Sepsis is a poorly understood and potentially life-threatening complication can occur as a result of infection. Early detection and treatment improves outcomes, and as such it poses an important challenge in medicine. In work, we develop a flexible classifier that leverages streaming lab, vitals, and medications to predict sepsis before it occurs. We model clinical time series with multi-output Gaussian processes, maintaining about the physiological state of a patient while also imputing values. The mean function takes into account the effects of medications on the trajectories of the physiological variables. Latent values from the Gaussian process are then fed into a deep recurrent network to classify patient encounters as septic or not, and the overall is trained end-to-end using back-propagation. We train and validate our on a large dataset of 18 months of heterogeneous inpatient stays from the University Health System, and develop a new "real-time" validation scheme simulating the performance of our model as it will actually be used. Our method substantially outperforms clinical baselines, and improves on a related model for detecting sepsis. Our model's predictions will be in a real-time analytics dashboard to be used by a sepsis rapid team to help detect and improve treatment of sepsis.
No takes yet. Share an insight, caveat, or question.
Futoma et al. (2017) studied this question.