Short term prediction of severe bradycardia in premature newborns

Premature newborns show frequent episodes of bradycardia due to the immaturity of their autonomic nervous system. There is a need for developing methods which may alert physicians as soon as early signs of bradycardia are detected. In this paper we studied the RR interval (RRI) series using data mining methods to detect early signs of bradycardia. We employed principal components analysis (PCA) and hierarchical ascending classification (HAC) according to the generalised Ward´s method. Time domain and frequency domain parameters as well as non-linear indices based on entropy were extracted from 13 stationary RRI series 3 minutes preceding the bradycardias. The projection of observations on the first factorial plan demonstrated a well defined path: clustering of observations appeared approaching the bradycardia (in 10/13 cases). These results suggest that the RRI contains information that can be employed to predict the onset of the bradycardia event.