Systematic review and meta-analysis of the diagnostic accuracy of ultrasonography for deep vein thrombosis

Background: Ultrasound (US) has largely replaced contrast venography as the definitive diagnostic test for deep vein thrombosis (DVT). We aimed to derive a definitive estimate of the diagnostic accuracy of US for clinically suspected DVT and identify study-level factors that might predict accuracy. Methods: We undertook a systematic review, meta-analysis and meta-regression of diagnostic cohort studies that compared US to contrast venography in patients with suspected DVT. We searched Medline, EMBASE, CINAHL, Web of Science, Cochrane Database of Systematic Reviews, Cochrane Controlled Trials Register, Database of Reviews of Effectiveness, the ACP Journal Club, and citation lists (1966 to April 2004). Random effects meta-analysis was used to derive pooled estimates of sensitivity and specificity. Random effects meta-regression was used to identify study-level covariates that predicted diagnostic performance. Results: We identified 100 cohorts comparing US to venography in patients with suspected DVT. Overall sensitivity for proximal DVT (95% confidence interval) was 94.2% (93.2 to 95.0), for distal DVT was 63.5% (59.8 to 67.0), and specificity was 93.8% (93.1 to 94.4). Duplex US had pooled sensitivity of 96.5% (95.1 to 97.6) for proximal DVT, 71.2% (64.6 to 77.2) for distal DVT and specificity of 94.0% (92.8 to 95.1). Triplex US had pooled sensitivity of 96.4% (94.4 to 97.1%) for proximal DVT, 75.2% (67.7 to 81.6) for distal DVT and specificity of 94.3% (92.5 to 95.8). Compression US alone had pooled sensitivity of 93.8 % (92.0 to 95.3%) for proximal DVT, 56.8% (49.0 to 66.4) for distal DVT and specificity of 97.8% (97.0 to 98.4). Sensitivity was higher in more recently published studies and in cohorts with higher prevalence of DVT and more proximal DVT, and was lower in cohorts that reported interpretation by a radiologist. Specificity was higher in cohorts that excluded patients with previous DVT. No studies were identified that compared repeat US to venography in all patients. Repeat US appears to have a positive yield of 1.3%, with 89% of these being confirmed by venography. Conclusion: Combined colour-doppler US techniques have optimal sensitivity, while compression US has optimal specificity for DVT. However, all estimates are subject to substantial unexplained heterogeneity. The role of repeat scanning is very uncertain and based upon limited data.