Using machine learning approaches to predict high-cost chronic obstructive pulmonary disease patients in China

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Public Health Economics

Background:

• Chronic Obstructive Pulmonary Disease (COPD) poses a significant economic burden.
• Accurate identification of high-cost COPD patients is crucial for effective resource management.
• Predictive modeling can help anticipate healthcare expenditures for COPD.

Purpose of the Study:

• To develop and compare machine learning models for identifying and predicting high-cost COPD patients.
• To explore key variables influencing the prediction of medical costs in COPD.
• To evaluate the predictive performance of different machine learning algorithms using distinct variable sets.

Main Methods:

• Utilized medical insurance data from a large city in western China.
• Employed machine learning algorithms: Logistic Regression, Random Forest (RF), and Extreme Gradient Boosting (XGBoost).
• Assessed model performance using ROC curves, precision, and accuracy metrics.

Main Results:

• All three models demonstrated good predictive performance.
• The Extreme Gradient Boosting (XGBoost) model exhibited superior predictive accuracy (AUC: 0.801) compared to Logistic Regression (AUC: 0.787) and Random Forest (AUC: 0.792).
• Precision and accuracy metrics confirmed the reliability of the developed prediction methods.

Conclusions:

• Machine learning, particularly XGBoost, is effective in identifying and predicting high-cost COPD patients.
• These findings support healthcare data analysts, policymakers, and planners in optimizing health service delivery.
• The study provides a reliable framework for managing the economic impact of COPD.