The risk of mortality is the most definitive disease severity measure. Understanding disease severity is key in Intensive Care Units (ICUs), where the sickest patients across all specialties are treated. Pioneering efforts have been made in the past 40 years to implement standardized and validated mortality risk scores for the ICU. However, most contemporary mortality models have limitations that hinder applicability at the level of the individual patient. Responding to these challenges requires a spectrum of interventions, some of which are addressed in this work. The scope of this PhD thesis was to investigate if/how machine learning can contribute to more accurate and transparent predictions of mortality over time applicable at the level of individuals. 

With the studies in this thesis we investigated how machine learning can be used for more accurate prognostication of patient outcome in the ICU. The application of machine learning technologies can help unleash the full potential of the vast amount of information that is continually collected in relation to an ICU admission. 

The project was carried out in collaboration with The Novo Nordisk Foundation, Center for Protein Research, University of Copenhagen.