Prediction of delirium in trauma patients using interpretable machine learning
Sujong Shin1, Seok Bum Lee2,3, Jung Jae Lee2,3
1AI-based Convergence, Dankook University, Yongin, 16890, Gyeonggi-do, Republic of Korea.
Scientific Reports
|June 4, 2026
Summary
This study identified key delirium risk factors in trauma patients using machine learning. Age, Injury Severity Score, and specific lab values predict delirium, enabling early intervention.
Area of Science:
- Trauma Care
- Machine Learning in Medicine
- Critical Care
Background:
- Delirium is a common complication in trauma patients, associated with adverse outcomes.
- Predicting delirium risk in trauma patients is crucial for timely intervention and improved patient management.
Purpose of the Study:
- To identify key risk factors for delirium in trauma patients.
- To develop an interpretable machine learning model for delirium risk prediction using routine admission data.
Main Methods:
- Analysis of data from 7,806 trauma patients admitted between 2015 and 2023.
- Construction and evaluation of an XGBoost-based prediction model.
- Assessment of model interpretability using Shapley Additive Explanations (SHAP).
Main Results:
- Delirium occurred in 7.3% of patients.
- The model achieved 92.0% accuracy and high AUC values (0.76 macro, 0.96 micro).
- Key predictors included age, Injury Severity Score (ISS), lactate dehydrogenase (LDH), and estimated glomerular filtration rate (eGFR), with identified threshold effects.
Conclusions:
- An interpretable machine learning model effectively predicts delirium risk in trauma patients using routine data.
- The model provides a transparent basis for individualized risk assessment and early prevention strategies.
- Identifying high-risk patients based on age, ISS, LDH, and eGFR can guide proactive delirium management in acute trauma care.
