Forecasting Trajectories of Physiological Mechanics with Sparse Clinical Data Using a Data Assimilation and Machine Learning Hybrid

Clinical decisions for determining optimal patient-specific interventions are complicated prediction tasks that rely on health care professionals’ understanding of physiological mechanisms and their dynamics. These decisions are challenged by (a) observational data sparsity and (b) patient heterogeneity. Here, we focus on estimating and forecasting specific physiological properties–that are not explicitly present in clinical observations–to provide additional features using only data available bedside at the time of decision-making.