Precision of digitized M-mode echocardiograms for clinical practice.

Computer digitization of M-Mode echocardiograms has proved a valuable research tool in the study of left ventricular function. It can be applied in clinical practice if the reproducibility of the technique is known so that imprecision can be minimised. Five normal echocardiograms of average clinical quality were selected and four cycles from each echocardiogram were digitized on two occasions by four different digitizers using an 'Apple II' computer. The study design allowed the sources of imprecision to be identified by analysis of variance. Assessment of the clinically most useful digitized measurement, peak rate of change of diastolic left ventricular dimension (dD/dt), was imprecise with a large within-subject variance and a 95% confidence interval of +/- 6.5 cm s-1; too wide for clinical use. The replication component accounted for 90% of this variance, with cycle to cycle and between-digitizer variance components being much less important. This large replication component will be reduced simply by repeating digitization; for dD/dt repeating digitization four times would theoretically reduce the confidence interval to a useful value of +/- 3.6 cm s-1. The 95% confidence interval for a single digitised measurement of diastolic left ventricular minor dimension was +/- 0.32 cm, for diastolic posterior wall thickness +/- 0.23 cm, and for the isovolumic relaxation time (A2-MVO) +/- 27 ms. Replication was again the largest component of variance so that the biggest improvement in precision will come from simply repeating digitization. Digitization on a cheap microcomputer system can provide clinically useful information provided reproducibility and sources of imprecision are known with clinical, as opposed to research, echocardiograms and conditions.