Hospital length of stay AI is too specific, says researchers

Artificial intelligence software that can predict a patient's length of stay is too ad-hoc and is in need of a unified common framework, according to a new study published in PLOS Digital Health April 14. 

Having accurate predictions for how long a patient will remain in hospital is a valuable asset that aids hospitals with resource allocation and forward planning which can reduce costs and improve patient care. Artificial intelligence tools can be used to formulate such predictions by taking into account complexity of care, patient characteristics and the diagnosis severity. 

A research team from Wales reviewed 93 papers on research and implementation of length of stay prediction models. They noted several shortcomings to the prediction models, including a lack of input and data from non-physician care team members such as nurses who spend more time with patients and are more likely to know social or economic determinants of a patient's health. 

The researchers named the most apparent problem though as the lack of a unified framework for approaching length of stay prediction. Almost all datasets used to train the algorithm include data specific to disease, condition or patient-groups and often use domain knowledge of the hospital itself, making the models unable to be generalized.

"Adopting a unified framework for the prediction of length of stay could yield a more reliable estimate of the length of stay," write the researchers.

Copyright © 2022 Becker's Healthcare. All Rights Reserved. Privacy Policy. Cookie Policy. Linking and Reprinting Policy.


Featured Whitepapers

Featured Webinars