Development and Internal Validation of Interpretable Machine Learning Models for Identifying Burnout Syndrome Among...

Burnout among intensive care unit (ICU) nurses threatens patient safety and healthcare quality. We aimed to develop and internally validate a machine learning model to identify current burnout and its key correlates in this population.

We surveyed 318 ICU nurses across four tertiary hospitals in three provinces of China (October 2024-November 2024), measuring 34 potential predictor variables. Data were partitioned into training (70%) and testing (30%) sets with downsampling addressing class imbalance. LASSO regression identified 12 significant predictors, which were evaluated using 10 machine learning algorithms. The final model was assessed using AUC, calibration, and SHAP analysis.

The random forest algorithm showed optimal performance, with the final nine-predictor model achieving an AUC of 0.983, with good calibration (Brier score 0.054). SHAP analysis revealed psychological resilience (0.197) and job satisfaction (0.152) as primary protective factors, while nursing stress (0.059), night shift frequency (0.016), and poor sleep quality (0.015) emerged as key risk factors. Marital status, commuting mode, children, and residential area contributed additionally to predictions.

The internally validated classification model developed in this study suggests that psychological resilience, job satisfaction, and nursing stress may play important roles in ICU nurse burnout. These findings can help nurse managers target organizational interventions-such as adequate staffing, recovery-protective scheduling, and support for resilience and job satisfaction-to prevent burnout, rather than placing responsibility on individual nurses. Further validation of this tool in diverse healthcare settings would be beneficial.