An Interpretable Machine Learning Approach to Prioritizing Factors Contributing to Clinician Burnout

Abstract

Clinician burnout is a multi-factorial problem, and there are limited studies utilizing a theoretical model to assess factors contributing to clinician burnout. A survey of demographic characteristics and work system factors was administered to 278 clinicians (participation rate: 55%). We compare four classifiers with four feature selection methods to predict clinician burnout. We used SHapley Additive exPlanations (SHAP) and permutation importance to prioritize key factors contributing to clinician burnout and interpret the predictions. Random forest had the highest AUC of 0.82 with work system factors only. Six work system factors (administrative burden, excessive workload, inadequate staffing, professional relationship, intrinsic motivation, and values and expectations) and one demographic factor (race) had the highest impact on predicting clinician burnout. Identifying and prioritizing key factors to mitigate clinician burnout is essential for healthcare systems to allocate resources and improve patient safety and quality of care.