SAN FRANCISCO — When the US Food and Drug Administration authorized marketing of the IDx-DR, the decision marked the first time regulators have ever allowed a patient to be diagnosed by a computer rather than a person.
"All of medicine is looking at us to see how we do this," said Michael Abramoff, MD, PhD, chief executive officer of IDx, the company that created the AI system for diagnosing diabetic retinopathy.
It is one thing to recognize that AI can diagnose an eye disease better than some ophthalmologists, Abramoff told Medscape Medical News. But it is another to figure out where that computer belongs in the healthcare system and to sort out workflow, reimbursement, and legal liability.
The IDx-DR software uses an AI algorithm to analyze images of the eye taken with the Topcon NW400 retinal camera. The operator uploads digital images to a cloud server. If the images are of adequate quality, the software provides the doctor with one of two options: "more-than-mild diabetic retinopathy detected: refer to an eye care professional"; or "negative for more-than-mild diabetic retinopathy; rescreen in 12 months."
A separate AI program guides the operator to correctly photograph the retina, and only a high-school education is required, Abramoff said.
But what happens when mistakes are made?
Abramoff said the legal liability is on IDx, not the operator of the system.
"If the physician overrules the AI, the liability should be with the physician," he added. "If the physician follows the recommendation, the liability should be with us."
In a pivotal trial, IDx-DR correctly identified the presence of more-than-mild diabetic retinopathy 87.4% of the time, and correctly identified patients without diabetic retinopathy 89.5% of the time (NPJ Digit Med. 2018;1:39).
In previous studies, board-certified ophthalmologists using indirect ophthalmoscopy correctly identified diabetic retinopathy only 33% to 73% of the time (Diabetes Care. 1993;16:889-895, Am J Ophthalmol. 2002;134:204-213, JAMA Ophthalmol. 2013;131:351-357).
One drawback of IDx-DR screening is that it doesn't catch other eye diseases, pointed out Michael Boland, MD, PhD, from Johns Hopkins University in Baltimore, who will chair a symposium on AI at the upcoming American Academy of Ophthalmology 2019 Annual Meeting.
However, IDx and others in the industry are working on AI systems for other diagnoses.
Glaucoma has proven particularly challenging because the diagnosis depends less on images than other eye conditions and because experts don't agree on a definition of the disease.
"Because it's multimodal and takes a long time to manifest, studies and databases or registries in glaucoma are hard to come by. We're trying to collect datasets that are big enough," explained Robert Chang, MD, from Stanford University in Palo Alto, California, who will present an analysis of AI at the meeting.
"A lot of patients don't undergo their annual recommended screening diabetic eye exam," he said.
Finding High-risk Patients
"If we can find people who are at high risk but who don't have access to care or never see a doctor, then screening them with an AI algorithm to discover disease at an earlier treatable stage can be quite valuable to the patient and society," Chang explained.
Although some ophthalmologists are purchasing the system, Abramoff sees the real market for this technology as endocrinology clinics, internal medicine clinics, diabetes education centers, diagnostic labs, community health clinics, and diabetes research groups.
Primary care clinics and endocrinologists, in particular, can use the system to screen a much larger proportion of patients at risk for diabetic retinopathy than is currently being screened. They can then refer patients identified as having the disease to retina specialists for further evaluation and potential treatment.
"I think it may actually help get more people who need treatment into ophthalmology offices," said Boland.
Johns Hopkins is already using the IDx-DR to screen pediatric patients. "In terms of the workflow, it's been great," he said. "You don't need to be seen by two doctors, just the endocrinologist."
Payers and providers are figuring out how the system works in their business models. The camera costs about $15,000 to purchase outright, but financing options are available, Abramoff said.
IDx charges $34 for each image it analyzes, he reported, but health systems can collect for the exam and then pay IDx the diagnostic fee.
In June, the company announced that the Current Procedural Terminology (CPT) panel at the American Medical Association had accepted a new category 1 CPT code for automated point-of-care retinal imaging that can be used to request reimbursement for using the device.
IDx customers are currently using a temporary code; however, the new code, which goes into effect January 2021, "truly captures the value of the autonomous AI making the diagnosis," said Abramoff.
No decision on whether the Centers for Medicare and Medicaid Services will reimburse for the system has been announced yet, he added.
Abramoff is chief executive officer of IDx. Boland reports a financial relationship with Carl Zeiss Meditec. Chang has disclosed no relevant financial relationships.
American Academy of Ophthalmology (AAO) 2019 Annual Meeting.
Medscape Medical News © 2019
Cite this: AI Sparks Legal Debate Over Machine-Generated Mistakes - Medscape - Oct 09, 2019.