In the time it takes to read this sentence, ChatGPT could have produced multiple discharge summaries on outgoing hospital patients.
The artificial intelligence tool that ushered in the AI explosion a little over a year ago was able to produce medical notes at a whopping ten times the speed of human doctors, according to a new study. Not only were ChatGPT’s rapid-fire notes written faster than humans, but they were rated “of comparable quality,” the authors said.
“Administrative demands in healthcare significantly contribute to surging healthcare expenses, low job satisfaction and, ultimately, burnout,” the ChatGPT study authors wrote. “The quality of discharge documents is crucial as it influences ongoing patient care and safety and even reimbursement schemes.”
So is ChatGPT ready to take the lead in referring patients to post-acute care?
Maybe not so fast: The study had strict limitations and caveats. The pilot did not involve any real study participants, but instead used six fictitious patient profiles. In addition, the discharge notes were limited to orthopedic cases.
In addition, a different study from last month found almost the opposite result: that AI continues to make significant, and dangerous, mistakes when simplifying post-discharge notes.
Before AI can be considered reliable, senior living operators need to be able to trust that these tools won’t underdiagnose or misdiagnose a senior coming to their facility.
More than a slight omission, the seriousness of many underlying conditions in older adults means that being too trusting in AI could lead to an emergency. In the latter study, the AI “hallucinated” chest pains in one patient and prescribed heart medication for another who didn’t need it.
The researchers behind the orthopedic note study are aware of the limitations of the results, and plan on conducting the study with 1,000 actual patient records as a follow-up.
In addition, further studies need to determine if the time it takes to review ChatGPT-generated notes counteract the rapid ability of the AI to generate them, the study authors said, adding that physicians would need relatively less time to review notes that they themselves produced.
There remains a lot of interest in the potential for AI to remove burdensome administrative tasks from the workload of healthcare staff, including in senior living settings.
While AI may still have difficulty creating accurate summaries from medical notes, AI tools are already being used to help transcribe doctor-patient conversations, and can be used for purposes like billing and scheduling, without staff needing to worry it will spark a health emergency.
