Abstract by David Kartchner
Predicting Disease Onset via Learned Patient Representations
With healthcare costs consuming an ever increasing proportion of the economy, hospital systems need to switch to preventative treatment to help patients and stay competitive. We provide a comprehensive survey of machine learning models on disease prediction tasks and identify novel means of learning joint representations of patient demographics and irregularly spaced time series representing encounters. We further demonstrate how using cost as a proxy for disease severity can notably improve disease prediction outcomes.