Optimal coding of electrocardiograms for epidemiological studies. The performance of human coders - astatistical model

A statistical model is introduced to characterize the ECG coding process when multiple coding is performed by human readers. Triplicate reading and arbitration of disagreements would considerably reduce the miss rate with a concomitant increase in cost and effort. The results suggest that the use of the majority rule (acceptance of the most common code from multiple, blinded individual codings) is inappropriate for ECG coding. The model proposed gives a plausible explanation to the observed high miss rate in coding of rare events in spite of duplicate reading and arbitration of disagreements. The main problem identified is the high individual miss rate in a fraction of "difficult" ECG records suggested that efficient use of computer programs and visual verification may be required to reduce coding errors to an acceptable level.