Artificial Intelligence Increasingly Used in Health Care, But Experts Warn of Potential for Misuse
Nov 08, 2023 11:41AM ● By Liv OsbySelf-driving cars, digital assistants, and chatbots are all familiar examples of artificial intelligence in today’s world.
But AI, as it’s known, is a growing technology that’s increasingly spreading into other areas of everyday life, among them health care, where it’s used in a variety of applications, including robotic surgery, billing, and diagnostic imaging.
Still, experts say that medicine is only at the beginning of how AI can be used in health care, and that it has the potential to change how care is delivered, to improve patient outcomes and speed new drug development – among other benefits – so long as it is carefully controlled.
“We are just starting to scratch the surface now of how this can be used to treat patients,” said Rich Rogers, senior vice president and chief information officer for Prisma Health. “There is tremendous potential if we can harness this the right way.”
AI refers to computer technology that can “essentially think for itself and make decisions based on the data it is being fed,” according to TechRadar, an online technology publication.
“AI systems are often hugely complex and powerful,” the publication reports, “with the ability to process unfathomable depths of information in an extremely quick time in order to come to an effective conclusion.”
AI is already integral to a lot of diagnostic testing – where it can pick up problems on an MRI image, for example, that the human eye may not see, said Dr. Georges Benjamin, executive director of the American Public Health Association.
It’s also used to triage ER patients and schedule appointments, he said.
Many tasks that physicians perform can be assisted by the technology, he said, such as documenting information about hospital stays, lab work, medical procedures, and discharge notes. Historically, he added, the physician took notes and dictated the information, which was then transcribed by a staffer, before being reviewed, verified and if necessary, corrected by the doctor.
Today, he said, technology allows information to be recorded at the bedside, which frees up the doctor to see more patients.
“It’s great to have a machine collect what’s happening at the bedside,” Benjamin said. “It’s an enormous time savings.”
Prisma started using AI in diagnostic imaging as a tool to help physicians identify areas on scans that specialists should focus on, Rogers said.
And through virtual assistants, it’s already helping patients navigate Prisma’s website to find a special provider, one who speaks a certain language, for example, he said.
“It can scan all that content and present options to the consumer,” he said. “And they can go in and schedule an appointment.”
Dr. Patrick Cawley, CEO of MUSC Health, said he’s intrigued by the possibility of AI being used to help diagnose conditions and find treatments on a larger scale.
Recently, he said, a patient was having a stroke at an MUSC community hospital and AI was used to review the CAT scan while it was still on the scanner. It immediately notified a neurosurgeon before the ER doctor had the report from the radiologist, he said.
“The radiologist read (the scan) and confirmed what the AI was seeing,” he said. “But you get it done a heck of a lot quicker and they were already moving that patient to Charleston for a surgical intervention.”
And given the shortage of radiologists, he said, AI use could be even more extensive.
“We do thousands of X-rays a day,” he said, “and if I could have AI prioritize them and give the serious ones to the radiologist first, that would help.”
AI can also be helpful in identifying the best treatments, Cawley said.
“If a patient comes in and is complaining about back pain and you do an assessment, history and physical, then if I could ask AI what’s the most likely diagnosis … it might help guide me in deciding what’s causing that back pain,” he said. “And once you get the diagnosis … then what’s the best therapeutic approach.”
Suggesting a diagnosis that a physician may not have thought about can improve care, Benjamin said.
“Somebody comes in with chest pain, and he’s 45 years old, and you’re going to think about heart attack. You may not think about some rare diagnosis that may cause chest pain,” he said. “Having a machine think about it and display those things … would help reduce missing things.”
It will also be useful for gathering patient information and presenting it to providers, Cawley said.
“These days, so much information is coming at nurses and physicians that it’s almost hard to catch it all,” he said. “If you could ask AI to scan a patient’s electronic medical record and present a summary, that would be helpful. There’s more information in there than any one person can recognize.”
Prisma is also focused on using AI to off-load documentation from physicians, which contributes to burnout, Rogers said.
“Physicians have a lot of documentation they are required to do legally, and it’s time consuming,” he said. “AI … has a tremendous opportunity to reduce the amount of time documenting.”
Nurses, who are also in short supply, have the same burden, Rogers said. Using AI to assist with documentation offers one way to reduce the amount of time they spend on a computer, he said.
And doctors must answer countless messages, he said, noting that AI can draft responses which the doctor can review and authorize, saving still more time.
Physicians now spend more than 50 percent of their time updating electronic health records, according to editorialists writing in the July 27 issue of the Journal of the American Medical Association, who said programs are being piloted around the country to reduce the burden.
“If these efforts are successful, natural language processing could turn unstructured data such as clinicians’ notes into the structured data needed for the EHR as well as for other uses, such as documenting quality metrics or filling in appropriate (medical) codes,” they wrote. “This application of AI would give clinicians more time to spend with patients and on tasks that require human judgment.”
Moreover, Benjamin said, it’s beyond anyone’s ability to be on the cutting edge of all medical knowledge, and AI can help with that.
Indeed, the JAMA editorialists said that the collective body of medical knowledge required to treat a patient doubled every seven years in 1980, but the doubling period was fewer than 75 days in 2010.
“Today, what medical students learn in their first three years would be only 6 percent of known medical information at the time of their graduation,” they wrote. “Their knowledge could still be relevant but might not always be complete, and some of what they were taught will be outdated.”
Franco Cardillo, executive director of digital strategy for MUSC Health, said AI could also facilitate consultations with other physicians to assist with a diagnosis, treatment plan, or referral.
“There are a lot of decision support tools out there for providers,” he said. “And more tools allow us to information-share with other providers.”
Electronic medical records are tremendous databases of patient information, which is generally deidentified, Rogers said. So, if a patient has a rare illness, a doctor might be able to find others with similar symptoms and consult with their physicians about the case and the best treatments.
In the future, he said, providers will also be able to input genomic profiles into these databases and get treatment plans based on a disease and what has worked, improving outcomes.
“We are at the start of some incredible advances in medicine,” he said. “AI can ingest so much more information and history and clinical trial data.”
AI also will be able to review medical charts and complex billing codes to produce accurate bills, Rogers said.
In addition, AI has great promise for speeding medical research, by identifying candidates for clinical trials, for example, Cawley said.
“Today, we hire nurses to go through a patient chart and learn the studies, and they call the patient. It’s very inefficient,” he said. “AI will know the patient. You could feed it data on the studies out there and automatically match up the patients.”
AI can also help ease the growing physician shortage, Benjamin said, by enabling physicians in urban settings to provide oversight to rural communities, to monitor vital signs remotely and allow a higher-trained physician to direct a lower-trained physician without having to travel to a different location, Benjamin said.
Will AI ever perform medical procedures?
Rogers said he thinks that’s coming someday, and Cardillo said he expects that AI use in precision medicine, while not there yet, will evolve over time.
The next generation may be able to do some tasks, but for now, AI enhances the tools clinicians need, said Cawley, adding that everyone in health care will be touched in some way by AI over next three to five years.
But AI is not without its challenges.
In a 2021 report, the World Health Organization said AI “holds great promise for improving the delivery of healthcare and medicine worldwide, but only if ethics and human rights are put at the heart” of implementation.
Concerns, according to WHO, include human control of health systems and medical decisions; patient privacy; requirements for safety, accuracy, efficacy, and quality control; safeguarding transparency and information accessibility; responsibility and accountability; and ensuring inclusiveness, equity, and access.
While it can improve the speed and accuracy of diagnosis and screening for diseases; strengthen health research and drug development, support public health interventions and bridge gaps in access to care in poor and rural communities, AI also has the potential for misuse, the WHO concluded.
Examples include “unethical collection and use of health data by powerful commercial interests of tech companies or governments, biased algorithms, and risks to patient safety, cybersecurity and the environment.”
It called on governments, providers, and designers to enshrine laws and policies that address ethics and human rights concerns “at every stage of the technology’s design, development and deployment.”
Cawley said the medical community needs to exercise care when using AI over issues of ethics, privacy, and bias.
“If a study has been done only … on Caucasians, AI might infer that that study is appropriate for everybody when that may not be the case,” he said.
“The other thing is AI … can be wrong,” he added. “You need to look at it with a learned eye.”
The data being input must be correct, Benjamin said, or it could result in errors that affect care.
“A human being needs to be in the loop on this at some level because technology can make a mistake,” he said. “Technology itself isn’t the issue. It’s the knowledge base of the algorithms. Anything can go wrong.”
“I view AI as a tool that could help guide, but could never be used to 100 percent,” Cawley said. “(It could) … make you think about things you might otherwise not have thought of. Other times, it will push something forward, that this is most likely diagnosis, and you have to say, ‘Not this patient. There’s no way that’s correct.’”
The medical community needs to know the strengths and weaknesses of all the technology and use it in an ethical way, Benjamin said.
“At least now, I’m not sure the technology has an ethical framework, and to the extent technology uses ethics, it’s doing so in a way that’s based on the logic of the person who put it in there,” he said.
“And we want to make sure everyone has equal access to the technology,” he added, “not just the very rich.”
Rogers said there is a potential for the technology to advance faster than it can be regulated and that much needs to be done in terms of oversight, especially with regard to malicious intent.
Moreover, he said, the medical community needs to provide input into any regulation.
“Those are the folks that have to drive the ethical use,” he said.
While some AI has been around for several years, society must be careful with the newer generative AI built on large language models, Cawley said.
And Rogers said that includes handling of private patient information, which must be secured.
AI must be implemented cautiously with checks and balances, including regulation, the experts said.
“We will start to see parameters and laws to make sure we don’t get too ambitious,” Cardillo said. “I do see that currently happening.”
And they all agree that it will never replace providers.
“You’re still depending on the doctor to make the diagnosis,” said Rogers. “AI is an assistant.”
“The provider controls the experience … but tools help,” Cardillo said.
“AI has an enormous opportunity to help us do a whole range of … tasks. But at the end of the day, people have to always be in the process,” Benjamin said.
“Technology should help us do our job. Not do our job for us.”