Selection of a reduced set of parameters for classification of ventricular conduction defects by cluster analysis

Currently used algorithms for classification of ventricular conduction defects (VCD) are fairly complex and employ a large number of mostly dichotomized ECG features or their combinations, based almost entirely on QRS morphology. The authors used disjoint clustering analysis in an effort to identify a smaller subset of ECG features to characterize various VCD categories using ECG data of 3,501 normal subjects and 887 with VCD. Their results with five parameters forming seven clusters suggest that the criteria for many VCD categories are based on arbitrary decision boundaries, resulting in fact in considerable overlap of normal conduction, fascicular and incomplete blocks even in multidimensional decision space. The same conclusion holds for the so-called undetermined type VCD, which in fact converges nearly completely into the clusters formed by LBBB and RBBB