Use and abuse of electrocardiographic classification systems in epidemiologic studies

An analytical study is performed to examine the utility of the Minnesota Code ECG classification for estimation of the prevalence and incidence of coronary and hypertensive heart disease. A statistical model is introduced which permits comparative analysis of the contribution of human coding errors to the overall fraction of misclassified records. It is demonstrated that the poor sensitivity and visual coding variability can seriously limit the usefulness and obscure the meaning of ECG coding results if the disease prevalence or annual event incidence is low. Gross over- or underestimation of prevalence and incidence data takes place frequently. Occasionally, reasonably good estimates are made on the basis of false evidence, for instance when false classifications 'compensate' for a high miss rate. Visual classification errors may contribute a substantial fraction to the overall misclassifications. It is pointed out that the traditional method of estimating classification errors by repeated coding of a given set of records gives a false impression of coding reliability. It is concluded that a substantial reduction of coding variation and an overall improvement of diagnostic accuracy will be required in order to justify continued use of ECG classification in population studies.