Sources of variability in echocardiographic measurements

Abstract

Experienced echocardiographers performed and interpreted echocardiograms of 16 normal volunteers in an experiment designed to quantify the sources of variability in echocardiographic measurements. The experimental factors were 16 subjects, two echocardiographic technicians, replication between 2 successive days, replication between two echocardiograms taken approximately 20 minutes apart, two subject positions and subject gender. The experimental factors in echocardiographic interpretation were four interpreters each interpreting two copies of each echocardiogram. The major conclusions of this study are: (1) Relatively little of the variability in echocardiographic measurements can be attributed to the technicians if the latter are experienced. (2) There appears to be relatively little variability in measurement with respect to days of replication within days, except for heart rate and the right ventricular internal dimension measurements. (3) The variability that resulted when an interpreter read unidentified copies of the same echocardiogram on different occasions was as large or larger than the variability that occurred when different interpreters read the same echocardiogram. (4) There is significant extraneous variability in the measurement of the thickness of the ventricular septum and posterior wall in normal subjects. (5) The position of the subject systematically influences the value of the measurements of the right ventricular internal dimension, but has relatively little or no effect on other measurements. It is concluded that each echocardiogram should be read at least twice by each of at least two interpreters. Minimally, each echocardiogram should be read either by one interpreter on two separate occasions or by two interpreters. The reported result for a measurement should be the mean of the values. If possible, an interpreter reading the same echocardiogram more than once should be “blinded” to the identity of the echocardiograms so that bias on subsequent readings is minimized.