Review: Noninvasive tests are useful only for symptomatic deep venous thrombosis
ACP J Club. 1998 Sep-Oct;129:46. doi:10.7326/ACPJC-1998-129-2-046
Kearon C, Julian JA, Newman TE, Ginsberg JS, for the McMaster Diagnostic Imaging Practice Guidelines Initiative. Noninvasive diagnosis of deep venous thrombosis. Ann Intern Med. 1998 Apr 15;128:663-77.
How accurate and safe is noninvasive testing for detecting deep venous thrombosis (DVT)?
Studies were identified by searching MEDLINE up to January 1997 using the terms thrombosis, thrombophlebitis, plethysmography, impedance, ultrasonography, randomized controlled trials, and cohort studies. Bibliographies and personal files were checked.
Accuracy studies were selected if they included ≥ 50 consecutive patients and the diagnostic standard was venography assessed in a blinded manner. Management studies were selected if they included ≥ 50 consecutive patients, if all patients who did not receive anticoagulants (i.e., those with negative serial test results) were followed, and if an objective measurement of venous thromboembolism (DVT and pulmonary embolism) was done during follow-up.
Data were extracted on study quality, test characteristics, patients, tests evaluated, venography, and outcomes.
The sensitivity and specificity were high for detection of symptomatic proximal DVT. In asymptomatic patients, both tests had low sensitivities but high specificities (Table). The management studies showed that withholding treatment in patients who had normal results on serial venous ultrasonography (4 studies) resulted in a mean frequency of 2.0% (95% CI 0.0% to 4.0%) for confirmed DVT within 6 months. For patients with normal results on serial impedance plethysmography (7 studies), the mean frequency of confirmed DVT within 6 months was 1.5% (CI 0.8% to 2.2%). Graded recommendations were made in the article.
Noninvasive tests are useful for the diagnosis of deep venous thrombosis in symptomatic patients.
Sources of funding: Heart and Stroke Foundation of Ontario and Cancer Care Ontario Practice Guidelines Initiative.
For correspondence: Dr. C. Kearon, McMaster Medical Unit, Henderson General Hospital, 711 Concession Street, Hamilton, Ontario L8V 1C3, Canada.
Table. Diagnosis of a first episode of deep venous thrombosis (DVT)
|Tests||Patients||DVT||Sensitivity (95% CI)||Specificity (CI)||+LR*||-LR*|
|Plethysmography||Symptomatic||All||83% (76 to 90)||92% (88 to 92)||10.3||0.2|
|Asymptomatic||All||15% (11 to 19)||95% (91 to 99)||3.0||0.9|
|Symptomatic||Proximal||93% (82 to 97)||90% (85 to 94)||9.3||0.08|
|Asymptomatic||Proximal||24% (15 to 36)||97% (91 to 99)||8.0||0.8|
|Ultrasonography||Symptomatic||All||89% (85 to 92)||94% (90 to 98)||14.8||0.1|
|Asymptomatic||All||47% (37 to 57)||94% (91 to 98)||7.8||0.5|
|Symptomatic||Proximal||95% (93 to 97)||98% (96 to 99)||47.5||0.05|
|Asymptomatic||Proximal||58% (49 to 66)||97% (95 to 98)||19.3||0.4|
*+LR = likelihood ratio for the presence of disease if the test is positive; -LR = likelihood ratio if the test is negative. Numbers for proximal DVT and likelihood ratios are calculated from data in article.
The diagnosis of DVT is complex. This is because several tests are available; 3 clinical states exist (proximal, distal, and no DVT); patients may or may not be symptomatic; and some characteristics, such as predictive values, vary by population. The array of possible combinations makes it difficult to evaluate tests in standard ways—that is, with sensitivity, specificity, and likelihood ratios (LRs). For example, we would need the following LRs for each test result: proximal compared with distal (if normal ruled out), proximal compared with distal and normal, proximal and distal compared with normal, proximal compared with normal (if distal ruled out), and distal compared with normal (if proximal ruled out).
Traditional test characteristics for diagnosing all DVT (proximal or distal) compared with no DVT, along with sensitivity data for proximal and distal DVT, are provided in the review by Kearon and colleagues. If proximal DVT and distal DVT mean the same thing in terms of clinical judgments, this may be acceptable. But if they do not, a positive test result for all DVT does not help the clinician because proximal DVT must be differentiated from distal DVT to make management decisions.
The evidence-based algorithm is a more useful way to present a test's performance for complex clinical disorders, such as DVT. The algorithm would start with 3 levels of prior probability (low, medium, and high), and the conditional probabilities of each clinical state would follow at each branch. Wells and colleagues (1) provide us with the first steps to such an algorithm.
But complexity begets complexity—we must carry a complex algorithm in our heads (or pockets). A compromise for simplicity is to present the +LR for proximal DVT and the -LR for all DVT. A positive test result for proximal DVT will allow us to treat if the prior probability is high or moderate and may lead to more testing if it is low. A negative test result for all DVT would allow us to rest easy (if the prior probability is low). All other combinations would lead to further testing either in the form of serial noninvasive testing or venography.
Robert A. McNutt, MD
Arthur T. Evans, MD, MPHCollaborative Research Unit, Cook County HospitalChicago, Illinois, USA