Related Experiment Video
Updated: Feb 17, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.6K
Mapping Patient Trajectories using Longitudinal Extraction and Deep Learning in the MIMIC-III Critical Care Database
Brett K Beaulieu-Jones1, Patryk Orzechowski, Jason H Moore
1Computational Genetics Lab, Institute for Biomedical Informatics, Perelman School of Medicine, University of Pennsylvania, 3700 Hamilton Walk, Philadelphia PA, 19104, United States of America, brettbe@med.upenn.edu.
Pacific Symposium on Biocomputing. Pacific Symposium on Biocomputing
|December 9, 2017
Summary
Researchers can now extract meaningful patient trajectories from Electronic Health Records (EHRs) using deep learning. This approach analyzes sequences of care events, offering a richer view than single data snapshots.
Area of Science:
- Biomedical Informatics
- Machine Learning in Healthcare
Background:
- Electronic Health Records (EHRs) offer vast patient data for research.
- Current data extraction and analysis often use single-time point snapshots.
- Patient care involves sequential events not fully captured by static data.
Purpose of the Study:
- To develop methods for analyzing sequential patient data from EHRs.
- To create patient trajectories from discrete care events.
- To compare deep learning with traditional methods for EHR data analysis.
Main Methods:
- Extracted sequences of patient care events from EHRs.
- Applied unsupervised autoencoders for learning data embeddings.
- Utilized long short-term memory networks for sequence analysis.
- Compared deep learning embeddings against traditional machine learning on static EHR snapshots.
Main Results:
- Deep learning methods (autoencoders, LSTMs) successfully learned meaningful embeddings from patient care event sequences.
- Sequential analysis provided a richer representation of patient interactions than single-time point snapshots.
- Learned embeddings captured patient trajectories and level of healthcare attention.
Conclusions:
- Sequential analysis of EHR care events using deep learning is feasible and effective.
- This approach offers a more comprehensive understanding of patient journeys.
- Future research can leverage these sequential embeddings for advanced biomedical research.
