Cost prediction for ischemic heart disease hospitalization: Interpretable feature extraction using network analysis.

Area of Science:

• Medical Informatics
• Machine Learning
• Health Economics

Background:

• Ischemic heart disease (IHD) presents a significant global health and economic challenge.
• Accurate prediction of healthcare costs is essential for efficient resource allocation in IHD management.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting IHD hospitalization costs.
• To identify specific diagnoses and procedures contributing to high healthcare expenditures in IHD patients.

Main Methods:

• Developed a network-enhanced machine learning model integrating network analysis and graph kernel techniques.
• Extracted explainable features from a diagnosis-procedure concurrence network to capture complex medical code relationships.
• Utilized temporal validation datasets to assess model performance.

Main Results:

• The network-enhanced model achieved an R² of 0.804 ± 0.008 and RMSE of 17,076 ± 420 CNY.
• Performance was comparable to less interpretable models, demonstrating a balance between accuracy and explainability.
• Identified key diagnoses (e.g., acute kidney injury) and procedures (e.g., temporary pacemaker placement) associated with elevated costs.

Conclusions:

• The proposed model offers a balance between predictive accuracy and interpretability for IHD hospitalization costs.
• It provides insights into cost drivers, supporting intelligent management of IHD.
• The interpretable nature aids in identifying specific high-cost medical events within IHD care pathways.