Clinicians, the Mount Sinai program and the Veterans' Administration program evaluated against clinico-pathological data derived independently of the electrocardiogram.

Abstract

221 electrocardiograms (ECGs) recorded from adult subjects with cardiac symptomatology were interpreted by clinicians and the Mount Sinai (MS) and Veterans' Administration (VA) computer programs. After the clinicians had eliminated their interobserver variability, their interpretations and those of the computer programs were compared with the corresponding clinico-pathological data that had been derived independently of the ECG. A measure of overall diagnostic accuracy was used which showed that the performance of the 2 computer programs was similar and somewhat worse than that of the clinicians. The clinicians and the computer programs were not as good at diagnosing the ECGs of women as at diagnosing those of men. As the VA program's interpretations may be altered by including information about the patient's provisional diagnosis, 6 different sets of these prior probabilities were used to analyse 141 of the original 221 ECGs. Different fixed sets of prior probabilities made only a small difference to overall diagnostic accuracy but a marked difference to the VA program's ability to differentiate Normal from Abnormal. Optimizing the prior probabilities for the individual subjects increased the VA program's overall diagnostic accuracy up to that of the clinicians. Both computer programs correctly diagnosed sinus rhythm as the dominant rhythm in 165 out of 177 subjects. The MS program diagnosed the dominant rhythm as Atrial Fibrillation in 35 out of 41 subjects and the VA program in 27 (X(2) = 3.24; not significant). ECG interpretation; Mount Sinai program; Veterans' Administration program; independent clinico-pathological data; sex differences in the ECG; prior probabilities.