Clues to physician-patient rapport might lay in EHR in-baskets

A full in-basket of messages from patients in a physician’s electronic health record doesn’t add to clinician burnout, but messages loaded with curse words might perhaps help point the way toward how to improve provider operations and also act as an early-warning alert to possible violent attacks.

Those are two of the findings in a recent study in JAMA Network Open. Researchers with the University of California, San Diego suggest that EHR portals be equipped with natural language processing (NLP) capabilities, an AI system that facilitates communication between humans and computer systems.

NLP has been applied in biometric applications, including electronic phenotyping and language generation for digital health chatbots. In addition, NLP has been used to analyze unstructured/free-text clinical notes, radiology reports and pathology findings.

Though the study focused on physician burnout and the use of EHRs, it did state that certain high-frequency words point to the possibility of their leading to violence.

“This is concerning, especially given documentation of patient-inflicted violence against physicians,” the study said.

Ways to preempt this problem include directions on the portals for users to employ courteous language and creating filters that track expletives or threatening words.

The cross-sectional study used NLP to analyze 1,453,245 messages received by 609 physicians representing multiple specialties. The data were collected from April to September 2020 and analyzed from September to December 2020, during the first waves of the COVID-19 pandemic.

Researchers note the 21st Century Cures Act encourages the use of the EHR in-basket, as does the increased use of telemedicine. They cite data showing that EHR in-basket messages increased by 157% during the pandemic.

The in-basket messages can include, aside from the physician/patient exchange, notes from other staff members and various other treating physicians. Information might include laboratory or imaging results.

Researchers compared 767,855 messages received by 307 physicians saying that they feel burned out, with 685,390 messages received by 302 physicians saying that they weren’t burned out. Though the physicians saying they felt burned out received more messages, researchers concluded that the difference was not statistically significant.

“There were no significant differences in the message sentiment score, proportion of positive messages, proportion of negative messages, or mean word count,” the study found. “Approximately one-half of messages were positive in both groups, and only approximately 5% of messages received were classified as negative in both groups, with the remainder being neutral.”

Furthermore, patient messages tended to be positive for both groups: 69% for physicians without burnout and 67% for physicians with burnout.

Though the study failed to find a link between physician burnout and how many messages might be in an EHR in-basket, researchers did find that NLP could capture negative messages.

“Although infrequent, these negative messages demonstrated a striking range of hostility toward physicians and health systems in general,” the study said.

In addition, “using NLP for analyzing EHR in-basket messages comprises a novel approach for furthering our understanding of physician burnout and for developing strategies to reduce risk of burnout, an important consideration as federal regulation and shifting models of health care delivery during the COVID-19 pandemic have increased the use of EHR in-basket messaging.”