A new AI tool from Washington University in St. Louis researchers aims to help identify burnout among physicians and could potentially prevent it in the future.
The end-to-end deep learning tool developed by Chenyang Lu, the Fullgraf Professor in the McKelvey School of Engineering, and his team combs through automatically generated electronic health record activity logs to predict physician burnout.
Burnout has long been an emerging issue for healthcare providers, but the COVID-19 pandemic pushed healthcare professionals to the brink. One recent alarming study found 14% of physicians admit to using alcohol or substances on the job to cope with stress, trauma and burnout from their jobs. Burnout, characterized by feelings of depersonalization and exhaustion, as well as a low sense of accomplishment at work, is up in 2022, with 47% of the healthcare workforce reporting feelings of burnout, according to a January study.
Typically, burnout is determined and measured through self-reporting surveys, and there are few solutions to prevent burnout through earlier identification and mitigation. However, this type of self-reporting could miss physicians and healthcare professionals who aren’t comfortable sharing that information or who simply don’t have time to take surveys, according to researchers.
Researchers tested their AI tool, known as the Hierarchical burnout Prediction based on Activity Logs (HiPAL), with data from the activity logs of 88 physicians of Barnes-Jewish Hospital and fellows at Washington University School of Medicine. They determined that electronic health records (EHRs) are a good barometer for physicians’ concentration, focus and behavior patterns thanks to the thousands of touchpoints that come with usage. For example, EHRs log all activity by users, from log in to reading reports, creating notes and reviewing labs. All in all, EHRs generate between 1,000 and 8,000 actions per shift, up to 90,000 actions per week. All those data points can lead to a sort of “signature” for physicians in their behavior patterns at work, researchers stated.
“These logs can be used to measure workload and tell us how much time one spends on various tasks in an electronic health record, when they perform those tasks, and more,” Lu said in a press release. “If we could predict burnout based on this information readily available in the electronic health record, our model could be deployed in most hospitals and predict physician burnout in an unobtrusive and timely fashion.”
The HiPAL model looked at raw activity log data to predict physician burnout. By comparison, a previous model relied on predefined features extracted from activity logs, such as hours spent working, how many notes were made, how many reports were read and more. That model required these features to be manually defined based on domain knowledge.
“We let the model learn the most important connections itself without human-defined features,” Hanyang Liu, a third-year doctoral student in Lu’s lab and first author on the paper, said in a statement. “The deep model can discover sophisticated patterns in the log and delivers higher predictive performance than standard machine learning models with defined features. The downside is that because the patterns the model learns are sophisticated, they require model interpretation techniques to explain the predictions.”
Washington University researchers plan to present their findings at the ACM SIGKDD Conference on Knowledge Discovery and Data Mining and are optimistic about the potential for HiPAL in healthcare, with plans to conduct further research and validation prior to clinical use.
Amy joined TriMed Media as a Senior Writer for HealthExec after covering home care for three years. When not writing about all things healthcare, she fulfills her lifelong dream of becoming a pirate by sailing in regattas and enjoying rum. Fun fact: she sailed 333 miles across Lake Michigan in the Chicago Yacht Club "Race to Mackinac."