Time-Sensitive Modeling of ICU/Critical Care at the Visit-Level: a Study of 48-Hour Mortality and ARDS Phenotyping

Presentation Time: 04:30 PM - 04:45 PM

Abstract Keywords: Natural Language Processing, Machine Learning, Deep Learning, Knowledge Representation and Information Modeling, Reproducibility
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Longitudinal NLP studies often focus on evidence gathered from visit-to-visit without including the time course of events within a visit. We hypothesize that a framework shown to work at a between-visit temporal granularity can be adapted to a within-visit granularity. We have found Pytorch_EHR a reasonable means for enriching NLP models to include within-visit temporality based on two use cases as compared against performance published by other researchers: 48-hour ICU mortality and ARDS phenotyping.

Speaker(s):
Paul Heider, PhD
Medical University of South Carolina

Author(s):
Dmitry Scherbakov, PhD - Medical University of South Carolina; Jihad Obeid, MD - Medical University of South Carolina; Charles Terry, MD - Medical University of South Carolina;