Models for Predicting Sepsis in the ICU Trained on True Expert Labels

Sepsis is a life-threatening, critical condition often encountered in a 
hospital's intensive care unit (ICU). Starting as a body's response to 
infection, sepsis causes injury to the patient's tissue and organs and 
can lead to severe sepsis or septic shock, resulting in multiple organ 
failure and high risk of mortality. However, sepsis can be treated 
effectively if identified early and medicated in a timely manner. We 
present ongoing work on modifications of the predictive model we 
described last time. Our new models are the first to utilize the expert 
labels we collect using our electronic questionnaire from supervisors 
of the intensive care unit of the UMM since last year. On the patient's 
side, we moved from the restricted polytrauma cohort to the mixed quest 
cohort with surprising results. Our new linear and non-linear models 
are able to predict sepsis with an AUROC of up to 87.5 before onset, 
and still show a considerable performance of AUROC 77.9 in a reasonable 
prediction window of 48-12h before sepsis onset.