Accuracy of clinical assessment of deep-vein thrombosis

The clinical diagnosis of deep-vein thrombosis is generally thought to be unreliable. From experience, we hypothesised that this widely held view might be incorrect. We developed a clinical model and prospectively tested its ability in three tertiary care centres to stratify symptomatic outpatients with suspected deep-vein thrombosis into groups with high, moderate, or low probability groups of deep-vein thrombosis. We evaluated our clinical model in combination with venous ultrasonography to determine the potential for an improved and simplified diagnostic approach in patients with suspected deep-vein thrombosis. All patients were clinically assessed to determine the probability for deep-vein thrombosis before they had ultrasonography and venography. All tests were performed and interpreted by independent observers. In 529 patients, the clinical model predicted prevalence of deep-vein thrombosis in the three categories: 85% in the high pretest probability category, 33% in the moderate, and 5% in the low category. There was no statistical difference in the performance of the model in the three centres. The model demonstrated excellent interobserver reliability (Kappa = 0.85). There were important differences with ultrasonography between the high and low pretest probability groups for both positive predictive values (100% (95% CI, 94-100%) vs (63% [35-85%], respectively). Thus, use of the clinical model combined with ultrasonography would decrease the number of false positive and negative diagnosis if venography were done when the ultrasound result and pretest probability were discordant. The diagnostic process could be simplified by excluding those patients with low pretest probability and normal ultrasound results from serial testing.