Pharmacological Venous Thromboembolism Prophylaxis in Hospitalized Medical Patients: A Meta-analysis of Randomized Controlled Trials

Background  There is uncertainty regarding which pharmacological agents most effectively prevent venous thromboembolism in hospitalized medical patients. We therefore performed a meta-analysis to determine this.

Methods  MEDLINE, EMBASE, and the Cochrane Central Register of Controlled Trials were searched from 1950, 1966, and 1800, respectively, through June 30, 2006, for randomized controlled trials that involved medical patients comparing unfractionated heparin (UFH) or low-molecular-weight heparin or heparinoid (LMWH) with a control, LMWH with UFH, or selective factor Xa inhibitors with a comparator. Study selection, validity assessment, and data abstraction were performed by 2 independent reviewers (L.W. and S.W.). Data synthesis was undertaken by 1 blinded investigator (S.J.H.).

Results  Thirty-six studies were included. Compared with the control, UFH was associated with a reduced risk of deep venous thrombosis (DVT) (risk ratio [RR], 0.33; 95% confidence interval [CI], 0.26-0.42) and pulmonary embolism (RR, 0.64; 95% CI, 0.50-0.82), as was LMWH (RR, 0.56; 95% CI, 0.45-0.70; and RR, 0.37; 95% CI, 0.21-0.64, respectively). A UFH dosage of 5000 U 3 times daily was more effective in preventing DVT than a UFH dosage of 5000 U twice daily when compared with the control (RR, 0.27; 95% CI, 0.20-0.36; vs RR, 0.52; 95% CI, 0.28-0.96). Neither UFH nor LMWH reduced mortality. When directly compared with UFH, LMWH was associated with a lower risk of DVT (RR, 0.68; 95% CI, 0.52-0.88) and injection site hematoma (RR, 0.47; 95% CI, 0.36-0.62), but no difference was seen between the 2 agents in the risk of bleeding or thrombocytopenia.

Conclusions  Both UFH and LMWH reduce venous thromboembolic risk in hospitalized medical patients, but neither agent alters mortality. When directly compared, LMWH is more effective in preventing DVT.

Venous thromboembolism (VTE), which consists of deep venous thrombosis (DVT) and pulmonary embolism (PE), is a major and often unrecognized cause of morbidity and mortality in hospitalized patients.^1-9 Approximately 10% of hospital deaths can be attributed to pulmonary emboli.¹⁰ It is widely accepted that reliance on the diagnosis and treatment of an established event is an inappropriate way to approach VTE because diagnosis is often difficult and massive PE may be the first clinical manifestation of the disease.^11,12 Prevention is therefore of paramount importance.

Thromboprophylaxis is routinely used in surgical patients. However, it is not as widely practiced in the medical setting,^13-15 even though medical patients represent most hospitalized patients and at least 75% of fatal PEs occur in this group.¹⁶ Unfractionated heparin (UFH), low-molecular-weight heparin or heparinoid (LMWH), and selective factor Xa inhibitors are all used for the prevention of VTE. Current International Consensus Statement¹¹ and American College of Chest Physicians¹⁷ guidelines recommend the use of UFH or LMWH in medical patients at risk for VTE. We performed a meta-analysis of randomized controlled trials to compare the efficacy and safety of the various agents available for thromboprophylaxis.

MEDLINE (via PubMed), EMBASE, and the Cochrane Central Register of Controlled Trials were searched from 1950, 1966, and 1800, respectively, through June 30, 2006. MEDLINEand the Cochrane Central Register of Controlled Trials were searched using the medical subject heading terms heparin, venous thrombosis, and pulmonary embolism. EMBASE was searched using the EMTREE keywords heparin and venous thromboembolism. These databases were also searched using the term factor Xa inhibitor. All searches were restricted to trials that involved humans and were published in English. In addition, the reference lists of all relevant trials were hand searched.

Only prospective randomized controlled trials were considered for inclusion in this meta-analysis. Studies were eligible if they compared (1) UFH with the control, (2) LMWH with the control, (3) LMWH with UFH, or (4) a selective factor Xa inhibitor with a placebo. Studies were considered appropriate for inclusion only if they involved hospitalized medical patients and reported the occurrence of DVT, PE, and/or mortality after the administration of therapy.

Studies with fewer than 30 patients were excluded. Studies were also excluded if they examined thromboembolism in surgical, trauma, or critical care patients only or if the study authors did not provide a subgroup analysis for medical patients regarding the occurrence of thromboembolism. In addition, trials that studied thromboembolism associated with central venous catheters in patients with cancer and trials that involved patients admitted to intensive care units were excluded.

Study selection, validity assessment, and data abstraction were performed by 2 independent reviewers (L.W. and S.W.) in an unblinded standardized manner. Two other investigators (H.K. and J.S.) were consulted whenever the need arose for further discussion about the eligibility of a trial for inclusion in this meta-analysis or about results reported by individual studies. This process was undertaken in accordance with the guidelines of the Quality of Reporting of Meta-analyses statement.¹⁸

We were able to conduct a meta-analysis for a particular outcome if it was investigated in 3 or more trials. DVT, PE, mortality, and total bleeding were analyzed in studies that compared LMWH with the control, UFH with the control, and LMWH with UFH. Results of all studies that compared UFH, LMWH, or factor Xa inhibitor with the control were combined to produce an analysis of prophylaxis vs no prophylaxis for these outcomes. Studies that compared LMWH with the control and LMWH with UFH had also provided further details regarding other outcome measures, which have been analyzed in this investigation: major bleeding, minor bleeding, thrombocytopenia, and injection site hematoma. In addition, UFH dosages of 5000 U twice daily and 5000 U 3 times daily were analyzed separately by comparison with the control for the outcome of DVT.

The definitions of outcome measures were considered to be those provided by the authors of the various studies with the exception of total bleeding and major bleeding. Intracerebral and intracranial hemorrhages and hemorrhagic transformations were included in the definition of total bleeding and major bleeding episodes. The outcome of total bleeding includes major, minor, and fatal bleeding episodes but not subcutaneous injection site hematomas.

The raw data results reported by the authors of the studies were used in this meta-analysis. Where the required values were not explicitly and unambiguously reported by the authors, the number of participants in the study was assumed to be the number who started treatment for all outcomes analyzed, except for DVT. In the case of DVT, the total number of patients was deemed to be the number who completed the trial as set out in the individual study protocols for that outcome. The reason for this discrepancy is that in some of the trials, several randomized patients did not undergo assessment for DVT. Therefore, the proportion of patients who underwent investigation who were found to have DVT would more closely reflect the true rate of DVT than the proportion of all patients who started treatment.

Formal quantitative data synthesis was undertaken in a blinded manner by 1 investigator (S.J.H.) in consultation with 2 other investigators (L.W. and S.W.). Data analysis was conducted with STATA statistical software, version 8.2 (StataCorp, College Station, Texas) using the Mantel-Haenszel fixed-effect method and the DerSimonian and Laird random-effects method^19,20 of meta-analysis for binary outcomes. The fixed-effects model calculates an average of the outcome statistic from each study, whereas the random-effects model additionally considers the variability among the studies analyzed. Results were presented as estimates of relative risk (RR) for each outcome measure, with 95% confidence intervals (CIs), in which relative weights were assigned to each study on the basis of treatment group size and number of observed events.^19,20 Heterogeneity χ² tests were performed in each analysis, and statistical significance was assumed at the .05 level via the use of statistical 2-way z tests. Sensitivity analyses were performed to assess the impact of each individual trial on the final pooled estimate for each outcome measure. Funnel plots were constructed via RevMan Analyses, version 1.0.5 (the Nordic Cochrane Centre, Copenhagen, Denmark) to investigate the potential for publication bias.

Our search for studies conducted in accordance with the Quality of Reporting of Meta-analyses statement¹⁸ is summarized in Figure 1. We identified 936 potentially suitable articles. Of these, 54 examined the relevant end points in appropriate patient groups. Nine of these trials^21-29 were excluded because they did not compare (1) UFH with the control, (2) LMWH with the control, (3) LMWH with UFH, or (4) a selective factor Xa inhibitor with a comparator. Four studies^30-33 were excluded because they were not randomized controlled trials, and 2 studies^34,35 were excluded because of potential biases from the study randomization process. One article³⁶ was excluded because it was an abstract that outlined research later published in another journal.³⁷ One additional study³⁸ was excluded because it was not possible to extract sufficient data from the published article for statistical analysis to be performed (the number of patients in each group who experienced an event and the number of patients randomized to each group were required). The remaining 36 trials^9,37,39-72 were included in this meta-analysis (Table 1).

Among the 36 trials included in this meta-analysis, 4 different comparisons of therapy were studied. Fourteen trials compared UFH with the control,^50-63 11 trials compared LMWH with the control,^39-49 10 compared LMWH with UFH,^9,37,64-71 and 1 compared fondaparinux sodium with placebo⁷² (Table 2).

Several different modes of investigation were used to diagnose DVT among the trials. Venography,^{9,37,40-42,47-49,66,67,69,70,72} iodine 125 fibrinogen scanning,^{39-41,50-60,65-68} impedance plethysmography,^40,64,66 and ultrasonography^{9,42,47,49,61,64,71} were all used. Alternatively, in some cases DVT was diagnosed either clinically^46,57,64 or at autopsy.^37,42,62,70 In 1 study,⁴³ the method used to diagnose reported DVT was not specified.

Methods used to diagnose PE were ventilation and perfusion lung scanning,^{9,37,40,47,49,57,58,68-72} pulmonary angiography,^{9,37,47,49,68-72} spiral computed tomography,^47,49,71,72 magnetic resonance imaging,⁴⁹ and chest radiography.⁷⁰ In some studies, PE was diagnosed clinically^{41,50,51,54,56,57} or at autopsy.^{37,39,47,49,59,62,68-70,72} The method used was not specified in some studies,^{42,45,46,48,53,60,63,67} whereas in others descriptions such as “radiographic evidence” or “radiological evidence” were used.^50,51,54

The way in which mortality was reported differed among trials. Although most studies^{37,39,41-44,46-49,53,54,63-66,68,70,71} stated the number of patients who died during the treatment or trial period, others were less clear in their definition of mortality or provided only data that incorporated deaths both during and outside this period.

Data on total bleeding, major bleeding, and minor bleeding episodes were reported in several different ways across the individual studies. Some trials reported the number of patients who experienced these end points,^{37,39-41,43,45-47,50,57,58,62,65-72} whereas others described the number of events or episodes observed.^{9,42,48,49,63} When possible, we used data for the number of patients who experienced the relevant outcomes. Furthermore, in their results sections, some authors did not provide data for the total number of patients who experienced bleeding yet reported subcategories of this outcome separately.^{9,37,42,43,46,49,66,67,72} When this occurred, we added the patients or episodes into the different subcategories to obtain a combined figure for the purposes of our analysis of total bleeding. Major and minor bleeding episodes were also occasionally reported in terms of subcategories.^43,46

Unless otherwise stated, reported figures are discussed in the context of the fixed-effects mode because of no major apparent difference with results obtained from the random-effects model.

Pooled results demonstrated reductions in the risk of DVT (RR, 0.33; 95% CI, 0.26-0.42) and PE (RR, 0.64; 95% CI, 0.50-0.82) among those receiving UFH. No mortality difference was seen between the UFH and control groups (RR, 0.95; 95% CI, 0.88-1.02). Therapy was associated with an increased risk of total bleeding (RR, 3.11; 95% CI, 2.44-3.96) (Figure 2).

A UFH dosage of 5000 U 3 times daily was associated with a greater reduction in the risk of DVT than a UFH dosage of 5000 U twice daily (RR, 0.27; 95% CI, 0.20-0.36; and RR, 0.52; 95% CI, 0.28-0.96, respectively). When the random-effects model was used, the decreased risk of DVT associated with a UFH dosage of 5000 U twice daily became statistically nonsignificant (RR, 0.41; 95% CI, 0.10-1.73).

Compared with the control, LMWH was associated with a reduced risk of DVT (RR, 0.56; 95% CI, 0.45-0.70) and PE (RR, 0.37; 95% CI, 0.21-0.64). An increased risk of total bleeding (RR, 1.51; 95% CI, 1.31-1.74), major bleeding (RR, 1.92; 95% CI, 1.32-2.78), minor bleeding (RR, 1.40; 95% CI, 1.17-1.67), and injection site hematoma (RR, 2.04; 95% CI, 1.06-3.93) was observed with therapy (Figure 3). When the random-effects model was used, the increased risk of major bleeding and injection site hematoma became statistically nonsignificant. No difference in mortality (RR, 1.02; 95% CI, 0.88-1.19) or thrombocytopenia (RR, 1.10; 95% CI, 0.69-1.77) was found between the 2 groups.

Compared with UFH, LMWH was associated with a reduced risk of DVT (RR, 0.68; 95% CI, 0.52-0.88) and injection site hematoma (RR, 0.47; 95% CI, 0.36-0.62). No statistically significant differences were observed between the 2 agents with respect to PE (RR, 0.57; 95% CI, 0.25-1.34), mortality (RR, 1.16; 95% CI, 0.85-1.59), total bleeding (RR, 0.83; 95% CI, 0.60-1.14), major bleeding (RR, 0.77; 95% CI, 0.50-1.20), minor bleeding (RR, 0.61; 95% CI, 0.34-1.10), or thrombocytopenia (RR, 0.25; 95% CI, 0.05-1.16) (Figure 4).

Prophylaxis with UFH, LMWH, or fondaparinux was associated with a reduced risk of DVT (RR, 0.45; 95% CI, 0.39-0.53) and PE (RR, 0.57; 95% CI, 0.45-0.72) and an increased risk of total bleeding (RR, 1.90; 95% CI, 1.69-2.14). Prophylaxis did not have an effect on mortality (RR, 0.95; 95% CI, 0.89-1.02).

When 2 trials^59,63 were singly removed in turn from the analysis of UFH vs the control for the outcome of PE using the random-effects method, the reduction in risk observed with UFH became statistically nonsignificant. Analysis of the UFH dosage of 5000 U twice daily vs the control for DVT was also influenced by the removal of 2 trials^53,60 when singly removed, with the association no longer significant in either the fixed-effects or random-effects method. Furthermore, for LMWH vs the control, removal of 1 study⁴³ made the result for major bleeding significant using the random-effects method, and removal of a different study⁴⁶ made the outcome for minor bleeding no longer significant when the random-effects method was applied.

Funnel plots showed some asymmetry, indicating the potential for publication bias (Figure 5).

The trial that met the eligibility criteria for inclusion in this meta-analysis found that fondaparinux was effective in the prevention of asymptomatic and symptomatic VTE events, with the frequency of major bleeding being similar for both fondaparinux- and placebo-treated patients.⁷²

This meta-analysis has shown that UFH and LMWH are both associated with a reduced risk of VTE in medical patients, with LMWH being more effective in preventing DVT than UFH when considering trials that directly compared the 2 agents. The UFH dosage of 5000 U 3 times daily was more effective than the UFH dosage of 5000 U twice daily in reducing the risk of DVT.

Despite the observed reduction in VTE events, thromboprophylaxis did not affect mortality. This result may be accounted for by several factors. The patients who participated in the trials were generally unwell, with multiple comorbidities and a high in-hospital mortality rate (up to 8.19% in our analysis). It is therefore likely that a large proportion of patient deaths were attributable to causes other than VTE events. Furthermore, it is likely that many of the DVTs detected by the investigators either did not embolize at all or did not do so during the study period. It is also possible that a considerable proportion of the reported PEs were not fatal events. Although UFH and LMWH were associated with an increased risk of bleeding, it is unlikely that these episodes, many of which were minor, led to an increase in fatalities that would offset the reduction in mortality due to VTE. This theory is substantiated by the minimal reporting of fatal hemorrhagic episodes in the individual trials.

In contrast to our study, a retrospective database analysis⁷³ found that thromboprophylaxis reduces mortality. Several possible explanations exist for this discrepancy between results. Notably, randomization was not undertaken in the database analysis, and although results were adjusted for age, sex, and severity of illness, many other factors that contribute to VTE risk were not taken into account. Furthermore, as acknowledged by the investigators of the database analysis, the results may have been affected by treatment bias. For example, as the authors explained, some patients may not have received prophylaxis because their prognosis was poor and physicians believed it would not provide any benefit. Additionally, unlike in the retrospective study, in many of the trials included in our meta-analysis, patients were routinely screened at an early stage of DVT. Treatment of detected DVTs, when undertaken, would likely have reduced the number of PEs and deaths that occurred, potentially minimizing a difference in mortality between the prophylaxis and control groups.

A limitation of our analysis is that the patient population is not homogeneous. Participants in the trials had a diverse range of medical conditions and risk factors for VTE. However, although not ideal, we believe it acceptable to pool results from the various trials given the natural diversity of patients in a general medical ward. One must nevertheless be wary of the heterogeneity of the patient group analyzed when applying the results of this analysis to particular patient groups in the clinical setting. A further limitation of this study is that the type of LMWH used was not consistent among the trials. Any potential differences in efficacy or safety among these agents would not be reflected in the pooled analyses.

A meta-analysis⁷⁴ published in 2000 investigated the efficacy and safety of pharmacological agents used for VTE prophylaxis in medical patients. The authors of that meta-analysis stated that their study lacked sufficient power to detect a difference in efficacy between LMWH and UFH. In contrast to our meta-analysis, the 2000 study found that LMWH reduced the risk of major bleeding compared with UFH. The inconsistency of definitions of major bleeding in the individual trials analyzed in both meta-analyses could potentially contribute to the contrasting results observed between the 2 analyses. Furthermore, our study differs from the previous meta-analysis because it considers patients with acute myocardial infarction and ischemic stroke, who represent an important cohort of medical patients, and includes several large trials that have been completed since the 2000 publication.

Our meta-analysis shows that UFH and LMWH reduce the risk of VTE, with LMWH being more effective in preventing DVT when the 2 agents are directly compared. Our results indicate that if UFH is to be used, a dose of 5000 U 3 times daily is preferable to 5000 U twice daily. We believe that routine prophylactic anticoagulation has an important place in the medical setting. Although such therapy may not necessarily decrease mortality among hospitalized medical patients, it will reduce the occurrence of DVT and PE and therefore the burden of illness currently caused by these events.

Correspondence: Henry Krum, MBBS, PhD, FRACP, National Health and Medical Research Council Centre of Clinical Research Excellence in Therapeutics, Department of Epidemiology and Preventive Medicine, Faculty of Medicine, Nursing, and Health Sciences, Monash University, Third Floor, Burnet Tower, Alfred Medical Research and Education Precinct, 89 Commercial Rd, Melbourne, Victoria 3004, Australia (henry.krum@med.monash.edu.au).

Accepted for Publication: February 2, 2007.

Author Contributions:Study concept and design: L. Wein, S. Wein, Haas, Shaw, and Krum. Acquisition of data: L. Wein and S. Wein. Analysis and interpretation of data: L. Wein, S. Wein, Haas, Shaw, and Krum. Drafting of the manuscript: L. Wein, S. Wein, Haas, Shaw, and Krum. Critical revision of the manuscript for important intellectual content: L. Wein, S. Wein, Haas, Shaw, and Krum. Statistical analysis: L. Wein, S. Wein, and Haas. Administrative, technical, and material support: Haas. Study supervision: Haas, Shaw, and Krum.

Financial Disclosure: None reported.

Funding/Support: This study was supported by a Centre of Clinical Research Excellence grant from the National Health and Medical Council of Australia.