An Improved Multi-Output Gaussian Process RNN with Real-Time Validation\n for Early Sepsis Detection

Abstract

Sepsis is a poorly understood and potentially life-threatening complication\nthat can occur as a result of infection. Early detection and treatment improves\npatient outcomes, and as such it poses an important challenge in medicine. In\nthis work, we develop a flexible classifier that leverages streaming lab\nresults, vitals, and medications to predict sepsis before it occurs. We model\npatient clinical time series with multi-output Gaussian processes, maintaining\nuncertainty about the physiological state of a patient while also imputing\nmissing values. The mean function takes into account the effects of medications\nadministered on the trajectories of the physiological variables. Latent\nfunction values from the Gaussian process are then fed into a deep recurrent\nneural network to classify patient encounters as septic or not, and the overall\nmodel is trained end-to-end using back-propagation. We train and validate our\nmodel on a large dataset of 18 months of heterogeneous inpatient stays from the\nDuke University Health System, and develop a new "real-time" validation scheme\nfor simulating the performance of our model as it will actually be used. Our\nproposed method substantially outperforms clinical baselines, and improves on a\nprevious related model for detecting sepsis. Our model's predictions will be\ndisplayed in a real-time analytics dashboard to be used by a sepsis rapid\nresponse team to help detect and improve treatment of sepsis.\n

Ähnliche Arbeiten