Title:
Learning Predictive Risk Scores to Improve Early Diagnosis of Sepsis

Abstract:
Sepsis is an inflammatory response of the whole body to infection. As a leading cause of death in intensive care unit patients, its early diagnosis is of major concern. I will present the Disease Severity Score (DSS) by Dyagilev and Saria (2015), a model that is able to predict the latent severity of diseases based on electronic health records, and show its connections to known IR ranking methods. In the context of discussing two sepsis related experiments, I will present planned extensions of the model and shed light on problems concerning the construction of training data.