Markov-based Modeling of Respiratory Patterns for Prediction of Extubation Readiness in Preterm Newborns

In this work, we applied Markov chain modeling to gain insight into the respiratory patterns of extremely preterm newborns being considered for weaning from breathing support (extubation). We show that semi-Markov chains provide more robust modeling capability than Markov chains, especially when dwell time durations in states are very long and are not distributed exponentially. We also introduce the use of multi-chain models for exposing non-stationary behaviour across time segments. The developed models revealed interesting characteristics between newborns who were ready to be extubated and those who were not. They were then applied to the predictive task of determining extubation readiness.