The University of Kansas Health System will test software that automatically generates notes from doctor-patient discussions, a move described as “the most important deployment of generative AI in healthcare to date.”
The system, built by Pittsburgh, Pennsylvania-based startup Abridge, seeks to cut clinicians’ workloads and enhance patient care. Shivdev Rao, the company’s CEO, and a cardiologist told The Register that doctors can spend hours outside of work filling up notes from past patient visits.
As per the Register, this has contributed significantly to the current public health crisis involving doctors and nurses abandoning the profession due to burnout and exhaustion. When drafting their notes, clinicians will frequently need to transcribe audio recordings or recall talks from memory, she added.
“As an example of the current status quo, I may see patients and then not document until the weekend, at which point I’ll use scribbles on paper to remember what we discussed. So, I am losing a great deal of information in this document “Rao said.
Using AI and natural language processing algorithms, Abridge’s software automatically produces summaries of medical conversations. In a brief demonstration, The Register pretended to be a patient suffering from shortness of breath, diabetes, and weekly consumption of three bottles of wine. The software from Abridge was able to record information such as symptoms, medications prescribed by the doctor, and steps the clinician should follow up on during future appointments.
The code operates by listening for keywords and classifying relevant data. As per Rao, “If I said to take Metoprolol twice per day, Metoprolol would be an entity, and twice per day would be an attribute. And if I mentioned it verbally, that is another characteristic. And we could do the same with the wine illustration. Three bottles of wine would be an attribute of the wine entity, and another attribute would be added nightly.”
With a flood of startups racing to build products on commercial APIs such as OpenAI’s ChatGPT, Big Tech is feverishly attempting to integrate generative AI into their offerings. Yet, it is well-known that these models originate erroneous information and struggle to generate reliable content, making their use in fields such as healthcare and law extremely problematic.
Patients can access the notes via an app, while physicians can amend the notes further. Rao compared Abridge’s technology like a copilot, emphasizing that doctors retain control and should review and alter the generated notes as needed. Patients and physicians have access to recordings of their meetings and can click on specific keywords to have the program play back portions of the audio when the keyword was said.
“We are retracing our steps from the summary presented to users all the way back to the essence of the interaction. Hence, if I have a conversation in which I cannot recollect something that transpired, I can always confirm that this was not a hallucination. There are models in the middle that take care not to reveal something that was not considered,” said Rao.
The software of Abridge has purportedly been evaluated by over 2,000 clinicians and has assisted over 200,000 patients. Now, the company will test it in a real hospital, beginning with a small group of physicians. The University of Kansas Health System’s Chief Medical Information Officer and Head and Neck Surgeon, Gregory Ator, anticipated that the experiment will commence soon.
“We must determine which clinicians are suitable for this type of service. First, we will check that the datasets utilized [to train the program] are accurate and relevant to healthcare in the Midwest, as well as the types of problems we face, and then we will gradually roll it out ” he stated to The Register.
