Abstract: UPhenome is a model which infers phenotypes from heterogeneous clinical data and summarizes patients’ clinical records in terms of low-dimensional, phenotypic profiles. In this work, we evaluate the utility of phenotypic profiles as features for patient mortality prediction. We expect our predictive models to effectively discriminate between patients with low and high mortality risk, and that these patients’ phenotypic profiles will be skewed toward phenotypes characterizing severe and benign pathologies respectively.

Learning Objective 1: Understand how data-driven phenotyping methods can be used to extract rich, clinically meaningful features from heterogenous, high-dimensional clinical data.


Victor Rodriguez (Presenter)
Columbia University

Adler Perotte, Columbia University
Noemie Elhadad, Columbia University

Presentation Materials: