Related Experiment Video
Updated: Jun 27, 2025
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
7.5K
Development and validation of a machine learning model to predict postoperative delirium using a nationwide database:
Manabu Yoshimura1, Hiroko Shiramoto1, Mami Koga1
1Department of Anesthesiology, Ube Industries Central Hospital, Ube City, Japan.
Journal of Clinical Anesthesia
|April 28, 2024
Summary
A new machine learning model effectively predicts postoperative delirium in elderly surgical patients using extensive healthcare data. This tool aids in early identification, potentially reducing prolonged hospital stays and improving patient outcomes.
Area of Science:
- Medical Informatics
- Machine Learning in Healthcare
- Geriatric Surgery
Background:
- Postoperative delirium is a significant complication in surgical patients, leading to increased morbidity, mortality, and hospitalization.
- Prompt identification of delirium signs is crucial for timely intervention and improved patient management.
- Existing methods for delirium prediction may lack accuracy or efficiency in large-scale clinical settings.
Purpose of the Study:
- To develop and validate a machine learning (ML) model for predicting postoperative delirium in elderly patients undergoing surgery.
- To leverage extensive Japanese Diagnosis Procedure Combination (DPC) inpatient data for model development and validation.
- To assess the predictive performance of the ML model using internal and temporal validation datasets.
Main Methods:
- A retrospective observational study design was employed.
- Japanese DPC inpatient data from 2016 to 2019 were utilized, including patients aged 65 years and older who underwent general anesthesia.
- A light-gradient boosting machine model was trained and optimized using 10-fold cross-validation, with performance evaluated by Area Under the Receiver Operating Characteristic Curve (AUC), recall, and precision.
Main Results:
- The light-gradient boosting machine model achieved a high AUC of 0.826 (95% CI: 0.822-0.829) in internal validation.
- The model demonstrated a recall of 0.124 and precision of 0.659.
- Performance was sustained in temporal validation (AUC: 0.815), with a specificity of 0.672 and a negative predictive value of 0.975 at 80% sensitivity.
Conclusions:
- A validated machine learning model for predicting postoperative delirium has been successfully developed using large-scale DPC data.
- The developed model shows promising performance and may serve as a valuable tool for facilitating the prediction of postoperative delirium in clinical practice.
- Further implementation and evaluation of this ML model could enhance early detection and management strategies for postoperative delirium.
