Clinical assessment plus ultrasonography accurately predicted deep venous thrombosis
ACP J Club. 1996 Jan-Feb;124:19. doi:10.7326/ACPJC-1996-124-1-019
Wells PS, Hirsh J, Anderson DR, et al. Accuracy of clinical assessment of deep-vein thrombosis. Lancet. 1995 May 27;345:1326-30. [PubMed ID: 7752753]
To assess the accuracy of a clinical model for diagnosing suspected deep venous thrombosis (DVT).
A blinded comparison of a clinical model with venous ultrasonography and contrast venography.
2 university-affiliated centers in Canada and 1 in Italy.
635 outpatients with suspected DVT who had symptoms for < 60 days were eligible for inclusion. Exclusion criteria were previous DVT or pulmonary embolism, concomitant clinically suspected pulmonary embolism, contraindication to contrast media, anticoagulant therapy for more than 48 hours, below-knee amputation, and pregnancy. 529 patients (83%) were evaluated.
Description of Test and Diagnostic Standard
All patients were clinically assessed before having ultrasonography and venography. On the basis of signs and symptoms of DVT, risk factors for DVT, and potential alternative diagnoses, a clinical model was developed that categorized patients as having a low, moderate, or high pretest probability of DVT. Ultrasonograms and venograms were interpreted by a panel of 3 or more observers blinded to other diagnostic test results and patients' clinical histories. An abnormal ultrasound result was a lack of full compressibility of the common femoral or popliteal veins. The diagnosis of DVT was confirmed by venography characterized by a constant intraluminal filling defect in ≥ 2 projections.
Main Outcome Measures
Sensitivity, specificity, and post-test probabilities of DVT with normal and abnormal ultrasonography results for each pretest probability classification.
DVT was confirmed in 72 of the 85 patients (85%, CI 75% to 92%; likelihood ratio [LR] 16.2) classified as having a high pretest probability; in 47 of 143 patients (33%, CI 25% to 41%; LR 1.4) classified as having a moderate probability; and in 16 of 301 patients (5%, CI 3% to 9%; LR 0.2) classified as having a low probability. For proximal DVT, LRs for the combination of the clinical model and ultrasonography ranged from > 50 (virtually diagnostic) for an abnormal ultrasonogram plus a high or moderate pretest probability, through 1 (no useful information) for a normal ultrasonogram and a high pretest probability, to < 0.1 (very unlikely) for a normal ultrasonogram with a low pretest probability. The numbers were similar for all DVT.
The stratification of patients into pretest probabilities using a clinical model combined with venous ultrasonography accurately predicted deep venous thrombosis and streamlined the diagnostic process.
Source of funding: Not stated.
For article reprint: Dr. P. Wells, Ottawa Civic Hospital, Ottawa, Ontario K1Y 1J8, Canada. FAX 613-761-5340.
Although some signs and symptoms are associated with lower-extremity DVT (1), we still admonish students and residents to confirm their clinical suspicions with objective testing. The prospective study by Wells and colleagues links historical and physical findings with ultrasonographic results to provide a simplified approach to confirming or excluding DVT. The methods were straightforward, and we are reassured that the pretest probability assignments are reproducible and easily applied.
The results serve as an important reminder of the limitations of test accuracy for DVT and many other conditions. At the extremes of disease prevalence or clinical probability, even for sensitive and specific tests, some patients will have positive results but not disease (positive predictive value < 100%), and others with negative results will have disease (negative predictive value < 100%).
In this study, the positive predictive value (post-test probability) of abnormal ultrasonography was only 63% in the low pretest probability group of patients, but the combination of abnormal ultrasonography and moderate or high pretest probability was virtually diagnostic. The negative predictive value of ultrasonography was only 68% in the high pretest probability group. Thus, patients with positive ultrasound results and low pretest probability should be investigated further, as should patients with normal ultrasound results and high or moderate pretest probability.
Can we now apply these clinical criteria for DVT to stratify our patients in advance of further testing? Not yet. As shown in the results of this study, when comparing DVT frequencies among the 3 centers, disease rates and severity at presentation are likely to vary from institution to institution. Although it is reassuring that the high pretest estimates predicted positive venography in at least 80% of patients at all 3 centers, the clinical criteria need to be validated by means of prospective re-evaluation at other centers (2).
Daniel M. Becker, MD
University of Virginia Medical SchoolCharlottesville, Virginia, USA
Since the publication of this 1995 paper by Wells and colleagues, 2 studies (3, 3) have been done that have asessed the value of pretest probablility of DVT management and application of a diagnosis model for the management of hospitalized patients with suspected DVT.
These articles stress the importance for clinicians of using pretest probability models in patients with DVT.