Influence of measurement parameter changes and statistical parameter characteristics on the stability of ECG classification in HES ECG

Increasing needs-and possibilities-for serial comparison of ECG computer analysis results bring up the question of their reproducibility and stability. The authors report here results from systematic measurement manipulation experiments and corresponding changes in diagnostic statements of the HES algorithm. Adding noise or filtering the original data resulted in totally less than 4% changes in any of the diagnostic categories NOR, HYP and INF. Shifting the relevant wave onsets or offsets creates changes in diagnostic statements as well. For classification based on discriminant analysis if is demonstrated in what way classification errors depend on the distributions of discriminant values. A discriminant function becomes as steeper as larger the group mean distances are and as smaller the dispersion of the discriminant values is. Consequently small changes in some measurements will result in large changes of the discriminant value and may cause a reallocation of the case under consideration