Abstract 4117479: INTERPRETABLE PREDICTION OF HOSPITALIZATION COSTS FOR PATIENTS WITH ISCHEMIC HEART DISEASE USING NETWORK ANALYTICS AND MACHINE LEARNING

Background: Ischemic heart disease (IHD) represents a significant global health challenge, imposing a considerable social and economic burden on individuals and healthcare systems worldwide. To optimize medical resource allocation and better serve populations, accurately predicting healthcare costs is paramount. Methods: This study proposes an interpretable feature extraction method based on network analytics to improve the prediction of IHD hospitalization costs. A diagnosis-procedure concurrence network is constructed, and a graph kernel-based approach is utilized to capture and encode information on diseases and treatments. The study introduces a hospitalization cost prediction model that utilizes a diagnosis-procedure concurrence network to extract features that strike a balance between predictability and interpretability. Results: Our model demonstrates superior performance, considering that the average cost for all patients in the HQMS dataset is approximately 20,000 CNY, whereas the average cost for our IHD patient dataset is 37,853 CNY. This indicates the effectiveness of our approach in predicting hospitalization costs for a specific patient population with complex treatment options such as IHD. Conclusion: This study aims to develop an interpretable and precise model for forecasting hospitalization costs of Ischemic Heart Disease (IHD) patients in China. It achieves this goal by integrating machine learning with network analytics and incorporating cost ratio features.