Related Experiment Video
Updated: Oct 8, 2025
07:31
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
7.3K
Postoperative delirium prediction using machine learning models and preoperative electronic health record data
Andrew Bishara1,2, Catherine Chiu1, Elizabeth L Whitlock1
1Department of Anesthesia and Perioperative Care, University of California, San Francisco, 521 Parnassus Avenue, San Francisco, CA, 94143, USA.
BMC Anesthesiology
|January 4, 2022
Summary
Machine learning models accurately predict postoperative delirium (POD) risk using electronic health record data. These models offer improved risk stratification for surgical patients, enhancing perioperative care.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Medicine
- Perioperative Medicine
Background:
- Accurate risk stratification for postoperative delirium (POD) is crucial for targeted preventative interventions.
- Machine learning (ML) presents a promising approach for predicting POD by utilizing electronic health record (EHR) data.
Purpose of the Study:
- To develop and validate an ML-derived POD risk prediction model using preoperative EHR data.
- To compare the performance of ML models against traditional logistic regression models and existing tools.
Main Methods:
- Retrospective analysis of EHR data from 24,885 adult patients undergoing procedures requiring anesthesia.
- Utilized 115 preoperative risk features to predict POD, defined by specific screening scales.
- Evaluated Neural Network, XGBoost, logistic regression models, and the AWOL-S tool using AUC-ROC and calibration metrics.
Main Results:
- ML models (XGBoost: 0.851, Neural Net: 0.841) demonstrated significantly higher predictive accuracy (AUC-ROC) than logistic regression and AWOL-S.
- All ML models and the ML hybrid regression model showed excellent calibration.
- Clinician-guided regression and AWOL-S models exhibited moderate calibration, tending to overestimate risk in high-risk patients.
Conclusions:
- ML models effectively predict POD using readily available EHR data in a diverse perioperative population.
- Automated, real-time POD risk stratification via ML can enhance perioperative management for at-risk surgical patients.
