Computer interpretation of pediatric orthogonal electrocardiograms: statistical and deterministic classification methods.

Abstract

Statistical multivariate and conventional deterministic methods of computerized interpretation of the electrocardiogram (ECG) were compared in the analysis of 1711 pediatric orthogonal ECGs validated by nonelectrocardiographic criteria on the basis of clinical and anatomic diagnoses. Among 642 children catheterized for the evaluation of congenital heart disease, there were 140 patients with left ventricular hypertrophy, 299 with right ventricular hypertrophy, and 203 with biventricular hypertrophy. A group of 1069 obviously healthy school children was studied as a control. The overall accuracy of multigroup ECG diagnosis was 85% and 79% for the statistical and deterministic methods, respectively. The diagnostic performances of both methods expressed in terms of sensitivity and predictive value were the highest for normal children and those with right ventricular hypertrophy and lowest for children with biventricular hypertrophy. The statistical method was more sensitive in the diagnosis of left ventricular hypertrophy (74% vs 64%), right ventricular hypertrophy (86% vs 83%), and biventricular hypertrophy (62% vs 50%). Mutual agreement for a correct diagnosis by the two methods was 83% for normal children and 82% for those with right ventricular hypertrophy but only 61% for children with left ventricular hypertrophy and 39% for those with biventricular hypertrophy. In conclusion, better classification results are obtained with statistical multivariate techniques as compared with conventional deterministic analysis, but both methods of ECG interpretation are complementary and their combination in the same electrocardiographic computer program can improve diagnostic accuracy.