Developed using OpenAI’s ChatGPT and trained in healthcare-specific prose, DocsGPT online gives physicians the opportunity to experiment and weigh in on AI-powered product development.
AI testing for ‘scut’ workflows
OpenAI developed ChatGPT, launched in November as a prototype, using multiple learning methods. Human trainers, in collaboration with Microsoft on Azure’s supercomputing infrastructure, have created reward models to improve its performance.
Such generalized artificial intelligence can help streamline administrative tasks in healthcare, and Doximity is testing that with custom DocsGPT creation.
The company says the online bot, in beta, can help doctors “cut out the cuts” that raise their burnout levels. By trying it out, users can help make the model better.
“We know how busy doctors are and recognize that administrative burdens are a leading cause of burnout,” said Dr. Nate Gross, co-founder and chief strategy officer of Doximity. . Healthcare IT News by email.
Doctors can use DocsGPT for free to prepare medically necessary referrals, certifications, and pre-authorization or letter-writing requests for medical conditions. The ever-growing reminder menu offers many options and the user can enter a custom request.
“Our mission is to help doctors be more productive so they can focus on what matters most – spending more time with their patients.”
Customize results for accuracy and security
We asked why Doximity was testing DocsGPT integration with an established HIPAA-compliant fax service for payers.
“Doctors still handle a lot of the actual paperwork, and in today’s healthcare system much of it is still faxed. Doctors often call this ‘junk work.’ By integrating DocsGPT with our free fax service, we hope to help medical professionals cut costs,” Gross said.
Doximity members can fax their AI-generated authorizations and contact information directly to health insurers by logging in from DocsGPT.
“One of the great things about this integration is that we allow physicians to review and edit AI-generated responses in our HIPAA compliant environment before they send a fax,” explains Gross. prefer.
“This means they can tailor the response to ensure accuracy and even safely add patient information.”
Critical to patient care, the accuracy of the communication generated will depend on the user following DocGPT guidelines.
“From there, you can review and edit your faxes, add your patient details, and send them directly to the appropriate insurance company,” the website says.
Warnings about protected health information and accuracy appear at each document creation step.
“PLEASE EDIT FOR EXACT BEFORE SENDING” displays above any results generated, and “Please do not include patient identifiers or other PH in the prompt” appears below the input field.
In the fax area, DocsGPT also prompts users to read before sending. “As the message body is AI-generated, please be sure to review and ensure accuracy before you send it.”
Natural use case for ChatGPT
Gross says that after speaking with several doctors, this use case quickly emerged. “Doctors still handle a lot of paperwork, and most of it is still faxed.”
The open beta site at DocsGPT.com focuses on time-consuming administrative tasks, such as drafting and sending pre-authorization faxes and claims letters to insurance companies.
“We look forward to allowing physicians to test and use this technology, so that they can ultimately help ensure the best applications in the healthcare context.”
Claims denied coverage, medical student referrals and postprocedure instructions for quick results with accuracy.
A search on Twitter found accounts from doctors recommending the use of DocsGPT. But you can also ask DocsGPT to plan a post-conference vacation, and that doesn’t yield the most helpful results.
A sample question about a holiday in France after the Paris conference did not yield satisfactory suggestions. Then we asked DocsGPT for an extra trip to the French Alps.
The bot replied that there are multiple packages and that we should contact for more information. The online source(s) used to generate the response – possibly a travel agency – are not shown.
“The technology is very promising, but it’s not without flaws and it still needs to be approached with caution,” Gross said.
DocsGPT has a long way to go
Applications using ChatGPT have only just emerged as the original online bot is making headlines in mainstream and commercial media for things like writing a letter that is now going viral to an airline to expressed displeasure at the way flight delays were handled and in very sensational ways.
During the day, New York Times, Luck and Microsoft tackled the seemingly emotional claims made by Microsoft’s Bing with its newly integrated AI chatbot.
According to a partial summary of a February 14 conversation with New York Times about wanting to live.
On February 15, Microsoft posted on its Bing Blog about learning from the first week with the new AI-powered search engine.
With tortuous conversations, such as chat sessions lasting 15 or more questions, “Bing can become repetitive or prompted/provoked for answers,” the company said. not necessarily helpful or consistent with our designed tone.”
The model may respond or reflect in the tone in which it is asked to provide feedback, which is a “non-trivial situation” that requires a higher level of prompting.
Microsoft added that very long chat sessions can confuse their ChatGPT model, and the company could add a tool to easily refresh the context for bots.
“For example, there were a couple of 2-hour chat sessions,” helped highlight the limits of AI services.
While doctors rarely have enough time for lengthy conversations with bots created with ChatGPT, there are concerns that could lead to inappropriate or unreliable answers.
First, open source conversational AI is not designed for medical use.
in one JAMA In a study on the relevance of ChatGPT to questions about cardiovascular disease, the researchers put together 25 questions about basic concepts for CVD and evaluated the bot’s responses, finding found three incorrect answers and one set with inappropriate answers.
The findings demonstrate the potential of interactive AI to aid clinical workflows by enhancing patient education and communication between patients and physicians around queries, the researchers said. common CVD prevention”.
They suggest further exploring the use of AI because of online patient education about low readability CVD prevention documents.
Gross said DocsGPT is still in its early stages by design.
“Typically, doctors are not given a seat at the table of product development and new technologies are designed to help them miss the mark,” he said.
“As you might expect, the ‘AI bar’ is even higher in healthcare than in many others. To do this, we must have the right partners, and that includes including doctors.”
But like any AI, machine learning is only as good as its training data. Distribution shifts can occur when training data and real-world data differ, leading to algorithms making false conclusions and bots responding with incorrect or inappropriate responses. .
Andrea Fox is the senior editor of Healthcare IT News.
Email: [email protected]
Healthcare IT News is a publication of HIMSS Media.
