Predicting the length of stay using Bayesian model selection for Neonatal heart rate monitoring

Although heart rate is commonly measured in various clinical settings the advanced algorithms for its prediction are rarely implemented in clinical settings and patient management. In neonatal intensive care timely prediction of dangerous levels of heart rate can lead to improved care, positive long-term effects and reduced morbidity. Successful development and implementation of time series prediction algorithms is often based on selection of adequate signal processing models. In practical applications this issue is addressed almost invariably using statistical hypothesis testing. In this paper we propose preliminary model selection algorithm based on Bayesian approach leading to calculation of posterior odds ratios. We evaluate the applicability of the proposed method using a real data set containing over 180 pre-term infants whose heart rates were recorded over the length of their stay in the Neonatal Intensive Care Unit (NICU).