퇴원손상심층조사 데이터를 활용한 장기입원 환자 예측 모델 개발 = Development of Predictive Model for Long-Stay Patients Using the Korean National Hospital Discharge In-depth Injury Survey Data

다국어 초록 (Multilingual Abstract)

Length of stay is a critical indicator that significantly impacts medical costs and health insurance finances, and long-term hospitalization is particularly important in the efficient allocation of healthcare resources. This study proposes a machine learning model to predict long-term hospitalization using Korean National Hospital Discharge In-depth Injury Survey data. Seven models were developed and validated using 3,633,134 cases from 2008 to 2022. Comparative analysis of Logistic Regression, Decision Tree, Random Forest, XGBoost (eXtreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine), CatBoost (Categorical Boosting), and MLP (Multi-Layer Perceptron) revealed that CatBoost demonstrated the highest predictive performance. Additionally, to enhance model interpretability, SHAP (SHapley Additive exPlanations), an XAI (eXplainable Artificial Intelligence) technique, was employed to visualize the key variables influencing long-term hospitalization. The analysis identified admission route, age, Charlson Comorbidity Index, and payment method as major factors affecting long-term hospitalization. The findings of this study are expected to contribute to efficient hospital bed management and patient care.