Computation of multifactorial receiver operator and predictive accuracy characteristics [ECG analysis]

The computation of the so-called receiver operator characteristics is simple when based on univariate data. On the contrary, the complexity of computation of multivariate characteristics increases exponentially with the dimension of the data. This study describes a new algorithm for computation of multivariate receiver operator characteristics. The algorithm is based on several concepts which increase its computational efficiency. The most important of them is a pre-sorting of the data in each dimension and the division of each dimension into groups in which the negative cases precede positive ones. The algorithm was tested in a risk stratification study that was aimed at identifying survivors of acute myocardial infarction at risk of early death. A cohort of 539 patients was stratified based on time-domain (3 variables) and spectral turbulence (6 variables) indices of signal averaged electrocardiogram. The computing requirements of this study are presented in the text and the efficiency of the algorithm is discussed in detail