Greg Maynard, MD, MSc, SFHM

Patients hospitalized for acute medical illness have more than a 10‐fold increased risk for venous thromboembolism (VTE),[1] with an undeniably dramatic, negative impact on the lives of those afflicted, including fatal pulmonary embolism (PE), which most commonly affects patients on the medical service.[2, 3, 4] Yet estimates for the overall rate of VTE in this population are relatively low, raising questions about which subsets of medical patients warrant the risk and cost of prophylaxis.

Recently, the American College of Physicians published guidelines (ACP‐1)[5] and a supporting review[6] addressing VTE prophylaxis in nonsurgical inpatients, followed by publication of the American College of Chest Physicians (ACCP) 9th Edition of the Chest Guidelines on Antithrombotic Therapy and Prevention of Thrombosis (AT9),[7] which divides VTE prevention into 3 articles,[8, 9, 10] including 1 on nonsurgical patients.[8] Both ACP‐1 and AT9 differ significantly from the 2008 ACCP guidelines (AT8),[11] but took different approaches to methodology, risk assessment, and several other aspects of thromboprophylaxis (Table 1). This narrative review summarizes and compares these recommendations and the methods used to arrive at them, with a final section focusing on implications for improvement teams designing order sets and system changes to address VTE prophylaxis.

WHY ARE THE NEW GUIDELINES DIFFERENT?

Major randomized controlled trials (RCTs)[12, 13, 14] of thromboprophylaxis used routine deep vein thrombosis (DVT) surveillance and included both symptomatic (S‐VTE) and asymptomatic VTE (A‐VTE) end points. These studies consistently demonstrated 44% to 63% reductions in VTE without increases in major bleeding.[11] Because of the strong relationship between A‐VTE and S‐VTE outcomes, and a paucity of studies using only S‐VTE outcomes, AT8 judged that A‐VTE outcomes were valid to include, whereas the new guidelines reject the use of asymptomatic VTE end points.[5, 8, 15] To minimize financial and intellectual conflicts of interest, AT9 also used methodologists rather than VTE experts as topic editors, excluded conflicted experts from voting on recommendations, and attempted to estimate patient values and preferences.[15] As a result, AT9 makes fewer strong recommendations (182 1A recommendations in 2008, but only 29 in 2012), replacing them with weak suggestions.

WHAT DO THE NEW GUIDELINES RECOMMEND?

AT8 recommended anticoagulant prophylaxis for acutely ill medical inpatients with known risk factors, but did not recommend routine thromboprophylaxis. However, because of well‐known problems with underprophylaxis,[16, 17, 18, 19] particularly in medical patients, the low risk of bleeding, and difficulties with explicitly defining low‐risk patients, many discounted the need for VTE risk stratification.

Both new guidelines recommend prophylaxis for many nonsurgical patients, but discourage routine thromboprophylaxis for nonsurgical inpatients. AT9 specifically recommends against any thromboprophylaxis for low‐risk medical inpatients, implying that many nonsurgical, non‐critical care patients belong in this category, citing lower estimates of benefit, lower estimates of VTE risk, and potential bleeding risks.

The guidelines[5, 8] agree that, when indicated and absent contraindications, anticoagulant prophylaxis is preferred over mechanical prophylaxis, and agree there is insufficient evidence to recommend 1 anticoagulant over another.

For patients at risk of both VTE and bleeding, ACP‐1 states that intermittent pneumatic compression (IPC) devices are a reasonable option, given the evidence showing benefit in surgical patients. However, ACP‐1 recommends against graduated compression stockings (GCS) in nonsurgical patients based on a meta‐analysis dominated by the CLOTS‐1 (Clots in Legs Or sTockings after Stroke) trial, which found that thigh‐high GCS increased the risk of skin breakdown without reducing VTE[20] in immobilized stroke patients. AT9 does not recommend against GCS for patients facing bleeding and VTE risk. AT9 notes the hazards of generalizing results from stroke patients, and also considers the somewhat contradictory results from the CLOTS‐2 trial in stroke patients, which found a lower rate of VTE with thigh‐high GCS than with knee‐high GCS.[21] AT9 designates a recommendation of 2C for either IPC devices or thigh‐high GCS for those at VTE risk when anticoagulants are contraindicated.

Combination mechanical‐pharmacologic prophylaxis has proven superior in some surgical populations, and many hospitals use combined prophylaxis in high‐risk medical patients. However, combination prophylaxis has not been studied in this population. ACP‐1 does not comment on the practice; AT9 does not recommend for or against it. Institutions that use combination prophylaxis should be aware that although it may seem logical to extrapolate estimates of benefit seen in selected surgical patients, this is not a recommended practice.

RCTs for thromboprophylaxis in nonsurgical inpatients provided prophylaxis for 6 to 21 days. Neither ACP‐1 or AT9 recommend routinely extending prophylaxis beyond the hospital stay, citing an RCT[22] in which the benefit of extended duration low molecular weight heparin was limited to selected subsets of patients and offset by bleeding complications. AT9 suggests prophylaxis for 6 to 21 days, until full mobility is restored, or until dischargewhichever comes first.[8] However, we know of no study that establishes a mobility level at which prophylaxis can be safely discontinued, especially in inpatients with multiple risk factors.

ESTIMATING RISK AND BENEFIT OF PROPHYLAXIS AND LIMITATIONS OF METHODS

Calculating risk/benefit ratios for thromboprophylaxis requires estimates of baseline VTE and bleeding risks, and estimates of the impact of prophylaxis on those baseline risks. Methods to estimate the impact of prophylaxis on S‐VTE from studies relying on A‐VTE all have limitations, as acknowledged by the AT9 introduction.[15]

The ACP‐1 review found the only significant effect of prophylaxis on medical inpatients was a modest reduction in PE and a modest increase in total bleeding events, without effects on major bleeding, DVT, or mortality.[6] The authors summarized the findings as indicative of little or no net benefit for the medical population as a whole. The ACP‐1 review derives estimates of S‐VTE risk, bleeding, and mortality from control (baseline) and interventional arms of RCTs that used routine VTE screening, and included A‐VTE end points. The baseline risk of VTE could potentially be overestimated, because the populations in the trials are not representative of the entire medical population.

On the other hand, pooling trials with screening‐detected VTE to estimate S‐VTE outcomes is a questionable practice that may falsely lower estimates of VTE prophylaxis benefit. Screening‐detected VTE may be treated or declared a study end point before it becomes symptomatic. MEDENOX (Medical Patients With Enoxaparin) is an illustrative example.[12] The 263 placebo recipients suffered 37 A‐VTEs and 4 S‐VTEs. The 272 enoxaparin recipients suffered 17 A‐VTEs and 3 S‐VTEs. Patients at the highest risk of S‐VTE were counted as reaching an end point before they could develop symptoms; this happened more than twice as often in the placebo arm. This decreases both estimates of baseline VTE risk and the measured benefit of prophylaxis for S‐VTE. Screening could conceivably reduce measured effects on mortality as well, because patients begin VTE therapy earlier. Per ACP‐1, the estimated risk for DVT is lower than for PE, running counter to literature experience[8, 16] and raising issues of face validity. The ACP‐1 review accepts all original definitions of major bleeding, including a 2 g/dL drop in hemoglobin,[12] which commonly occurs without any bleeding or clinical consequence, and bleeding events were ascribed to heparins up to 120 days after randomization, long after they could have been responsible.

Previous meta‐analyses of thromboprophylaxis studies[23, 24] shared many of these same limitations, but did not ascribe bleeding complications to heparins for this extended duration, and had point estimates that suggested a larger impact from prophylaxis than ACP‐1. Dentali et al., for example, showed statistically significant impact on PE (relative risk [RR] 0.43), fatal PE (RR 0.38), and a nearly statistically significant large impact on DVT (RR 0.47, 95% confidence interval [CI]: 0.22‐1.00),[24] whereas ACP‐1 estimated a smaller significant impact on PE (RR 0.69), no significant difference in fatal PE, and a much smaller estimate of the impact on DVT (RR 0.78, 95% CI: 0.45‐1.35) (Table 2).

AT9 used a variety of methods to estimate each component of the risk/benefit equation. Critical care and non‐critical care estimates were generated independently, but because of limited data, the critical care estimates were highly imprecise. In non‐critical care patients, as in ACP‐1, treatment effects were estimated from RCTs that routinely screened for A‐VTE, and they adapted the Dentali et al. estimate of DVT risk reduction. The baseline risk for bleeding and mortality were derived from the control population of the same meta‐analysis.[24]

Using a novel approach, AT9 estimated baseline nonsurgical VTE risk from a prospective observational cohort study of 1180 medical inpatients divided into high‐ and low‐risk groups by a point‐scoring system.[25] Deriving risk estimates from an observational cohort has theoretical advantages. Many patients did not receive prophylaxis, allowing for unadjusted risk estimates; they represented a cross‐section of medical inpatients rather than a selected trial population, and risk estimates were not reduced by the culling of screen‐detected A‐VTE.

The Padua risk‐assessment model (RAM) (Table 3) defines high VTE risk as a cumulative score 4. There were 60.3% of patients at low risk and 39.7% at high risk using this threshold. Among unprophylaxed patients, VTE occurred in 11% of high‐risk patients versus 0.3% of low‐risk patients (hazard ratio 32.0, 95% CI: 4.1251.0).

PADUA: A CLOSER LOOK

In the Padua study, 60% of the population appeared to be at such low risk for VTE that prophylaxis would seem unnecessary, but closer scrutiny should raise concern about generalizing these results. Of the 711 low‐risk patients, <1% were immobile, only 6% had cancer, 6% were obese, and only 12% had any acute infection or inflammatory condition, yet their mean length of stay was 7.9 days. These characteristics do not apply to 60% of American inpatients. Furthermore, 964 of 2208 eligible patients (44%) were excluded because they required therapeutic anticoagulation.[25]

Correspondence with the authors revealed that the 2 PEs in patients with Padua scores 4 occurred among 192 patients with a risk score of 3 (Figure 1), a 1% (2/192) risk of PE. This is a very small sample, and the true risk of VTE for medical inpatients with a risk score of 3 may be lower or significantly higher. In the Padua population, a risk score of 3 equated to a VTE risk of 6.9%, whereas those with a score of 0 to 2 had no VTE. For those adapting the Padua model, careful consideration of using a cutoff of 3, versus 4, is warranted.

Figure 1

IMPLICATIONS FOR VTE PROTOCOL IMPLEMENTATION AND IMPROVEMENT TEAMS

AT9 and ACP‐1 sought to focus on S‐VTE, remove bias from recommendations, and highlight potential risks of unnecessary prophylaxis in low‐risk patients. They have largely succeeded in these important goals. However, the complexity of the new guidelines and lack of consensus about VTE risk assessment pose significant challenges to improvement teams tasked with implementing the guidelines in real‐world settings.

CHOOSING A VTE RAM

The fundamental question is: How can hospitals assess VTE risk, assure adequate prophylaxis for patients who need it, while minimizing excess prophylaxis, in a practical, efficient way?

BOTTOM LINE IN CHOOSING A VTE RAM

Many medical inpatients are at high risk for VTE, but others are not at sufficient risk to warrant prophylaxis. VTE risk assessment should be embedded in admission, transfer, and perioperative order sets and may need a hard stop to insure completion. There is a trend to favor individualized point‐based models over models that place patients in groups of risk, but evidence is insufficient to recommend 1 type of RAM over another, and more complex point‐based models often require extensive local customization and algorithmic clinical decision support to effectively implement them. Centers without advanced capability may find the bucket models more effective. We urge improvement teams to trial their RAM with common patient case scenarios, and to make a choice based on an effort‐benefit analysis, feedback from their clinicians, and the level of customization in clinical decision support available to them.

OTHER IMPLEMENTATION STRATEGIES

VTE and bleeding risk change during hospitalization. We have used ongoing daily surveillance and measurement of patients on no prophylaxis to prompt concurrent intervention (ie, measure‐vention) to increase prophylaxis for patients at risk.[28] Improvement teams should focus not only on increasing prophylaxis for those at risk, but should also use measure‐vention, checklists, or other techniques to identify low‐risk (eg, ambulating) patients for cessation of overly aggressive prophylaxis. Efforts to improve early progressive ambulation, limit central venous catheters to those who truly need them, and improve adherence to mechanical prophylaxis can also reduce VTE, as well as benefitting patient populations in other ways.

We recognize there are several approaches to close the implementation gap in delivering thromboprophylaxis judiciously but reliably, and encourage research and publication of varied strategies. Last, we hope efforts to limit unnecessary prophylaxis and challenges inherent in implementing new and complex guidelines do not increase the morbidity and mortality of hospital‐acquired VTE, by derailing the delivery of prophylaxis to those in whom the benefits outweigh the risks.

Patients hospitalized for acute medical illness have more than a 10‐fold increased risk for venous thromboembolism (VTE),[1] with an undeniably dramatic, negative impact on the lives of those afflicted, including fatal pulmonary embolism (PE), which most commonly affects patients on the medical service.[2, 3, 4] Yet estimates for the overall rate of VTE in this population are relatively low, raising questions about which subsets of medical patients warrant the risk and cost of prophylaxis.

Recently, the American College of Physicians published guidelines (ACP‐1)[5] and a supporting review[6] addressing VTE prophylaxis in nonsurgical inpatients, followed by publication of the American College of Chest Physicians (ACCP) 9th Edition of the Chest Guidelines on Antithrombotic Therapy and Prevention of Thrombosis (AT9),[7] which divides VTE prevention into 3 articles,[8, 9, 10] including 1 on nonsurgical patients.[8] Both ACP‐1 and AT9 differ significantly from the 2008 ACCP guidelines (AT8),[11] but took different approaches to methodology, risk assessment, and several other aspects of thromboprophylaxis (Table 1). This narrative review summarizes and compares these recommendations and the methods used to arrive at them, with a final section focusing on implications for improvement teams designing order sets and system changes to address VTE prophylaxis.

WHY ARE THE NEW GUIDELINES DIFFERENT?

Major randomized controlled trials (RCTs)[12, 13, 14] of thromboprophylaxis used routine deep vein thrombosis (DVT) surveillance and included both symptomatic (S‐VTE) and asymptomatic VTE (A‐VTE) end points. These studies consistently demonstrated 44% to 63% reductions in VTE without increases in major bleeding.[11] Because of the strong relationship between A‐VTE and S‐VTE outcomes, and a paucity of studies using only S‐VTE outcomes, AT8 judged that A‐VTE outcomes were valid to include, whereas the new guidelines reject the use of asymptomatic VTE end points.[5, 8, 15] To minimize financial and intellectual conflicts of interest, AT9 also used methodologists rather than VTE experts as topic editors, excluded conflicted experts from voting on recommendations, and attempted to estimate patient values and preferences.[15] As a result, AT9 makes fewer strong recommendations (182 1A recommendations in 2008, but only 29 in 2012), replacing them with weak suggestions.

WHAT DO THE NEW GUIDELINES RECOMMEND?

AT8 recommended anticoagulant prophylaxis for acutely ill medical inpatients with known risk factors, but did not recommend routine thromboprophylaxis. However, because of well‐known problems with underprophylaxis,[16, 17, 18, 19] particularly in medical patients, the low risk of bleeding, and difficulties with explicitly defining low‐risk patients, many discounted the need for VTE risk stratification.

Both new guidelines recommend prophylaxis for many nonsurgical patients, but discourage routine thromboprophylaxis for nonsurgical inpatients. AT9 specifically recommends against any thromboprophylaxis for low‐risk medical inpatients, implying that many nonsurgical, non‐critical care patients belong in this category, citing lower estimates of benefit, lower estimates of VTE risk, and potential bleeding risks.

The guidelines[5, 8] agree that, when indicated and absent contraindications, anticoagulant prophylaxis is preferred over mechanical prophylaxis, and agree there is insufficient evidence to recommend 1 anticoagulant over another.

For patients at risk of both VTE and bleeding, ACP‐1 states that intermittent pneumatic compression (IPC) devices are a reasonable option, given the evidence showing benefit in surgical patients. However, ACP‐1 recommends against graduated compression stockings (GCS) in nonsurgical patients based on a meta‐analysis dominated by the CLOTS‐1 (Clots in Legs Or sTockings after Stroke) trial, which found that thigh‐high GCS increased the risk of skin breakdown without reducing VTE[20] in immobilized stroke patients. AT9 does not recommend against GCS for patients facing bleeding and VTE risk. AT9 notes the hazards of generalizing results from stroke patients, and also considers the somewhat contradictory results from the CLOTS‐2 trial in stroke patients, which found a lower rate of VTE with thigh‐high GCS than with knee‐high GCS.[21] AT9 designates a recommendation of 2C for either IPC devices or thigh‐high GCS for those at VTE risk when anticoagulants are contraindicated.

Combination mechanical‐pharmacologic prophylaxis has proven superior in some surgical populations, and many hospitals use combined prophylaxis in high‐risk medical patients. However, combination prophylaxis has not been studied in this population. ACP‐1 does not comment on the practice; AT9 does not recommend for or against it. Institutions that use combination prophylaxis should be aware that although it may seem logical to extrapolate estimates of benefit seen in selected surgical patients, this is not a recommended practice.

RCTs for thromboprophylaxis in nonsurgical inpatients provided prophylaxis for 6 to 21 days. Neither ACP‐1 or AT9 recommend routinely extending prophylaxis beyond the hospital stay, citing an RCT[22] in which the benefit of extended duration low molecular weight heparin was limited to selected subsets of patients and offset by bleeding complications. AT9 suggests prophylaxis for 6 to 21 days, until full mobility is restored, or until dischargewhichever comes first.[8] However, we know of no study that establishes a mobility level at which prophylaxis can be safely discontinued, especially in inpatients with multiple risk factors.

ESTIMATING RISK AND BENEFIT OF PROPHYLAXIS AND LIMITATIONS OF METHODS

Calculating risk/benefit ratios for thromboprophylaxis requires estimates of baseline VTE and bleeding risks, and estimates of the impact of prophylaxis on those baseline risks. Methods to estimate the impact of prophylaxis on S‐VTE from studies relying on A‐VTE all have limitations, as acknowledged by the AT9 introduction.[15]

The ACP‐1 review found the only significant effect of prophylaxis on medical inpatients was a modest reduction in PE and a modest increase in total bleeding events, without effects on major bleeding, DVT, or mortality.[6] The authors summarized the findings as indicative of little or no net benefit for the medical population as a whole. The ACP‐1 review derives estimates of S‐VTE risk, bleeding, and mortality from control (baseline) and interventional arms of RCTs that used routine VTE screening, and included A‐VTE end points. The baseline risk of VTE could potentially be overestimated, because the populations in the trials are not representative of the entire medical population.

On the other hand, pooling trials with screening‐detected VTE to estimate S‐VTE outcomes is a questionable practice that may falsely lower estimates of VTE prophylaxis benefit. Screening‐detected VTE may be treated or declared a study end point before it becomes symptomatic. MEDENOX (Medical Patients With Enoxaparin) is an illustrative example.[12] The 263 placebo recipients suffered 37 A‐VTEs and 4 S‐VTEs. The 272 enoxaparin recipients suffered 17 A‐VTEs and 3 S‐VTEs. Patients at the highest risk of S‐VTE were counted as reaching an end point before they could develop symptoms; this happened more than twice as often in the placebo arm. This decreases both estimates of baseline VTE risk and the measured benefit of prophylaxis for S‐VTE. Screening could conceivably reduce measured effects on mortality as well, because patients begin VTE therapy earlier. Per ACP‐1, the estimated risk for DVT is lower than for PE, running counter to literature experience[8, 16] and raising issues of face validity. The ACP‐1 review accepts all original definitions of major bleeding, including a 2 g/dL drop in hemoglobin,[12] which commonly occurs without any bleeding or clinical consequence, and bleeding events were ascribed to heparins up to 120 days after randomization, long after they could have been responsible.

Previous meta‐analyses of thromboprophylaxis studies[23, 24] shared many of these same limitations, but did not ascribe bleeding complications to heparins for this extended duration, and had point estimates that suggested a larger impact from prophylaxis than ACP‐1. Dentali et al., for example, showed statistically significant impact on PE (relative risk [RR] 0.43), fatal PE (RR 0.38), and a nearly statistically significant large impact on DVT (RR 0.47, 95% confidence interval [CI]: 0.22‐1.00),[24] whereas ACP‐1 estimated a smaller significant impact on PE (RR 0.69), no significant difference in fatal PE, and a much smaller estimate of the impact on DVT (RR 0.78, 95% CI: 0.45‐1.35) (Table 2).

AT9 used a variety of methods to estimate each component of the risk/benefit equation. Critical care and non‐critical care estimates were generated independently, but because of limited data, the critical care estimates were highly imprecise. In non‐critical care patients, as in ACP‐1, treatment effects were estimated from RCTs that routinely screened for A‐VTE, and they adapted the Dentali et al. estimate of DVT risk reduction. The baseline risk for bleeding and mortality were derived from the control population of the same meta‐analysis.[24]

Using a novel approach, AT9 estimated baseline nonsurgical VTE risk from a prospective observational cohort study of 1180 medical inpatients divided into high‐ and low‐risk groups by a point‐scoring system.[25] Deriving risk estimates from an observational cohort has theoretical advantages. Many patients did not receive prophylaxis, allowing for unadjusted risk estimates; they represented a cross‐section of medical inpatients rather than a selected trial population, and risk estimates were not reduced by the culling of screen‐detected A‐VTE.

The Padua risk‐assessment model (RAM) (Table 3) defines high VTE risk as a cumulative score 4. There were 60.3% of patients at low risk and 39.7% at high risk using this threshold. Among unprophylaxed patients, VTE occurred in 11% of high‐risk patients versus 0.3% of low‐risk patients (hazard ratio 32.0, 95% CI: 4.1251.0).

PADUA: A CLOSER LOOK

In the Padua study, 60% of the population appeared to be at such low risk for VTE that prophylaxis would seem unnecessary, but closer scrutiny should raise concern about generalizing these results. Of the 711 low‐risk patients, <1% were immobile, only 6% had cancer, 6% were obese, and only 12% had any acute infection or inflammatory condition, yet their mean length of stay was 7.9 days. These characteristics do not apply to 60% of American inpatients. Furthermore, 964 of 2208 eligible patients (44%) were excluded because they required therapeutic anticoagulation.[25]

Correspondence with the authors revealed that the 2 PEs in patients with Padua scores 4 occurred among 192 patients with a risk score of 3 (Figure 1), a 1% (2/192) risk of PE. This is a very small sample, and the true risk of VTE for medical inpatients with a risk score of 3 may be lower or significantly higher. In the Padua population, a risk score of 3 equated to a VTE risk of 6.9%, whereas those with a score of 0 to 2 had no VTE. For those adapting the Padua model, careful consideration of using a cutoff of 3, versus 4, is warranted.

Figure 1

IMPLICATIONS FOR VTE PROTOCOL IMPLEMENTATION AND IMPROVEMENT TEAMS

AT9 and ACP‐1 sought to focus on S‐VTE, remove bias from recommendations, and highlight potential risks of unnecessary prophylaxis in low‐risk patients. They have largely succeeded in these important goals. However, the complexity of the new guidelines and lack of consensus about VTE risk assessment pose significant challenges to improvement teams tasked with implementing the guidelines in real‐world settings.

CHOOSING A VTE RAM

The fundamental question is: How can hospitals assess VTE risk, assure adequate prophylaxis for patients who need it, while minimizing excess prophylaxis, in a practical, efficient way?

BOTTOM LINE IN CHOOSING A VTE RAM

Many medical inpatients are at high risk for VTE, but others are not at sufficient risk to warrant prophylaxis. VTE risk assessment should be embedded in admission, transfer, and perioperative order sets and may need a hard stop to insure completion. There is a trend to favor individualized point‐based models over models that place patients in groups of risk, but evidence is insufficient to recommend 1 type of RAM over another, and more complex point‐based models often require extensive local customization and algorithmic clinical decision support to effectively implement them. Centers without advanced capability may find the bucket models more effective. We urge improvement teams to trial their RAM with common patient case scenarios, and to make a choice based on an effort‐benefit analysis, feedback from their clinicians, and the level of customization in clinical decision support available to them.

OTHER IMPLEMENTATION STRATEGIES

VTE and bleeding risk change during hospitalization. We have used ongoing daily surveillance and measurement of patients on no prophylaxis to prompt concurrent intervention (ie, measure‐vention) to increase prophylaxis for patients at risk.[28] Improvement teams should focus not only on increasing prophylaxis for those at risk, but should also use measure‐vention, checklists, or other techniques to identify low‐risk (eg, ambulating) patients for cessation of overly aggressive prophylaxis. Efforts to improve early progressive ambulation, limit central venous catheters to those who truly need them, and improve adherence to mechanical prophylaxis can also reduce VTE, as well as benefitting patient populations in other ways.

We recognize there are several approaches to close the implementation gap in delivering thromboprophylaxis judiciously but reliably, and encourage research and publication of varied strategies. Last, we hope efforts to limit unnecessary prophylaxis and challenges inherent in implementing new and complex guidelines do not increase the morbidity and mortality of hospital‐acquired VTE, by derailing the delivery of prophylaxis to those in whom the benefits outweigh the risks.

Patients hospitalized for acute medical illness have more than a 10‐fold increased risk for venous thromboembolism (VTE),[1] with an undeniably dramatic, negative impact on the lives of those afflicted, including fatal pulmonary embolism (PE), which most commonly affects patients on the medical service.[2, 3, 4] Yet estimates for the overall rate of VTE in this population are relatively low, raising questions about which subsets of medical patients warrant the risk and cost of prophylaxis.

Recently, the American College of Physicians published guidelines (ACP‐1)[5] and a supporting review[6] addressing VTE prophylaxis in nonsurgical inpatients, followed by publication of the American College of Chest Physicians (ACCP) 9th Edition of the Chest Guidelines on Antithrombotic Therapy and Prevention of Thrombosis (AT9),[7] which divides VTE prevention into 3 articles,[8, 9, 10] including 1 on nonsurgical patients.[8] Both ACP‐1 and AT9 differ significantly from the 2008 ACCP guidelines (AT8),[11] but took different approaches to methodology, risk assessment, and several other aspects of thromboprophylaxis (Table 1). This narrative review summarizes and compares these recommendations and the methods used to arrive at them, with a final section focusing on implications for improvement teams designing order sets and system changes to address VTE prophylaxis.

WHY ARE THE NEW GUIDELINES DIFFERENT?

Major randomized controlled trials (RCTs)[12, 13, 14] of thromboprophylaxis used routine deep vein thrombosis (DVT) surveillance and included both symptomatic (S‐VTE) and asymptomatic VTE (A‐VTE) end points. These studies consistently demonstrated 44% to 63% reductions in VTE without increases in major bleeding.[11] Because of the strong relationship between A‐VTE and S‐VTE outcomes, and a paucity of studies using only S‐VTE outcomes, AT8 judged that A‐VTE outcomes were valid to include, whereas the new guidelines reject the use of asymptomatic VTE end points.[5, 8, 15] To minimize financial and intellectual conflicts of interest, AT9 also used methodologists rather than VTE experts as topic editors, excluded conflicted experts from voting on recommendations, and attempted to estimate patient values and preferences.[15] As a result, AT9 makes fewer strong recommendations (182 1A recommendations in 2008, but only 29 in 2012), replacing them with weak suggestions.

WHAT DO THE NEW GUIDELINES RECOMMEND?

AT8 recommended anticoagulant prophylaxis for acutely ill medical inpatients with known risk factors, but did not recommend routine thromboprophylaxis. However, because of well‐known problems with underprophylaxis,[16, 17, 18, 19] particularly in medical patients, the low risk of bleeding, and difficulties with explicitly defining low‐risk patients, many discounted the need for VTE risk stratification.

Both new guidelines recommend prophylaxis for many nonsurgical patients, but discourage routine thromboprophylaxis for nonsurgical inpatients. AT9 specifically recommends against any thromboprophylaxis for low‐risk medical inpatients, implying that many nonsurgical, non‐critical care patients belong in this category, citing lower estimates of benefit, lower estimates of VTE risk, and potential bleeding risks.

The guidelines[5, 8] agree that, when indicated and absent contraindications, anticoagulant prophylaxis is preferred over mechanical prophylaxis, and agree there is insufficient evidence to recommend 1 anticoagulant over another.

For patients at risk of both VTE and bleeding, ACP‐1 states that intermittent pneumatic compression (IPC) devices are a reasonable option, given the evidence showing benefit in surgical patients. However, ACP‐1 recommends against graduated compression stockings (GCS) in nonsurgical patients based on a meta‐analysis dominated by the CLOTS‐1 (Clots in Legs Or sTockings after Stroke) trial, which found that thigh‐high GCS increased the risk of skin breakdown without reducing VTE[20] in immobilized stroke patients. AT9 does not recommend against GCS for patients facing bleeding and VTE risk. AT9 notes the hazards of generalizing results from stroke patients, and also considers the somewhat contradictory results from the CLOTS‐2 trial in stroke patients, which found a lower rate of VTE with thigh‐high GCS than with knee‐high GCS.[21] AT9 designates a recommendation of 2C for either IPC devices or thigh‐high GCS for those at VTE risk when anticoagulants are contraindicated.

Combination mechanical‐pharmacologic prophylaxis has proven superior in some surgical populations, and many hospitals use combined prophylaxis in high‐risk medical patients. However, combination prophylaxis has not been studied in this population. ACP‐1 does not comment on the practice; AT9 does not recommend for or against it. Institutions that use combination prophylaxis should be aware that although it may seem logical to extrapolate estimates of benefit seen in selected surgical patients, this is not a recommended practice.

RCTs for thromboprophylaxis in nonsurgical inpatients provided prophylaxis for 6 to 21 days. Neither ACP‐1 or AT9 recommend routinely extending prophylaxis beyond the hospital stay, citing an RCT[22] in which the benefit of extended duration low molecular weight heparin was limited to selected subsets of patients and offset by bleeding complications. AT9 suggests prophylaxis for 6 to 21 days, until full mobility is restored, or until dischargewhichever comes first.[8] However, we know of no study that establishes a mobility level at which prophylaxis can be safely discontinued, especially in inpatients with multiple risk factors.

ESTIMATING RISK AND BENEFIT OF PROPHYLAXIS AND LIMITATIONS OF METHODS

Calculating risk/benefit ratios for thromboprophylaxis requires estimates of baseline VTE and bleeding risks, and estimates of the impact of prophylaxis on those baseline risks. Methods to estimate the impact of prophylaxis on S‐VTE from studies relying on A‐VTE all have limitations, as acknowledged by the AT9 introduction.[15]

The ACP‐1 review found the only significant effect of prophylaxis on medical inpatients was a modest reduction in PE and a modest increase in total bleeding events, without effects on major bleeding, DVT, or mortality.[6] The authors summarized the findings as indicative of little or no net benefit for the medical population as a whole. The ACP‐1 review derives estimates of S‐VTE risk, bleeding, and mortality from control (baseline) and interventional arms of RCTs that used routine VTE screening, and included A‐VTE end points. The baseline risk of VTE could potentially be overestimated, because the populations in the trials are not representative of the entire medical population.

On the other hand, pooling trials with screening‐detected VTE to estimate S‐VTE outcomes is a questionable practice that may falsely lower estimates of VTE prophylaxis benefit. Screening‐detected VTE may be treated or declared a study end point before it becomes symptomatic. MEDENOX (Medical Patients With Enoxaparin) is an illustrative example.[12] The 263 placebo recipients suffered 37 A‐VTEs and 4 S‐VTEs. The 272 enoxaparin recipients suffered 17 A‐VTEs and 3 S‐VTEs. Patients at the highest risk of S‐VTE were counted as reaching an end point before they could develop symptoms; this happened more than twice as often in the placebo arm. This decreases both estimates of baseline VTE risk and the measured benefit of prophylaxis for S‐VTE. Screening could conceivably reduce measured effects on mortality as well, because patients begin VTE therapy earlier. Per ACP‐1, the estimated risk for DVT is lower than for PE, running counter to literature experience[8, 16] and raising issues of face validity. The ACP‐1 review accepts all original definitions of major bleeding, including a 2 g/dL drop in hemoglobin,[12] which commonly occurs without any bleeding or clinical consequence, and bleeding events were ascribed to heparins up to 120 days after randomization, long after they could have been responsible.

Previous meta‐analyses of thromboprophylaxis studies[23, 24] shared many of these same limitations, but did not ascribe bleeding complications to heparins for this extended duration, and had point estimates that suggested a larger impact from prophylaxis than ACP‐1. Dentali et al., for example, showed statistically significant impact on PE (relative risk [RR] 0.43), fatal PE (RR 0.38), and a nearly statistically significant large impact on DVT (RR 0.47, 95% confidence interval [CI]: 0.22‐1.00),[24] whereas ACP‐1 estimated a smaller significant impact on PE (RR 0.69), no significant difference in fatal PE, and a much smaller estimate of the impact on DVT (RR 0.78, 95% CI: 0.45‐1.35) (Table 2).

AT9 used a variety of methods to estimate each component of the risk/benefit equation. Critical care and non‐critical care estimates were generated independently, but because of limited data, the critical care estimates were highly imprecise. In non‐critical care patients, as in ACP‐1, treatment effects were estimated from RCTs that routinely screened for A‐VTE, and they adapted the Dentali et al. estimate of DVT risk reduction. The baseline risk for bleeding and mortality were derived from the control population of the same meta‐analysis.[24]

Using a novel approach, AT9 estimated baseline nonsurgical VTE risk from a prospective observational cohort study of 1180 medical inpatients divided into high‐ and low‐risk groups by a point‐scoring system.[25] Deriving risk estimates from an observational cohort has theoretical advantages. Many patients did not receive prophylaxis, allowing for unadjusted risk estimates; they represented a cross‐section of medical inpatients rather than a selected trial population, and risk estimates were not reduced by the culling of screen‐detected A‐VTE.

The Padua risk‐assessment model (RAM) (Table 3) defines high VTE risk as a cumulative score 4. There were 60.3% of patients at low risk and 39.7% at high risk using this threshold. Among unprophylaxed patients, VTE occurred in 11% of high‐risk patients versus 0.3% of low‐risk patients (hazard ratio 32.0, 95% CI: 4.1251.0).

PADUA: A CLOSER LOOK

In the Padua study, 60% of the population appeared to be at such low risk for VTE that prophylaxis would seem unnecessary, but closer scrutiny should raise concern about generalizing these results. Of the 711 low‐risk patients, <1% were immobile, only 6% had cancer, 6% were obese, and only 12% had any acute infection or inflammatory condition, yet their mean length of stay was 7.9 days. These characteristics do not apply to 60% of American inpatients. Furthermore, 964 of 2208 eligible patients (44%) were excluded because they required therapeutic anticoagulation.[25]

Correspondence with the authors revealed that the 2 PEs in patients with Padua scores 4 occurred among 192 patients with a risk score of 3 (Figure 1), a 1% (2/192) risk of PE. This is a very small sample, and the true risk of VTE for medical inpatients with a risk score of 3 may be lower or significantly higher. In the Padua population, a risk score of 3 equated to a VTE risk of 6.9%, whereas those with a score of 0 to 2 had no VTE. For those adapting the Padua model, careful consideration of using a cutoff of 3, versus 4, is warranted.

Figure 1

IMPLICATIONS FOR VTE PROTOCOL IMPLEMENTATION AND IMPROVEMENT TEAMS

AT9 and ACP‐1 sought to focus on S‐VTE, remove bias from recommendations, and highlight potential risks of unnecessary prophylaxis in low‐risk patients. They have largely succeeded in these important goals. However, the complexity of the new guidelines and lack of consensus about VTE risk assessment pose significant challenges to improvement teams tasked with implementing the guidelines in real‐world settings.

CHOOSING A VTE RAM

The fundamental question is: How can hospitals assess VTE risk, assure adequate prophylaxis for patients who need it, while minimizing excess prophylaxis, in a practical, efficient way?

BOTTOM LINE IN CHOOSING A VTE RAM

Many medical inpatients are at high risk for VTE, but others are not at sufficient risk to warrant prophylaxis. VTE risk assessment should be embedded in admission, transfer, and perioperative order sets and may need a hard stop to insure completion. There is a trend to favor individualized point‐based models over models that place patients in groups of risk, but evidence is insufficient to recommend 1 type of RAM over another, and more complex point‐based models often require extensive local customization and algorithmic clinical decision support to effectively implement them. Centers without advanced capability may find the bucket models more effective. We urge improvement teams to trial their RAM with common patient case scenarios, and to make a choice based on an effort‐benefit analysis, feedback from their clinicians, and the level of customization in clinical decision support available to them.

OTHER IMPLEMENTATION STRATEGIES

VTE and bleeding risk change during hospitalization. We have used ongoing daily surveillance and measurement of patients on no prophylaxis to prompt concurrent intervention (ie, measure‐vention) to increase prophylaxis for patients at risk.[28] Improvement teams should focus not only on increasing prophylaxis for those at risk, but should also use measure‐vention, checklists, or other techniques to identify low‐risk (eg, ambulating) patients for cessation of overly aggressive prophylaxis. Efforts to improve early progressive ambulation, limit central venous catheters to those who truly need them, and improve adherence to mechanical prophylaxis can also reduce VTE, as well as benefitting patient populations in other ways.

We recognize there are several approaches to close the implementation gap in delivering thromboprophylaxis judiciously but reliably, and encourage research and publication of varied strategies. Last, we hope efforts to limit unnecessary prophylaxis and challenges inherent in implementing new and complex guidelines do not increase the morbidity and mortality of hospital‐acquired VTE, by derailing the delivery of prophylaxis to those in whom the benefits outweigh the risks.