Detection of left ventricular thrombus by two-dimensional echocardiography: sensitivity, specificity, and causes of uncertainty.

To define the sensitivity, specificity and predictive accuracy of two-dimensional echocardiographic detection of left ventricular thrombus, the echocardiograms of 78 patients who had independent proof of the presence or absence of a left ventricular thrombus were interpreted without knowledge of any clinical data. The presence of thrombus was established by autopsy in four patients, by aneurysmectomy in three, and by indium-111 platelet imaging in 15; the absence of thrombus was proved by autopsy in 55 patients and by aneurysmectomy in one patient. The characteristics of true-positive and false-positive echocardiograms, interobserver variability, and clinical features associated with proved thrombus were also defined. The echocardiogram was positive for thrombus in 22 patients, equivocal in seven and negative in 49. For detection of thrombus, a positive or equivocal echocardiogram had a sensitivity of 95% (21 of 22), a specificity of 86% (48 of 56), and a predictive value of 72% (21 of 29); the predictive value of a negative study was 98% (48 of 49). Considering positive and equivocal studies separately, the predictive value of a positive study was 86% (19 of 22), while that of an equivocal study was only 29% (two of seven). Compared with patients who had no thrombus, patients with proved thrombus had a higher prevalence of electrocardiographic transmural anterior infarction (86% vs 13%), left ventricular aneurysm (73% vs 5%), and clinical systemic emboli (36% vs 7%) (all p less than 0.05). These clinical features help to identify a subset of patients most likely to have left ventricular thrombi who may benefit from echocardiography. Two-dimensional echocardiography is highly sensitive in detecting left ventricular thrombus, but false-positive studies are relatively common. Several echocardiographic criteria derived from analysis of the true and false positives in this study may help minimize diagnostic errors.