Preventing Hospital-Associated Venous Thromboembolism

Chapter 4. Choose the Model To Assess VTE and Bleeding Risk

This chapter provides an overview of the major categories and characteristics of VTE risk assessment models. Once barriers are identified and the team has analyzed its facility’s care delivery process related to VTE prevention, a risk assessment model can be adopted.

A venous thromboembolism (VTE) prevention protocol is a standardized VTE risk assessment, linked to a menu of appropriate VTE prophylaxis options for each level of risk, which provides guidance for management of patients with contraindications to pharmacologic prophylaxis. Bleeding risk tools and guidance for the timing of administering anticoagulant prophylaxis around surgical procedures or other high bleeding risk intervals should also be part of a protocol. Protocols define best practice at the local level based on the best evidence available, with operational definitions that drive order set design, measurement tools, and other aspects of the quality improvement process.

The ideal VTE prevention protocol would have these characteristics:

  • Accurately detect all patients at risk of developing deep vein thrombosis (DVT).
  • Reliably exclude patients who would be unlikely to develop DVT, minimizing inappropriate over-prophylaxis in those of lower risk.
  • Provide actionable recommendations for permutations of VTE and bleeding risk.
  • Be simple to use in routine clinical practice, with minimal need for laboratory investigations or complex calculations.
  • Have predictors of VTE risk available to ordering provider at the point of care.
  • Provide decision support regarding those who would benefit from combination mechanical and anticoagulant prophylaxis.
  • Integrate into clinical practice results in a way that decreases hospital-associated VTE without any increase in bleeding.
  • Lend itself to automation, and even to dynamic ongoing reevaluations.

Unfortunately, there is no consensus regarding the preferred VTE risk assessment tool. VTE risk assessment is essentially a tool. Patients are targeted for interventions to prevent VTE (anticoagulant or mechanical prophylaxis and efforts to improve mobility) based on the assessment of risk of a VTE event. The positive potential to reduce VTE must be balanced with the discomfort, bleeding, expense, and other adverse effects that could result from the prophylactic measures. There is no consensus on the answer to the fundamental question, "How can hospitals assess VTE risk, then ensure adequate prophylaxis for patients who need it, while minimizing excess prophylaxis, in a practical, efficient way?"1

Several reviews of risk assessment models are available in the literature.2-4 These reviews tend to focus on the rigor of model derivation and predictive value. This guide focuses on the practical issues of implementation and utility in clinical practice. Risk assessment models that are in wide use, that are featured in guidelines, or that have demonstrated efficacy in actual practice or clinical trials will be reviewed. While this chapter will not provide definitive guidance on the fundamental question posed above, it will give VTE improvement teams the context under which to make a reasonable and thoughtful decision about what will work best in their setting.

Overview—Major Categories and Characteristics of VTE Risk Assessment Models


In the absence of consensus on the best risk assessment model, one approach is to avoid this issue altogether and simply present a prompt to consider prophylaxis. A list of options for prophylaxis is presented in the following example (Figure 4.1), but no clinical decision support (CDS) is offered to sway the judgment of the individual provider.

Figure 4.1: DVT Prophylaxis Orders

__ Anti-Thromboembolism Stockings

__ Sequential Compression Devices

__ UFH 5,000 units SubQ q 12 hours

__ UFH 5,000 units SubQ q 8 hours

__ LMWH (Enoxaparin) 40 mg SubQ q day

__ LMWH (Enoxaparin) 30 mg SubQ q 12 hours

__ No Prophylaxis, Ambulate

__ Other

Key: UFH = unfractionated heparin; SubQ = subcutaneous; LMWH = low-molecular-weight heparin.

The Hierarchy of Reliability (Table 1.1) and published experience suggests this approach produces only very modest improvement insufficient to make a meaningful reduction in hospital-associated VTE (HA-VTE) rates.5,6 Widespread, well documented under-prophylaxis7-10 is largely the result of relying on physician judgment, imperfect human memory, and relatively passive interventions such as educational sessions and pocket cards.11 Basic tenets of quality improvement also caution against this approach as it offers no opportunity for measurement, standardization, or even definition of best practice, and this approach would generally not meet meaningful use criteria or help institutions meet The Joint Commission's standards for VTE prevention.1

"Opt Out"

A second approach is the "opt out" approach (Appendix B.2). This approach has an automatic default of anticoagulant prophylaxis and assumes the great majority of inpatients are candidates for it. Ordering providers can "opt out" if they specify the patient is at low risk, on therapeutic anticoagulation, or has contraindications to prophylaxis. While this approach is appealing for the simplicity and effectiveness in inducing high rates of anticoagulant prophylaxis, it can easily result in over-prophylaxis, which is a particular concern in medical populations.12

Both ACP and AT9 guidelines13,14 discourage the universal prophylaxis approach for this population. On the other hand, opt-out mechanisms can be appropriate for some services with uniformly high VTE risk. For example, an orthopedic surgery service focused on total hip replacement might have default orders for their preferred anticoagulant and mechanical prophylaxis in place, or colorectal surgeons with high volumes of cancer surgery might have combination prophylaxis as a default.

Qualitative Models Versus Quantitative Risk Models

Qualitative models ascribe groups of patients to broad risk categories or "buckets" of risk that are linked to appropriate prophylaxis options for each group, without going through individualized point scoring. These models tend to be relatively easy to use and have demonstrated success in the literature and unpublished experience in reducing HA-VTE. They have sometimes been criticized for being too simplistic and for setting too low a threshold for initiating prophylaxis. This threshold varies among the different models, however, and can be adjusted to be more discriminating.

AT815 and most major international guidelines incorporate qualitative models, whereas AT9 now implicitly endorses the individualized, quantitative approach, which requires summing a cumulative point score over multiple risk factors.16 The risk factors are often weighted to reflect the variable impact of each risk factor. These quantitative, or point-based, scoring systems may be devised by expert opinion and review of the literature; they can also be derived empirically. External validation in other populations, while desirable, has only been performed on a few models.

Ideally, empirically derived models are scientifically sound and preferable to expert models, but the expert-derived models (Caprini and Padua, for example) are in more common use, and at least some of them have anecdotal evidence of effectiveness in clinical practice. The complexity of the scoring systems varies, but in general, these models have often been criticized for being difficult to implement and use—and, to date, effectiveness in reducing HA-VTE has not been demonstrated. Some of these models incorporate risk factors (e.g., length of stay, intensive care unit [ICU] days) that are not available to the provider on initial assessment and are better suited for reassessments during the stay or for raising the issue of extended duration prophylaxis. Others use only risk factors available at the time of admission to the hospital.

The next section looks at selected qualitative and quantitative models.

Qualitative/Grouping Models

The most widely used qualitative model in the United States is the "3 bucket" or University of California (UC) San Diego model, which is derived directly from tables in the AT8 guideline.15 It was disseminated widely in the earlier version of this AHRQ VTE prevention guide.17,18

In the classic "3 bucket" model (Figure 4.2), observation patients, patients with an expected hospital stay of 2 days or less, most same-day surgery patients, and patients with no acute HA-VTE risk factors are designated low risk, with a recommendation for ambulation and education. On the other end of the spectrum, patients with major, high-risk surgeries qualify for combination anticoagulant and mechanical prophylaxis. Most medical and surgical patients fall into the middle category, qualifying for anticoagulant thromboprophylaxis, unless they have bleeding risk factors.

In the original demonstration project at UC San Diego, this model was chosen after considering and rejecting more complicated individualized point-scoring systems that proved unpopular and had poor inter-observer agreement in pilot testing. In contrast, this risk assessment model was considered intuitive and easy to use. Direct observations revealed that it could be filled out in a few seconds, and there were high levels of inter-observer agreement. Integration into order sets, coupled with multifaceted interventions, resulted in marked improvements in protocol-defined adequate prophylaxis (from 58 percent to 98 percent) and reduced HA-VTE by 40 percent in medical and surgical populations without any increase in detectable bleeding or heparin-induced thrombocytopenia.19,20

Figure 4.2: Classic "3 Bucket" Model Derived From AT8

Low Risk: Minor surgery in mobile patients. Medical patients who are fully mobile. Observation patients with expected hospital stay <48 hours. No prophylaxis; reassess periodically, ambulate.
Moderate Risk: Most general, thoracic, open gynecologic, or urologic surgery patients. Medical patients, impaired mobility from baseline or acutely ill. UFH or LMWH prophylaxis*
High Risk: Hip or knee arthroplasty, hip fracture surgery, multiple major trauma, spinal cord injury or major spinal surgery, abdominal-pelvic surgery for cancer. IPCD AND LMWH or other anticoagulant*

* For those at moderate or high risk and contraindications to anticoagulation, use intermittent pneumatic compression device (IPCD).

A wide variety of other hospitals have enjoyed improved prophylaxis and reduced HA-VTE with a multifaceted approach that included variants of this VTE risk assessment model. This includes published results19,21 and many unpublished results. Some of these site success stories are available (PDF File, 764.4 KB). Large-scale VTE prevention collaborative efforts from SHM, AHRQ/QI organization partnerships, and many others have reported similar positive results, but these efforts did not have a standardized method to monitor outcomes.22,23

This model was updated (Figure 4.3) to be more discriminating in terms of a higher threshold for who receives thromboprophylaxis, in a manner more consistent with AT9 guidance to avoid prophylaxis in those at low risk. Note that medical patients without active cancer or past history of VTE must have reduced mobility and an acute illness to qualify for prophylaxis. This version offers more granular guidance at the expense of being slightly more complex.

Figure 4.3: Updated "3 Bucket" Model In Use at UC San Diego

Low Risk: Observation status, expected LOS <48 hours. Minor ambulatory surgery unless multiple strong risk factors. Medical patients ambulatory in hall and not moderate or high risk. Ambulatory cancer patients admitted for short chemotherapy infusion. No prophylaxis; reassess periodically, ambulate.
Moderate Risk (most general medical/surgical patients): Most general, thoracic, open gynecologic, or urologic surgery patients. Active cancer or past VTE/known thrombophilia in medical patient with LOS >48 hours. Medical patients with decrease in usual ambulation AND VTE risk factors (myocardial infarction, stroke, congestive heart failure, pneumonia, active inflammation/infection, dehydration, age >65). UFH or LMWH prophylaxis*
High Risk: Hip or knee arthroplasty, hip fracture surgery, multiple major trauma, spinal cord injury or major neurosurgery, abdominal-pelvic surgery for cancer. IPCD AND LMWH or other anticoagulant*

* For those at moderate or high VTE risk and contraindications to anticoagulation, use IPCD alone until bleeding risk subsides.

A model at the University of California, Davis, deserves mention as an innovative approach that has lent itself to ongoing, dynamic risk assessment and active surveillance and has been associated with a significant decrease in HA-VTE (unpublished data as of yet). Select for Model (PDF File, 298.23 KB).

Note the approach to ambulation taken in these models. Ambulation is not judged to be so protective as to eliminate the need for inpatient prophylaxis in patients with strong risk factors, such as active cancer, history of VTE/thrombophilia, and moderate to major surgery in the prior 7 days. On the other hand, most other medical conditions require reduced mobility and an acute illness to qualify for prophylaxis.

For all of these grouping variants, the following points should be kept in mind for implementation:

  • Many include critically ill ICU patients in high-risk groups (this is reasonable but not directly supported by clinical trials).
  • Patients on therapeutic anticoagulation can either be categorized as a low-risk population or be included in the contraindications to prophylaxis.
  • Note that selected populations, such as elective cardiac surgery and some OB-GYN surgery, that may have IPCDs as a preferred first choice for prophylaxis would have their own order sets and be "carved out." Alternatively, a fourth bucket for those who can have IPCD as a first choice could be added.
  • Specific options for anticoagulant choices, dosing, and timing are presented in the actual order sets. They can be presented to the provider more simply if separate order sets are provided to selected services. For example, major orthopedic surgery patients have agents no one else uses in some hospitals, and the start time for anticoagulant prophylaxis will be different in medical and surgical patients. Having different versions for these patient populations can simplify the order sets and increase acceptance.

Many other variants of grouping VTE risk assessment models are in use across the globe,24-35 including models from Australia and New Zealand,24-26 Italy,27 United States (Johns Hopkins),28-30 and Great Britain (the NHS 2010 National Institute for Health and Clinical Excellence, or NICE, guideline).31-35 Many of these models have shown clinical utility. Most are available for review in Appendix B.

Expert-Derived Quantitative (Point-Scoring) Models

Caprini pioneered individualized quantitative risk assessment models for both medical and surgical patients in the 1980s and 1990s, reasoning that a detailed and individualized risk assessment would be more accurate than those that describe broad categories of risk.36,37 The model has been revised multiple times over the years, with the most recent version depicted in Figure 4.4, and with a computerized physician order entry (CPOE) implementation example.38-41 Each individual weighted risk factor is designed to be checked off by the provider, with the cumulative score being used to place each patient into one of four risk categories, with different recommendations for each level.

The Caprini model is embedded in AT9 recommendations for VTE prophylaxis in the nonorthopedic surgical population.42 It is not mentioned in the AT9 guideline for VTE prophylaxis in medical inpatients, but it is a commonly used point-based model for medical inpatients.

The model includes a scoring system with several sets of risk factors. One set is scored as 1 point for each risk factor, the second as 2 points, the third as 3 points, and the fourth as 5 points. Each set is scored to produce a subtotal, and the four subtotals are summed to yield the total risk factor score.ii Scoring and recommended prophylaxis are noted in an article in Chest.

Figure 4.4: Scoring and Recommended Prophylaxis

Caprini Score Risk VTE Incidence Recommended Prophylaxis
0-2 Very low-low <1.5%1 Early ambulation, IPC
3-4 Moderate 3%1 LMWH; UFH; or IPC
If high bleeding risk, IPC until bleeding risk diminishes.
5-8 High 6%1 LMWH + IPC; or UFH + IPC
If high bleeding risk, IPC until bleeding risk diminishes.
>8 Very high 6.5-18.3% LMWH + IPC; or UFH + IPC
If high bleeding risk, IPC until bleeding risk diminishes.
Consider extended duration prophylaxis.

* Abdominal or pelvic surgery for cancer should receive extended VTE prophylaxis with LMWH x 30 days.1
IPC = intermittent pneumatic compression
LMWH = low-molecular-weight heparin
UFH = unfractionated heparin

Source: Based on information in Gould MK, Garcia DA, Wren SM, et al. Prevention of VTE in nonorthopedic surgical patients: antithrombotic therapy and prevention of thrombosis, 9th ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines. Chest 2012;141(2)(Suppl):e227S-e277S.

Until recently, the Caprini model was the only quantitative point-based model that had been externally validated as being predictive of VTE risk in general surgery, plastic surgery, and—in a modified form—in a sample of Jordanian oncology patients.40,43,44 Recently, the Caprini model was found to be more sensitive to VTE risk in a retrospective cohort of Chinese patients with HA-VTE than the Padua or Kucher models.45

The Caprini model also has one published report of success in clinical practice, resulting in a reduction in HA-VTE. A layered combination of provider education, provider reminders with decision support, audit and feedback, and deployment of the Caprini tool resulted in an increase in appropriate prophylaxis from 63 percent to 96 percent, with an associated reduction in HA-VTE rate in a medicine department at a tertiary care hospital center.46 In addition, the University of Michigan and University of Wisconsin both have unpublished records of success (the University of Michigan case study is presented in Chapter 5).

In spite of these impressive credentials, there are several caveats to those considering the use of individualized point-based models such as the Caprini model. First and foremost is the relative complexity of the tool and the difficulty many sites have integrating the risk assessment into order sets. Experience from collaborative improvement efforts suggests that, for many hospitals, the model is too complex to be used reliably.22,23 Clinicians often simply bypass the CDS offered in the tool rather than checking off all risk factors, adding up the point total, and identifying the appropriate prophylaxis choices based on the point total.22

A Michigan Hospital Medicine Safety Consortium funded by Blue Cross Blue Shield of Michigan and the Blue Care Network enrolled 43 hospitals in an effort to reduce HA-VTE in medical inpatients. The great majority of hospitals used the Caprini risk assessment model (RAM). The effort failed to reduce HA-VTE in a large cohort of noncritically ill patients, even in centers with relatively high adherence. It is unclear if this lack of progress is attributable to most hospitals using the Caprini model. In fact, hospitals with high rates of prophylaxis in this cohort did not have significantly lower rates of HA-VTE.47

This disconnect between higher rates of prophylaxis and VTE rates could stem from several factors. The population in this study represented a relatively low-risk patient group. Patients with any ICU days, VTE in 6 months prior to admission, and admissions that represented readmissions from the registry were excluded, and the definition of "at risk for VTE" required a score of ≥2 points, a relatively low threshold for inclusion. Median length of stay was just 4 days; although most VTE events were diagnosed postdischarge, the surveillance bias reported in surgical populations might also play a role.48

Complex point-based RAMs can suffer from poor inter-observer agreement when users attempt to apply them toward patient case scenarios in the literature; this proved the case in the pilot testing at UC San Diego.19,49 There may also be a limited discriminatory ability for low-risk patients. In an external validation study performed in surgical patients, only 0.9 percent of patients were defined as low risk not requiring prophylaxis; 10.4 percent were classified as moderate risk, in whom anticoagulation was optional.40

A closer look at sites that have documented success also raises some important caveats. There is only one published report of clinical success with reduced HA-VTE, even though the tool has been widely available for more than 30 years. The successful published site used a multifaceted approach and limited its efforts to general medicine residency teaching teams. This effort enjoyed the support and "authority gradient" from faculty attending physicians, who cosigned the VTE risk assessments.46

In the unpublished experience at the University of Michigan, success with the Caprini RAM hinged on skillful deployment of a number of CPOE techniques outlined in more detail in Chapter 5. Although electronic health records and CPOE are becoming more and more common, this level of CDS capability remains the exception rather than the rule. At the University of Wisconsin, a safety net of pharmacy providers specifically tasked with double checking the accuracy of admission VTE risk assessment ensured otherwise poor compliance with the tool. Dr. Caprini has also suggested that it is possible to capture the VTE risk information from history and physical forms, especially for elective surgical procedures.41

In summary, the Caprini VTE RAM was the first quantitative model to enjoy wide use, and until recently was the only model to be externally validated for prediction of VTE risk. The relative complexity of the model has been overcome with closely supervised environments that enjoy an authority gradient, intelligent use of sophisticated CDS, or a safety net of nonphysician providers who redundantly check accuracy of scoring. These strategies can augment the success of any VTE RAM, but they may be more of a necessity for this model. Sites considering the Caprini VTE RAM may want to carefully consider the relative strengths and limitations and consider whether they have the environment and tools demonstrated to minimize the model's limitations.

The Brigham and Women's Hospital Model (aka the Kucher model; go to Appendix B11) is a weighted scoring system with eight risk factors. Patients with a cumulative score of ≥4 points are considered to be at high risk. This model was not designed as a screening tool to be embedded in admission order sets. Rather, it was designed to define a known high-risk population to target with computerized alerts. It is not a sensitive instrument to capture all patients at risk. In a randomized trial, an increase in prophylaxis and a decrease in VTE by 41 percent resulted when computerized alerts were sent to providers of patients with scores ≥4 but not on prophylaxis.50 Physicians had to acknowledge the computer alert but could hold prophylaxis at their discretion. Similar results were obtained in an environment without the capacity for a computerized alert (in which a human alert was used instead).51 The Kucher model has not been tested as a VTE RAM embedded in order sets.

The Padua VTE RAM (Appendix B12) is derived from the Kucher model, and it is designed to address medical inpatients.52 Like the Kucher model, active cancer, previous VTE, and known thrombophilia patients receive a weighted score of 3 points, but patients with bathroom privilege level of ambulation or less are also given 3 points, along with a few other modifications of Kucher. A score of ≥4 was associated with an HA-VTE risk of more than 11 percent in patients without prophylaxis in this Italian cohort study, while those with a score of <4 (approximately 60 percent of the Italian cohort) had a VTE risk of only 0.3 percent. The high predictive value of the model in the Padua population led the AT9 guidelines to prominently highlight the Padua VTE RAM, which many have taken as an implicit endorsement of the model.14

There are several limitations and caveats to consider. The Padua results have not been externally validated. The high predictive value of this model seen in this small Italian cohort seems almost too good to be true and is not consistent with the results of much larger observational studies described later in this chapter. More than 1 percent of patients with a Padua score of 3 suffered from pulmonary embolism, raising questions about the adequacy of sensitivity in the model.1 A recent study found the Padua model inferior in predictive ability compared with the Caprini model.46

The Padua RAM has never been tested or shown to be effective as a VTE RAM in order sets. Since it is designed specifically for medical inpatients, medical centers wishing to use the Padua model require an entirely different VTE RAM for surgical populations.

Empirically Derived Quantitative (Point-Scoring) Models

Typically, in empirically derived qualitative models, a VTE risk stratification tool is developed by applying multiple logistic regression modeling to a large inpatient population. Ideally, in the next step, the model is applied to a validation sample, and the predicted VTE incidence from the model is compared with the actually observed VTE incidence. The predictive accuracy of the model is summarized in a c-statistic, the area under the receiver operating curve (ROC curve), with the best scores approaching 1.0 and the worst being 0.5.

In an ideal world, the model would go through external validation in different patient populations to assess the generalizability of the model, and then an assessment of the clinical utility of the VTE RAM would be carried out.2-4,53 To date, external validation has only been performed on one of these models (modified IMPROVE model with seven factors) and the clinical utility step has not been accomplished with any of them.

The Rogers risk assessment model was derived from more than 183,000 surgical patients.54 This complex model with 15 weighted risk factors has never been used in clinical practice and is mentioned only because the AT9 guideline recommendations for nonorthopedic surgery patients mention the Rogers model within its recommendations, along with the Caprini model.55

The Intermountain model found prior VTE, known thrombophilia, bed rest orders, and placement of a peripherally inserted venous catheter (PICC) to be the most powerful predictors of VTE in medical inpatients.56 Other risk factors, such as cancer, obesity, age >70, and other commonly reported risk factors, did not add significantly to the c-statistic score of 0.74 (originally published as 0.84, corrected in later erratum). The authors did not specifically report on how often the model would have missed VTE cases, and there is no experience reported using the model clinically.

The IMPROVE investigators leveraged a VTE registry to derive two kinds of VTE RAMs (Appendix B14) in medical patients.57 One model identified four factors available at admission that were most predictive of VTE during and up to 3 months after hospitalization. Patients with a score of 2 or 3 had a VTE risk of 1.9 percent, while those with a score of ≥4 had a risk of 5.0 percent. The authors proposed that patients with scores ≥2 (just 11 percent of the cohort) could benefit from prophylaxis with data available on admission, while the majority of patients with lower scores might not.

The predictive value of this model was relatively low with a c-statistic of 0.65. Also, setting the threshold for prophylaxis this high would essentially be giving up on preventing two-thirds of VTE in medical inpatients. A large proportion (56 percent) of the population with an IMPROVE score of 1 had a VTE risk of 1 percent, generating half of the VTE in the cohort, and this moderate threshold for prophylaxis may be appropriate for patients without significant bleeding risks. Patients with a score of zero, representing one-third of the cohort, had an observed VTE risk of 0.5 percent and suffered 17 percent of the VTE in the cohort. This study and the other large studies used to empirically derive RAMs likely portray a more realistic distribution of VTE risk than the smaller Padua study.

A second model that included three more factors that evolved over the course of hospitalization (lower limb paralysis, immobilization ≥7 days, admission to ICU or CCU during the hospital stay) was marginally better, with a c-statistic of 0.69. Patients with a score of 0 or 1 (69 percent of the medical cohort) had a 3-month VTE rate of <1/person, while those with higher scores had rates of 1.5 percent and up.

Recently, there have been two external validation studies of the 7-factor IMPROVE VTE RAM to predict VTE risk at 90 days posthospitalization. The first reported an improved c-statistic of 0.773.58 In the validation cohort, the incidence of VTE was 0.20 percent, 1.04 percent, and 4.15 percent in the low- (score 0-1), moderate- (score 2-3), and high-risk (score ≥3) groups, respectively. In the second external validation study, 68 percent of the cohort with a score of 0 to 2 had a VTE event rate of 0.42, while patients with a score ≥3 had a VTE event rate of 1.29.59 The vast majority of the VTE events occurred in the 90 days postdischarge rather than during the index admission. A length of hospital stay ≥7 days served as a proxy for prolonged immobility. The c-statistic was 0.702.

Modified versions of this second model are being deployed in clinical trials to identify potential high-risk medical patients for extended duration prophylaxis. While this approach to stratify patients for extended duration prophylaxis with the 7-factor variant is promising, it has not yet been shown to improve clinical care. Because the 7-factor VTE risk model includes some things (such as prolonged immobilization and critical care days) that are not always apparent on admission, utility as an admission VTE RAM may be limited. The IMPROVE model also provides a bleeding risk calculation juxtaposed with VTE risk (refer to  Assessing Bleeding Risk, below).

The Premier VTE Risk Model was derived through analysis of a very large database representing all regions of the United States.60 Age, sex, and 10 additional risk factors were associated with VTE during and up to 30 days after the hospital stay. They included risk factors that developed during the hospital stay as well as factors present on admission. The strongest risk factors identified were known thrombophilia, hospital stay ≥6 days, inflammatory bowel disease, central venous catheter placement, and cancer (among adults <65 years). The c-statistic for the validation set was 0.75. Their captured rate of VTE was lower than similar studies. The authors did not provide a practical weighted scoring system and, like the preceding models, this model has not been applied in clinical practice.

Table 4.1: Characteristics of Four Empirically Derived Models for VTE Risk

Characteristics Rogers53 Intermountain55 IMPROVE56 Premier58
Hospitals 142 22 52 374
Country U.S. U.S. 12 countries U.S.
Design Prospective Retrospective Prospective Retrospective
Participants Surgical Medical Medical Medical
Derivation n
Validation n
15,156 194,198
% with VTE
1.2% 0.4%
% with cancer <3% 44% 22% 14%
Followup (days) 30 90 92 30
Prophylaxis NS NS 44% 30%
Predictors 15 4 4 17
c-statistic 0.73 0.74 0.65 0.75

Table 4.1 summarizes the characteristics of these models and helps to illustrate the continuing reasons for controversy and lack of agreement among the models. There is a tenfold variation in the incidence of HA-VTE. There is variability in the proportion of patients on prophylaxis, and how this potential confounder is controlled for—or, in some cases, the proportion of patients on prophylaxis is not specified (NS). Methods to identify cases and the duration of followup after discharge varies. The cohorts used for validation vary for the distribution of important risk factors such as cancer and age. Upper extremity DVT and distal DVT are included in some models, but not others, and some include risk factors known only after a considerable length of time in the hospital.

Risk factors that are potent predictors in one model are seemingly inconsequential in the next. External validation and reports of the clinical utility of the models, with demonstrated reduction in HA-VTE, are not available for any of them. Some models, particularly the IMPROVE model, show some promise for beneficial clinical use in medical patients, especially for reevaluation of risk during hospitalization or to risk stratify for potential extended duration prophylaxis.

Assessing Bleeding Risk

Bleeding risk is weighed along with a concurrent VTE risk assessment. Bleeding risk may be increased by surgery, medications, or factors inherent to the patient. A recent observational study by the IMPROVE investigators reported on factors found to be most predictive of in-hospital bleeding in medical patients.61 Active gastroduodenal ulcer, active bleeding within 3 months prior to admission, and a platelet count <50,000 were the strongest independent risk factors. Age ≥85 years, hepatic failure with an INR >1.5, GFR <30mL/min/m2, ICU or CCU admission, central venous catheter, rheumatic disease, cancer, and male gender rounded out the list in order of descending importance.

A point-scoring quantitative model was built to predict bleeding risk, analogous to the quantitative VTE risk models. One half of bleeding episodes occurred in the 10 percent of patients with a high (≥7) score. This model has not been externally validated, and the scoring model is cumbersome to integrate into clinical practice. However, the AT9 panel considered bleeding risk to be excessive if patients had any one of the top three risk factors or multiple other risk factors.14 Note that several of these risk factors are also frequently listed as risk factors for VTE. A patient age 86 and with cancer, for example, may still be considered for prophylaxis, even though both are considered risk factors for bleeding. Most hospitals avoid complicated scoring systems for bleeding risk and instead provide lists of bleeding risk factors to consider. Explicit definitions of "leeway" times for short-lived bleeding risk factors can also guide assessment of prophylaxis in audits, as well as guide therapy at the point of care. Table 4.2 depicts one example; several others are available in Appendix B.

Table 4.2: Bleeding Risk Factors and Conditions To Consider With Pharmacologic VTE Prophylaxis

Active bleeding (last 3 months unless low risk profile on endoscopy) Intracranial bleeding within last year or until cleared by neurological services
Active gastroduodenal ulcer Intraocular surgery within 2 weeks
Platelet count <50,000, or <100,000 and downtrending Untreated inherited bleeding disorders
Therapeutic levels of anticoagulation Hypertensive urgency/emergency
Advanced liver disease with INR >1.5 Postoperative bleeding concerns*
Heparin induced thrombocytopenia (no heparinoids; consider consultation) Epidural/spinal anesthesia within previous 4 hours or expected within next 12 hours

* Leeway times:

  • 24 hours maximum for most general surgery, orthopedic surgery.
  • Status posttransplant or multiple major trauma to clear bleeding risk: 48 hours.
  • Status post spinal cord open surgery: 5 days leeway.

iiThe scoring system can be found in Bahl V, Hu MH, Henke PK, et al. A validation of a retrospective venous thromboembolism risk scoring method. Ann Surg 2010;251(2):344-50.

Next Page
Return to Contents

Page last reviewed May 2016
Page originally created February 2016
Internet Citation: Chapter 4. Choose the Model To Assess VTE and Bleeding Risk. Content last reviewed May 2016. Agency for Healthcare Research and Quality, Rockville, MD.