A summary of the results with a strength of evidence rating of low, moderate, or high for Key Questions 1 through 8 of our CER can be found in Table 4. Evaluations for Key Questions 9 through 11 had insufficient strength of evidence and are not included. To see how our strength and applicability of evidence ratings were derived, please see Appendices H and I. For more detailed analysis of our results or to see results for comparisons with an insufficient strength of evidence rating, please see the results section for that Key Question. Although major orthopedic surgery is inclusive of total hip or knee replacement surgery and hip fracture surgery, the vast majority of literature evaluated hip or knee replacement surgery with very little evaluation of hip fracture surgery. No literature was found evaluating health related quality of life or post thrombotic syndrome as outcomes while harms such as bleeding leading to infection, bleeding leading to transfusion, readmission, and reoperation were rarely reported. No trials or studies were found to evaluate the comparative effectiveness of inferior vena cava filters with mechanical prophylaxis in major orthopedic surgery whereas comparative data of prophylaxis to no prophylaxis or between injectable and mechanical prophylaxis in other nonmajor orthopedic surgeries was very limited.

For several outcomes, although trials were designed to report events that occurred during the period of followup, many times there were no events which occurred. Therefore, the trial would be excluded from the pooled analysis in this report. Three Key Questions were more affected by the lack of outcomes occurring, including Key Question 4, 5, and 8. In Key Question 4, 25 percent of analyses with at least one randomized controlled trial had half or more excluded because no events occurred in the groups compared. The outcomes for which this occurred included fatal pulmonary embolism, pulmonary embolism, mortality due to bleeding, and major bleeding. In Key Question 5, the outcomes with the most comparisons which had half or more trials excluded because of no events included fatal pulmonary embolism, mortality due to bleeding, and major bleeding leading to reoperation. In Key Question 8, 21 percent of outcomes had half or more trials excluded because of no events, including fatal pulmonary embolism, mortality, mortality due to bleeding and bleeding leading to transfusion. Overall, in the majority of cases, although the trials were adequately designed to detect outcomes, the followup period was likely inadequate to capture the occurrence of events. Additionally, these outcomes were not commonly primary outcomes of the trials therefore underpowered to detect differences, which were not overcome by pooling since the events were rare.

Where applicable, we compare our results to those derived from previous meta-analyses as typified in Table 5 and Table 6. We used more recent search dates and generally used more restrictive inclusion and exclusion criteria than other meta-analyses including more stringent criteria for diagnosing deep venous thrombosis and pulmonary embolism endpoints. We also limited our evaluation to medications or mechanical devices available for use within the United States.

In Key Question 1 we limited our inclusion criteria to placebo or control arms of trials for pulmonary embolism and deep vein thrombosis outcomes and placebo, control, or mechanical prophylaxis arms for bleeding outcomes. However, there are some limitations arising from having excluded other study designs. In contemporary surgical practice, the native incidence of deep vein thrombosis events is still relatively high but pulmonary embolism and bleeding events are rarer. In total hip replacement, total knee replacement, and hip fracture surgery, respectively, the incidence of deep vein thrombosis (39 percent, 53 percent, 47 percent), proximal deep vein thrombosis (32 percent, 17 percent, --), distal deep vein thrombosis (30 percent, 30 percent, --), pulmonary embolism (6 percent, 1 percent, 3 percent), major bleeding (1 percent, 3 percent, 8 percent), minor bleeding (6 percent, 5 percent, --), major bleeding leading to reoperation (0 percent, 0 percent, --), and bleeding leading to transfusion (0 percent, 0 percent, --) are reported in clinical trials.

There was high statistical heterogeneity between trials for most endpoints which likely reflects several study characteristics. The majority of trials did not specifically define the duration of followup and implied an immediate postoperative followup, although this could vary between studies. Additionally, the followup period may not reflect the period of highest risk for venous thromboembolic events after major orthopedic surgery. The countries and ethnicities where the trials were conducted in and when or how rigorously the endpoints were assessed for also varied. Our results are similar to that of previous pooled-analyses of patients with total hip and total knee arthroplasty conducted between 1993 and 2001 where 23 to 60 percent of patients had deep vein thrombosis, 23 to 26 percent had proximal deep vein thrombosis, 2 percent had pulmonary embolism, 1 percent had major bleeding, and 3 percent had minor bleeding as shown in Table 5.^{153,159,164,169,171} Using our stringent inclusion and exclusion criteria, including more stringent definitions of outcomes, and allowing trials and studies right up to the present day are strengths of our comparative effectiveness review.

In Key Question 2, several randomized controlled trials identified through our literature search evaluated different surgical characteristics on outcomes of interest including anesthetic regimen, cemented arthroplasty, tourniquet use, limb positioning, and fibrin adhesive use. However, few trials evaluated each characteristic and subsequently did not address all major orthopedic procedures. Additionally, most trials evaluated intermediate health outcomes and did not address final health outcomes and only one trial evaluated bleeding outcomes. As such, pooling was not possible. The surgical comparison with the most identified data was general anesthesia versus regional anesthesia. The impact of general versus regional anesthesia on several measures of deep vein thrombosis (overall, asymptomatic, proximal) was favorable to neutral for regional anesthesia while distal and symptomatic deep vein thrombosis, pulmonary embolism, and major bleeding were neutral. In a previous meta-analysis of 21 studies, regional anesthesia was associated with a nonsignificantly reduced odds of pulmonary embolism and a significantly reduced odds of deep vein thrombosis with no real impact on mortality versus general anesthesia.¹⁶² We did not pool the results from our six included trials because many of the trials did not maintain similar prophylaxis regimens between groups or between genders reducing our confidence in the similarity between the groups.

Although one trial compared spinal versus epidural anesthesia on the risk of deep vein thrombosis, no events occurred in the groups compared. The other surgical characteristics were too limited to make any determinations.

Patient characteristics were primarily evaluated in multivariate regression analyses of observational studies. Few characteristics were evaluated in multiple studies and often times when a significant finding was observed, the magnitude or direction of the effect was not reported. There were no data regarding harms. Patient characteristics that were found to significantly increase the odds of symptomatic objectively confirmed venous thromboembolism in all available studies included congestive heart failure (two studies), inactive malignancy (one study), hormone replacement therapy (one study), living at home (one study), intertrochanteric fracture (one study), subtrochanteric fracture (one study), increased hemoglobin (one study), personal or familial history of venous thromboembolism, (one study), and varicose veins (one study). Patient characteristics consistently found to increase the odds of pulmonary embolism (evaluated in one study each) included age and genitourinary infection while cardiovascular disease was found to decrease the odds of pulmonary embolism. The following characteristics showed a mixed effect on deep vein thrombosis: age (two studies showed a significant increase while one study showed no effect), obesity (one study showed a significant increase while one study showed no effect), and gender (one study showed a significant increase in females while one study showed gender had no effect). Metabolic syndrome increased the odds of symptomatic deep vein thrombosis while congestive heart failure increased the odds of proximal deep vein thrombosis in the single study that evaluated each covariate.

In Key Question 3, no trials or studies were available assessing whether DVT was correlated with, or a multivariate predictor of, PE. This data may be limited because the routine use of prophylaxis may have reduced the occurrence of deep vein thrombosis and the scheduled anticoagulant treatment for deep vein thrombosis once it was detected may have diminished the percentage that developed into pulmonary embolism. In one observational study in total knee replacement surgery, the overall occurrence of pulmonary embolism and the subset with symptomatic pulmonary embolism occurred more frequently in those with deep vein thrombosis. However the data were not adjusted for confounders and we cannot discern whether these variables are correlated or colinear. While we could have allowed inclusion of other literature types to describe the relationship between DVT and PE, we felt, as did our expert panel, that what our a priori defined inclusion criteria would provide the most compelling data should that literature exist. We feel that the other types of literature would not allow us to adequately answer this question.

In Key Question 4, the comparative balance of benefits to harms for providing pharmacologic prophylaxis versus no prophylaxis is favorable. There is high evidence that pharmacologic prophylaxis versus no prophylaxis significantly decreases proximal deep vein thrombosis and moderate evidence that prophylaxis decreases the risk of proximal or distal deep vein thrombosis in patients undergoing major orthopedic surgery. There is low evidence that pharmacologic prophylaxis versus no prophylaxis significantly decreases major venous thromboembolism in patients undergoing major orthopedic surgery. Pharmacologic prophylaxis did not significantly impact pulmonary embolism in the base case analysis, although it was trending in that direction, and significantly reduced the risk of pulmonary embolism in the most stringent trials where they did not allow any background prophylaxis (such as compressions stockings) in the experimental groups. There is moderate evidence that pharmacologic prophylaxis versus no prophylaxis significantly increases minor bleeding and in a single observational study, pharmacologic prophylaxis increased the risk of reoperation. Pharmacologic prophylaxis versus no prophylaxis does not significantly impact nonfatal pulmonary embolism, mortality, symptomatic deep vein thrombosis or major bleeding in patients undergoing major orthopedic surgery. We cannot determine the impact of pharmacologic prophylaxis on other endpoints either due to a lack of data or because there were no events in either experimental group. Our results are in general agreement with the six previous meta-analyses of trials comparing pharmacologic prophylaxis versus placebo/control in patients with major orthopedic surgery (Table 6).^{160,161,168,179,184,188} Four assessed low molecular weight heparin versus placebo/control, one evaluated low molecular weight heparin or unfractionated heparin versus placebo/control, and the last compared vitamin K antagonists versus placebo. In the most recent meta-analysis comparing low molecular weight heparins versus placebo, there was a significant reduction in the odds of nonfatal pulmonary embolism with nonsignificant reductions in mortality and major bleeding. Deep vein thrombosis was not assessed in this meta-analysis but in the previous three meta-analyses was found to be significantly reduced with low molecular weight heparin use versus placebo/control. When either low molecular weight heparins or unfractionated heparin was compared with placebo/control deep vein thrombosis (overall, proximal, and deep) were significantly reduced but no significant impact on pulmonary embolism was found or death was found. In another meta-analysis, vitamin K antagonists significantly reduced pulmonary embolism and deep vein thrombosis but nonsignificantly increased major bleeding versus placebo.

The comparative balance of benefits to harms for providing mechanical prophylaxis versus no prophylaxis is possibly favorable but more data are needed to support this assumption. While mechanical prophylaxis versus no prophylaxis was not found to significantly impact proximal or distal deep vein thrombosis in patients undergoing major orthopedic surgery, the power to detect these differences is low and the strength of evidence for all outcomes was insufficient. We could not adequately assess the other outcomes because there were either no trials or the available trials had no events in either group. Given the mechanism of action for these devices, bleeding should not result from their use so benefits would likely overwhelm the risk of harms. In the only meta-analysis comparing mechanical prophylaxis (intermittent pneumatic compression and venous foot pump) versus control, the risk of deep vein thrombosis (any, proximal, and distal) and pulmonary embolism were significantly reduced and the risk of fatal pulmonary embolism and mortality were nonsignificantly reduced (Table 6).¹⁶⁰

In Key Question 5, we sought to determine the impact of therapy on numerous outcomes but were only able to discern significant differences between classes for relatively few outcomes. For the other outcomes, there was either a lack of evaluable data or no significant differences. We cannot determine if this means that there is a lack of effect versus a lack of power to show that it is significant.

Low molecular weight heparin agents, as a class, have a better comparative balance of benefits to harms versus unfractionated heparin with significantly fewer pulmonary embolisms, deep vein thromboses, proximal deep vein thromboses, major bleeding, and heparin induced thrombocytopenia events. The comparative balance of benefits to harms for low molecular weight heparins to other classes cannot be readily determined. Low molecular weight heparin agents are also superior to vitamin K antagonists at reducing measures of deep vein thromboses (any, asymptomatic, proximal, and distal) but increase major, minor, and surgical site bleeding. Since no significant differences were found in important final health outcomes, the relevance of these reductions in deep vein thrombosis needs to be considered. Low molecular weight heparin agents may be inferior to factor Xa inhibitors in terms of any, proximal, and distal deep vein thromboses but have a lower risk of major and minor bleeding. Observational data suggested low molecular weight heparin agents had decreased mortality although this was not supported by data pooled from randomized trials which showed no significant difference. The comparison of low molecular weight heparins agents to direct thrombin inhibitors is difficult because the occurrence of deep vein thrombosis is greater but the occurrence of distal deep vein thrombosis is less with low molecular weight heparin agents and while surgical site bleeding is higher with low molecular weight heparin therapy, the overall risk of serious bleeding was not significantly altered. Finally, when low molecular weight heparin agents are compared versus mechanical prophylaxis, the only significant difference is the lower occurrence for patient discomfort in the group receiving low molecular weight heparin agents. The results of previous meta-analyses are in general agreement with the findings of our comparative effectiveness review. In six previous meta-analyses, low molecular weight heparins were compared with unfractionated heparin in patients with major orthopedic surgery (Table 6) ^{152,160,161,170,187,188} There were significant reductions in deep vein thrombosis (five of six meta-analyses), proximal deep vein thromboses (two of three meta-analyses with the third showing a nonsignificant reduction), and major bleeding (one of three meta-analyses with a nonsignificant reductions in the second and third) with low molecular weight heparins versus unfractionated heparin. There was a nonsignificant reduction in mortality (two of two meta-analyses), mixed effects on pulmonary embolism (one meta-analysis showing a significant reduction, one showing a nonsignificant increase the other showing a nonsignificant decrease), and a nonsignificant increase in any bleeding with low molecular weight heparins versus unfractionated heparin. In two previous meta-analyses, the impact of dabigatran versus enoxaparin in total hip and total knee arthroplasty outcomes was assessed.^182,185 No significant differences in venous thromboembolism or major bleeding were seen. Four meta-analyses comparing vitamin K antagonists to low molecular weight heparins.^{161,168,170,181} The three meta-analyses evaluating deep vein thrombosis found a significant increase while the one of two meta-analyses evaluating major bleeding found a significant decrease and another found a nonsignificant decrease with vitamin K antagonist use versus low molecular weight heparins. One meta-analysis found a nonsignificant increase in pulmonary embolism and death while another found a nonsignificant increase in symptomatic pulmonary embolism. One meta-analysis compared the factor Xa inhibitor fondaparinux to the low molecular weight heparin enoxaparin and the odds of venous thromboembolism and proximal deep vein thrombosis was significantly reduced.

It is difficult to discern the comparative balance of benefits to harms for oral antiplatelet therapy versus mechanical prophylaxis or vitamin K antagonists. Oral antiplatelet therapy had significantly greater occurrence of any and distal deep vein thrombosis versus mechanical prophylaxis. In a controlled observational study, oral vitamin K antagonists had significantly fewer fatal pulmonary embolism events versus oral antiplatelet agents. In the only available randomized trial comparing vitamin K antagonists to oral antiplatelet agents, the same direction of effect was found suggesting vitamin K antagonist superiority but this was not significant. Mortality was higher in patients receiving aspirin versus warfarin in one observational study, was nonsignificantly trending in that direction in another observational study but showed no difference in a clinical trial. These results are characteristically similar to a previous meta-analysis which found that vitamin K antagonists nonsignificantly decreased overall deep vein thrombosis but nonsignificantly increased proximal deep vein thrombosis with no real difference in major hemorrhage (Table 6).¹⁶⁸

Unfractionated heparin, which was found to be inferior to low molecular weight heparin agents in the balance of benefits to harms, had a greater occurrence of death and major bleeding versus factor Xa inhibitors in an observational study (with no clinical trial data to support or refute the findings), had a greater occurrence of any and proximal deep vein thrombosis versus direct thrombin inhibitors, and had a greater occurrence of deep vein thrombosis versus mechanical prophylaxis. As such, it is likely inferior to factor Xa inhibitors in the balance of benefits to harms as well.

Patients receiving vitamin K antagonists had less occurrence of proximal deep vein thrombosis versus mechanical prophylaxis but with no other differences in other health outcomes or bleeding, it is hard to discern a difference in the balance of efficacy to harms between them. This is consistent with a previous meta-analysis that found nonsignificant decrease in proximal deep vein thrombosis but nonsignificant increases in mortality, asymptomatic deep vein thrombosis, and major hemorrhage (Table 6).¹⁶⁸

In Key Question 6, there were no significant differences in pulmonary embolism (any, fatal, and nonfatal), mortality, and mortality due to bleeding between modalities within low molecular weight heparin and mechanical device classes but these evaluations were based on one or two trials with either no events or a very low number of events.

The balance of benefits to harms from using enoxaparin versus another low molecular weight heparin within the class (dalteparin or tinzaparin) is similar although the amount of literature available is very limited. No difference in the occurrence of deep vein thrombosis or proximal deep vein thrombosis occurred between low molecular weight heparins (enoxaparin versus either tinzaparin or dalteparin). No significant difference was seen in asymptomatic deep vein thrombosis between enoxaparin and dalteparin, symptomatic deep vein thrombosis between enoxaparin and tinzaparin, or distal deep vein thrombosis between enoxaparin and tinzaparin. For major bleeding, two trials compared low molecular weight heparins and found no differences between enoxaparin and either dalteparin or tinzaparin. For minor bleeding, one trial found no significant difference between enoxaparin and tinzaparin for this outcome. For surgical site bleeding, two trials compared low molecular weight heparins and found no differences between enoxaparin and either dalteparin or tinzaparin. For heparin induced thrombocytopenia, one trial found no significant difference between enoxaparin and tinzaparin for this outcome.

The balance of benefits to harms for different mechanical modalities within a class cannot be determined with the current literature base. The Venaflow pneumatic compression device significantly reduced the occurrence of deep vein thrombosis or distal deep vein thrombosis versus the Kendall pneumatic compression device in the only trial but did not significantly reduce proximal deep vein thrombosis.

Intermittent compression stockings significantly reduced the occurrence of deep vein thrombosis or distal deep vein thrombosis versus graduated compression stockings but did not significantly reduce proximal deep vein thrombosis.

In the only observational study, two intermittent compression devices were compared (ActiveCare system versus Flowtron excel pump) and found to have a similar occurrence of deep vein thrombosis. Harms were not assessed in these trials or observational studies.

In Key Question 7, the balance of benefits to harms for combining a pharmacologic and mechanical strategy versus using either strategy alone in patients undergoing major orthopedic surgery cannot be determined. The use of a pharmacologic plus mechanical strategy versus either pharmacologic or mechanical prophylaxis does not significantly impact nonfatal pulmonary embolism, mortality, or deep vein thrombosis subsets (asymptomatic, symptomatic, proximal, or distal). The comparative impact of pharmacologic plus mechanical prophylaxis versus pharmacologic or mechanical prophylaxis on major or minor bleeding could not be determined. There is moderate evidence that the use of pharmacologic plus mechanical prophylaxis significantly decreases the occurrence of deep vein thrombosis versus pharmacologic prophylaxis alone. The impact of dual prophylaxis versus single modality on other outcomes cannot be determined.

For Key Question 8, the balance of benefits to harms for prolonged versus shorter term prophylaxis in patients undergoing major orthopedic surgery is favorable. The impact of prolonging prophylaxis for 28 days or longer on events was compared with prophylaxis for 7 to 10 days in patients who had major orthopedic surgery. In base case analyses, prolonged prophylaxis reduced the occurrence of symptomatic objectively confirmed venous thromboembolism, pulmonary embolism (overall and nonfatal), and deep vein thrombosis (overall, symptomatic, asymptomatic, and proximal) versus shorter term prophylaxis. While higher heterogeneity was found for symptomatic venous thromboembolism and deep vein thrombosis, the direction of effect was consistent between all of the trials. In base case analyses, prolonged prophylaxis increases the occurrence of minor bleeding and surgical site bleeding versus shorter term prophylaxis. Four previous meta-analyses compared the impact of longer versus shorter duration pharmacologic prophylaxis and are in general agreement with our present comparative effectiveness review (Table 6). ^{156,163,174,186} Decreases in deep venous thrombosis (overall, symptomatic, asymptomatic, and proximal) and venous thromboembolism were found with nonsignificant increases on minor bleeding but not major bleeding in these meta-analyses.

For Key Questions 9 and 11, there were no trials or studies that met our inclusion criteria. For Key Question 10, one trial was available for Achilles tendon rupture and for knee arthroscopy but no literature met inclusion criteria for elective spine surgery. Both of the available trials were small in size leading to limited power to detect differences between the groups compared. The comparative balance of benefits to harms for dalteparin therapy versus placebo or control is difficult to discern based on the scant data. In patients who had surgical repair of Achilles tendon rupture, the use of dalteparin versus placebo for six weeks did not significantly impact the incidence of total or proximal deep vein thrombosis. No patients developed a pulmonary embolism or major bleeding. In patients who had arthroscopic knee surgery, the use of dalteparin versus control led to significantly fewer patients with total or distal deep vein thrombosis. One patient in the dalteparin group developed a pulmonary embolism and also had a deep vein thrombosis. No patients had major bleeding and the occurrence of minor bleeding was not significantly different between the two groups.

In summary, in major orthopedic surgery, the incidence of DVT is appreciable but the risk of PE, major and minor bleeding is smaller. The balance of benefits to harms is favorable for providing prophylaxis to these patients and to extend the period of prophylaxis beyond the standard 7-10 days. The comparative balance of benefits to harms for LMWHs are superior to unfractionated heparin. Other interclass comparisons either could not be made due to lack of data, showed similarities between classes on outcomes, or had offsetting effects where benefits of one class on efficacy was tempered by an increased risk of bleeding. The balance of benefits to harms for dual pharmacologic plus mechanical prophylaxis versus either strategy alone could not be determined. We could not determine the impact of IVC filters on outcomes or the impact of prophylaxis on the nonmajor orthopedic surgeries evaluated. We present the limitations of current research as well as future research needs which may help to shape future practice and policy in this clinical area.