Anticoagulants have been available since around 1940 and have become the standard of treatment for venous thromboembolism (VTE) for over four decades. However, as with other treatments which became established before the evidence-based era, there is a paucity of evidence from randomized controlled trials validating their effectiveness in preventing the most feared complication of VTE, recurrent fatal pulmonary embolism (PE). Only two such trials have been performed, the results of which conflict. The bulk of data supporting their use are derived from three sources. First, studies of thromboprophylaxis, and comparisons of shorter and longer courses of anticoagulants in high-risk patients with established VTE have clearly demonstrated their effectiveness in primary and late secondary prevention. Given that heparin has an immediate onset of action, anticoagulants should therefore also be effective in early secondary prevention, the proposed mechanism of action in the acute treatment of VTE. Secondly, studies of inadequately treated patients have consistently shown higher recurrence rates than in those adequately treated. Finally, comparisons of outcomes in untreated and treated historical series, and of untreated historical series to treated series in the modern era have shown substantially lower rates of fatal PE in anticoagulated patients. Because these differences are so marked, harmonize with our current understanding of the mechanism of action of anticoagulants and are supported by other evidence, it is much more likely that they at least partly reflect the effectiveness of anticoagulants as opposed to being explicable purely in terms of accumulated biases and a changing distribution of disease severity.
The annual incidence of recognized symptomatic deep vein thrombosis (DVT) and pulmonary embolism (PE) in the western world is 1.0 and 0.5 per thousand population, respectively  and has not declined in recent years . Moreover, many other symptomatic episodes go unrecognized, so that the true incidence is probably far greater [3, 4]. Many millions of patients with venous thromboembolism (VTE) worldwide have therefore been treated with anticoagulants over the past few decades in the expectation that they will improve survival. Only two small randomized controlled trials (RCTs) have assessed the effectiveness of this approach, the results of which conflict [5–7]. Given that anticoagulants are themselves associated with morbidity and mortality , the paradigm of unselective anticoagulation in patients with symptomatic VTE, whilst accepted by the great majority of clinicians, has been questioned [9–11].
This review aims to examine available evidence with respect to the following question: are anticoagulants effective in preventing the most feared complication of VTE – fatal PE? Whilst secondary aims of treatment include amelioration of symptoms and a reduction in the risk of post-thrombotic syndrome, we have focused on the issue of mortality.
The proposed mechanism of action of anticoagulants in VTE
Heparin and oral anticoagulants (OACs) have no significant thrombolytic action, so can be expected to have no immediate effect on clot burden. These drugs act by altering the dynamic balance between inherent thrombogenic and fibrinolytic processes, so preventing or reducing propagation of DVT (although thrombus progression may still occur in a significant minority despite treatment [12–14]). The rate of propagation is uncertain but may be up to several centimetres per day in susceptible patients. Hence, they are thought to reduce the likelihood of PE in patients presenting with DVT, and recurrence in those presenting with PE, whilst existing clot undergoes organization and/or fibrinolysis via endogenous mechanisms. These mechanisms are much more active in the pulmonary arterial circulation than the lower-limb venous system: half of patients with treated DVT continue to have an abnormal ultrasound 1 year later , whilst angiographic studies show that resolution of PEs may begin after 1 day and be complete after only 2 weeks in some patients . In most patients with PE, therefore, early resolution of acute symptoms is probably independent of the initiation of anticoagulation, the main goal of treatment being the prevention of PE recurrence [3, 16, 17]. In haemodynamically unstable PE, a further aim of treatment is an immediate reduction in clot burden which may be achieved either by thrombolysis or embolectomy. In summary, then, the proposed action of anticoagulants in VTE is via early secondary prevention.
Proximal DVT versus below-knee DVT
Proximal DVT (PDVT) and below-knee DVT are subgroups with distinct prognoses . Symptomatic PE occurs in up to half of patients with untreated PDVT , but whilst PEs can arise from nonextending below-knee DVTs  they are very unlikely to be life-threatening in the presence of adequate cardiorespiratory reserve [17, 21–26] (Table 1), and the incidence of clinically important PE is probably around 1% in these patients [19, 23, 27–31]. Hence, the main concern in patients with isolated below-knee DVTs, which rarely produce symptoms, is the 20% risk of proximal propagation in untreated cases [27, 32]. It is considered safe to withhold treatment in patients with below-knee DVT and adequate cardiorespiratory reserve (Table 1), enabling them to survive a potential small PE from a below knee source, as long as serial noninvasive imaging over the next fortnight excludes proximal propagation, though treatment would usually be given in symptomatic cases [21, 24, 26, 32, 33].
Table 1. Definition of inadequate cardiopulmonary reserve in patients with suspected pulmonary embolism in the study of Hull et al. 
Inadequate cardiopulmonary reserve defined by any of the following:
In this review, therefore, we are primarily interested in the effectiveness of anticoagulants in patients with PDVT as opposed to isolated below-knee DVT. Although only around one-third of asymptomatic DVTs identified by screening in high-risk populations are proximal , PDVTs account for 80–90% of all symptomatic cases [33, 35]. Hence, when we refer to clinically apparent DVT, we are referring to DVT that is predominantly proximal.
RCTs of anticoagulants in symptomatic VTE
Unfractionated heparin (UFH) was first used clinically in 1939  and dicoumarol, the first coumarin, was introduced in 1942 , but anticoagulants were not consistently used for the treatment of VTE until publication of the classic trial by Barritt and Jordan in 1960 , the first of only two RCTs of anticoagulants in VTE. Thirty-five patients with PE were randomized either to UFH (10 000 units intravenously six-hourly for six doses without laboratory monitoring) followed by the OAC nicoumalone (target prothrombin time two to three times normal) for a total of 2 weeks or no anticoagulants, both groups also receiving 10 days of bed rest. Of the 19 untreated patients, five (26.5%) died from recurrent PE and five (26.5%) developed nonfatal recurrence. None of the 16 treated patients developed PE, although one patient died of a combination of pneumonia and bleeding from a duodenal ulcer (probability for recurrent fatal and nonfatal PE: P = 0.0005). Subsequently, it was felt unethical to withhold anticoagulants in a further 38 recruits, all of whom were treated. Of the 54 treated patients, nonfatal recurrent PE occurred in only one (P = 0.0000014), and deaths from all causes were significantly reduced in the treated group (P = 0.01).
This study has been widely cited as supporting the effectiveness of anticoagulation for VTE, but several caveats should be borne in mind. First, like all studies performed before the advent of widespread use of venography and pulmonary angiography, VTE was diagnosed clinically. Subsequent studies using objective diagnostic methods found that VTE was confirmed in less than half of patients in whom it had been suspected [21, 37, 38], so that early series contain a proportion of patients with diagnoses other that VTE. It should be noted, however, that alternative diagnoses in patients presenting with symptoms suggestive of DVT are rarely life threatening, whilst those in patients presenting with symptoms suggesting PE, such as pneumonia and myocardial infarction, may well be. Therefore, early data probably do not overestimate the mortality associated with untreated DVT, but the situation regarding PE is uncertain.
Secondly, historical series probably contain a higher proportion of patients with more severe disease than is seen in the modern era of declining tolerance for diagnostic uncertainty , so may overestimate the current mortality of untreated VTE. Thirdly, the trial was not conducted in a double blind fashion and the comparability of the two groups was not reported. Finally, whether or not all deaths in the untreated patients were due to PE has been questioned [9, 10, 40].
The other RCT, which is not widely cited, is that of Nielsen et al. [6, 7] who randomized 90 patients with venographically confirmed symptomatic DVT (proximal in 74) but no clinical evidence of PE to treatment with UFH (given by continuous intravenous infusion aiming for an activated partial thromboplastin time of 1.5–2.5 times normal) followed by the OAC phenprocoumon (target international normalized ratio: 2.0–4.3) for 3 months, or to a 10-day course of the nonsteroidal anti-inflammatory phenylbutazone with no anticoagulants. Both groups were mobilized from the day of diagnosis and wore graded compression stockings. Baseline ventilation–perfusion (VQ) scanning was performed, and repeated on days 10 and 60, whilst venography was repeated on day 30. There were no significant differences in progression or regression of DVT between the two groups, or in the appearance of new PEs. Three treated patients developed clinical evidence of DVT recurrence, compared with nine untreated patients. In the latter group, one patient developing PE and another developing phlegmasia cerulea dolens were anticoagulated. In the remaining seven, symptoms settled without the use of anticoagulants. At 3 months, two clinical PEs had occurred in the treated patients, one of which was fatal, compared with one nonfatal PE in the untreated patients.
It should be noted that patients unable to mobilize on the day of diagnosis were excluded, and no patients had severe cardiovascular disease so that the study consisted of relatively fit patients. Further, the recently demonstrated effectiveness of antiplatelet agents in the primary prevention of VTE [41, 42] means that phenylbutazone cannot be regarded as a placebo. Finally, confidence intervals for differences in outcomes between the two groups were wide.
A prima facie interpretation of these RCTs is that untreated major PE is associated with a high mortality and that such patients should be treated, whilst ambulatory patients with symptomatic DVT have a relatively benign prognosis and may not require treatment. However, as we will see, early reports of the natural history of untreated DVT do not support this notion, which is also discordant with the widely accepted view that DVT and PE are part of a disease spectrum rather than separate entities.
Non-RCT derived data
Several alternative lines of evidence are relevant to the issue of whether or not anticoagulants prevent fatal PE. Historical data are available from an era when anticoagulation was not considered mandatory in patients with VTE, providing information on outcomes in treated and untreated patients. Some problems with the interpretation of such data have been discussed. Furthermore, all of these series were retrospective, no information was given on how patients were selected for anticoagulant or conservative treatment, and treatment regimes varied widely (in some cases consisting only of a few days treatment with heparin). Moreover, nonanticoagulated patients were kept on bed rest for periods of up to 40 days  which may have worsened outcomes. It is recognized that selection biases tend to weight outcomes in historical controls in favour of new therapies , so such data should be interpreted with great caution, although may still be important if a consistent picture of improved outcome with anticoagulants emerges. Limited additional data on outcomes of untreated patients from highly selected subgroups in the modern era are also available. The outcome of patients with treated VTE in recent years is clearly established, so that comparison can be made with outcomes of untreated patients from the modern and preanticoagulant era, but again, for the reasons outlined, caution is required.
Data are also available regarding the effectiveness of anticoagulants in the primary and late secondary prevention of VTE: if anticoagulants are effective in these circumstances, they should also be effective in early secondary prevention as the onset of action of heparin is immediate. If there is reasonable evidence to suggest that untreated VTE in the current era is associated with a significant risk of recurrent fatal PE, then anticoagulants should improve outcomes. Finally, a number of studies have compared recurrence rates in patients with VTE treated with regimes now known to be inadequate, to regimes now known to be adequate. If higher rates of recurrence were consistently shown in inadequately treated patients, this would provide further evidence for the effectiveness of anticoagulants.
Historical series of patients with treated and untreated VTE
Before reviewing these series, it should be appreciated that the ‘mortality of untreated clinical PE’ (caused by incident or recurrent PE) could be interpreted in two different ways. First, it could mean the subsequent mortality from incident or recurrent PE from the point at which any clinical PE initially occurs. This figure would include immediately fatal or misdiagnosed PE found at postmortem (PM), and other deaths as a result of incident or recurrent PE. It is well recognized that PE is the commonest unsuspected diagnosis at PM , with 30%  or fewer  of fatal events suspected during life, and no sign of improvement in the current era [48, 49] Hence, this group defines the true mortality of PE, and we will refer to it hereafter as ‘recognized and unrecognized clinical PE’. Information on outcomes in this group requires capture of data from multiple sources including PM examinations.
Secondly, the ‘mortality of untreated clinical PE’ could refer to the selected subgroup of patients who survive to the point of diagnosis. Given that around one-fifth of deaths from massive fatal PE are immediate , and two-thirds occur within 2 h of onset [3, 48, 50–52], outlook in these patients is clearly considerably better. We will refer to this subgroup as ‘recognized clinical PE’, and it is in these patients that most studies provide data. These observations do not apply to patients with clinical DVT, who, by definition, do not present with sudden death. Furthermore, we are only concerned with deaths in patients with VTE caused by PE: it is estimated that one- to two-thirds of PEs found at PM have either caused or contributed to death, the rest being incidental [48, 50], and the majority of deaths over the subsequent months in recognized and appropriately treated cases are attributable to comorbidities rather than recurrent PE [53, 54].
Outcomes of recognized and unrecognized clinical PE in the presence or absence of treatment of recognized cases are available from two historical series (Table 2). Pooled data show a total risk of fatal PE of 65% in untreated versus 35% in treated cases. Up to one-third of all PEs were diagnosed for the first time at PM [43, 55].
Table 2. Mortality from incident and recurrent pulmonary embolism (PE) in untreated/treated patients with recognized and unrecognized clinical PE in the preanticoagulant era
Tables 3 and 4 show pooled data on outcomes in historical series of untreated and treated patients with DVT and recognized PE. The overall risks of fatal PE in untreated recognized PE and treated recognized PE were 26.6 and 2.6%, respectively [43, 55–67]; corresponding figures for PDVT were 16.2 and 0.7% [58–60, 62, 65, 67–74]. The series reported by Hermann et al.  of 107 untreated cases of recognized PE is of particular interest. Patients who died within 48 h of diagnosis were excluded, and a proportion of this group was composed of patients diagnosed 4–5 days after the event who had improved clinically to the extent that they were felt not to require anticoagulants. This was therefore a selected group who had already survived their initial PE. Overall, recurrent PE occurred in 20% of 249 treated patients versus 60% of untreated, with recurrent fatal PE in 4.7% vs. 35.5%. In patients with no prior cardiac disease, recurrent fatal PE occurred in 0.8% vs. 18.5%, and in 7.7% vs. 45% in those with pre-existing cardiac disease. Hence, this study shows that patients without cardiac disease who have clearly survived an initial PE are still at substantial risk of recurrent fatal PE.
Table 3. Mortality from incident and recurrent pulmonary embolism (PE) in untreated/treated patients with recognized clinical PE in the preanticoagulant era
Apart from the trial of Nielsen et al. [6, 7], data are available from two prospective studies. Kakkar et al.  studied the incidence and natural history of untreated DVT in 132 postoperative general surgical patients using 125I radiolabelled fibrinogen scanning. 40 DVTs were detected, nine of which extended to the popliteal vein or higher. Four (45%) patients with PDVT developed clinically-diagnosed PE compared with none of those with nonextending below-knee DVT. The proportion of patients with clinical as opposed to subclinical PDVT was not stated, although half of all DVTs in the study were symptomatic. This study had been widely cited as supporting the concept that patients with untreated PDVT have a high risk of developing symptomatic PE.
In a Korean study, Kim et al.  venographically screened 227 patients 1 week following knee arthroplasty and found PDVT in 35 cases. These patients had received no form of prophylaxis and were not treated, and no cases of clinical PE occurred subsequently. These results are difficult to reconcile with data derived from patients undergoing major lower-limb orthopaedic surgery not given prophylaxis in whom PDVT occurred in 20–30% with a rate of fatal PE of 2–3%, indicating a 10% rate of fatal PE for subclinical PDVT, assuming all fatal PEs arose from PDVT [18, 23, 76], and may reflect ethnic differences in thrombophilia.
Retrospective data are available from a study by Pellegrini et al. , who performed bilateral venography 7–10 days post-hip arthroplasty in 174 patients, with anticoagulation in all recognized cases of DVT. On retrospective review several weeks later, 13 cases of calf vein DVT were identified which had been overlooked and left untreated. Four (31%) of these patients had subsequently developed clinical PE (P = 0.014), fatal in two cases. PDVT was identified in the two survivors. PMs were not performed in the two fatal cases, but PE was thought highly probable on clinical grounds. It is highly likely that proximal extension of calf vein DVT had also occurred prior to fatal PE in the two patients that died. This study is at odds with that of Kim et al. , clearly demonstrating that untreated subclinical PDVT in the modern era can lead to fatal PE.
Finally, data are available on outcomes in moderate to high-risk general surgical patients found to have subclinical PDVT who acted as controls in trials of thromboprophylaxis in the 1970s and 1980s. Pooled data from 1206 patients show prevalences of venographically confirmed PDVT and fatal PE of seven and 0.9%, respectively , giving an estimated mortality associated with subclinical PDVT of 12–13%. In most of these studies, it was not explicitly stated whether or not PDVTs were treated, however, given that the earliest of these was published in 1972, it is likely that most were. As fatal PE is rare in patients with treated DVT , most cases of fatal PE probably occurred early on before screening or in patients with known PDVT who were untreated. The figure of 12–13% must therefore represent a minimum for the risk of fatal PE in untreated subclinical PDVT as the natural history of PDVT was modified by screening and anticoagulation. It should also be noted that the mortality of untreated clinical PDVT may be greater than subclinical PDVT as thrombus load tends to be lower in the latter ; indeed, the more severe the PE suffered by a patient, the more likely they are to have associated clinically apparent DVT .
Heit et al. [53, 81] conducted a retrospective, population-based study identifiying all cases of first in a lifetime VTE from an inception cohort of residents of the Olmsted County region between 1966 and 1990. These data include the full-spectrum of disease in all clinical settings (community, nursing homes and hospitals) and PM data. For 2218 confirmed cases of VTE, These data showed that 36.4% of patients with PE died on the first day, the great majority of whom presented as sudden death. Hence, the proportion of patients developing PE who die before recognition and treatment does not appear to have changed from the preanticoagulant era.
Subsequent to Barritt and Jordan's study , no prospective data are available on the mortality of untreated recognized PE. However, other strands of evidence are relevant. Stein et al.  analysed the 3-month outcomes in 20 patients with PE in the prospective investigation of pulmonary embolism diagnosis (PIOPED) study , In which 755 patients with suspected PE underwent both VQ scanning and pulmonary angiography, who escaped treatment because their pulmonary angiograms were initially incorrectly interpreted as normal. One patient died from the original PE and/or recurrence and another suffered nonfatal recurrence, giving a total recurrence rate of 10%. Untreated patients had segmental or subsegmental PE on angiography in 84% of cases compared with 36% in treated patients (P < 0.001), so these data reflect the outcome of untreated mild PE in the modern era and probably underestimate the overall mortality from untreated PE in unselected patients.
Johnson et al.  reviewed outcomes in 603 post-hip arthroplasty patients from 1962 to 1973 in a prospective registry with clinically diagnosed PE which was not fatal at presentation. A total of 295 patients were treated with anticoagulants, whilst 308 cases were treated conservatively. The first group contained patients treated with UFH alone for 2–7 days, patients treated with warfarin only for 3–6 weeks and patients treated with 24–48 h of UFH followed by warfarin, so fewer than one-third of these patients were treated adequately. Recurrent PE occurred in 13% of the treated patients including one fatality compared with 3.2% in the untreated group, with no fatalities. This trial suffers from many of the problems of the preanticoagulant era series, in particular the absence of objective testing for PE, absence of information about how a decision was made to give or withhold anticoagulants, and absence of information about matching between groups.
Two recent major studies do, however, suggest that it may be possible to identify a subgroup of good-prognosis patients with PE who may have a relative benign outlook in the absence of treatment, and highlight the prognostic importance of residual PDVT and cardiorespiratory reserve. Hull et al.  withheld treatment in 627 patients with indeterminate VQ scans, adequate cardiorespiratory reserve (Table 1), and serially negative impedance plethysmography for residual PDVT over a 2-week period. Recurrent VTE occurred in only 1.9% at 3 months, including one fatal PE. By contrast, in a separate analysis of 77 of the patients excluded because of inadequate cardiopulmonary reserve who also had low probability VQ scans, most of whom were left untreated, nine (11.7%) developed recurrent VTE and six (7.8%) died from PE . Wells et al.  investigated a cohort of 665 patients with nonhigh clinical suspicion for PE, indeterminate VQ scans and negative serial US examinations for PDVT over 2 weeks and showed that the incidence of clinical nonfatal VTE over the next 3 months was only 0.5%.
Extrapolating data from the PIOPED study , and assuming a 50% prevalence of residual PDVT following PE , it can be estimated that the prevalence of genuine PE in the subgroups of patients in these studies from whom treatment was withheld was around 10%. Furthermore, Stein  retrospectively combined data from these two studies to show that patients with nonmassive PE, adequate cardiopulmonary reserve and negative serial US who were left untreated could be expected to have a 3-month incidence of fatal and nonfatal recurrent PE of 0 and 3%. Hence, good-risk patients with PE who are at high risk of bleeding, for example postoperatively, might safely be left untreated, but this approach requires prospective validation .
Outcome of treated VTE in the modern era
Douketis et al.  overviewed 25 high-quality prospective studies published between 1966 and 1997 reporting on outcomes in patients with objectively diagnosed recognized clinical VTE treated with UFH (target activated partial thromboplastin time: 1.5–2.0) or low molecular weight heparin (LMWH) followed by OACs (target international normalized ratio: 2.0–4.5) for 3 months. Data were available on 4221 patients presenting with DVT, and 1302 with PE. During 3 months of treatment, 0.4 and 1.5%, respectively, suffered fatal PE. Patients were excluded if they had been thrombolysed (3%) or had a life expectancy of less than 3 months because of severe comorbidities (7%), so these data may underestimate the true risk of recurrent fatal PE in unselected patients, although are likely to be representative for the majority. In addition, 3.8% of patients presenting with DVT developed nonfatal recurrent VTE over this period, although comparable data were not available for those presenting with PE.
In a more recent study  of 1021 patients with VTE treated either with UFH or LMWH followed by OACs for 3 months, recurrent fatal PE occurred in 2.2% of patients presenting with PE versus 0 in those presenting with DVT, whilst recurrent nonfatal events occurred in 4 and 5%, respectively. The same exclusions applied. All six patients dying of recurrent PE had cardiorespiratory disease. Pooled data from these studies are shown in Tables 5 and 6.
Table 5. Mortality from incident and recurrent pulmonary embolism in treated patients with clinical deep vein thrombosis in the modern era
Data from the unselected patients recruited to the PIOPED study [54, 83] show a similar outcome in patients presenting with recognized PE attributable to the PE itself. A total of 399 treated patients (9.5%) died during the index admission, and total mortality at 1 year was 23.8%. However, 90% of deaths were related to comorbidities, particularly cancer and cardiorespiratory disease, so the absolute risk of death as a result of PE was only 2.5%. Nine of 10 deaths as a result of PE were caused by recurrence, eight occurring within a week of diagnosis. In other prospective studies, the mortality caused by PE itself has been up to 8%, most deaths occurring on a background of cardiorespiratory disease [88, 89].
Whilst these studies show that most fatal PEs occur in patients with severe comorbidities, fatal PE may also occur in previously fit subjects: in two PM series together containing 1144 cases of fatal PE [50, 90], PE was the sole cause of death in 22% of patients who otherwise would have had an excellent long-term outlook. Furthermore, a proportion of patients with significant comorbidities dying of PE may still otherwise have survived to the point of discharge [50, 90].
In summary, mortality from the initial PE remains high and probably has not changed over the years as most cases are either diagnosed at PM, or die before treatment can be administered. In haemodynamically stable patients surviving to the point of diagnosis and treatment, outcome is generally good. Most deaths occurring in these patients are attributable to comorbidities, and most deaths attributable to PE are due to PE recurrence, predominantly affecting those with cardiorespiratory disease. In haemodynamically stable patients without major comorbidities, the outlook once treatment is started is excellent.
Anticoagulants in the primary and late secondary prevention of VTE
In a meta-analysis of 70 trials  in general surgical, orthopaedic and urological surgery patients, heparin reduced the absolute risk of fatal PE from 0.9 to 0.3% (64% reduction, P < 0.0001). Overall mortality was also significantly reduced (P < 0.02). A more recent meta-analysis of seven studies of prophylaxis in high-risk medical  patients has shown a 52% reduction in clinical PE (P < 0.001). Therefore, anticoagulants are clearly effective in the primary prevention of VTE in a broad range of patients.
Studies comparing longer and shorter courses of anticoagulants in patients presenting with VTE at high risk of recurrence have generally shown a substantial reduction in recurrences with more prolonged courses of treatment [93–98]. For example, Kearon et al.  compared 3 months of anticoagulant therapy plus continued treatment for a further 24 months to 3 months anticoagulant therapy plus placebo for a further 24 months in patients with a first episode of idiopathic VTE. The trial was stopped after an interim analysis showed recurrence rates at 10 months of 1.3% vs. 27.4% per patient-year (P < 0.001); and Shulman et al.  compared 6 months to indefinite anticoagulation in 227 patients with a second episode of VTE and found 4 year incidences of recurrence of 2.6% vs. 20.7% (odds ratio 8; 95% confidence interval 2.5–25.9). Given that longer courses of anticoagulants are more effective than shorter courses in these patients, it can be concluded that anticoagulants are highly effective in the late secondary prevention of VTE.
Studies of recurrence rates in inadequately treated patients
Hull et al.  randomized 115 patients with symptomatic PDVT to initial treatment either with intermittent full-dose subcutaneous or intravenous UFH, followed by OAC therapy for 3 months. A satisfactory anticoagulant response failed to occur in most of the patients given subcutaneous heparin, but did occur in most patients given intravenous heparin. Recurrent symptomatic VTE occurred in 11 of 57 patients (19.3%) vs. three of 58 (5.2%) (P = 0.024), often in the first week, including one fatal PE in the first group. Recurrences were limited to patients with an initial subtherapeutic anticoagulant response, occurring in 13 of 53 (24.5%) patients inadequately treated in the first 24 h versus one of 62 (1.6%) patients who did achieve an initial adequate response (P < 0.001).
In another study, Hull et al.  randomized 68 patients with clinical or subclinical DVT to treatment with intravenous UFH for 2 weeks, followed either by low-dose subcutaneous UFH (5000 units b.i.d.) or warfarin, with either regime given for 6 weeks (below-knee DVT) or 3 months (PDVT). Nine of 19 (47%) patients with PDVT in the subcutaneous UFH group suffered recurrent VTE during the study period, compared with none of 17 receiving warfarin (P < 0.001).
Brandjes et al.  compared 3 months of OACs alone to intravenous heparin plus 3 months of OACs in 120 patients with symptomatic PDVT, with follow-up for 6 months. The trial was stopped early after an interim analysis showed that symptomatic recurrence occurred in 20% vs. 6.7% (P = 0.06). Further, repeat venography 1 week after initiation of treatment showed asymptomatic thrombus extension in 39.6% vs. 8.2% (P < 0.001).
These and other studies [102–104] clearly establish that recurrence rates are lower in adequately than inadequately treated patients and so strongly support the effectiveness of anticoagulants.
Morbidity and mortality associated with anticoagulant use in patients with VTE
No discussion of the effectiveness of anticoagulants would be complete without considering the morbidity and mortality associated with these drugs. In a retrospective study of 21 250 patients hospitalized in California with a diagnosis of DVT between 1992 and 1994 , 1.4% were readmitted principally for bleeding in the subsequent 3 months, and a further 1% were readmitted with a new medical diagnosis during this period with bleeding as a secondary problem. Most bleeding episodes occurred within the first month of treatment. These figures include an absolute 0.3% risk of intracranial bleeding, and a 0.4% risk of fatal bleeding. By contrast, matched controls originally admitted with cellulitis or pneumonia had a 0.7% risk of readmission with bleeding in the 3 months following discharge so that not all episodes can be attributed to anticoagulation. These data do not include bleeding episodes occurring during the index admission. Although such bleeding occurred in 2.9% of patients, it is not known what proportion of these were major. Other prospective studies have generally shown a 2–3% risk of major haemorrhage associated with 3–6 months of anticoagulation in patients with VTE [93, 95, 96, 98].
An appreciation of the importance of residual PDVT in determining recurrence risk in patients with PE [16, 22] suggests that a capability for comprehensive imaging in patients with suspected VTE, allowing simultaneous ascertainment of both DVT and PE and hence thrombus load, might be the next step in the evolution of the diagnosis and management of this condition, potentially facilitating a more titrated approach to treatment. A novel application of magnetic resonance (MR) imaging technology known as MR direct thrombus imaging (MRDTI) is proving promising in this regard. It is noninvasive and provides a direct image of thrombus by detecting methaemoglobin in clot, unlike previous imaging techniques which have demonstrated clot either as a defect in flowing blood or in terms of surrogate markers. Early studies show that it is highly accurate for both DVT  and PE [107, 108], providing unprecedented information about thrombus characteristics such as age and volume. For example, MRDTI has already been used to show that a thrombus volume threshold of 18 mL in patients with DVT is more predictive of PE than the proximal extent of thrombus . The further evaluation of this technique will be awaited with interest.
Studies comparing outcomes in patients with VTE according to the duration and adequacy of therapy strongly suggest that anticoagulants are effective in preventing VTE recurrence, their proposed mechanism of action in improving outcomes in patients presenting with DVT or PE. Hence, the 10–20-fold lower risk of fatal PE in patients with treated, recognized VTE in the current era in comparison with untreated, recognized VTE in historical series is likely at least partly to reflect the effectiveness of anticoagulants, over and above the likely higher prevalence of milder disease in the modern era (see Figs 1 and 2). Furthermore, Barritt and Jordan's trial , despite its shortcomings, is fairly persuasive direct evidence that anticoagulants improve the outcome in more severe forms of PE. Although the studies of Johnson , Nielsen [6, 7] and Kim  are apparently at odds with this body of evidence, when viewed in the Bayesian context of the total edifice of evidence, they probably should not greatly shift our degree of belief in the precept that anticoagulants are effective in the prevention of recurrent fatal PE. And whilst the majority of PE-related deaths occur in patients with comorbidities and/or inadequate cardiorespiratory reserve, the fact that a significant minority occur in previously fit patients [50, 90], and that recurrent fatal PE occurred in up to one-fifth of untreated patients without cardiac disease who had clearly survived the initial event in early series , suggest that VTE should not be regarded as benign in unselected, previously fit patients.
One small study has suggested, however, that the outcome in more minor forms of PE in the absence of treatment may be better than is generally thought , although the risk of fatal PE was still substantially greater that the risk of fatal bleeding associated with a 3-month course of anticoagulants . Furthermore, there is evidence that it may be possible to identify a subgroup of low-risk patients with PE on the basis of their cardiorespiratory reserve and likelihood of recurrence, determined by the presence or absence of residual PDVT over the next fortnight, in whom the prognosis may be relatively benign without treatment [16, 22, 25]. In a subgroup of such patients judged to be at high risk of bleeding with anticoagulants, a study might now be justified in which treatment was withheld and outcomes carefully monitored.
In summary, whilst the prognosis of untreated VTE forms a spectrum determined by factors such as thrombus load, comorbidities and cardiorespiratory reserve, an overview of all available evidence supports the concept that anticoagulants are effective.