Although our study was carefully executed, the above letters question our finding that in patients suspected of having pulmonary embolism, compression ultrasonography has a false-positive rate for deep venous thrombosis of 13% to 26%.

Drs. Perrier and Bounameaux mention that the specificities of ultrasonography and lung scanning are the same and that the number of inappropriately treated patients would therefore not differ. Our theoretical population was intended to clarify the point that although the specificity for ultrasonography was 97%, the positive predictive value of an abnormal ultrasonography result is only 74% when the lung scan is nondiagnostic. This occurred because of the low prevalence of pulmonary embolism in these patients (26%), combined with the low sensitivity of ultrasonography.

In asymptomatic patients in whom a gold standard test excluded pulmonary embolism, we are confident that no venous thromboembolism was present and thus considered the abnormal ultrasonography results to be false-positive. This was confirmed by follow-up of these patients, during which no venous thromboembolism occurred [1, 2]. We agree that venography would have been the optimal test for determining whether the abnormal ultrasonography results were false-positive. Whether false-positive ultrasonography results occur in clinical practice remains uncertain. The overall positive predictive value of an abnormal ultrasonography result varies from 95% (95% CI, 93.6% to 96.7%) in symptomatic patients with deep venous thrombosis to 82% (CI, 71.7% to 89.8%) in asymptomatic patients with deep venous thrombosis [3]. False-positive results are well known, especially in asymptomatic patients.

We fully agree with Drs. Weg and Froehlich that patients with deep venous thrombosis need anticoagulation, regardless of symptoms. Our study did not contest this fact; we merely reported the presence of false-positive ultrasonography results. We disagree that "imaging below the knee is an essential component of ultrasonography of the lower extremity" in patients suspected of having pulmonary embolism. Well-designed studies need to show that investigation of the calf veins is not associated with an unacceptable decrease in positive predictive value.

In response to Drs. Bernard and Yves, the combination of a low clinical probability and a high-probability lung scan occurs in approximately 7% of patients with a high-probability scan [4]; the resulting overall sensitivity remains high (87%) [4]. Thus, it is unlikely that this affected our findings. Anticoagulation is generally started in all patients with a high-probability lung scan [5]. Bernard and Yves were also concerned that a normal lung scan does not exclude pulmonary embolism. In our study sample, however, anticoagulation was withheld in patients with normal lung scans. No episodes of pulmonary embolism occurred during follow-up [1]. In patients with an uncertain diagnosis, treatment was instituted in four of eight patients with an abnormal ultrasonography result. A sensitivity calculation in this group is not valid, and comparison with patients who have a high-probability lung scan is not prudent.

Perrier and colleagues observed that ultrasonography had a sensitivity of 51% (CI, 44% to 58%) in outpatients and approximately 30% in inpatients. In our sample of outpatients and inpatients with a high-probability lung scan, the sensitivity of ultrasonography was 39% (23 of 59 patients; CI, 27% to 53%) and 21% (12 of 57 patients; CI, 11% to 34%), respectively. A trend toward a higher sensitivity in outpatients is present, albeit not a statistically significant one. Contrary to our findings, Perrier and colleagues found that a diagnostic approach including ultrasonography in outpatients suspected of having pulmonary embolism was marginally cost-effective. Further studies are necessary to elucidate these differences.