The flow of articles through the literature search and screening process is illustrated in Figure 2. Our search for systematic reviews (SRs) identified 162 unique citations from a combined search of MEDLINE via PubMed (n=117), Embase (n=42), and the Cochrane Database of Systematic Reviews (n=3). Manual searching of included study bibliographies and review articles added 20 more citations for a total of 182 unique citations. After applying inclusion and exclusion criteria at the title-and-abstract level, 47 full-text articles were retrieved and screened. Of these, 38 were excluded at the full-text screening stage, leaving 9 articles (representing 9 unique studies) for data abstraction. After further review, we excluded three systematic reviews3234 because they reviewed only one drug of interest and all of the primary studies included in these systematic reviews were already represented in another, more comprehensive included review. Thus, the final set of articles used in this evidence report comprises six systematic reviews.

Figure 2. Literature flow diagram.

Figure 2

Literature flow diagram. a Search results for systematic reviews from PubMed (117), Embase (42), Cochrane (3), previous database (14), and manual (6) were combined. b Cao, 2010; Huang, 2011; and Turun, 2011.

Appendix B provides a complete listing of published articles excluded at the full-text screening stage, with reasons for exclusion. We did not search for randomized controlled trials (RCTs) currently underway, as we relied on methods in the included systematic reviews to ascertain publication bias. We grouped the studies by key question (Figure 2).


We identified six recent, good-quality SRs9,2023,35 that were relevant to our KQs (Table 5). All of the SRs compared newer oral anticoagulants with other drug classes used for thromboprophylaxis in THR or TKR (KQ 1), although one considered only safety outcomes such as major bleeding.35 Two of the six SRs compared one newer oral anticoagulant with another (KQ 3) though all results were based on indirect comparisons; i.e., through common comparison with enoxaparin.21,23 Only one SR9 compared a pharmacological agent plus mechanical modality versus mechanical modality alone (KQ 2). Five of the six SRs included trials examining thromboprophylaxis for both THR and TKR, while one also included hip fracture surgery.9 Characteristics of the SRs are summarized in Table 5; detailed quality assessments are presented in Appendix D.

Table 5. Characteristics of included systematic reviews.

Table 5

Characteristics of included systematic reviews.

Search dates ranged from May 2009 to December 2011. All literature search strategies included MEDLINE, and all but one21 included some aspect of the Cochrane Library. Other databases or sources of information were meeting abstracts (5), Embase (3), regulatory Web sites (4), clinical trial registries (3), and the Center for Reviews and Dissemination (1). Four studies9,2022 also involved a manual search of the bibliographies of exemplary primary articles. The searches were limited only to RCTs in four of the SRs. One included SRs,22 and one included observational studies of more than 750 subjects9 in addition to RCTs. Language limits were used in only two of the studies.

All reviews assessed the quality of included trials. Overall trial quality was judged to be good, with the most common quality problems being unclear allocation concealment and incomplete reporting of outcome data. Publication bias was assessed most commonly with funnel plots, which did not indicate any publication bias that would favor newer oral anticoagulants. All studies conducted random-effects meta-analyses, but specific strategies varied. Two SRs compared drug classes as a whole (for example, FXa inhibitors versus LMWH9,20), while the other four SRs compared individual drugs, some analyzing THR and TKR studies separately. All of the SRs performed meta-analysis using direct comparisons, and two also provided indirect comparisons.21,23 Every SR except one21 evaluated major bleeding using the same definition: bleeding that was fatal, involved a critical organ, required reoperation, or where bleeding was associated with a fall in hemoglobin level of at least 2 g/dL or required infusion of 2 or more units of whole blood or packed cells. Our other prespecified primary outcomes—all-cause mortality, symptomatic DVT, and nonfatal PE—were reported in three of the SRs.9,20,22 It was difficult to evaluate the other SRs21,23,35 due to the individuality of the definitions given for the outcomes reported, many of which were composite outcomes. For example, in the study by Loke et al.,21 the authors state the composite primary outcome as “total VTE” and define it as “DVT, non-fatal PE and all-cause mortality” (emphasis added). The study also reports bleeding as a combination of major bleeding (using the standard definition given above) and “clinically relevant non-major bleeding.”

Three of the SRs were unfunded and reported no conflicts of interest.21,23,35 One SR was unfunded but did report a conflict of interest for one author.20 Two SRs were funded by government agencies.9,22


Information on the populations studied was very limited in all of the included SRs (Table 6). The number of primary articles included ranged from 5 to 45; total sample size ranged from just over 19,000 to almost 240,000, but two articles did not report the total number of subjects. Females made up approximately 50 to 75 percent of the population when reported. Mean age ranged from 55 to 68 years in the 5 studies reporting age. Weight or body mass index was reported in four studies and indicated most subjects were moderately overweight to slightly obese. Risk factors for VTE were limited to a prior history of VTE (two studies) or history of cancer (one study); only a small proportion of patients had one of these risk factors. No other risk profiles or population characteristics were reported, and no study reported whether or not Veterans were included in the sample. However, our review of the primary studies found that no studies specifically included Veterans or were conducted at VA medical centers.

Table 6. Characteristics of patient samples.

Table 6

Characteristics of patient samples.

KEY QUESTION 1. For patients undergoing total hip or total knee replacement, what is the comparative effectiveness of newer oral anticoagulants and standard drug classes (low molecular weight heparin, injectable FXa inhibitors, unfractionated heparin, warfarin, aspirin) on the incidence of symptomatic, objectively confirmed venous thromboembolism (VTE), other VTE events, total mortality, and bleeding outcomes?

Key Points

  • For all-cause mortality and nonfatal PE, there were no important differences between oral FXa inhibitors and LMWH (high strength of evidence). Using a base rate of 9 events per 1000 patients with LMWH, FXa inhibitors were associated with lower symptomatic DVT (4 fewer events per 1000 patients; 95% CI, 3 to 6). Overall, FXa inhibitors were associated with an increased risk of major bleeding, but major bleeding did not differ importantly at low to moderate doses (moderate strength of evidence). Based on subgroup analyses, there was not a consistent pattern of differences in treatment effects for THR and TKR.
  • There were fewer studies evaluating oral DTIs than oral FXa inhibitors; all trials compared dabigatran with enoxaparin. Although estimates of effect were often imprecise, there were no significant differences between oral DTIs and enoxaparin for any major outcome.
  • Neither oral FXa inhibitors nor DTIs have been compared directly with adjusted-dose warfarin, oral antiplatelet drugs, or unfractionated heparin in existing SRs.

We identified six good-quality SRs9,2023,35 that evaluated thromboprophylaxis using newer oral anticoagulants versus LMWH. For each comparison, we focus our discussion on the review having the most recent search date and comprehensive analysis, and which reported our prespecified outcomes of interest. Other reviews are described briefly when findings differed importantly or additional analyses provided relevant results.

Effects of Oral FXa Inhibitors Compared With Low Molecular Weight Heparin

A good-quality SR20 (search date December 2011) included 22 RCTs and a total of 32,159 patients that compared FXa inhibitors with LMWH for surgical thromboprophylaxis. Eleven of the included studies were on hip replacement, 10 were on knee replacement, and 1 was on either procedure. FXa inhibitors included apixaban (four studies), rivaroxaban (eight studies), edoxaban (four studies), YM150 (two studies), and LY1517717, TAK442, razaxaban, and betrixaban (one study each). Of these drugs, only rivaroxaban is currently available in the United States. In the majority of trials, the European-approved dose of enoxaparin, 40 mg daily, was the comparator instead of the U.S.-approved dose, 30 mg twice daily. The duration of thromboprophylaxis was 14 days or less in all but 4 trials. Patients were followed for less than 14 days in 9 trials, 30 to 70 days in 12 trials, and up to 90 days in one trial. In addition to a random-effects meta-analysis of drug class comparisons, this sophisticated review performed a multiple-treatment-comparison meta-analysis to evaluate effects of FXa dose, and sensitivity analyses to examine the effects of missing outcomes. Pooled estimates of effect were presented as summary odds ratios and summary risk differences. In addition, risk differences were estimated by applying the relative risk reduction from meta-analysis to the baseline risk estimated from a large cohort study.

This SR by Neumann et al.20 found high strength of evidence suggesting no important difference between oral FXa inhibitors and LMWH for all-cause mortality (OR 0.95; 95% CI, 0.55 to 1.63; I2=43%) and nonfatal PE (OR 1.07; CI, 0.65 to 1.73; I2=35%) in patients undergoing THR or TKR. However, high strength of evidence indicated that the risk of symptomatic DVT is decreased by 4 events for every 1000 patients treated using FXa thromboprophylaxis compared with LMWH (OR 0.46; CI, 0.30 to 0.70; I2=0%). There was moderate strength of evidence because of inconsistency, suggesting that the risk of major bleeding may be increased with oral FXa inhibitors compared with LMWH (OR 1.27; CI, 0.98 to 1.65; I2=55%). This finding represents an increase of 2 major bleeding events per 1000 patients treated with FXa, for 1 to 5 weeks compared with LMWH. The pooled effect estimate of bleeding that led to reoperation also was increased (OR 1.62; CI, 0.82 to 3.19; I2=1%), but the confidence interval included the possibility of a chance association.

In a subgroup analysis, higher doses of FXa inhibitors, but not intermediate or lower doses, were associated with increased risk of bleeding (OR 2.50; 95% CI, 1.38 to 4.53; p=0.02). The authors did not report the drug doses used for this subgroup analysis. Total daily doses were reported for the primary studies and ranged from 5 to 20 mg for apixaban and 5 to 60 mg for rivaroxaban. In an analysis that adjusted for FXa dose, there was no significant difference in thrombotic or bleeding outcomes for different FXa inhibitors. Sensitivity analyses that accounted for missing outcomes did not differ appreciably from the main analyses. Thus this SR concluded that while there is no important difference between low-dose oral FXa inhibitors and LMWH for the outcomes of all-cause mortality, nonfatal PE, and major bleeding, there is a small absolute reduction in symptomatic DVT events (4 fewer events per 1000 patients treated). However, most studies included in this SR reported bleeding as a composite outcome and did not include details; this introduces uncertainty about the importance of the reported bleeding events. Other limitations of the included trials in this SR were (1) missing outcomes for 3 to 41 percent of randomized patients, (2) the short duration of followup in many trials, (3) the nonstandard dosing of enoxaparin, and (4) the short duration of prophylaxis in patients undergoing THR.

The other SRs9,2123,35 were generally in agreement with the results and conclusions of Neumann et al. Where disagreements occurred, they were mainly due to different outcomes (e.g., composite outcomes), differences in approach to data analysis (separate analyses for each drug), and fewer included studies due to earlier search dates and more restrictive inclusion criteria (e.g., only FDA-approved drugs). Also, most SRs reported on outcomes by individual new oral anticoagulants, whereas those by Neumann et al. and Sobieraj et al. reported on outcomes by drug class.

We summarize below the notable findings from these other SRs:

  • In a good-quality review23 that separately analyzed the effects of apixaban (4 trials) and rivaroxaban (8 trials), both drugs were associated with lower symptomatic DVT than enoxaparin (apixaban, RR 0.41; 95% CI, 0.18 to 0.95, and rivaroxaban, RR 0.40; CI, 0.22 to 0.72). Symptomatic VTE (DVT or PE) was decreased with rivaroxaban (RR 0.48; CI, 0.31 to 0.75; I2=5%) but not apixaban (RR 0.82; CI, 0.41 to 1.64; I2=40%). All-cause mortality was not reported as a separate outcome. Symptomatic PE and major bleeding did not differ significantly from LMWH, but confidence intervals for these estimates were wide and included the potential for clinically important differences. Notably, to increase the consistency of outcome definitions and results, major bleeding rates for the RECORD studies of rivaroxaban3639 were analyzed using data reported to the FDA—a definition that included wound bleeding. Subgroup analyses showed no differences in treatment effect by type of surgery (THR vs. TKR) for symptomatic VTE or clinically relevant bleeding.
  • In a review limited by the exclusion of rivaroxaban,9 the pooled effect from four RCTs comparing LMWH with FXa inhibitors did not show a significant difference in major bleeding leading to reoperation (OR 0.67; 95% CI, 0.28 to 1.61).
  • In a review limited by the exclusion of apixaban,21 the risk of hemorrhage (major and clinically relevant nonmajor bleeding) did not differ significantly for rivaroxaban compared with LMWH (RR 1.26; 95% CI, 0.94 to 1.69; I2=28%). Hemorrhage was defined as major bleeding leading to death, reoperation, blood transfusion of two or more units, a drop in hemoglobin level of more than two g/dL, or bleeding into a critical organ. In contrast to the review by Gómez-Outes et al., published rates of bleeding rather than rates reported to the FDA (that included wound bleeding) were used for these analyses.
  • A report on adverse outcomes by type of surgery compared two oral FXa inhibitors with enoxaparin.35 There was a lower risk of major bleeding with apixaban compared with LMWH in TKR (RR 0.56; 95% CI, 0.32 to 0.96) but not in THR (RR 1.22; 95% CI, 0.65 to 2.26). Major bleeding events were not different with rivaroxaban treatment compared with LMWH in both types of surgeries. Subgroup analysis showed an increased risk of bleeding with the 30-mg twice-daily dosing regimen of LMWH compared with the 40- mg once-daily dose.

Effects of Direct Thrombin Inhibitors Compared with Low Molecular Weight Heparin

Only four SRs9,2123 included comparisons of dabigatran—the only available DTI—with standard thromboprophylaxis using LMWH. A good-quality SR (search date April 2011) included 4 trials involving 12,897 patients that compared dabigatran with enoxaparin for thromboprophylaxis of THR or TKR.23 The surgical procedure was THR and TKR in two trials each. In three trials, the comparator was enoxaparin at 40 mg daily, and in one trial the dose was 30 mg twice daily. The duration of thromboprophylaxis was 15 days or less in the two TKR studies and 28 to 35 days in the two THR studies. The duration of followup was approximately 3 months. Three studies used a three-arm design; the dabigatran 150 mg and dabigatran 220 mg treatment arms were combined for meta-analysis. The two-arm trial evaluated dabigatran 220 mg, a dose that is not approved by the FDA. All-cause mortality was not reported as a separate outcome.

In a random-effects meta-analysis, the risk of symptomatic PE (RR 0.69; 95% CI, 0.31 to 1.54; I2=NR) and symptomatic DVT (RR 0.82; CI, 0.17 to 3.99; I2=NR) did not differ between dabigatran and enoxaparin.23 Similarly, there was no statistically significant difference in symptomatic VTE (DVT and PE), but treatment effects differed substantially across studies (I2=73%). Clinically relevant bleeding events (major bleeding and clinically relevant nonmajor bleeding) were not different with dabigatran treatment (RR 0.94; CI, 0.58 to 1.52; p=0.79). In subgroup analyses, there was no statistically significant interaction between type of surgery and effects on symptomatic VTE or clinically relevant bleeding.

Three other SRs9,21,22 reported additional outcomes, including mortality, major bleeding, and bleeding leading to rehospitalization, summarized below:

  • The SR by Ringerike et al.22 included an additional study (BISTRO-II)40 of patients undergoing either THR or TKR, but anticoagulation was given for 7 days only. The SR by Sobieraj et al.9 evaluated an injectable DTI (desirudin) but did not include the most recent trial of oral DTI,40 which was also omitted in the SR by Loke et al.21 Despite these differences in approach, mortality did not differ significantly for DTIs compared with enoxaparin in any of these analyses. Consistent with the findings by Gómez-Outes et al.23 for clinically relevant bleeding, major bleeding did not differ between drug classes when analyzed by surgical procedure22 or in aggregate.9,21
  • Sobieraj et al.9 found no significant difference between LMWH and dabigatran for bleeding leading to rehospitalization (RR 1.27; 95% CI, 0.43 to 3.75; moderate strength of evidence).

Other Comparisons of Interest

Only one good-quality SR by Sobieraj et al.9 (search date May 2011) addressed drug class comparisons between older antithrombotics. We summarize results for key drug class comparisons and outcomes below.

Low molecular weight heparin versus vitamin K antagonists. Sobieraj et al.9 reported on the comparative effects of LMWH thromboprophylaxis versus adjusted-dose warfarin. LMWHs included enoxaparin (30 mg every 12 hours) and logiparin. Other details such as duration of treatment and duration of followup were not reported uniformly for the included trials. Depending on outcomes, 3 to 7 trials were included in the meta-analyses. There was no significant difference in mortality (OR 0.79; 95% CI, 0.42 to 1.50; I2=0%), nonfatal PE (OR 1.00; CI, 0.20 to 4.95; I2=NR), or symptomatic DVT (OR 0.87; CI, 0.61 to 1.24; I2=28.4%). The risk of major bleeding was significantly higher in the LMWH treatment group (OR 1.92; CI, 1.27 to 2.91; I2=0; high strength of evidence).

Oral FXa inhibitors versus unfractionated heparin. Sobieraj et al.9 reported on the comparative effects of oral FXa inhibitors versus unfractionated heparin. There were no RCTs comparing oral or injectable FXa inhibitors with unfractionated heparin. One observational study compared an injectable FXa inhibitor (fondaparinux) with unfractionated heparin; drug doses were not reported. The injectable FXa inhibitor was associated with lower mortality compared with unfractionated heparin. The risk of major bleeding was found to be increased in the unfractionated heparin treatment group compared with the injectable FXa inhibitor group (OR 1.27; 95% CI, 1.06 to 1.52). Effects on symptomatic DVT and nonfatal PE were not reported.

Low molecular weight heparin versus oral antiplatelet agents. Sobieraj et al.9 reported on the comparative effects of LMWH versus oral antiplatelet agents but identified no studies comparing these drug classes.

Antiplatelet agents versus vitamin K antagonists. Sobieraj et al.9 reported on the comparative effects of antiplatelet agents versus vitamin K antagonists, identifying a single RCT. Among patients undergoing hip fracture surgery, the risk of mortality was similar in both treatment arms (RR 0.98; 95% CI, 0.32 to 3.05). Nonfatal PE was evaluated, but there were no events in either treatment arm. The risk of major bleeding was also reported in this trial and did not show a statistically significant difference (RR 0.20; CI 0.03 to 1.23). In addition to this RCT, two observational studies compared aspirin prophylaxis with vitamin K antagonists in patients undergoing THR or TKR. One study showed higher mortality with aspirin prophylaxis (0.3 percent vs. 0 percent; p=0.013); the other study found no significant difference in mortality. There were no reports on symptomatic DVT or symptomatic VTE.

KEY QUESTION 2. For patients undergoing total hip or total knee replacement, what are the effects of combined pharmacological and mechanical modalities versus pharmacological treatment alone on the incidence of symptomatic, objectively confirmed VTE, other VTE events, total mortality, and bleeding outcomes?

Key Points

  • In the included SRs, no studies were identified that compared the combination of newer oral anticoagulants and mechanical thromboprophylaxis with pharmacological treatment alone.
  • Few studies have compared older antithrombotics (LMWH, oral antiplatelet agents, or unfractionated heparin) combined with mechanical prophylaxis to pharmacological or mechanical prophylaxis alone.
  • The strength of evidence is insufficient to determine the comparative effectiveness for combined pharmacological and mechanical prophylaxis compared with pharmacological prophylaxis alone for all major outcomes prioritized for this report.

One good-quality SR9 (search date May 2011) included 6 trials and a total of 995 patients that compared a combined-modality thromboprophylaxis (pharmacological and mechanical agents) with a single modality and found a paucity of data. Four of the included studies were on hip replacement, one was on knee replacement, and one included both surgeries. No trial evaluated the combination of a newer oral anticoagulant together with mechanical prophylaxis. Combination treatments included LMWH, aspirin, or unfractionated heparin together with mechanical prophylaxis. Of the six trials, the comparator was pharmacologic prophylaxis alone (4 trials), mechanical prophylaxis alone (1 trial), and both pharmacological and mechanical comparators (1 trial). Duration of followup ranged from the postoperative period to 90 days.

Three trials reported effects on mortality, but treatment effects were not pooled. Two of these trials had no mortality events; the third trial comparing the combination of aspirin plus pneumatic compression to aspirin alone found no effects on mortality (OR 7.72; 95% CI, 0.15 to 389.59), but the trial was underpowered for clinically significant differences. Two trials evaluated the effects on nonfatal PE, but there were no events in either trial. A single older trial evaluated the effects of sequential unfractionated heparin for 3 days, then aspirin together with a venous foot pump versus sequential pharmacologic prophylaxis alone or a venous foot pump alone. The combined modality had a lower risk of symptomatic DVT compared with pharmacologic prophylaxis only (RR 0.14; 95% CI, 0.01 to 1.42), but there were few events, and the confidence interval included no effect. Only one trial reported the effects of combined thromboprophylaxis (aspirin plus venous foot pump) compared with aspirin alone for major bleeding. However, there were no major bleeding events in either treatment arm. The authors concluded that there was insufficient evidence for all outcomes when comparing pharmacologic plus mechanical prophylaxis to pharmacologic prophylaxis alone, with the exception of overall DVT (including asymptomatic DVT). For overall DVT, combined treatment was more effective than pharmacologic prophylaxis alone.

KEY QUESTION 3. For patients undergoing total hip or total knee replacement, what is the comparative efficacy of individual newer oral anticoagulants on the incidence of symptomatic, objectively confirmed VTE, other VTE events, total mortality, and bleeding outcomes?

Key Points

  • No clinical trials directly compared newer oral anticoagulants with each other for thromboprophylaxis of TKR or THR.
  • The included SRs did not estimate the comparative efficacy of newer oral anticoagulants for symptomatic DVT, nonfatal PE, all-cause mortality, or surgical site bleeding.
  • Based on indirect comparisons, there were few differences between newer oral anticoagulants for the outcomes examined. Rivaroxaban was associated with more major bleeding than apixaban (RR 1.59; 95% CI, 0.84 to 3.02). In contrast, the risk of symptomatic VTE was lower for rivaroxaban than apixaban or dabigatran, but confidence intervals included the possibility of a chance association.

In the absence of direct comparisons between the newer oral anticoagulants, two good-quality SRs used indirect comparisons21,23 to analyze these drugs.

A good-quality comprehensive review23 (search date April 2011) evaluated apixaban (4 trials), dabigatran (4 trials), and rivaroxaban (8 trials) against a common comparator (enoxaparin). These indirect comparisons utilized pooled risk ratios and yielded an unbiased estimate of effect “when there is no interaction between covariates defining subgroups of patients (reflected, for instance, in different inclusion criteria in different studies) and the magnitude of the treatment effect.”41 Of the 16 trials (total of 38,747 patients), 8 were of total hip replacement and 8 of total knee replacement. Outcomes reported in the indirect comparisons included symptomatic venous thromboembolism (DVT or PE), clinically relevant bleeding (major bleeding or clinically relevant nonmajor bleeding), major bleeding, and a net clinical endpoint—defined as a composite of symptomatic VTE, major bleeding, and all-cause death. Overall, the primary trials were rated low risk of bias. Individual drug comparisons across these 4 outcomes (12 comparisons) showed only one statistically significant difference: rivaroxaban resulted in more clinically relevant bleeding compared with apixaban (RR 1.52; 95% CI, 1.19 to 1.95). The risk of major bleeding was also increased with rivaroxaban compared with apixaban (RR 1.59; CI, 0.84 to 3.02), but the difference was not statistically significant. Rivaroxaban was associated with lowest risk of symptomatic VTE compared with dabigatran (RR 0.68; CI, 0.21 to 2.23) and apixaban (RR 0.59; CI, 0.26 to 1.33), but neither comparison was statistically significant. Overall, differences in the number of VTE events were offset by the number of major bleeding episodes. Thus, there was no difference on the net clinical endpoint among apixaban, dabigatran, and rivaroxaban. The review concluded that “higher efficacy of new anticoagulants was generally associated with higher bleeding tendency. The new anticoagulants did not differ significantly for efficacy and safety.”23

The other SR provided few additional relevant findings. Similar to the review described above, Loke et al.21 (search date May 2009) used indirect analysis methods but excluded studies of apixaban, yielding a less informative analysis. In addition, a dabigatran trial published after 2009 and three rivaroxaban studies were excluded due to more restrictive eligibility criteria. Despite these differences, findings regarding rivaroxaban compared with dabigatran were generally similar. The authors concluded that rivaroxaban was superior to dabigatran in preventing VTE (RR 0.50; 95% CI, 0.37 to 0.68) although with an increased risk of bleeding (RR 1.14; CI, 0.80 to 1.64). The decreased risk of VTE with rivaroxaban was consistent across different doses of dabigatran (150 mg vs. 220 mg), different dosing regimens of enoxaparin in the control groups (30 mg twice daily vs. 40 mg once daily), and the type of surgery (THR vs. TKR).