Predicting Radiology Burnout via Machine Learning: A Cohort Study
1MOE Key Laboratory of Geriatric Diseases and Immunology, School of Public Health, Suzhou Medical College of Soochow University, Suzhou 215123, Jiangsu Province, China (C.M., T.Z.).
Rationale And Objectives:
Burnout is prevalent among radiology staff, negatively impacting job satisfaction and patient care quality. This study aims to develop and validate machine learning-based predictive models for burnout in radiology staff, focusing on the role of artificial intelligence (AI) software utilization.
Materials And Methods:
This prospective longitudinal cohort study enrolled 496 radiologists and technicians from 68 public hospitals in China. Data were collected at baseline and at a 3-month follow-up using questionnaires that covered demographic characteristics, work-related factors, AI software usage, and burnout assessed by the Maslach Burnout Inventory-Human Services Survey (MBI-HSS). Five machine learning algorithms were trained on a training set (n = 402) and validated on an independent validation set (n = 94). SHapley Additive exPlanations (SHAP) values were utilized to enhance model interpretability. The predictive performance of the models was evaluated using the area under the receiver operating characteristic curve (AUC-ROC).
Results:
The XGBoost model exhibited the best predictive performance on the external validation set, achieving an AUC-ROC of 0.719. Salary satisfaction and the use of AI for coronary CTA were positively associated with burnout risk, whereas the use of AI for chest CT was negatively associated. SHAP analysis uncovered complex interactions between work-related stressors, AI utilization patterns, and burnout risk. Machine learning models, particularly the XGBoost model, effectively predicted burnout risk among radiology staff.
Conclusion:
Machine learning models, especially the XGBoost model, demonstrated robust predictive capabilities for identifying burnout risk in radiology staff. Future research should incorporate multi-modal data sources and validate the model in diverse healthcare settings to further enhance its generalizability and applicability.