Construction of performance score dynamic prediction system for clinical departments using explainable machine...

Area of Science:

• Healthcare Management
• Machine Learning in Medicine
• Public Health Informatics

Background:

• Accurate evaluation of clinical departmental performance is crucial for public hospital management.
• Existing methods rely on static, retrospective data, limiting early intervention and informed decision-making.
• There is a need for dynamic, interpretable frameworks for real-time performance assessment.

Purpose of the Study:

• To develop a dynamic framework for predicting annual departmental performance scores.
• To utilize explainable machine learning for enhanced interpretability of performance indicators.
• To provide a practical tool for supporting managerial decisions in public hospitals.

Main Methods:

• Collected 24 months of performance data from 27 clinical departments.
• Trained six machine learning models (LR, DT, RF, Gradient Boosting, XGBoost, CatBoost) to predict annual performance scores.
• Applied SHapley Additive exPlanations (SHAP) for model interpretability and identified key performance indicators.

Main Results:

• The Linear Regression (LR) model demonstrated superior performance with high R² values and low error metrics (RMSE, MAE, MSE).
• SHAP analysis identified consistent top 5 performance indicators across all time windows, including revenue proportions and average inpatient expense.
• A web-based prediction tool, ClinDeptPredictor, was developed using the best-performing LR model.

Conclusions:

• The study presents a novel dynamic framework for clinical departmental performance prediction using explainable machine learning.
• The framework enables progressive performance monitoring and timely insights for managerial decision-making.
• The ClinDeptPredictor tool facilitates practical application and offers methodological support for healthcare performance management.