Diagnostic Utility of Ultrasonography of Leg Veins in Patients Suspected of Having Pulmonary Embolism

Methods

Patients

Patients were eligible for the study if they were 18 years of age or older and underwent perfusion-ventilation lung scanning for a diagnostic work-up of pulmonary embolism at the Academic Medical Center in Amsterdam, the Netherlands. All patients were primarily referred because pulmonary embolism was clinically suspected (outpatients) or because they developed signs or symptoms of pulmonary embolism during hospitalization for another illness (inpatients). All patients were scheduled to undergo ultrasonography as soon as possible; this test was done independently of the other tests. All patients were prospectively followed for 6 months. The study protocol was approved by the institutional review board, and informed consent was obtained for all patients.

Diagnostic Methods

Perfusion lung scanning was done in all patients after the administration of 100 MBq of ^99mTechnetium-labeled macroaggregates of albumin. Six views were routinely obtained: anterior, posterior, left and right lateral, and left and right posterior oblique. Lung scans were interpreted by using an anatomic-segment lung chart [25] and were considered normal if no perfusion defects were seen in any of the six projections. If segmental or larger defects were seen, ventilation lung scanning was done using ^81mKrypton gas. Pulmonary embolism was considered to be excluded if the lung scan was normal and was considered to be proven if a high-probability scan (that is, a scan showing at least one segmental perfusion defect with locally normal ventilation [1, 7]) was obtained.

Selective pulmonary angiography was attempted in all patients who had a nondiagnostic lung scan. Angiography involved a modified Seldinger approach with a 6.7F-braded, multiple side-holed, Grollman-type pig-tail catheter. The angiogram was classified according to standard definitions as normal, indicative of pulmonary embolism, or inadequate for interpretation [1, 2, 9].

B-mode gray-scale compression ultrasonography was done with a 7.5-MHz linear-array sonographic scanner. While the patient was in the supine position, the common femoral vein was visualized at the inguinal ligament; the adjacent artery was used as a reference point. The popliteal vein was scanned while the patient was in the prone or lateral decubitus position, and the transducer was placed posteriorly in the mid-popliteal fossa. For evaluation of the distal popliteal vein, the transducer was moved slowly from the popliteal fossa along the calf until the trifurcation of the calf veins was visualized. No attempt was made to visualize the calf veins. Ultrasonographic results were considered abnormal (that is, consistent with the presence of deep venous thrombosis) if a venous segment could not be completely compressed [19-21]. All patients underwent bilateral compression ultrasonography, which was done by an independent investigator who was not aware of the results of lung scanning or pulmonary angiography. Results of compression ultrasonography were not forwarded to the referring physician, and decisions about anticoagulant treatment were based on the results of lung scanning or pulmonary angiography.

Statistical Analysis

The rate of abnormal compression ultrasonography in patients in whom pulmonary embolism was proven (sensitivity) was calculated for all patients and for patients with the diagnosis of pulmonary embolism by using as a conjoint gold standard a high-probability lung scan or a non-high-probability lung scan plus a subsequent abnormal angiogram. We also determined the rate of abnormal ultrasonographic results in patients in whom pulmonary embolism was excluded by either a normal lung scan or a normal angiogram (1 −specificity). Finally, the rate of abnormal results on compression ultrasonography was calculated for patients whose diagnosis of pulmonary embolism was uncertain because angiography could not be performed or because the result could not be interpreted.

The possible contribution of compression ultrasonography to the diagnostic management of symptomatic patients was assessed by 1) calculating the number of lung scans and angiograms that could be avoided if compression ultrasonography yielded abnormal results and 2) determining the number of patients who would be inappropriately treated with anticoagulation because of false-positive ultrasonographic results. These calculations were done by applying the sensitivity and specificity of compression ultrasonography obtained in our study to a hypothetical population of 1000 patients suspected of having pulmonary embolism; the proportional distribution of lung scanning and angiography results were the same as those seen in our study.

To minimize bias, all 397 patients (including the 40 patients in whom ultrasonography was not done) were used to calculate the necessary figures. We assumed that the prevalence of pulmonary embolism in the 30 patients without a diagnosis (those whose lung scan was nondiagnostic but who did not undergo pulmonary angiography) was 27%, as was seen in the remaining patients who had a nondiagnostic lung scan. Furthermore, the sensitivity and specificity of ultrasonography in these 30 patients were assumed to be similar to those obtained in the cohort of patients with a nondiagnostic lung scan in whom angiography was done. This resulted in an overall calculated prevalence of pulmonary embolism of 41.1%, with a high-probability lung scan in 30.2% of these patients and a nondiagnostic lung scan in 40.6%. Pulmonary embolism was considered to be present in 27% of patients who had a nondiagnostic lung scan (if angiography had been performed). Finally, we assumed that all patients with abnormal ultrasonographic results would be treated with anticoagulant agents without further testing.

Results

A total of 397 consecutive patients who were clinically suspected of having pulmonary embolism were enrolled (Figure 1). The mean age was 56 years (range, 18 to 92 years); 223 patients (56%) were women, and 206 (52%) were outpatients. Twenty-four percent of patients had cancer, 21% had recently had surgery, 12% had congestive heart failure, and 12% had a history of venous thromboembolism. No risk factor was seen in 26% of patients. The median interval between the onset of symptoms and diagnostic investigations was 2 days. Compression ultrasonography could not be done in 40 patients. Thirty of these patients had a normal perfusion lung scan; 22 of the 30 were outpatients for whom compression ultrasonography could not be arranged before they left. Compression ultrasonography was not performed in 4 patients who had a high-probability lung scan and 6 patients who had a nondiagnostic lung scan (angiography results were abnormal in 1 patient and normal in 2; angiography was not done in 3 patients). Treatment decisions in these 40 patients were made on the basis of lung scans and angiography results, as was done for the 357 patients in whom ultrasonography was performed.

View larger version:

• In this window
• In a new window

Figure 1. In patients with a normal lung scan (Q scan) and in patients with a normal angiogram, no anticoagulant therapy was given. All patients in whom pulmonary embolism was proven according to a high-probability lung scan (VQ scan) or abnormal angiogram were given long-term anticoagulation. CUS = compression ultrasonography; PE = pulmonary embolism. Flow diagram of test outcomes of 397 patients suspected of having pulmonary embolism.

The study sample therefore consisted of 357 patients (Figure 1). All of these patients underwent lung scanning that yielded an interpretable result. In 86 patients (24%), lung scans were normal; as a result of these normal findings, anticoagulation was not started or heparin was withdrawn. The lung scan was consistent with a high probability of pulmonary embolism in 116 patients (33%); because of these findings, anticoagulation was either started or continued for the long term. In the remaining 155 patients (43%), the lung scan was nondiagnostic. As specified in the study protocol, all patients with a nondiagnostic scan were scheduled to have pulmonary angiography. This test could not be performed in 28 patients (18%) because of such conditions as manifest heart failure, renal insufficiency, and pulmonary hypertension. In 2 other patients who had a nondiagnostic lung scan, the angiogram could not be interpreted. In all of the 30 patients who had an inadequate angiogram, decisions about anticoagulation were made on clinical grounds; as a result, 8 patients were treated. During 6 months of follow-up, 1 of these 8 patients had a fatal hemorrhage; none of the 8 had an episode of recurrent venous thromboembolism. Of the 22 untreated patients, 1 had confirmed recurrent pulmonary embolism (the ultrasonographic results were initially normal). Five of these 22 patients died of underlying illnesses.

Of the 125 patients with an interpretable angiogram, 33 (26%) had an abnormal result; anticoagulation was therefore started or continued. Pulmonary angiograms were normal in the other 92 patients; thus, no anticoagulation was given.

Forty-three of 149 patients with proven pulmonary embolism had abnormal results on ultrasonography of the leg veins (sensitivity, 29% [95% CI, 22% to 37%]) (Table 1). The sensitivity of ultrasonography was 30% in the patients who had a high-probability lung scan and 24% in the patients with angiographically proven emboli (in whom the lung scan was nondiagnostic). Of the 178 patients in whom pulmonary embolism was excluded by a normal lung scan or normal angiogram, 5 patients had abnormal results on compression ultrasonography (specificity, 97% [CI, 94% to 99%]). Two of these patients had recently had surgery, 1 patient had been hospitalized for septicemia and had a history of pulmonary embolism, and 2 patients were outpatients with no known risk factors. None of these 5 patients had symptoms suggesting deep venous thrombosis, and none received anticoagulant therapy; their test results were considered to be false positive. No venous thromboembolic complications occurred in these 5 patients during the 6-month follow-up period. In the remaining group of 30 patients in whom angiography could not be performed or in whom the result of angiography could not be interpreted, 4 (13%) had abnormal results on compression ultrasonography.

The potential role of compression ultrasonography in the diagnostic management of patients suspected of having pulmonary embolism was analyzed in terms of reduction in the need for performing other diagnostic tests and in terms of the proportion of patients who received unnecessary anticoagulant therapy because of false-positive results on compression ultrasonography (Table 2).

Compared with a strategy of perfusion-ventilation lung scanning followed by angiography if the lung scan is nondiagnostic, performance of compression ultrasonography before lung scanning in the diagnostic work-up would reduce the number of lung scans by approximately 14% and the number of angiograms by about 9%. However, because some results of compression ultrasonography were falsely abnormal, approximately 4% of all patients treated with anticoagulant agents would receive these drugs unnecessarily. When calculated as a proportion of patients with an abnormal ultrasonographic result, this strategy results in unnecessary treatment for 18 of 137 patients (13.1%). When ultrasonography is done only in patients whose lung scan before angiography was nondiagnostic, the number of angiograms required is reduced by 9%; with this strategy, about 2% of all treated patients would receive unnecessary anticoagulant therapy. However, the proportion of patients with abnormal ultrasonographic results who receive unnecessary anticoagulation increases to 25.7% (9 of 35 patients).

A sensitivity analysis in the hypothetical cohort of 1000 patients was done using various assumptions about the presence of pulmonary embolism in the two groups of patients who did not undergo ultrasonography (40 patients) or who had an inadequate angiogram (30 patients) showed that the number of false-positive results ranged from 15 to 18 when ultrasonography was the first test performed and from 8 to 11 when lung scanning was done first and ultrasonography was done only if the scan was nondiagnostic. Thus, the results of this analysis did not change the conclusions about the potential role of compression ultrasonography in the diagnostic management of patients suspected of having pulmonary embolism.

Discussion

We determined the accuracy of compression ultrasonography in consecutive patients who were suspected of having pulmonary embolism, and we evaluated the potential clinical utility of this method in the diagnostic work-up of these patients.

Among the patients with proven pulmonary emboli, only 29% (CI, 22% to 37%) had abnormal results on compression ultrasonography that indicated deep venous thrombosis. This sensitivity is substantially lower than that found in patients in whom deep venous thrombosis is clinically suspected; many previous studies have documented a sensitivity exceeding 95% [21]. The 29% sensitivity that we observed is also lower than the overall sensitivity of approximately 50% seen in nonsymptomatic patients at high risk for venous thromboembolism (for example, those who have recently had major surgery or stroke) [22-24].

How can this lower sensitivity in patients with proven pulmonary embolism be explained? The difference in the sensitivity of compression ultrasonography between symptomatic and nonsymptomatic patients at high risk for thrombosis may be attributable to a difference in characteristics of the thrombi [21]. Venous thrombi in nonsymptomatic patients tend to be smaller and less organized and therefore softer. They are frequently nonocclusive and are often limited to the calf. The low sensitivity found in our study may be partially explained by the same phenomena because patients with pulmonary embolism usually do not have symptoms of venous thrombosis in the legs. Furthermore, the thrombi in patients with pulmonary embolism may have been completely or partially dislodged, leaving only small residual thrombi (especially in the calves) for detection. This hypothesis is supported by the observation that bilateral contrast venograms in patients with proven pulmonary embolism may be normal in as many as 10% to 30% of patients [17, 26]. Moreover, thrombi are limited to the calf in approximately 30% to 60% of patients who have an abnormal venogram [26, 27].

As a consequence, it may not be unexpected that an abnormal result on compression ultrasonography would be observed in only 20% to 40% of patients with proven pulmonary embolism. This figure agrees with our findings, which in turn confirm those of earlier retrospective studies [28-30]. In our study, a similar sensitivity was seen in the two subgroups of patients with pulmonary emboli (those in whom the diagnosis was based on a high-probability lung scan and those in whom the diagnosis was based on a nondiagnostic lung scan and a subsequently abnormal angiogram). Other studies in patients with pulmonary embolism have reported a higher sensitivity for ultrasonography [31-33]. One of these studies, however, had a small sample size [31]; in the other two studies [32, 33], ultrasonography was performed with knowledge of the outcome of other tests, which introduces the potential for diagnostic suspicion bias. In our study, the ultrasonographer was blinded to the results of other diagnostic tests for pulmonary embolism.

It is unlikely that the use of compression ultrasonography rather than color or duplex ultrasonography explains the relatively low sensitivity because the sensitivity of these techniques is similar in both symptomatic and asymptomatic patients [21]. Finally, it could be argued that the sensitivity of compression ultrasonography in patients with pulmonary embolism would be increased if the calf veins are also examined. However, we decided not to extend the examination of the popliteal vein beyond the trifurcation in the calf because previous studies have shown that extension to the individual calf veins results in a substantial decrease in specificity and, therefore, a decrease in the positive predictive value of an abnormal test result [21]. In patients in whom the diagnosis of pulmonary embolism was excluded, the rate of false-positive results on ultrasonography of the leg veins was 3% (CI, 0.9% to 6.4%; specificity, 97%). This low rate is probably the result of the ultrasonographic examination strategy. The high specificity of compression ultrasonography that we saw in patients with clinically suspected pulmonary embolism is similar to the specificity of this diagnostic technique in patients with symptomatic deep venous thrombosis [21]. Even with this low false-positive rate, a policy of initiating anticoagulation in each patient whose ultrasonographic results are abnormal [34, 35] would result in the unnecessary treatment of approximately 13% to 26% of all patients with abnormal ultrasonographic results or 2% to 4% of all anticoagulation recipients in this population with suspected pulmonary embolism.

Should compression ultrasonography be included in the diagnostic work-up of patients suspected of having pulmonary embolism, either as a first test or only in patients with a nondiagnostic lung scan? If initial compression ultrasonography is done in all 1000 patients before lung scanning and if pulmonary angiography is done only when the lung scan is nondiagnostic, the number of lung scans and angiograms is reduced by 137 (approximately 14%) and 36 (9%), respectively. On the other hand, the consequences of this strategy include 1000 compression ultrasonographies and the unnecessary treatment of 18 patients with initial short-term heparin therapy followed by 3 months of anticoagulation. When lung scanning is the first test done in all patients and compression ultrasonography is then done only in patients who have a nondiagnostic lung scan, the reduction in the number of angiograms is similar. The consequences would then include 400 ultrasonographies and 9 patients receiving unnecessary heparin and oral anticoagulant treatment. Given the relatively low prevalence of pulmonary embolism in symptomatic patients (149 of 357 patients in our series [41%]) and the potential for a substantial number of false-positive results on compression ultrasonography (the latter depends on the extent of the investigation, which in turn determines the specificity of the test [30, 32, 36]), the gain in testing efficiency achieved by avoiding lung scanning and pulmonary angiography may be counterbalanced by the loss caused by unnecessary anticoagulant therapy in patients without pulmonary embolism.

In hospitals in which angiography is not available, venography of the legs could be considered because, as noted, the sensitivity of this test for the presence of deep venous thrombosis exceeds that of ultrasonography (70% to 90%) [17, 26]. In addition, serial testing of the deep veins of the legs using impedance plethysmography has been shown to be safe in the management of patients with a nondiagnostic lung scan [18]. Serial compression ultrasonography may yield results similar to those of impedance plethysmography, but this has not yet been proven in direct clinical trials.

We conclude that the introduction of compression ultrasonography of the leg veins early in the diagnostic strategy for pulmonary embolism can lead to a correct diagnosis in most patients while reducing the number of lung scans and pulmonary angiograms required. However, these improvements in efficiency are achieved at the cost of unnecessary anticoagulant therapy in some patients. Until we understand the reasons for the low sensitivity of compression ultrasonography and improve the predictive value of abnormal results on ultrasonography for the presence of pulmonary embolism, compression ultrasonography should be used with great caution in patients suspected of having pulmonary embolism.

Dr. Kuijer: Department of Internal Medicine, Academic Medical Center, Meibergdreef 9, 1105 AZ Amsterdam, the Netherlands.

Dr. van Beek: Department of Radiology, Academic Medical Center, Meibergdreef 9, 1105 AZ Amsterdam, the Netherlands. Dr. Brandjes: Department of Internal Medicine, Slotervaart Hospital, Louwesweg 6, 1066 EC Amsterdam, the Netherlands.