Predicting outcomes after hospitalisation for COPD exacerbation using machine learning.

Area of Science:

• Pulmonary Medicine
• Health Informatics
• Artificial Intelligence in Healthcare

Background:

• Exacerbation of chronic obstructive pulmonary disease (ECOPD) hospitalizations carry high risks of early readmission and mortality.
• Accurate prediction of these adverse outcomes is crucial for patient management and healthcare resource allocation.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting 30-day readmission and death after ECOPD hospitalization.
• To compare the performance of ML models against the established LACE index.
• To identify key predictors influencing readmission and mortality.

Main Methods:

• Utilized a nationwide database including index ECOPD hospitalization and prior year data.
• Developed prediction models using logistic lasso regression, random forest, extreme gradient boosting (XGBoost), and neural networks.
• Assessed model performance using ROC curves and calibration plots; identified predictors via SHapley Additive exPlanations.

Main Results:

• The study analyzed 101,011 hospitalizations (development) and 17,565 (validation).
• 30-day readmission and death rates were 29.1% and 4.3%, respectively.
• XGBoost demonstrated superior performance (AUC 0.721 for readmission, 0.809 for death), outperforming the LACE index (AUC 0.651 and 0.641).
• Prior hospitalizations and lowest hemoglobin levels were top predictors for readmission and death, respectively.

Conclusions:

• Machine learning models effectively enhance prediction of early readmission and death post-ECOPD hospitalization.
• Identifying critical prognostic factors allows for improved, targeted post-discharge care strategies for high-risk ECOPD patients.