Evaluation of Patients With Suspected Acute Pulmonary Embolism: Best Practice Advice From the Clinical Guidelines Committee of the American College of Physicians

Although pulmonary embolism (PE) due to thrombotic occlusion of the main or branching pulmonary arteries is common (1), it remains difficult to diagnose owing to the nonspecific signs, symptoms, and risk factors with which it is associated (2, 3). Acute PE can lead to significant morbidity and mortality (4, 5), and patients presenting to their physicians or to an emergency department (ED) with cardiopulmonary symptoms are often evaluated for the disease.

Because no individual risk factor, patient symptom, or clinical sign can definitively diagnose or exclude PE (6), clinical decision tools have been developed to help guide clinicians during their evaluation of patients with suspected acute PE. These decision tools (discussed below) are meant to help physicians stratify patients into groups for whom different diagnostic strategies are appropriate: those for whom PE is so unlikely that they need no further testing, those for whom plasma d-dimer testing can provide additional risk stratification, and those who are at high enough risk that imaging is indicated.

Highly sensitive plasma d-dimer tests (those that measure the level of this fibrin degradation product by using enzyme-linked immunosorbent assays) can be used to rule out PE in patients with low or intermediate pretest probability of PE, whereas older latex or erythrocyte agglutination assays can only rule out PE in patients with low pretest probability (7, 8). For the purposes of these guidelines, we will assume that highly sensitive d-dimer assays are being used.

Computed tomography (CT) has become the predominant imaging modality used for the diagnosis of PE. Although the use of CT for the evaluation of patients with suspected PE is increasing in the inpatient, outpatient, and ED settings (9–14), no evidence indicates that this increased use has led to improved patient outcomes. In fact, evidence suggests that many of the PEs diagnosed with increasing use of CT may be less severe (15–17). As a result, although the incidence of PE has risen significantly with the use of CT, there has been minimal or no associated change in mortality (9, 10). This questionable benefit of increased testing, in combination with the significant expense of PE evaluations and the unintended costs of follow-up imaging needed for incidental findings discovered on these potentially inappropriate CTs (5, 18), has led some to conclude that current practice patterns for the evaluation of PE are not cost-effective (5, 19–21).

Given this lack of clear benefit, the potential risks from CT make its increasing use even more concerning. Radiation from CT is thought to be a risk factor for cancer. In one recent study, a cohort of children with CT exposure was found to have a significantly higher incidence of leukemia and brain tumors later in life (22–24). Given the radiosensitive thoracic and breast tissue imaged during the evaluation of patients with suspected PE, this potential risk is concerning, especially in women. In addition, the contrast dye used in CTs for the evaluation of PE may cause nephropathy (25, 26). These risks are compounded by the fact that repeated imaging may be common: In one study performed at a large academic center, at least one third of ED patients who had CT for the evaluation of PE underwent another CT for the same reason within 5 years (27).

With the rising cost of PE evaluations, along with increasing awareness of potential harm and doubts about mortality benefits (5), a more focused strategy is needed. This report aims to present an evidence-based and high-value diagnostic strategy for the diagnosis of PE. Its goal is to help clinicians understand the potential hurdles to such an approach and outline performance improvement strategies to overcome them.

Methods

The literature search encompassed studies indexed by MEDLINE (1966–2014; English-language only) by using the search terms ((pulmonary embol* or pulmonary thromboembol*) and (diagnosis or diagnostic)), limited to meta-analyses; clinical trials; and randomized, controlled trials. This resulted in 1752 articles, including studies of decision rules, laboratory tests, and imaging studies for the diagnosis of PE. One author reviewed all titles and selected relevant abstracts. Articles found to be germane to this Best Practice Advice publication were independently reviewed for incorporation into this manuscript by 3 authors, to ensure that all Best Practice Advice statements were based on the highest-quality evidence; disagreements were resolved via discussion. Notably, however, this document is not based on a formal systematic review. Instead, it seeks to provide practical advice based on the best available evidence.

The target audience for this publication is all clinicians; the target patient population is all adults, both inpatient and outpatient, suspected of having acute PE.

Results

What Are the Evidence-Based Recommendations for Use of Laboratory and Imaging Tests in Patients With Suspected Acute PE?

Clinical guidelines advocating for the focused evaluation of patients with suspected PE have been published by professional societies, including the American College of Physicians/American Academy of Family Physicians (28), the American College of Emergency Physicians (29), and the European Society of Cardiology (30). These guidelines are all based on the use of Bayesian analysis, in which pretest probability is combined with elements from the history, physical examination, and laboratory results to identify patients at such low risk for PE that further testing is both unnecessary and may lead to false-positive results. These analyses involve the use of clinical decision tools or clinician gestalt to determine whether individual patients require additional testing (either plasma d-dimer measurement or diagnostic imaging) on the basis of risk stratification (3).

Although clinician gestalt varies and quality is probably dependent on expertise and familiarity with pathophysiology and presentation of PEs (31, 32), the overall accuracy of an experienced clinician's gestalt seems to be similar to that of structured decision tools (33). However, a benefit of decision tools is that they help standardize the evaluation for clinicians who find themselves only infrequently evaluating for PE.

Most of these decision tools—including the original Wells criteria, the dichotomized Wells criteria, and the simplified Wells criteria (Appendix Table 1) (34, 35), as well as the revised Geneva score and the simplified Geneva score (Appendix Table 2) (36, 37)—use d-dimer testing for patients at lower risk for PE, with the aim of avoiding unnecessary CT if d-dimer levels are normal. Of note, the specificity of an elevated level may be lower in inpatients than in outpatients or ED patients, probably owing to comorbidities in the inpatient population (38, 39). However, d-dimer testing as an initial step for inpatients suspected of having PE is still appropriate, because it remains highly sensitive for the disease and a normal level, in combination with appropriate pretest risk stratification, can prevent unnecessary imaging (40, 41). Both the Wells and Geneva tools have been externally validated, but neither has been found to be superior to the other or to risk stratification by using clinician gestalt (6, 32, 42).

Appendix Table 1. Wells Prediction Rule for Pretest Probability of PE*

Appendix Table 2. Revised Geneva Score for Predicting Pretest Probability of PE*

Earlier data had suggested that d-dimer testing was appropriate only for risk stratification of the lowest-risk patients, and that patients at intermediate risk for PE need imaging (34). However, 3 more recent studies have demonstrated that a normal high-sensitivity d-dimer level can be used to further risk-stratify patients at both low and intermediate risk for PE. The first study, by Perrier and colleagues (43), enrolled 674 non–high-risk patients (at either low or intermediate risk for PE). Those with normal d-dimer levels were followed for 3 months, and no thromboembolic events were noted.

The latter 2 studies both looked specifically at intermediate-risk groups: Warren and Matthews (44) used the Wells criteria, and Gupta and colleagues (45) used the revised Geneva score. They evaluated 1679 and 330 patients, respectively, who were determined to be at intermediate risk for PE and found that a normal d-dimer level was 99.5% and 100% sensitive, respectively, for excluding PE on CT.

The most recent decision tool was developed in response to growing use of d-dimer testing (a test with known low specificity) among patients with the wide range of signs and symptoms potentially suggestive of PE. The Pulmonary Embolism Rule-Out Criteria (PERC) (Table 1) were specifically developed to help guide clinicians in identifying low-risk patients in whom the risks of any testing, including a plasma d-dimer level, outweigh the risk for PE (about 1%) (46–49). The PERC are not a screening tool for all patients, but rather are meant to be applied to patients in whom a clinician has a genuine concern about PE and whose initial risk stratification identifies them as being at very low risk. When used in this manner, the PERC should decrease the use of d-dimer testing only in patients who would have otherwise been tested, rather than increase d-dimer testing in patients in whom PE is not reasonably suspected. A recent large meta-analysis of 12 studies determined that the overall proportion of missed PEs by using PERC was only 0.3% (44 of 14 844 total cases) (49). The pooled sensitivity of PERC for all 12 studies was 97% (95% CI, 96% to 98%), and the pooled specificity was 22% (CI, 22% to 23%), indicating that 22% of d-dimer tests could have been safely avoided had the PERC been universally applied.

Table 1. Pulmonary Embolism Rule-Out Criteria for Predicting Probability of Pulmonary Embolism in Patients With Low Pretest Probability*

The low specificity of d-dimer testing has also resulted in changes to the acceptable normal ranges of the plasma test. To date, most recommendations have considered any value above 499 ng/mL as elevated. However, some studies have used age-adjusted d-dimer cutoffs, and a recent meta-analysis of 13 studies and 12 497 patients without high pretest probability found that the use of age-adjusted d-dimer cutoffs for patients older than 50 years (age × 10 ng/mL) maintained a sensitivity for PE above 97% while significantly increasing specificity (50).

Taking into account all of this evidence, below are the current evidence-based, high-value approaches to the diagnosis of PE (Figure 1).

• Download figure
• Download PowerPoint

Diagnostic Approach for Patients With Low Pretest Probability of PE

In patients believed to be at low risk for PE, the PERC criteria should be applied. In those who meet all 8 PERC criteria (Table 1), the risk for PE is lower than the risks of testing; do not order a plasma d-dimer test (49). Those who do not meet all of the criteria should be further stratified by using a plasma d-dimer test. A normal plasma d-dimer level (ideally, age-adjusted [age × 10 ng/mL] but otherwise <500 ng/mL) provides sufficient negative predictive value for PE; do not order imaging studies (28). An elevated plasma d-dimer level should lead to imaging studies.

Diagnostic Approach for Patients With Intermediate Pretest Probability of PE

For patients at intermediate risk for PE, d-dimer testing is warranted. As for patients at low pretest probability, a normal plasma d-dimer level (ideally, age-adjusted [age × 10 ng/mL] but otherwise <500 ng/mL) provides sufficient negative predictive value for PE; no imaging studies are indicated (43–45). An elevated plasma d-dimer level should prompt imaging studies (43–45).

Diagnostic Approach for Patients With High Pretest Probability of PE

For patients with high pretest probability of PE according to either clinician gestalt or a clinical prediction tool, imaging studies should be performed. Computed tomographic pulmonary angiography (CTPA) is the preferred method of diagnosis when it is available and there is no contraindication to radiographic contrast dye. Ventilation–perfusion (V/Q) lung scanning should be used when CTPA is unavailable or contraindicated. Of note, a d-dimer assay should not be obtained in patients with a high pretest probability of PE because a negative value will not obviate the need for imaging (Figure 1).

Does Practice Follow the Evidence?

Although professional society guidelines and well-validated decision tools exist to determine which patients should undergo work-up for suspected PE, current practice does not follow guidelines (51). Retrospective chart reviews of ED, inpatient, and outpatient data have demonstrated that a substantial proportion of patients with suspected acute PE who are risk-stratified as low or intermediate either have no plasma d-dimer value obtained or go on to have CT despite normal d-dimer levels, both of which are contrary to guidelines for these patients (51–55). Conversely, many patients who have elevated d-dimer levels (which, if obtained, should be used to determine the need for additional imaging) do not have follow-up CT, again contrary to evidence-based guidelines (52–55).

What Factors Promote the Overuse of Imaging in Patients With Suspected Acute PE?

Overuse of imaging is driven by physician-, patient-, and systems-level factors. Several issues may underlie physicians' tendencies to overuse imaging tests in the evaluation for PE. Some physicians, especially those who do not evaluate for PE regularly, may assume that each epidemiologic risk factor for PE contributes to an individual patient's predicted risk for PE when using a validated prediction tool (for example, the Wells criteria). Many population-level risk factors do not add meaningfully to those that have been included in a validated risk prediction tool. For example, the analyses leading to the Wells criteria found that adding family history, the postpartum period, or lower-extremity fracture to the risk factors that were included in the final tool did not add to its performance (34). Similarly, pregnancy, the postpartum period, and a history of congestive heart failure or stroke are not included in the Geneva score because they did not add to this tool's predictive performance. Notably, a recent large meta-analysis confirmed that pregnancy itself does not confer a greater risk for acute PE (56, 57). Thus, physicians may note the presence of an epidemiologic risk factor without realizing that it is not part of validated decision algorithms.

In addition, because PEs may be life-threatening events, physicians may believe that they need to rule out the condition even if its likelihood is extremely low. This is compounded by the fact that some physicians may be less than comfortable with the Bayesian analyses described above, either owing to a lack of training or because they simply feel more comfortable with tests that they perceive as giving a dichotomous true/false answer, such as CT (58). Furthermore, even when physicians are aware of the low likelihood of PE, the fear of litigation from missing a potentially fatal diagnosis may lead them to order imaging tests anyway (59, 60). The use of the well-validated decision tools outlined above can help avoid these issues by providing clinicians who infrequently evaluate for PE with pretest probability estimates that are as accurate as the gestalt of experts, and by providing risk-averse clinicians with evidence-based guidance.

Patients themselves may prefer to have their history and physical examinations supplemented by laboratory tests and imaging for certain symptom presentations (61). This may be a result of insufficient patient–physician communication about their individual risk factors and how they fit into validated decision algorithms; although this patient-centered communication does take more time, it has also been found to be associated with less use of diagnostic testing (62).

Finally, several trends in the health care system have resulted in physicians being more likely to use imaging, especially CT, for patient evaluation. First is the growth in availability and use of CT scanners in EDs in general, where many patients with suspected PEs are evaluated (12, 63). In addition, CTPA for suspected PE can be ordered and carried out quickly, and the tests interpreted rapidly, especially with the growth of after-hours remote radiology services.

Second is the phenomenon of supply-sensitive care: There is a documented connection between the availability of health care technologies, including imaging technology, and their use (64). Thus, the growth in availability of imaging modalities not only makes their use easier, but may also promote overuse. Reasons for this connection may include cultural tendencies, as more providers are trained in environments with rapid access to newer imaging tests, as well as the financial incentives given to providers and hospitals from the use of imaging.

Finally, the desire to determine the cause of symptoms may prompt unwarranted imaging. Although the advice given here is specifically related to the evaluation of patients with suspected acute PE, evidence-based reasoning should be applied when chest imaging for the evaluation of any cardiopulmonary symptoms is being considered.

How Can Physicians Reduce Overuse of Imaging for Patients With Suspected Acute PE?

Physicians can reduce the use of imaging for PE by focusing on several critical decisions in the work-up of PE. First, identify which patients require any diagnostic testing at all. The mere presence of 1 or more of multiple symptoms that could be consistent with PE does not always indicate that testing is needed. Clinical judgment is needed to determine whether a patient requires evaluation, and if the decision is made to do so, the clinician should determine the patient's pretest probability of PE by using a validated decision rule or their gestalt.

In patients with low pretest probability, application of PERC can safely identify patients for whom diagnostic testing is not necessary (48, 65, 66). If a patient with low pretest probability of PE meets all 8 PERC criteria, their likelihood of PE is 0.3% and no further testing is required (49). In a meta-analysis of 12 studies, which included 14 844 patients, the PERC were found to have a sensitivity of 97% (49). By avoiding d-dimer testing in these low-risk patients, physicians can avoid false-positive d-dimer results and subsequent CT, which is unnecessary. Of note, the PERC should not be applied to patients at intermediate or high risk for PE.

Second, the diagnostic testing strategy for patients with low pretest probability who do not meet all of the PERC (as well as all patients with intermediate pretest probability) should begin with d-dimer testing. Studies have shown that approximately one third of ED patients who receive CT for PE either did not have a d-dimer test performed or had a negative d-dimer result (51). On the other hand, implementation of pathways to standardize the appropriate use of d-dimer in the evaluation of patients with suspected PE decreased use of CT in one Australian tertiary care hospital by 27% (67).

Third, age-adjusted d-dimer thresholds should be used to determine whether imaging is warranted. Although it is highly sensitive, plasma d-dimer testing is nonspecific, and false-positive results can lead to unnecessary imaging. In the meta-analysis mentioned above, the use of an age-adjusted threshold of age × 10 ng/mL (rather than a generic value of 500 ng/mL) resulted in maintenance of sensitivities greater than 97% in all age groups (50). In addition, specificities increased significantly in all age groups, from 57.6% to 62.3% in patients aged 51 to 60 years, 39.4% to 49.5% in patients aged 61 to 70 years, 24.5% to 44.2% in patients aged 71 to 80 years, and 14.7% to 35.2% in patients older than 80 years. These findings were confirmed in a recent large multicenter, multinational prospective trial in which the use of age-adjusted d-dimer testing resulted in maintenance of sensitivity and a significant increase in specificity for the diagnosis of acute PE (68). Given these results, we recommend using age-adjusted d-dimer thresholds to determine d-dimer elevation in patients older than 50 years.

Finally, physicians, hospitals, and EDs should develop diagnostic and treatment pathways for patients with a history of multiple CTs for PE. Typically, such patients have a history of PE and have recurrent symptoms that are suspicious for PE, such as chest pain. Each time they present to their physician or the ED, CT might be performed because their pretest probability may be high. At least 1 study has demonstrated that patients evaluated for PE by using CT had a significant probability of having another CT performed for PE within 5 years. In fact, 5% of the patients in this study had 5 or more CTs for PE (27).

Preventing this frequent use of repeated CT requires thoughtful planning. Clinicians should educate these patients about the risk of radiation from multiple CTs. When such patients develop symptoms, providers should review them in the context of their prior symptoms and discuss testing strategies with the patients and their primary care providers. An individualized approach to testing is reasonable, including lower-extremity venous ultrasonography or V/Q scanning when appropriate (although V/Q scanning may not be useful in patients with chronic obstructive pulmonary disease, pneumonia, or pulmonary edema). For patients in whom V/Q scanning cannot be done, lower-extremity venous ultrasonography can be used; magnetic resonance imaging should not be done, because it has not been found to have the sensitivity necessary to detect segmental or subsegmental PEs (69, 70).

Several alternative approaches to the work-up of PE may also be beneficial. One approach is to perform lower-extremity venous ultrasonography before CT (71, 72). In hemodynamically stable patients with lower-extremity symptoms, identifying deep venous thrombosis can eliminate the need to perform CT, because the need for anticoagulation will have already been established. For the patients in this group who have cardiothoracic symptoms, the need for long-term anticoagulation (for example, in cases of unprovoked or recurrent PE) can be determined after the initial treatment period. This approach has particular utility in pregnant patients in the first trimester with suspected PE, in whom the risks and benefits of CT should be weighed even more carefully. Although CT exposes these patients to less radiation than V/Q imaging does (73), it may have teratogenic effects, making the use of lower-extremity ultrasonography in patients with lower-extremity symptoms a valid strategy. A summary of imaging suggestions for patients with various clinical scenarios in whom PE is suspected is provided in Table 2.

Table 2. Suggestions for Imaging in Patients With Suspected PE

Another approach is to actively engage patients in the diagnostic process and use informed decision making to help reduce testing. Studies have shown that the use of decision tools and quantitative estimates to educate patients with chest pain about their risk for acute coronary syndrome result in both lower health care resource utilization and higher patient satisfaction scores (75, 76). Similarly, at least one study has shown that the use of evidence-based decision aids (to demonstrate to patients the comparative risks for PE and of any diagnostic tests) may reduce imaging in patients with suspected acute PE (77). Ideally, this kind of shared decision-making model would allow patients to weigh their options and decide, with their physicians, whether to pursue laboratory testing or CT. Nevertheless, some patients may prefer to have CT even if it is inappropriate (61), necessitating that physicians continue to act as the final decision makers regarding diagnostic testing.

Physician practices and EDs can also reduce the use of CT for PE by implementing systems-based processes to monitor utilization and appropriateness of CT for PE. A promising intervention to improve appropriateness is integrated computerized clinical decision support. One version of this prompts the ordering clinician to document the pretest probability using one of the validated clinical decision rules and d-dimer results (when appropriate). If the patient does not fall into the appropriate testing group, clinical decision support can offer alternatives, such as ordering a d-dimer test, and can offer resistance to test ordering. This resistance can range from requiring the clinician to attest to the indication to requiring formal authorization from a utilization review clinician on call.

Although computerized point-of-care clinical decision support has been shown to reduce ordering of CTs for PE in EDs using electronic health records (63, 78–80), it is also possible to add basic decision support without an electronic health record by using a simple radiology order form that requires the clinician to document the pretest probability and d-dimer result (for an example, see reference 67) (67, 74). The radiologist can then review this form before beginning the study and discuss any inappropriate imaging orders with the ordering clinician. These interventions may result in true patient-oriented outcomes; a recent prospective, randomized, controlled trial of clinical decision-support effectiveness determined that both radiation dose and the cost of medical care decreased for patients with chest pain and dyspnea for whom decision support regarding PE and acute coronary syndrome were provided (81).

Of course, decision support does not have to focus on risk stratification alone; it can also present the risks and costs of alternative tests. In fact, a recent survey of U.S. physicians found that the majority would agree with the statement that “Decision support tools that show costs would be helpful in my practice” (82).

Given physician variation in use of imaging (83–85), quality improvement approaches to utilization and appropriateness of CT for PE can be used to identify and address physician practice variation and track trends in use of CT (84). If a practice has computerized decision support, it is also possible to track and give feedback on appropriateness; however, administrative data alone can be used to track utilization and diagnostic yield (80). Although there is no “ideal” or “correct” level of utilization or diagnostic yield, comparisons within a group of providers in a similar practice or ED allow the measurement of meaningful variation. Physicians with higher utilization and lower diagnostic yield can then receive focused interventions, such as education, feedback, and chart review. It is possible that such measures may be adopted by payers or health systems into public reporting or pay-for-performance programs in the near future, especially because the use of CT for PE has been identified as a test whose potentially inappropriate use can be improved upon (21).

Conclusion

The first step when evaluating a patient with suspected acute PE is to establish his or her pretest probability of PE. The Wells and Geneva rules have been validated and are considered equally accurate in predicting the probability of PE, as is clinician gestalt when used for risk stratification. The PERC were specifically developed to help clinicians identify patients with low pretest probabilities of PE in whom the risks of any testing outweigh the risk for PE.

ACP Best Practice Advice

Best Practice Advice 1: Clinicians should use validated clinical prediction rules to estimate pretest probability in patients in whom acute PE is being considered.

Best Practice Advice 2: Clinicians should not obtain d-dimer measurements or imaging studies in patients with a low pretest probability of PE and who meet all Pulmonary Embolism Rule-Out Criteria.

Best Practice Advice 3: Clinicians should obtain a high-sensitivity d-dimer measurement as the initial diagnostic test in patients who have an intermediate pretest probability of PE or in patients with low pretest probability of PE who do not meet all Pulmonary Embolism Rule-Out Criteria. Clinicians should not use imaging studies as the initial test in patients who have a low or intermediate pretest probability of PE.

Best Practice Advice 4: Clinicians should use age-adjusted d-dimer thresholds (age × 10 ng/mL rather than a generic 500 ng/mL) in patients older than 50 years to determine whether imaging is warranted.

Best Practice Advice 5: Clinicians should not obtain any imaging studies in patients with a d-dimer level below the age-adjusted cutoff.

Best Practice Advice 6: Clinicians should obtain imaging with CTPA in patients with high pretest probability of PE. Clinicians should reserve V/Q scans for patients who have a contraindication to CTPA or if CTPA is not available. Clinicians should not obtain a d-dimer measurement in patients with high pretest probability of PE.

Figure 2 summarizes the recommendations and clinical considerations.

• Download figure
• Download PowerPoint