South SFV Today

The issue of clinician burnout is gaining attention, with nearly half of U.S. doctors affected, resulting in an estimated annual cost of $4.6 billion due to turnover and reduced work. Mohsen Bayati, a professor at Stanford Graduate School of Business, emphasizes the importance of understanding and addressing this problem.

A recent study led by Bayati and colleagues from Stanford University School of Medicine, Harvard Medical School, and Washington University School of Medicine explores the potential of electronic medical records (EMRs) to predict physician burnout. The research utilized artificial intelligence to analyze data from EMRs, aiming to identify signs of burnout as they occur rather than retrospectively.

"The current way we identify and respond to burnout is always retrospective," says Daniel Tawfik, an assistant professor of pediatrics involved in the study. The researchers sought a method to prospectively pinpoint conditions that elevate the risk for healthcare worker burnout.

Surveys have traditionally been used to assess burnout levels among clinicians. However, Bayati points out that while surveys indicate who has burnout, EMRs can help predict which clinics are most likely to experience it.

The study involved machine learning analysis on data from 233 physicians across 60 clinics over 18 months. The team identified about 200 relevant predictors from an initial set of over 1,500 variables related to clinical workload and efficiency measures.

One significant finding was that the number of automated messages received by physicians was a strong predictor of burnout. "All sorts of messages come into the clinician’s inbox in the electronic record," Tawfik explains, creating an "inbox burden."

Another unexpected predictor was having team members write medical notes for clinicians. Interviews revealed that this practice might not reduce workload as expected because physicians often need to edit these notes themselves.

The research suggests that EMR-based predictors are more effective at identifying clinic-level risks rather than individual ones. This approach could help target interventions at high-risk clinics without relying solely on survey responses.

Potential interventions include customizing inbox messages for clinicians and allowing them to choose their preferred note-taking strategies. These individualized approaches could alleviate some stressors contributing to burnout.

Looking forward, Tawfik highlights the need for real-time risk indicators that do not add extra burdens on clinicians' workloads. "Real-time risk indicators could help us identify high-risk settings with less survey burden for clinicians," he says.

This research provides a foundation for developing methods to mitigate clinician burnout through targeted interventions based on EMR data analysis.