Approach to Outcome Measurement in the Prevention of Thrombosis in Surgical and Medical PatientsApproach to Outcome Measurement: Antithrombotic Therapy and Prevention of Thrombosis, 9th ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines FREE TO VIEW

This article provides the rationale for the Antithrombotic Therapy and Prevention of Thrombosis, 9th ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines (AT9) approach to estimating the effect of antithrombotic prophylaxis on patient-important outcomes of pulmonary embolism (PE) and symptomatic venous thrombosis.

Although some studies have limitations of lack of concealment and blinding, evidence from meta-analyses of randomized controlled trials (RCTs) strongly suggests that prophylaxis with an anticoagulant or aspirin reduces symptomatic VTE and fatal PE in medical and surgical patients (Table 1). In patients undergoing orthopedic, general, or urological surgery, unfractionated heparin (UFH) reduces the risk of fatal PE by about two-thirds¹; in patients undergoing hip or knee arthroplasty or hip fracture surgery, vitamin K antagonists (VKAs) reduce the risk of symptomatic VTE by about four-fifths²; in patients undergoing hip or knee arthroplasty, extended-duration low-molecular-weight heparin (LMWH) or warfarin reduces the risk of symptomatic VTE by about three-fifths³; in medical patients at highest risk, UFH, LMWH, danaparoid, or fondaparinux reduces the risk of PE by about two- to three-fifths^4,5; and in patients undergoing abdominal or pelvic surgery, LMWH reduces the risk of symptomatic VTE by about four-fifths.⁶ Antiplatelet therapy also is effective for the prevention of VTE in the highest-risk surgical or medical patients, reducing the risk of PE by about one-half and DVT by about three-fifths.⁷ Similar relative risk reductions are seen in trials comparing the efficacy of anticoagulant prophylaxis with placebo or no treatment based on a surrogate outcome; compared with placebo or no treatment, prophylactic anticoagulants reduce the relative incidence of silent DVT diagnosed through screening venography by 30% to 70%.⁸

The effects of antithrombotic prophylaxis on bleeding were inconsistent. In some meta-analyses, major or total bleeding was increased, whereas in others, there was no excess of bleeding (Table 1).

Collectively, the meta-analysis data indicate that prophylactic anticoagulants are effective for the prevention of patient-important VTE and that the benefit-risk trade-off justifies their use in patients who are at sufficiently high risk of symptomatic VTE. Identifying characteristics of surgical procedures and individual patients that predict VTE risk is necessary to wisely select patients for consideration of prophylaxis.

Mechanical methods of thromboprophylaxis include graduated compression stockings, intermittent pneumatic compression (IPC) devices, and the venous foot pump. Relative to anticoagulants and aspirin, these methods have the advantage of not increasing bleeding. Moreover, comparisons of these agents against no prophylaxis suggest that they are effective in reducing thrombosis.

Unfortunately, these studies and others comparing mechanical and pharmaceutical agents are relatively few and small in size.⁹ Therefore, the compelling evidence of a decrease in fatal PE that exists for anticoagulants and for aspirin does not exist for mechanical methods.

All the RCTs comparing two different antithrombotic agents have used asymptomatic DVT by mandatory venography as the primary outcome measure or as a component thereof. In general, any improvement in efficacy with any one agent was accompanied by an increase in bleeding. These studies are difficult to interpret because they do not provide information on the trade-off between benefits and risks in patient-important events.

The use of asymptomatic DVT detected by venography as a surrogate for a patient-important event is based on two premises. The first is that most thrombi start as small calf vein thrombi, which often remain asymptomatic but can grow to form symptomatic venous thrombi, which in turn can break off to cause asymptomatic PE, symptomatic PE, or fatal PE. The second is that a reduction in asymptomatic venous thrombosis by antithrombotic prophylaxis is paralleled by a similar reduction in symptomatic VTE and fatal PE. Pooled data from RCTs show that on average, the relative reduction is similar for silent DVT and patient-important VTE.¹⁰ This is true for placebo-controlled (or no-treatment) trials as well as active-controlled trials. It does not necessarily, however, mean that the relative effects of asymptomatic and symptomatic events will be similar.

Even if we assume that the reduction in asymptomatic DVT is associated with a similar relative risk reduction in symptomatic VTE, this alone would not be sufficient reason to recommend the use of one antithrombotic agent over another. Rather, the recommendation of a particular anticoagulant over another should be based on the trade-off between the absolute reduction in patient-important VTE and the absolute increase in patient-important bleeding. To complicate matters further, there is controversy (based on uncertainty) regarding the relative importance of symptomatic VTE and bleeding to the patient, to the physician, and to the health-care system.

The potential consequences of DVT are the discomfort associated with symptomatic DVT, the postthrombotic syndrome, the inconvenience and side effects of treatment, and the development of PE. The risk of postthrombotic syndrome following silent DVT is likely to be low, whereas up to 40% of patients with symptomatic DVT develop postthrombotic syndrome, which is debilitating in ∼15% of those who develop the syndrome.¹¹ The most important consequence of PE is death, although thromboembolic pulmonary hypertension, which occurs in up to 4% of patients with PE, is also important.¹²

Bleeding associated with the use of antithrombotic agents varies widely in severity. Apart from intracranial bleeding, little is known about the long-term patient-important consequences of bleeding. Of particular relevance to major orthopedic surgery, in patients receiving anticoagulants for the prevention of VTE, the incidence of bleeding into joints and the impact of wound and joint bleeding on functional patient outcomes are unknown. Specifically, there are no published data addressing the relationship between wound or joint bleeding and either wound infection or long-term joint function.

Based on our lack of knowledge about the consequences of asymptomatic thrombi and wound bleeds, it is difficult to formulate a trade-off between risk and benefits among antithrombotic agents. Nevertheless, clinicians must make decisions regarding thromboprophylaxis, and guidelines serving clinicians must provide guidance. Estimates of the frequency of symptomatic VTE and bleeding and their consequences are necessary for making appropriate recommendations.

In AT9, we considered possible strategies for estimating the absolute difference in the frequency of VTE associated with alternative approaches to antithrombotic management. To estimate the absolute benefit of one antithrombotic regimen over another in reducing VTE requires an estimate of the control group event risk and the relative risk reduction associated with the alternative or experimental antithrombotic regimen. In this section, we review the strategies we considered and discuss their relative merits. Table 2 summarizes the strategies and their key limitations.

One approach to estimating the absolute difference among VTE across management strategies is to simply use the number of symptomatic and fatal events observed (ie, the number of fatal and nonfatal PEs, the number of symptomatic DVTs). Using this strategy, estimates of both control event risk and relative effect come from symptomatic events. The implication of using this approach is that unless there is convincing evidence of a difference in symptomatic events, panelists recommend the strategy with the lowest bleeding rates.

In dealing with prophylaxis for nonsurgical patients in this guideline, Kahn et al¹³ initially considered this approach in addressing the impact of anticoagulant prophylaxis (LMWH, UFH, fondaparinux) in hospitalized medical patients.¹ They found moderate-quality evidence (rated down for imprecision) from a systematic review of four RCTs (risk ratio [RR], 0.47; 95% CI, 0.22-1.0).

This approach has the merit of simplicity. In addition, it uses events that have been observed and, thus, requires no questionable assumptions. Taking a perspective of primum non nocere, it is conservative, which some may perceive as an advantage.

The approach, however, does have a number of limitations. First, even in populations labeled as high risk, symptomatic thrombotic events in patients treated with antithrombotic agents typically are uncommon. Estimates of effect are, therefore, imprecise, and CIs are wide (as in the example from Kahn et al¹³). These wide CIs could, however, be seen as an advantage: They capture the true uncertainty associated with estimates of the effect of the intervention on patient-important events.

Second, the approach may have limitations of generalizability. Patients enrolled in RCTs are likely to be at a different risk than the populations to whom the results will be applied. If this is so, estimates of absolute difference may be misleading.

Third, estimates of absolute symptomatic events from studies that included mandatory venography may be misleading. If mandatory venography is positive, patients typically will receive antithrombotic treatment. Had venography not been undertaken and anticoagulant therapy not been administered, some of these patients likely would have developed symptomatic VTE. To the extent that this is the case, the number of symptomatic VTE and the benefit of any treatment that reduces VTE will be underestimated.

There is also the possibility of bias in the opposite direction. What might otherwise have been considered minor swelling or pain in the leg not warranting investigation could be labeled, after a positive mandatory venogram, a symptomatic leg event. What might otherwise have been considered a minor transient episode of dyspnea not warranting further investigation might be considered, after a positive mandatory venogram (with or without direct testing for pulmonary emboli), a symptomatic pulmonary embolus. To the extent that these phenomena occur, the number of symptomatic VTE and the benefit of any treatment that reduces VTE will be overestimated.

There is no evidence that allows confident estimates of the relative impact of the downward and upward biases on the estimates of symptomatic events that result from mandatory venography. One might speculate that the downward bias as a result of the treatment of asymptomatic events is likely to be greater than the upward bias as a result of labeling of clinically trivial events as symptomatic. To the extent that this is true, estimates of benefit of one treatment over another would be conservative. Depending on one’s perspective, one might view this conservatism in estimates as desirable or undesirable.

Considering all these issues, estimates of reduction in patient-important events using this strategy are likely, if they err, to underestimate the true effect. Therefore, despite their limitations, if well-conducted RCTs have been sufficiently large to yield precise estimates suggesting a reduction in symptomatic VTE, this constitutes high-quality evidence of benefit. Smaller studies with wider CIs will yield, at best, moderate confidence in estimates.

Eventually, Kahn et al¹³ determined that the baseline risk of medical patients enrolled in RCTs was too heterogeneous and used stratified baseline risks from symptomatic events from a risk assessment model.

A second approach would be to use estimates of the relative effect of the interventions under comparison from the combined end point of symptomatic and asymptomatic VTE or, if this is not available, the relative effect from asymptomatic events. One would use this relative risk estimate (often reasonably precise) rather than that derived from symptomatic events alone. One would apply this estimate of relative effect to the proportion of the absolute number of symptomatic events in the control intervention to generate a best estimate of the absolute number of events in experimental intervention.

For instance, assume that the relative risk of the combined end point of asymptomatic and symptomatic VTE pooled across relevant RCTs was 0.6, a relative risk reduction of 40%. Of the control patients, 1% experienced a symptomatic pulmonary embolus and 4% experienced symptomatic DVT (again, for both estimates, pooling across relevant studies). The absolute reductions in pulmonary embolus with the experimental treatment would therefore be 0.4% (40% of 1%, or one in 250 patients) for pulmonary embolus and 1.6% (40% of 4%, or one in ∼63 patients) for DVT.

For example, in the article addressing thromboprophylaxis in orthopedic surgical patients, Falck-Ytter et al¹⁴ used the event rate in trials of LMWH, adjusted for the effect of LMWH from previous trials of LMWH against placebo or no treatment, to generate a control group risk.² To estimate the absolute reduction in symptomatic DVT event rates with UFH, they applied the RR from the relevant RCTs for the combined end point of symptomatic and asymptomatic events (of which most were asymptomatic) to the control group risk derived from the LMWH trials. Their estimate of RR was 0.42 (95% CI, 0.36-0.5), their control group risk over the first 14 days after surgery was 1.8%, and their estimate of absolute benefit was 10 fewer events per 1,000 (95% CI, from nine fewer to 12 fewer). The very tight CI leaves no need to rate down for imprecision, but using asymptomatic events to generate that tight CI requires rating down for indirectness, and the quality evidence is, as a result, only moderate.

The advantage of this approach over the first, as the example from the orthopedic prophylaxis article¹⁴ shows, is that it provides a more precise estimate of relative risk than one would obtain from the much fewer symptomatic events. It requires, however, the assumption that the relative effect on symptomatic and asymptomatic events is the same (thus, the need to rate down for indirectness). Furthermore, it shares the following disadvantages with the first approach: It uses RCT populations that are likely to be unrepresentative of real-world practice; because the number of symptomatic events is small, the precision of the estimates of baseline control group risk of symptomatic events will be imprecise; and estimates of baseline control group risk of an asymptomatic event are subject to both the upward and the downward biases presented in the discussion of strategy 1. Because of these limitations and the uncertainty resulting from the substitution of the relative effect on symptomatic and asymptomatic VTE for the relative effect on symptomatic VTE, this approach can yield, at best, moderate confidence in estimates.

In the third approach, one would begin by identifying observational studies of risks of symptomatic VTE from relevant medical or surgical populations in which venographic or ultrasound surveillance was not undertaken. One would then apply the relative risk reductions from the combined end point of symptomatic and asymptomatic DVT from RCTs to those baseline estimates of risk. The approach is similar to the second strategy, but baseline risk estimates now come from observational studies without surveillance rather than from RCTs with surveillance. If satisfactory observational studies were not available, one could use control event risks from RCTs that did not perform venographic or ultrasound surveillance.

The advantages of this approach are threefold: Baseline risk estimates would, if one used observational studies, come from real-world populations; they would not be subject to the biases associated with treatment of asymptomatic events or with attribution of questionable symptoms to VTE after discovery of a venographic DVT; and if sample sizes were large enough, estimates of risk would be reasonably precise. These are compelling advantages.

One disadvantage would be the persisting assumption that relative effects of treatment on symptomatic and asymptomatic events are identical. A second would arise if the observational studies were conducted only on patients not receiving thromboprophylaxis. When applying risks from observational populations to comparisons of two active treatments, one would have to model results from RCTs of controlled active interventions vs no thromboprophylaxis and then apply the resulting risk estimates to the experimental VTE prophylaxis strategy.

Our judgment is that the advantages of this third strategy outweigh its disadvantages, given one crucial proviso. It requires the existence of large observational studies with a low risk of bias. To achieve a low risk of bias, observational studies would need to enroll representative populations; avoid exclusions; minimize loss to follow-up; and institute a rigorous, reproducible, and comprehensive strategy for ascertaining symptomatic VTE. The authors would also specify the number of patients receiving some sort of prophylaxis and ideally report results separately for those receiving and not receiving prophylaxis. Unfortunately, studies meeting all these criteria are rarely available. In situations in which there are large, high-quality observational studies available, if one is ready to believe that it is very likely that the relative effect on asymptomatic VTE in the particular context is very similar to the relative effect on symptomatic VTE, one could argue that it is possible for this approach to yield high confidence in effect estimates. Greater skepticism regarding application of the relative effect on asymptomatic events to symptomatic events would suggest that this strategy yields, at best, moderate confidence in effect estimates.

In the article on nonorthopedic surgery thromboprophylaxis, Gould et al¹⁵ used this strategy to address the effect of IPC on symptomatic VTE.³ The control group risk of VTE came from an observational study of 8,216 general, vascular, and urological surgery patients, most of whom received either mechanical or pharmacologic prophylaxis.⁴ The observational study enrolled a representative population, had minimal loss to follow-up, provided information about use of prophylaxis (which allowed adjustment for the effect of treatment in estimating untreated risk), and used the best available risk stratification strategy. It was limited in that it relied on recorded clinical diagnosis abstracted by trained nurses (ie, the approach to diagnosis of VTE was not systematic). Gould et al¹⁵ estimated that 26 of 1,000 moderate-risk patients would experience VTE if they received no prophylaxis. IPC lowered the risk of asymptomatic DVT substantially, with a narrow CI (OR, 0.48; 95% CI, 0.22-0.74), resulting in an estimate of 13 fewer VTE per 1,000 patients (95% CI, 7-20). Unfortunately, in addition to the consequent indirectness on estimates of relative effect coming from asymptomatic events, the RCTs also suffered from high risk of bias; therefore, the overall confidence in estimates is low.

In the nonorthopedic surgical prophylaxis article, Gould et al¹⁵ used a variant of this strategy that could potentially yield high-quality evidence. In calculating the likely effect of LMWH on reducing symptomatic VTE, their estimate of baseline risk came from the relatively high-quality observational study described in the previous paragraph and their estimate of relative effect from three RCTs that reported symptomatic events. Unfortunately, the studies were judged as suffering from a high risk of bias, and so the overall confidence in estimates was only moderate.

The fourth strategy provides an estimate of the frequency with which asymptomatic events will become symptomatic, and uses this estimate to add some proportion of the asymptomatic events to symptomatic events in generating the control group risk. It uses available data to provide this estimate.

What data should drive an informed estimate? Intuitively, one might estimate the proportion of asymptomatic events that would become, in the absence of antithrombotic therapy, symptomatic by looking at the ratio between symptomatic and asymptomatic events. Using this approach, we estimate that about one in 10 patients with asymptomatic DVT detected by mandatory venography develop symptomatic DVT.

To see how one would apply this estimate, let us take a hypothetical example in which 1% of the control patients experience symptomatic VTE and 10% experience asymptomatic VTE. If we estimate that of the asymptomatic events, one in 10 will become symptomatic, then the total symptomatic events in the control group will be 1% + (10%/10) = 1% + 1% = 2%. One then would apply the relative risk reduction associated with the composite of symptomatic and asymptomatic events to this control group risk. For example, if the relative risk reduction were 40%, the absolute reduction in events would 2% × 4/10 = 0.8%

This approach is fraught with limitations, five of which are particularly challenging:

One merit of this approach is that to the extent that the downward bias is greater than the upward bias when one uses symptomatic events alone, the approach may bring us closer to the true number of VTE events prevented by an experimental treatment. The most important limitation is the uncertainty of the best estimate, and because of this uncertainty, this approach can provide only low confidence in estimates.

We also considered a variation on strategy 4 that acknowledges the uncertainty associated both with the upward bias resulting from clinically trivial events being labeled as important because of a positive venogram and with the uncertainty in the estimate of venographic events that would become symptomatic. Instead of a single best estimate of the proportion of asymptomatic events that ultimately will be symptomatic, this approach generates upper and lower boundaries of the estimate, that is, the lowest plausible ratio and the highest plausible ratio. The available evidence suggests that this range varies from 2:1 to 30:1.

Thus, rather than using a conventional CI around the magnitude of effect, one might generate what could be called an “uncertainty interval.” We illustrate how this might work using the hypothetical example we present here for strategy 4. Of the control patients, 1% experienced symptomatic VTE, and 10% experienced asymptomatic VTE. We estimated that of the asymptomatic events, one in 10 would become symptomatic; the control event rate would be 2%; and with a relative risk reduction of 40%, the absolute reduction in events would be 0.8%. The point estimate of the relative risk reduction is associated with a CI of, say, between 10% and 70%. Assuming one boundary, 10%, the absolute reduction in risk would be 10% of 2%, or 0.2%. Assuming the other boundary of 70%, the absolute reduction in risk would be 70% of 2%, or 1.4%.

What are the consequences if we acknowledge the uncertainty around our estimate of a 10:1 ratio between asymptomatic and symptomatic events (as low as 2:1 and as high as 30:1)? Without belaboring the arithmetic, this would generate an uncertainty interval that suggests that the absolute benefit could be as low as 1.03% or as high as 4.5%.

This approach has the same fundamental advantages and disadvantages of the basic strategy 4. It has the advantage, however, of fully acknowledging the uncertainty around estimates of the ratio between asymptomatic and symptomatic events. Its disadvantage, relative to strategy 4, is that some may find the approach conceptually challenging, and others may be uncomfortable with the width of the uncertainty interval. We would argue, however, that this range accurately reflects the limited inferences one can make on the basis of the present evidence and the low confidence in estimates that this approach generates.

Kahn et al¹³ used this approach (almost) in only one recommendation. To estimate the impact of LMWH on PE in critically ill patients, they calculated the point estimate of the absolute reduction in asymptomatic DVT from two RCTs (22 per 1,000). For their point estimate (a reduction of four events per 1,000), they estimated that for each of five asymptomatic DVTs there would be one fewer PE. In generating a CI around this estimate (from eight fewer to one more PE), they used both 1:5 and 1:10 ratios and presented the widest CI consistent with these estimated ratios. Because of concerns about, in addition to the uncertainty about, the ratio of asymptomatic DVT to clinically significant PE, risk of bias, inconsistency, and imprecision, the data provided only very low confidence in the effect estimate. Ultimately, the authors abandoned the approach, preferring strategy 3 in which they obtained baseline risk from observational studies and relative risk reduction from asymptomatic events in RCTs.

Under different circumstances, with different bodies of evidence, any of strategies 1 to 3 might generate the highest-quality evidence. In the presence of rigorous observational studies and in the absence of large RCTs that show significant differences in symptomatic events between competing antithrombotic agents, strategy 3 will yield the highest-quality evidence. When large RCTs show convincing differences in symptomatic events between agents, strategy 1 will yield the highest-quality evidence. When neither of these conditions exist, it is likely that strategy 2 will yield the highest-quality evidence. It is unlikely that strategy 4 will ever yield the highest confidence in estimates.

In general, AT9 uses the best available evidence to generate estimates of the magnitude of intervention effects on patient-important outcomes. This is the approach we have adopted for estimating the reduction in symptomatic VTE with competing antithrombotic agents. Depending on the evidence in a particular situation, AT9 guideline panels have chosen the approach, among strategies documented here, that yields the highest confidence in estimates.

Having established best estimates of VTE and bleeding, making recommendations requires deciding on whether net benefit is optimized by administering or withholding antithrombotic prophylaxis. The relevant nonfatal events in medical and surgical prophylaxis are pulmonary embolus, DVT, and GI and surgical site bleeding. AT9 panelists¹⁷ rated the disutility (aversiveness or importance) of these events and judged them to be of similar importance to patients (DVT slightly less important; PE, GI, and perioperative bleeding very similar). Thus, considering nonfatal events alone, if an antithrombotic regimen prevents more VTE events than it causes bleeding events compared with an alternative, recommendations will favor that regimen; if therapy causes more bleeding events than it prevents VTE events, recommendations will favor withholding (or administering less aggressive) antithrombotic prophylaxis. Given a constant relative risk reduction in VTE for patient groups at relatively high risk of symptomatic VTE, the former situation is likely to hold, and for those at very low risk, the latter.

The approach to trading off presented thus far has two limitations. First, it ignores the most important events—fatalities. As we have noted, antithrombotic prophylaxis, relative to no prophylaxis, reduces deaths from pulmonary embolus in both medical and surgical patients enrolled in RCTs and may reduce all-cause mortality in surgical patients (Table 1), providing a compelling rationale for recommending prophylaxis in patients who meet the profile of those enrolled in the trials that demonstrated the reduction in mortality from pulmonary embolus.

Once one has decided to administer antithrombotic prophylaxis to patients at sufficiently high risk of VTE, the issue of lower mortality from pulmonary embolus becomes less compelling. The administration of any effective antithrombotic prophylaxis is likely to substantially reduce the number of fatal pulmonary emboli, which already is very rare. Thus, any further reduction in absolute risk from more-intensive antithrombotic therapy will be, in absolute terms, very small. Furthermore, an inference that any additional reduction in fatal pulmonary emboli will occur at all is based on indirect evidence from the impact on nonfatal and often largely asymptomatic thrombi and on the evidence from comparisons of antithrombotic prophylaxis vs no antithrombotic prophylaxis.

In this article, we have outlined the limitations associated with evidence quality in ascertaining best estimates of treatment impact on VTE. In the individual articles that present recommendations, we describe the limitations regarding the evidence quality for differences in bleeding. Our judgment is that in comparisons of alternative antithrombotic regimens, inferences regarding fatalities are even more tenuous. Therefore, for decisions regarding which antithrombotic agent or regimen to use, we largely base inferences on evidence in nonfatal VTE prevention vs increases in nonfatal bleeding.

The second limitation is that the judgment that bleeding events have more or less the same disutility as VTE events is fraught with uncertainty. Studies of patient values and preferences that would lead to confidence in this judgment do not exist; therefore, recommending one antithrombotic agent over another requires a substantial benefit-risk gradient between therapies, with a high level of confidence in that gradient.

None of the approaches we have suggested for estimating the effect of antithrombotic prophylaxis is without problems arising from the conduct of screening ultrasound and venography. Subsequent studies should rely on clinical surveillance to detect symptomatic events.

Author contributions: As Topic Editor, Dr Guyatt oversaw the development of this article, including any analysis and subsequent development of the information contained herein.

Dr Guyatt: served as a Topic Editor and was responsible for drafting the entire article, with the exception of the first part on evidence regarding the impact of antithrombotic prophylaxis on mortality, and for final editing and content.

Dr Eikelboom: served as a panelist and was responsible for drafting the first part of this article dealing with the evidence regarding the impact of antithrombotic prophylaxis on mortality.

Dr Gould: served as a panelist and provided intellectual input, critique, and revisions.

Dr Garcia: served as a panelist and provided intellectual input, critique, and revisions.

Dr Crowther: served as a panelist and provided intellectual input, critique, and revisions.

Dr Murad: served as a panelist and provided intellectual input, critique, and revisions.

Dr Kahn: served as a panelist and provided intellectual input, critique, and revisions.

Dr Falck-Ytter: served as a panelist and provided intellectual input, critique, and revisions.

Dr Francis: served as a panelist and provided intellectual input, critique, and revisions.

Dr Lansberg: served as a panelist and provided intellectual input, critique, and revisions.

Dr Akl: served as a panelist and provided intellectual input, critique, and revisions.

Dr Hirsh: served as a panelist and was responsible for drafting the first part of this article dealing with the evidence regarding the impact of antithrombotic prophylaxis on mortality.

Financial/nonfinancial disclosures: In summary, the authors have reported to CHEST the following conflicts of interest: Dr Eikelboom has received consulting fees and honoraria from AstraZeneca; Boehringer Ingelheim GmbH; Bristol-Myers Squibb; Corgenix; Daiichi-Sankyo, Inc; Eisai Co, Inc; Eli Lilly and Company; GlaxoSmithKline plc; Haemonetics Corp; McNeil Consumer Healthcare; and Sanofi-Aventis LLC and grants and in-kind support from Accumetrics, Inc; AspirinWorks; Bayer Healthcare Pharmaceuticals; Boehringer Ingelheim GmbH; Bristol-Myers Squibb; Corgenix; Dade Behring Inc; GlaxoSmithKline plc; and Sanofi-Aventis LLC. Dr Crowther has served on various advisory boards, has assisted in the preparation of educational materials, and has sat on data safety and management boards. His institution has received research funds from the following companies: Leo Pharma A/S, Pfizer Inc, Boehringer Ingelheim GmbH, Bayer Healthcare Pharmaceuticals, Octapharm AG, CSL Behring, and Artisan Pharma. Personal total compensation for these activities over the past 3 years totals less than US $10,000. Dr Kahn has received peer-reviewed and investigator-initiated industry research funding for projects related to venous thrombosis and postthrombotic syndrome prevention and treatment. She has received honoraria for industry-sponsored talks pertaining to venous thrombosis. Dr Francis received research grant support from the National Heart Lung and Blood Institute and Eisai Co. Ltd, and served as a steering committee member for a clinical trial sponsored by Eisai Co, Ltd. Dr Francis has also worked on projects supported by Eisai Co, Ltd, and sanofi -aventis. Dr Guyatt is co-chair of the GRADE Working Group and Drs Murad, Falck-Ytter, and Akl are contributing members. Drs Gould, Garcia, Lansberg, and Hirsh have reported that no potential conflicts of interest exist with any companies/organizations whose products or services may be discussed in this article.

Role of sponsors: The sponsors played no role in the development of these guidelines. Sponsoring organizations cannot recommend panelists or topics, nor are they allowed prepublication access to the manuscripts and recommendations. Guideline panel members, including the chair, and members of the Health & Science Policy Committee are blinded to the funding sources. Further details on the Conflict of Interest Policy are available online at http://chestnet.org.

Endorsements: This guideline is endorsed by the American Association for Clinical Chemistry, the American College of Clinical Pharmacy, the American Society of Health-System Pharmacists, the American Society of Hematology, and the International Society of Thrombosis and Hematosis.

Author contributions: As Topic Editor, Dr Guyatt oversaw the development of this article, including any analysis and subsequent development of the information contained herein.

Dr Guyatt: served as a Topic Editor and was responsible for drafting the entire article, with the exception of the first part on evidence regarding the impact of antithrombotic prophylaxis on mortality, and for final editing and content.

Dr Eikelboom: served as a panelist and was responsible for drafting the first part of this article dealing with the evidence regarding the impact of antithrombotic prophylaxis on mortality.

Dr Gould: served as a panelist and provided intellectual input, critique, and revisions.

Dr Garcia: served as a panelist and provided intellectual input, critique, and revisions.

Dr Crowther: served as a panelist and provided intellectual input, critique, and revisions.

Dr Murad: served as a panelist and provided intellectual input, critique, and revisions.

Dr Kahn: served as a panelist and provided intellectual input, critique, and revisions.

Dr Falck-Ytter: served as a panelist and provided intellectual input, critique, and revisions.

Dr Francis: served as a panelist and provided intellectual input, critique, and revisions.

Dr Lansberg: served as a panelist and provided intellectual input, critique, and revisions.

Dr Akl: served as a panelist and provided intellectual input, critique, and revisions.

Dr Hirsh: served as a panelist and was responsible for drafting the first part of this article dealing with the evidence regarding the impact of antithrombotic prophylaxis on mortality.

Financial/nonfinancial disclosures: In summary, the authors have reported to CHEST the following conflicts of interest: Dr Eikelboom has received consulting fees and honoraria from AstraZeneca; Boehringer Ingelheim GmbH; Bristol-Myers Squibb; Corgenix; Daiichi-Sankyo, Inc; Eisai Co, Inc; Eli Lilly and Company; GlaxoSmithKline plc; Haemonetics Corp; McNeil Consumer Healthcare; and Sanofi-Aventis LLC and grants and in-kind support from Accumetrics, Inc; AspirinWorks; Bayer Healthcare Pharmaceuticals; Boehringer Ingelheim GmbH; Bristol-Myers Squibb; Corgenix; Dade Behring Inc; GlaxoSmithKline plc; and Sanofi-Aventis LLC. Dr Crowther has served on various advisory boards, has assisted in the preparation of educational materials, and has sat on data safety and management boards. His institution has received research funds from the following companies: Leo Pharma A/S, Pfizer Inc, Boehringer Ingelheim GmbH, Bayer Healthcare Pharmaceuticals, Octapharm AG, CSL Behring, and Artisan Pharma. Personal total compensation for these activities over the past 3 years totals less than US $10,000. Dr Kahn has received peer-reviewed and investigator-initiated industry research funding for projects related to venous thrombosis and postthrombotic syndrome prevention and treatment. She has received honoraria for industry-sponsored talks pertaining to venous thrombosis. Dr Francis received research grant support from the National Heart Lung and Blood Institute and Eisai Co. Ltd, and served as a steering committee member for a clinical trial sponsored by Eisai Co, Ltd. Dr Francis has also worked on projects supported by Eisai Co, Ltd, and sanofi -aventis. Dr Guyatt is co-chair of the GRADE Working Group and Drs Murad, Falck-Ytter, and Akl are contributing members. Drs Gould, Garcia, Lansberg, and Hirsh have reported that no potential conflicts of interest exist with any companies/organizations whose products or services may be discussed in this article.

Role of sponsors: The sponsors played no role in the development of these guidelines. Sponsoring organizations cannot recommend panelists or topics, nor are they allowed prepublication access to the manuscripts and recommendations. Guideline panel members, including the chair, and members of the Health & Science Policy Committee are blinded to the funding sources. Further details on the Conflict of Interest Policy are available online at http://chestnet.org.

Endorsements: This guideline is endorsed by the American Association for Clinical Chemistry, the American College of Clinical Pharmacy, the American Society of Health-System Pharmacists, the American Society of Hematology, and the International Society of Thrombosis and Hematosis.