There is a lot of hype surrounding AI in healthcare and the rush of digital health companies looking to make money.
But experts aren’t sure how general AI applications like ChatGPT and GPT-4 will affect clinical diagnosis and decision-making. Most say the first wave of adoption will take place in areas with administrative redundancies.
Related: Microsoft, Nuance Introduce ChatGPT successor to healthcare
“There’s obviously a lot of energy and a lot of concern,” said Dr. Greg Ator, director of health informatics at the University of Kansas Health System. “People just need to get out in front of their skateboards on some of these technologies.”
Instead, the early adoption of general AI in healthcare is happening in the less flashy clinical arena. records. performer is part of a team that is implementing general AI technology at the academic health system to support clinicians records. The system is working with Abridge, a medical AI company, to summarize clinical conversations from audio recorded during patient visits.
Abridge’s general AI technology is similar to Nuance Communications, a clinical documentation software company owned by Microsoft. Last Monday, Nuance said it was adding OpenAI’s ChaptGPT successor GPT-4 to its latest application, which will be used in electronic health record systems.
In both cases, users must describe what they see for the software to function properly. For example, if a patient has a sore throat, specific comments about what the clinician is seeing must be verbally shared in order for the program to enter the information.
In addition to entering relevant information into the EHR, both apps delete conversations that don’t apply to a plan of care.
“They are powerful tools,” said Shiv Rao, co-founder and CEO of Abridge. “[Generative AI is] a powerful tool in the context of a much larger set of technologies that work together to create a solution that can create value in the workflow.”
Ator says medical records are a logical place to start because clinicians can quickly identify the source of AI-generated results. Clinicians can easily listen back to the exam recording if the AI misses valuable information.
“What you build next to, under, and above these foundational models like GPT-4 is the secret recipe,” says Rao. “There is a certain layer of technology that is available to all of us right now, but how we integrate those tools into larger solutions will make the difference between a truly magical experience for the physician and the patient. their patients with solutions that feel like they are available. toy.”
Investor interest remains strong
Healthcare AI investments total $4.4 billion by 2022, according to data from Rock Health, a digital health and research venture firm. While last year’s total was down more than 50% from 2021, it’s in line with 2020.
The same data revealed in 2021 has set a high mark with 224 transactions for companies using AI technology. While 2022 isn’t as productive, it’s still higher than 2020. While experts say levels from 2021 won’t return anytime soon, the space is still of interest.
While investments in AI remain strong, not many of those technologies are ready for widespread adoption, skeptics say.
“I think that for some time to come, we will continue to need human intervention,” said Erik Brynjolfsson, director of the digital economy lab at Stanford University’s Human-Centered AI Institute. Cards because AI is far from perfect. . “It can’t do many things.”
Brynjolfsson says trained medical professionals can quickly eliminate abnormalities on scans or medical images while Who can misdiagnosis. While it is likely to eventually replace some of the clinician’s roles, experts say human input is still crucial.
Innovative AI also takes a long time to install, and even in promising areas it’s not ready for primetime. Nuance will roll out the GPT-4 feature in the summer.
Ator said the University of Kansas Health System is rolling out the technology for several months. He is optimistic that it will be completed by the end of the year, but does not want to give a specific timeline. This is largely due to the clinician training time and required integration with the provider’s Epic EHR platform.
“Whenever you’re working with a complex system, such as Epic, which is our basic health system, we have to interact with them. Some [the implementation] motivated by their timetable,” said Ator.
Another potential barrier to adoption could be patient acceptance. A Pew Research Center survey conducted in December found that 60% of US adult patients would feel uncomfortable if their healthcare provider relied on AI for their care. medical care. Less than a third feel the quality of their care will increase as AI is deployed.
Although the study did not specifically ask respondents about analyzing audio recordings of their visits, the report’s authors found that “concerns about the speed of AI adoption” were widely shared. widely in medicine.
Brynjolfsson says medical spelling and imaging are areas where providers can improve the process. But he said the future of healthcare will continue to require clinician and human oversight.
Even so, others are more optimistic about future adoption.
Dr Robert Pearl, former CEO of Oakland, CA-based Kaiser Permanente and now a professor at Stanford University, said: “What we see today is just an indication of what’s going on,” said Dr. going to happen. “People are focusing on the mistakes of the day or the shortcomings of today. They are not related.”
This story first appeared in Digital Health Business & Technology.
