A novel dynamic model to predict abnormal oxygen desaturations in blood

This paper presents a novel approach for modeling and predicting the dynamics of blood oxygenation signals (SpO2). A performance metric based on a dynamically adjusted threshold is optimized to fit the SpO2 predictive model by maximizing its ability to predict abnormal desaturation events. While we use a predictive model structure that is discrete time autoregressive, we formulate fitting this model structure as a mixed integer programming problem (MIP) and we optimize the models to maximize multi-step ahead predictions of abnormal events. Our results show that the proposed model is better able to capture the dynamic behavior for abnormal changes of SpO2 compared to standard autoregressive models. This study provides a starting point for further needed investigation in prediction of physiological systems behavior.