How the rise of AI could lead to efficiency issues and lower reimbursement payments
Artificial intelligence (AI) tools in medical imaging are often promoted as a way to streamline care, reduce clinician burden and lower healthcare costs. But in practice, the economics and workflow implications may be more complicated, according to Friederike Keating, MD, professor of medicine at the Larner College of Medicine and director of nuclear cardiology at University of Vermont Health.
Keating, who serves on an AI task force for the American Society of Nuclear Cardiology (ASNC), said she is strongly supportive of AI innovation. However, she cautioned that real-world implementation has revealed challenges that health systems and policymakers need to consider — particularly around workflow efficiency and reimbursement.
She explained that many imaging AI tools function as pre-read systems, flagging potential abnormalities before a physician reviews the study. But these systems are not autonomous and still require full physician oversight.
“The accuracy is not such that you can blindly trust it, nor should you. You are legally responsible for rechecking what the AI wrote,” she said.
Because physicians must verify the AI’s findings, the promised time savings may not always materialize.
“It might discover something that you didn't see, which would be great for the patient," she said. "But in terms of time and effort, it sometimes is more than if you had just read the study yourself, because you have to recheck the AI.”
More data is needed to determine if AI is helpful of wasteful
The Trump administration wants to incorporate more AI in healthcare and make it easier to get FDA approvals because of the promise to increasing workflow efficiency and cost savings. But radiologists and cardiac imagers who use it say the level of accuracy and helpfulness varies between different algorithms.
The U.S. Food and Drug Administration has cleared more than 1,200 clinical algorithms, with about 80% of those in medical imaging, but only a small handful have reimbursement. The two prime examples are in cardiology, with newer category 1 CPT codes that can be reimbursed for computed tomography fractional flow reserve (CT-FFR) and CT coronary plaque assessments. These CPT codes exist because of the efforts of AI companies such as Heartflow and Cleerly, which invested in large clinical studies to show their technology improved patient care, increased efficiency and saved money. That empirical data is often lacking with the majority of AI models, which is the primary reason why they lack reimbursement.
Reimbursements for AI means cuts to physician pay elsewhere
While reimbursements for using AI are helpful for adoption and expanding AI usage, Keating raised concerns about how AI is being incorporated into federal reimbursement models. Some imaging AI vendors that are seeking dedicated CPT payment codes are positioning their software as a reimbursable service rather than a one-time software purchase. She warns that under the Centers for Medicare and Medicaid Services (CMS) budget neutrality, cuts will need to occur somewhere in the budget to accommodate these new payments.
“So what does that do? It reduces the Medicare conversion factor for all the physicians, because now you have one more player in the pie,” Keating said. “And where does that money go? Well, a lot of it goes to the company that developed the AI.”
She noted that there is limited transparency in how pricing is set for these new reimbursable AI services. While market competition may eventually shape pricing, she suggested the financial impact on physician reimbursement deserves closer scrutiny. But reimbursement can also help shape the AI field and boost competition. There are now several companies offering FDA-cleared CT plaque and FFR-CT AI algorithms.
“I’m not anti-AI at all. I'm on an AI task force for ASNC, so we're very pro-AI. But the payment and reimbursement models, they deserve a lot of thinking as well,” she said.
Although a handful of AI applications in cardiology have secured reimbursement, most imaging algorithms have not. Keating said the reimbursement pathway the cardiology vendors set should serve as a model.
“This is a model for them to emulate and it's great for innovation. But, it's not good for physician reimbursement. It would be nice if it came out of a different pot, because it's just in the same pot,” Keating explained.
Dave Fornell has covered healthcare for more than 17 years, with a focus in cardiology and radiology. Fornell is a 5-time winner of a Jesse H. Neal Award, the most prestigious editorial honors in the field of specialized journalism. The wins included best technical content, best use of social media and best COVID-19 coverage. Fornell was also a three-time Neal finalist for best range of work by a single author. He produces more than 100 editorial videos each year, most of them interviews with key opinion leaders in medicine. He also writes technical articles, covers key trends, conducts video hospital site visits, and is very involved with social media. E-mail: [email protected]