Panel: Human oversight still necessary when using AI in patient care
Click Here to Manage Email Alerts
ATLANTA — An expert panel at SECO 2024 agreed that artificial intelligence will have a role in patient care but will not replace the human component.
AI’s role in day-to-day eye care will be as a diagnostic assistant, predicted panelist Jorge Cuadros, OD, PhD, founder, owner and CEO of EyePACS and director of clinical informatics research at University of California, Berkeley, School of Optometry.
“It’s important to focus on what we’re good at and use AI as a tool,” he said. “We’re going to adapt and change the way we do things and how medical care is provided. That excites me. What scares me is letting AI take its own course without our input.”
Recent studies have used AI to analyze images to predict disease risk or progression. Cuadros stressed the importance of clinicians analyzing the data collected.
“You can’t rely on the automated answer,” he said. “It doesn’t give you enough granularity on the diagnosis; just red light or green light. It doesn’t distinguish between those who are severe and those who are not so severe.”
Good images are required, and false positives are common, he said.
“Let me offer this flow: Take the images in primary care, screen with the AI, but instead of putting it at a low threshold, find the ones who are most severe,” Cuadros said. “Talking about what humans bring to the table, especially optometrists, we can engage with patients to make sure they follow the path to get treatment.
“This is a call to action for organized optometry to come up with a way to share our data,” he continued, “similar to the IRIS Registry in ophthalmology and the UK Biobank and the Veterans Administration.”
When training the AI, the data being used must represent the people who will eventually be treated, Cuadros said.
“Those who are underrepresented skew the performance algorithm when you take it out,” he said. “We see this in clinical trials. Trials are often done in posh clinics in urban areas. How will rural folks be represented in clinical trials if they have to travel 5 hours to get these? We get all these guidelines that don’t appropriately represent the patients being treated.”
Cuadros suggested doing a “reality check.”
“I’m always a proponent of evidence-based medicine,” he said. “But what I think is better than that is reality-based medicine. AIs plus humans is the winning combination.”
Moderator Carl Spear, OD, MBA, senior vice president of eye care at EssilorLuxottica, polled the audience, asking if they feel like they are behind when it comes to AI, and 68% said “yes.”
“We’re all getting a bit of AI fatigue,” he said. “We can’t figure out what to do with it, but we’re hearing about it all the time.
Panelist Walter Mayo, OD, SECO technology director, said in 2015 there were 16 AI programs available to the public; in 2023 there were 36,000.
“AI interacts with you every day already — when you’re using Waze, searching on Amazon,” he said.
Mayo said that between 500 and 1,000 AI tools are being released every month.
Spear again polled the audience, asking how they will embrace AI in their daily life, and 60% said they would use it for some specific tasks. He also asked if they saw AI improving health care outcomes, and 91% said, “yes.”
Panelist Naama Hammel, MD, a clinical research scientist at Google who also is a glaucoma specialist, explained why Google has become involved in health care.
“We think of ourselves as an information company,” she said. “Our mission is to organize the world’s information and make it universally accessible and useful. When it comes to health, we think of information as a determinant of health. When people search for health information, we want to make sure it’s accurate, vetted, relevant and useful.”
Training AI to use fundus photos to predict disease progression, “takes a tremendous amount of resources and data,” she added.
“AI can tell someone’s HbA1c level by looking at a photo of the eye,” Hammel said. “We don’t know how. Now we have to prove if it’s correct.”
She shared a quote from one of her supervisors at Google, chief clinical officer Michael Howell, MD, MPH: “AI won’t replace clinicians, but clinicians who use AI may replace those who do not.”
When polled, 85% of the audience said they would be interested in using an AI-based platform in their clinic to help them diagnose and manage disease.
Mayo noted that AI needs to get to the point where it says, “according to your current treatment regimen, imaging, etc., this patient is going to lose vision, so your treatment plan should change.”
“But we first need to agree on what glaucoma is,” Hammel said. “We don’t have consensus on definition and diagnostic criteria.”
Collapse
Spear C, et al. Keynote session, Putting the eye in AI: The future of artificial intelligence in optometry. Presented at: SECO; Feb. 28-March 3, 2024; Atlanta.
Read more about
Click Here to Manage Email Alerts