Enhancing Delirium Prediction and Prevention in Elderly Patients Through Machine Learning-Based Analysis

Area of Science:

• Geriatric Medicine
• Computational Medicine
• Clinical Informatics

Background:

• Delirium is a common complication in elderly hospitalized patients, associated with adverse outcomes.
• Early identification of patients at risk for delirium is crucial for timely intervention.
• Existing methods for delirium prediction may lack accuracy and efficiency.

Purpose of the Study:

• To identify predictors of delirium within 24 hours of admission in elderly patients.
• To evaluate the performance of machine learning (ML) models in predicting early-onset delirium.
• To determine the most influential factors contributing to delirium development in this population.

Main Methods:

• A prospective cohort study involving 327 elderly patients (≥65 years) admitted to a general medical unit.
• Analysis of clinical and demographic data using four ML models: logistic regression, random forest, gradient boosting, and support vector machine.
• Model performance assessed using accuracy, precision, recall, F1 score, and AUC-ROC; robustness confirmed via cross-validation and feature importance analysis.

Main Results:

• The random forest model achieved the highest performance, with 96.9% accuracy, 97.2% F1 score, and 98.4% AUC-ROC.
• Cross-validation indicated stable and robust model performance.
• Key predictors identified include acute kidney injury, respiratory failure, dementia, stroke, and decompensated heart failure.

Conclusions:

• Machine learning models, particularly random forest, demonstrate significant potential for accurately predicting delirium in elderly patients within 24 hours of admission.
• These findings support the integration of ML tools for enhanced early delirium detection and targeted preventive strategies.
• External validation of the developed model is recommended for broader applicability across diverse healthcare settings.