Support
Contact us
English
Back to lobby 
Gil Ben-Horin Published: Jun 16, 2016
In a study done at Beth Israel Deaconess Medical Center, the authors developed novel deep learning models based on Gated Recurrent Units (GRU) to effectively exploit two types of informative missingness patterns, i.e., masking and time duration. They demonstrated the performance of their proposed models on one synthetic and two real-world healthcare datasets (MIMIC-III, PhysioNet) and compared them to several strong machine learning and deep learning approaches in classification tasks, achieving promising results. In the supplementary section about MIMIC-III preprocessing details, the authors note that they used 19,714 admission records collected from 2008-2012 by MetaVision, which is still used at the hospital, and point out that “The data collection and organization in MetaVision system is much neater than the earlier Philips CareVue system [2001-2008].” Moving forward, the authors would like to explore deep learning approaches to characterise missing-not-at-random data and to conduct theoretical analysis to understand the behaviors of existing solutions to handle missing values.
Gil Ben-Horin Jacob Krygier is a Clinical Application Specialist at iMDsoft with 20 years of hands-on healthcare experience. His background spans operating rooms, intensive care units, emergency medicine, and medivac air ambulances, with specialized expertise in anesthesia and emergency care.
Book a demo 
Login to Support Portal 
