31 October 2023
Bosco Seong Kyu Yang,
Sung Yoon Lim,
Proposes REMed, a new deep learning model for EHR prediction tasks
REMed can handle unlimited input records with no feature selection needed
Demonstrates strong performance across 27 prediction tasks and 2 datasets
Shows REMed's event selections align with clinical knowledge
Minimizes bottlenecks of manual input curation and selection
Predicting patient outcomes from electronic health records
This paper proposes a new deep learning model called REMed that can process virtually unlimited electronic health records to make predictions about patient outcomes like mortality and length of stay. REMed eliminates the need for manual feature selection or limiting the observation window, which are common bottlenecks. It works by retrieving the most relevant events from a patient's history and using those to make accurate predictions across many clinical tasks.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper