Effective communication between patients and healthcare providers is crucial to ensuring high-quality, patient-centered care.
With a substantial rise in use of web-based patient portals, the time required to engage with these communications has contributed to professional burnout in physicians.
To increase efficiency and accuracy of the process, researchers from the Vanderbilt Clinical Informatics Center are developing the Patient Artificial Intelligence Guided E-messages (PAIGE) system, which uses large language model computer programs to formulate follow-up questions that guide patients to the most relevant information needed by their healthcare providers.
A study evaluating the feasibility of the system was recently published in the Journal of the American Medical Informatics Association.
“We are trying to strike a balance, where doctors can be easily reached without experiencing overwhelming burnout,” said Adam Wright, Ph.D., a professor of Biomedical Informatics and Medicine at Vanderbilt University Medical Center.
Siru Liu, Ph.D., an assistant professor of Biomedical Informatics and Computer Science and co-author of the publication, noted that today it is common for patients and healthcare providers to engage in messaging exchanges through online messaging.
“Our goal is to improve the patient-physician messaging experience,” Liu said.
Eliminating Inefficient Messaging
Unsurprisingly, this type of back-and-forth messaging has placed an additional burden on healthcare providers. Delays or breakdowns in communication on either side may prevent patients from progressing to the next stage of their treatment.
“It can take days to resolve issues through electronic communication, due to its asynchronous nature,” Wright said. “Patients may be unresponsive to follow-up questions from their providers, resulting in untreated conditions and barriers to care.”
In an example of exchanges about a urinary tract infection, Wright said a patient might send the message: “I think I have another UTI. Do I need antibiotics?”
In this case, it is likely the provider will reply by asking whether the patient has a fever, back pain, blood in the urine or prior UTIs to rule out serious issues that merit an in-person evaluation. The follow-up via asynchronous messages will not only add to the providers’ obligations but cause delays in care, Wright explained.
Testing the PAIGE System
Large language model programs (LLMs) can improve the efficiency of patient-provider messaging, especially in drafting replies from clinicians to patients, according to studies.
In the most recent study, the Vanderbilt team compared two LLMs that engage with patients drafting questions for their healthcare providers. After an initial inquiry, the systems generate relevant follow-up questions to be addressed when the patient communicates again.
The team found that follow-up questions from one of the two models, GPT4, were more useful and complete – but less clear – than those written by healthcare providers.
The second model, known as Comprehensive Large Language Model Artificial Intelligence Responder (CLAIR), generated follow-up questions equally clear and succinct as a healthcare professional’s response. However, while CLAIR’s questions were less complete than those of GPT4, they were judged to be more useful than either that of healthcare providers or GPT4 models.
Overall, the researchers demonstrated that LLMs were capable of generating follow-up messages that compared favorably to those generated by healthcare providers.
Generalizability Across Practice Settings
In light of these findings, Wright and Liu believe that the PAIGE system holds great promise for improving patient-provider communication across practice settings.
The team is currently working to refine the program’s capabilities and train the system on real patient-provider messaging exchanges, as well as triage guidelines to ensure that its follow-up questions are most relevant and useful. Both patients and providers have been involved in the co-design process.
“We hope to assist patients in delivering a clear and comprehensive message, which in turn will reduce wait times and alleviate the administrative burden on providers,” Wright said.
