vein thrombosis and pulmonary embolism in acute ischaemic stroke
In patients with presumed or confirmed ischaemic stroke, allocation to immediate anticoagulation was associated with a highly significant 79% reduction in the odds of deep vein thrombosis during the treatment period, similar to that seen with the use of prophylactic heparin in patients undergoing different types of surgery (Collins 1988). In this review, the reductions in deep vein thrombosis with acute anticoagulation were substantial, with 281 deep vein thromboses prevented per 1000 patients treated. However, this estimate is based on relatively small numbers of patients, and most of the deep vein thromboses detected were asymptomatic. In addition, there was significant heterogeneity in the effects of treatment which renders the overall estimate less reliable. In addition to deep vein thrombosis prevention, the risk of pulmonary embolism was also reduced significantly with the use of anticoagulants (OR = 0.60; CI 0.44 to 0.81), with an additional 4 pulmonary emboli avoided per 1000 patients treated. The overall risk of pulmonary embolism appeared to be low, and the absolute benefit was small, and so the apparent reduction in deep vein thrombosis may have little clinical relevance if there is not a correspondingly large reduction in pulmonary embolism. However, there may well have been under-ascertainment of pulmonary embolism in all of the trials, since pulmonary emboli were not sought systematically. In addition, deep vein thrombosis can lead to morbidity (e.g. post-phlebitic leg and varicose ulcers), but data on these outcomes were not available from the trials. Finally, it is possible that once anticoagulants are stopped, rebound thrombosis could occur, and deep vein thromboses may begin to develop. We were unable to exclude this possibility because no trials sought deep vein thrombosis systematically after the treatment period. If anticoagulants result in no net increase or decrease in long-term death or disability, but do lead to a reduction in the number of deep vein thromboses and pulmonary emboli (albeit in immobile patients at higher risk), then the benefit of fixed heparin regimens associated with a low risk of bleeding (e.g. low fixed-dose unfractionated heparin) may yet outweigh the increased risk of haemorrhage. Unfortunately, there were insufficient randomised data comparing low-dose heparin to aspirin, or to non-pharmacological interventions such as compression stockings or early mobilisation, to determine what the most effective and safe antithrombotic regimen for deep vein thrombosis prophylaxis might be.

The stroke patients included in the trials of anticoagulants to prevent deep vein thrombosis generally had quite severe strokes, and paralysis of one leg (with the attendant high risk of deep vein thrombosis) was almost invariably present at randomisation. If however, one accepts the estimate of treatment effect from these trials in the 1980s and 1990s, it is then difficult to assess the extent to which it may be generalisable to clinical practice from 2000 onwards. In current practice, the risk of deep vein thrombosis may well be low, since many patients are admitted to Stroke Units, receive aspirin, many are given graded compression (anti-embolism) stockings, most maintain good hydration, and are generally mobilised early. A low DVT risk reduces the justification for unselective thromboprophylaxis with heparin. In IST (IST 1997) the frequency of fatal and non-fatal symptomatic pulmonary embolism (perhaps a surrogate for the occurrence of deep vein thrombosis) was very similar among patients allocated low dose subcutaneous heparin alone (0.8%) and aspirin alone (0.7%). Aspirin alone may therefore be an adequate antithrombotic agent to be used for routine deep vein thrombosis prophylaxis in many patients with acute ischaemic stroke. There is substantial evidence to support the use of antiplatelets in deep vein thrombosis and pulmonary embolism prophylaxis in other categories of high-risk patients (ATC 2002).

Different anticoagulant agents, doses and routes of administration
The present randomised evidence does not suggest that any one anticoagulant regimen is superior to any other. Indirect comparisons of unfractionated heparin, low molecular weight heparin, heparinoids and specific thrombin inhibitors have shown no significant net benefit in terms of reducing death during the treatment period, or death or dependency after follow-up greater than one month. Direct comparisons of different anticoagulants show no clear benefit of heparinoids versus unfractionated heparin (Counsell 2003), and an analysis of the effect of different dosages of anticoagulants in acute ischaemic stroke is presently being undertaken (Gubitz, in preparation). However, the available evidence does not support the routine use of adjusted-dose intravenous heparin (or heparinoid) regimens, or of more intensive fixed-dose regimens.

Different categories of patient
This systematic review provides information about the use of anticoagulants in stroke patients in general, as well as limited information about various subgroups. It is recognised that the evaluation of subgroups should only be undertaken with caution due to the increased potential for error due to the smaller numbers of patients and outcomes being evaluated. With that qualification in mind, the small amounts of (randomised) sub-group data evaluated here do not provide any evidence to support the routine use of anticoagulants in any specific category of stroke patient, including those with: cardioembolic stroke, 'stroke in progression', vertebrobasilar territory stroke, or following thrombolysis for acute ischaemic stroke to prevent re-thrombosis of the treated cerebral artery. We were not able to test the hypothesis generating subgroup analysis from the TOAST study (TOAST 1998) that anticoagulants might be of benefit in large-artery strokes. The data required for such an analysis were not available; furthermore, the subtype classifications used in the two main trials, IST and TOAST, were different and hence a subgroup analysis restricted to patients with a 'comparably defined large-artery stroke' between the two trials was not possible. Anticoagulants are also sometimes advocated for the treatment of acute carotid dissection and cerebral venous thrombosis. Separate Cochrane reviews have been prepared for these topics (Lyrer 2003; Stam 2003). This review is based on an analysis of tabular data, which limits the extent to which effects in subgroups can be explored. We were also reluctant to pursue further subgroups, since it is also hazardous to explore subgroup effects when there is no significant overall effect of an intervention on major outcomes. A more detailed assessment of the effects of anticoagulants in other categories of patient (e.g. patients treated within 3 hours, patients with large-artery strokes, patients with carotid stenosis) would however, be possible with an individual patient data meta-analysis.

As noted previously, since the last version of this systematic review (1999), there has been increasing acceptance of the concept that there is no good evidence to support the routine use of anticoagulants for the treatment acute ischaemic stroke. A number of guideline statements have been developed; the most recent American Stroke Association Guidelines (ASA 2002) include five specific recommendations which are supported by the evidence provided in this systematic review. These are as follows.

Subcutaneous unfractionated heparin is not recommended for decreasing the risk of death or stroke-related morbidity or for preventing early stroke recurrence.

Dose-adjusted, unfractionated heparin is not recommended for reducing morbidity, mortality, or early recurrent stroke in patients with acute stroke (i.e. in the first 48 hours) because the evidence indicates that it is not efficacious and may be associated with increased bleeding complications.

High-dose low molecular weight heparin/heparinoids have not been associated with either benefit or harm in reducing morbidity, mortality, or early recurrent stroke in patients with acute stroke, and are, therefore, not recommended for these goals.

Intravenous unfractionated heparin or high-dose low molecular weight heparin/heparinoids are not recommended for any specific subgroup of patients with acute ischemic stroke that is based on any presumed stroke mechanism or location (e.g. cardioembolic, large vessel atherosclerotic, vertebrobasilar, or "progressing stroke") because the data are insufficient.

Subcutaneous unfractionated heparin, low molecular weight heparins and heparinoids may be considered for DVT prophylaxis in at-risk patients with acute ischemic stroke, recognizing that non-pharmacologic treatments for DVT prevention also exist. A benefit in reducing the incidence of PE has not been demonstrated.

Despite the increasing numbers of consensus statements recommending against the general use of anticoagulants in acute ischaemic stroke, recent work from North America (Al-Sadat 2002) suggests that the message may not be getting through, and that clinical practice patterns are slow to change. This study was designed to determine current usage patterns of IV heparin for patients with acute ischaemic stroke by neurologists in the United States and Canada, hypothesizing that "US neurologists would utilize heparin more frequently" than Canadian neurologists, supposedly because of fear of litigation, should they not act. Two hundred and eighty neurologists from the United States and 270 Canadian neurologists were presented with five clinical scenarios, including: stroke in evolution, atrial fibrillation-related stroke, vertebrobasilar stroke, carotid territory stroke, and multiple TIAs. Each group was asked to respond with yes, no, or maybe to whether they would use IV heparin in each scenario. The effect of medicolegal factors as a potential influence on their decision was also ascertained. Interestingly, American neurologists were significantly more likely to use IV heparin for patients with stroke in evolution (51% versus 33%), vertebrobasilar stroke (30% versus 8%), carotid territory stroke (31% versus 4%), and multiple TIAs (47% versus 9%). Both American and Canadian neurologists tended to use IV heparin for patients with atrial fibrillation and stroke (88% and 84%, respectively).

Reviewers' conclusions

Implications for practice

Evidence from this systematic review indicates that anticoagulants in acute ischaemic stroke have no effect in terms of death, or death or dependency after follow-up of at least one month. A reduction in recurrent ischaemic stroke during the treatment period is exactly offset by an increase in intracranial haemorrhage. There is a dose dependent increase in both intra- and extracranial haemorrhage, although low dose regimens decrease deep vein thrombosis and pulmonary embolus, which are rare.

The data do not support the routine use of immediate high dose intravenous or subcutaneous anticoagulants in any form for patients with acute ischaemic stroke. Low dose subcutaneous regimens will prevent deep vein thrombosis, but with a small but definite increased risk of major haemorrhage. It may therefore be advisable to consider safer alternatives in immobile patients (such as aspirin, compression stockings, or early mobilisation).

The data do not support the use of low molecular weight heparins, heparinoids or thrombin inhibitors in the treatment of acute ischaemic stroke.

The analysis performed did not identify any category of patient where there was clear net benefit. Clinicians who feel compelled to use immediate anticoagulants for specific categories of patients following acute ischaemic stroke should weigh any potential theoretical benefits with the known risk of bleeding. Aspirin is an effective antithrombotic alternative to anticoagulation which is safe when used in the acute phase of ischaemic stroke.

Implications for research

Further large-scale trials comparing immediate anticoagulation with control in patients with acute ischaemic stroke are probably not warranted. Those who wish to continue to use IV dose-adjusted heparin regimens routinely to treat specific categories of stroke patient should provide convincing evidence from new randomised controlled trials to support such practices.

This review has not provided clear evidence about the optimum antithrombotic regimen for the prevention of deep vein thrombosis and pulmonary embolism in stroke patients. Aspirin alone, low-dose subcutaneous heparin, the combination of the two or the use of compression stockings are all promising possibilities, but a very large scale randomised trial with several tens of thousands of patients would be required to determine which has the most favourable balance of risk and benefit on overall clinical outcome.

Acknowledgements

We would like to thank individual trialists for supplying additional information: Professor B Boneu for providing us with a copy of Dr Pince's thesis; Dr Duke, Dr R Kay (FISS 1995), Dr G Pambianco, Dr H Magnani from Organon International for supplying additional information on the trial using danaparoid (Turpie 1987); Dr L Antonutti and Dr M Zorzon (Cazzato 1989); Dr Ewa Lindenstrom from Leo Pharmaceuticals for data on tinzaparin (Vissinger 1995); H Willems, who helped identify many of the early trials; Dr E Dick for supplying us with a copy of the MedStrategy document 'Stroke: A focus on opportunity'; Dr AGM van den Belt and Dr RI Lindley who helped produce the previously published version of this review (Sandercock 1993); and Hazel Fraser for supplying regular lists of trials identified by the Cochrane Stroke Group's search strategy. David Signorini was a co-author of an earlier version of the review. We also acknowledge the help given by the secretariat of the Antithrombotic Trialists' Collaboration (Dr C Baigent, Dr C Sudlow).

Ongoing trials
Any clinician who knows of additional trials that we have omitted please write to Dr Peter Sandercock.

Potential conflict of interest

Dr Sandercock was the principal investigator of the International Stroke Trial and Dr Counsell was also on the Steering Committee of this trial. Dr Gubitz enrolled several patients in the IST.