Predicting physician departure with machine learning on EHR use patterns: A longitudinal cohort from a large...

Area of Science:

• Healthcare Management
• Medical Informatics
• Predictive Analytics

Background:

• Physician turnover incurs significant costs for healthcare systems, patients, and medical professionals.
• Proactive identification of physicians at risk for departure is essential for implementing targeted retention strategies.

Purpose of the Study:

• To develop and validate a predictive model for identifying physicians at high risk of departure within a 6-month timeframe.
• To pinpoint key physician characteristics, electronic health record (EHR) usage patterns, and clinical productivity metrics that predict departure.

Main Methods:

• Utilized gradient boosted trees to predict physician departure probability.
• Trained, validated, and tested the model using data from a large ambulatory practice, including physician characteristics, EHR use, and productivity.
• Employed SHAP (SHapley Additive exPlanation) values to determine the influence of variables on model predictions.

Main Results:

• Key predictors of physician departure identified include tenure, panel complexity, physician demand, age, inbox volume, and documentation time.
• SHAP analysis revealed significant interactions between these top variables, underscoring their importance in departure prediction.
• The model demonstrated the ability to predict the likelihood of a physician leaving within 6 months.

Conclusions:

• The developed predictive model effectively identifies physicians at risk of departure.
• Factors such as tenure, workload, and EHR usage patterns are significant predictors of physician turnover.
• These findings can guide the development of targeted interventions to improve physician retention in healthcare settings.