An Interpretable Machine Learning Approach for Predicting Hospital Length of Stay and Readmission

Yuxi Liu, Shaowen Qin

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Length of stay (LOS) and risk of readmission of patients are critical indicators of the quality and operation efficiency of hospitals. Various machine learning (ML) approaches have been applied to predict a patient’s hospital LOS and risk of readmission, but those with more accurate predictions are often the so called ‘black-box’ approaches. This study aims to add interpretability in predicting LOS and the risk of readmission in 30-day among all-cause patients admitted through the emergency department (ED) while improving the accuracy and parsimony of the ML approach. Several state-of-the-art ML models were applied to our prediction tasks and their predictive power reported and compared. The CatBoost model outperformed the rest, hence is chosen as the baseline for this study. For interpretability, we introduced Shapley values and analyzed, at both aggregated and individual levels, the prediction results from the CatBoost model. Lower dimension models were further developed following the guidance of Shapley values. Our results show that the lower dimension model can robustly predict hospital LOS and risk of readmission, indicating that Shapley values are not only useful for adding model interpretability, but also effective for creating a lower-dimensional model amenable to implementation.

Original languageEnglish
Title of host publicationAdvanced Data Mining and Applications
Subtitle of host publication17th International Conference, ADMA 2021 Sydney, NSW, Australia, February 2–4, 2022 Proceedings, Part I
EditorsBohan Li, Lin Yue, Jing Jiang, Weitong Chen, Xue Li, Guodong Long, Fei Fang, Han Yu
Place of PublicationSwitzerland
PublisherSpringer Science and Business Media Deutschland GmbH
Pages73-85
Number of pages13
ISBN (Electronic)9783030954055
ISBN (Print)9783030954048
DOIs
Publication statusPublished - Feb 2022
Event17th International Conference on Advanced Data Mining and Applications, ADMA 2021 - Sydney, Australia
Duration: 2 Feb 20224 Feb 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13087 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference17th International Conference on Advanced Data Mining and Applications, ADMA 2021
Country/TerritoryAustralia
CitySydney
Period2/02/224/02/22

Keywords

  • Emergency department
  • Interpretability
  • Length of stay
  • Machine learning
  • Readmission

Fingerprint

Dive into the research topics of 'An Interpretable Machine Learning Approach for Predicting Hospital Length of Stay and Readmission'. Together they form a unique fingerprint.

Cite this