EHR Data Fails to Predict Physician Burnout, May Flag High-Risk Clinics

A new study published in Mayo Clinic Proceedings found that electronic health record (EHR) activity data offer limited value in predicting which individual primary care physicians are at risk for burnout, though they may help identify clinics where burnout is more prevalent. The study analyzed EHR use and clinical workload data from 233 physicians across 60 primary care clinics affiliated with a large academic medical center. Physicians completed 396 surveys measuring burnout during a 2-year period, using the Stanford Professional Fulfillment Index, which captures emotional exhaustion and interpersonal disengagement.

Researchers examined more than 1500 potential predictors derived from EHR activity, including time spent on documentation, In Basket messaging, order entry, and use of efficiency tools. Machine learning models, including gradient boosting and random forest algorithms, were used to assess whether these data could accurately identify physicians experiencing burnout, defined as a score of 3.325 or higher on a 10-point scale. The best-performing model, a gradient boosting classifier, achieved an area under the receiver operating characteristic curve (AUC) of only 0.59, well below what would be considered a clinically useful tool. Other models performed similarly poorly, with AUCs ranging from 0.56 to 0.66.

At the clinic level, however, EHR-based models performed modestly better, identifying clinics in the highest quartile of burnout rates with 56% sensitivity and 85% specificity. Among the most predictive features were physician age, the extent of team member contributions to clinical notes, and the number of administrative messages received through the EHR. Notably, mid-career physicians were at higher risk for burnout, and higher team involvement in notes, rather than reducing physician burden, was associated with greater burnout symptoms.

Although EHR activity reflects some aspects of clinical workload, it does not capture the broader organizational, interpersonal, and personal factors that contribute to burnout. The authors caution that using EHR data alone to identify individual physicians for interventions could be ineffective and ethically problematic. Limitations of the study include its single-institution focus, the use of vendor-specific EHR metrics, and the impact of early COVID-19 pandemic workflows, all of which may affect the generalizability of the findings.

“Future work may improve predictions to provide actionable results,” the study authors concluded. “If the predictive value of these measures can be improved with inclusion of additional characteristics, such unit-level measures may help organizations identify clinics to prioritize for system-level interventions.”

Reference:
Tawfik D, Bayati M, Liu J, et al. Predicting Primary Care Physician Burnout From Electronic Health Record Use Measures. Mayo Clin Proc. 2024;99(9):1411-1421. doi:10.1016/j.mayocp.2024.01.005