David Albers (University of Colorado, USA)
In this talk I will present some recent results estimating and forecasting physiology in an intensive care unit (ICU) using sparse, non-stationary data collected while administering care. While I will focus on blood glucose forecasting, I will treat this context as a prototype for other physiologic subsystems. The ensemble Kalman Filter (EnKF) as a starting point for data assimilation within this context and demonstrate how using an EnKF alone often fails to produce useful estimates due to the complex nature of the data and physiologic dynamics. I will then introduce several partial solutions that reduce identifiability problems, including an approach for combining data assimilation with machine learning to reduce the dimension of the parameter space estimated and constraining the EnKF. I will also discuss current outstanding problems that remain despite the application of these approaches and present a path forward for both achieving a deeper understanding of systems physiology and applying DA-based forecasting to help manage health in complex settings.
invited by Jana de Wiljes
