Research Article: An attention based deep learning model of clinical events in the intensive care unit

Date Published: February 13, 2019

Publisher: Public Library of Science

Author(s): Deepak A. Kaji, John R. Zech, Jun S. Kim, Samuel K. Cho, Neha S. Dangayach, Anthony B. Costa, Eric K. Oermann, Ilya Safro.

http://doi.org/10.1371/journal.pone.0211057

Abstract

This study trained long short-term memory (LSTM) recurrent neural networks (RNNs) incorporating an attention mechanism to predict daily sepsis, myocardial infarction (MI), and vancomycin antibiotic administration over two week patient ICU courses in the MIMIC-III dataset. These models achieved next-day predictive AUC of 0.876 for sepsis, 0.823 for MI, and 0.833 for vancomycin administration. Attention maps built from these models highlighted those times when input variables most influenced predictions and could provide a degree of interpretability to clinicians. These models appeared to attend to variables that were proxies for clinician decision-making, demonstrating a challenge of using flexible deep learning approaches trained with EHR data to build clinical decision support. While continued development and refinement is needed, we believe that such models could one day prove useful in reducing information overload for ICU physicians by providing needed clinical decision support for a variety of clinically important tasks.

Partial Text

The Society for Critical Care Medicine (SCCM) estimates that 55,000 patients per day are treated in intensive care units (ICUs) throughout the United States at an annual cost of approximately $82B USD in 2005 [1]. These patients have an average length of stay of 3.8 days with a mortality rate of 10-29% [1, 2]. One of many significant challenges faced by physicians managing these patients is the need to deal with a tremendous amount of real-time information. It is important to prevent information overload to ensure safe and efficient delivery of patient care. Reducing information overload is associated with more rapid care with fewer errors [3]. To aid in clinical care and provide high level supportive analytics, numerous attempts have been made to develop and implement predictive models and computer assisted diagnostic (CAD) solutions that are interpretable, generalizable, and able to predict important clinical outcomes [4–8].

Improving detection and/or optimizing management of each of the three targets we chose to model—sepsis, MI, and the need for starting medications such as vancomycin—are critical steps towards improving ICU outcomes [3–6, 38]. Sepsis is an important cause of morbidity and mortality for critically ill patients. In-hospital MI increases the risk of mortality and length of stay. Timely initiation of vancomycin can help improve patient outcomes in patients who suffer from MRSA infections.

 

Source:

http://doi.org/10.1371/journal.pone.0211057

 

Leave a Reply

Your email address will not be published.