Subjects with ischemic dilated cardiomiopathy tend to suffer episodes of sudden cardiac death, thus risk stratification is essential to establish an adequate therapy for the patients. In this work, a new methodology was proposed for the study of the heart rate variability by using a multiscale analysis based on the concept of entropy rates, for improving risk prediction in cardiac patients. Symbolic dynamics were applied to RR time series and sets of words in several scales were constructed. The multiscale regularity analysis was proposed by comparing the entropies, calculated using Shannon and Renyi definitions, of the series of words in different scales. The study considered the selection of the best parameters for the length of the words (l) and the order of the entropies (q). Statistical analysis with repeated measures and discriminant analysis revealed statistically significant differences (p-value<0.05) and a high percentage of well classified subjects in their different risk groups, with sensitivity, specificity and positive predictive values of 100%.