On Monday, August 16, the Chicago-based HIMSS (Healthcare Information & Management Systems Society) presented a webinar entitled “The Economics of Artificial Intelligence.” It was streamed live and sponsored by the New York City-based Ambra Health. The speakers were Greg Nicola, M.D., a practicing radiologist and vice president of the 50-radiologist Hackensack Radiology Group in Hackensack, N.J., and also chief medical officer of the New York City-based NeuTigers AI platform development company, and chair of the Commission on Economics of the American College of Radiology (OCR); Lauren Parks Golding, M.D., a practicing radiologist and CEO of the 49-radiologist, Winston Salem, N.C.-based Triad Radiology Associates; and Morris Panner, the CEO of Ambra Health.
The webinar focused on questions around how the application of artificial intelligence (AI) might impact radiology practice in the future. As the speakers noted, there are many potential paths forward, as algorithms could be developed to help radiologists diagnose specific findings in their diagnostic studies; could be used to help radiologists assess quality metrics in radiological practice; and also potentially to assess quality outcomes for reimbursement purposes.
Panner noted that there are three broad areas: “accepted procedures and services; supplemental tracking/performance measures; and emerging technologies.” But each area involves complications, and the panelists agreed that advances in the development and leveraging of AI algorithms will be complicated, with challenges.
Early on, Dr. Nicola brought up an analogy, noting that the invention of the European printing press in 1439 by Johannes Gutenberg created immense disruption, in unexpected ways. Yes, some traditional scribes lost their jobs, as the need for people to handwrite books no longer existed, except in rare cases. On the other hand, many, many former scribes ended up in the printing industry, where new jobs were created, with their creation somewhat offsetting job loss in traditional scribe roles. Nicola said that healthcare leaders, including practicing radiologists and radiology leaders, need to consider that analogy, as he does not believe that there will be any mass loss of jobs for practicing radiologists; instead, he said, radiology practice will inevitably change over time, as a variety of different AI algorithms are adopted and implemented.
Meanwhile, Dr. Golding noted that the RVU-based (relative value unit-based) reimbursement system is ultimately based on a finite total of funding. And, with regard to the three elements involved in every RVU-based payment, there are the technical component (which encompasses the cost of operations, of running a medical office or clinic, etc.); the professional component, or “work RVU”—which accounts for the amount of time spent in delivering patient care and the intensity of readings; and a malpractice element. She stated that she believes it will be a long time before AI is successfully leveraged in a way that would lead to changes in radiologist reimbursement, because of the complexity involved, along multiple dimensions.
Nicola noted that “Imaging is not an easy transaction. Also, think about an AI algorithm and what it does to the value of radiology. You can think about AI algorithms on both the technical and professional sides. AI might help physicians get through a reading faster,” but it will be a long time before AI algorithms are robustly used in radiology.
Said Golding, “What we’ve really seen is on the RVU side, detection algorithms for very specific elements. The impact on the total time isn’t much. It doesn’t really reduce your time that much. So it doesn’t have a lot of downward pressure on the RVU side.” In other words, the leveraging of AI algorithms will take a long time to be incorporated into radiology practice. Still, he said, “At some point, we will probably see AI algorithms that are much more disruptive to radiology operations; we haven’t seen anything yet at all.”
Obstacles seen to clinical operations adoption of AI
Then again, Nicola said, “When we do” see AI algorithms begin to be incorporated into clinical practice, “and I suspect we will, that AI algorithms will be packages of algorithms that remove a significant amount of time or liability from the radiologist, and then we’ll start seeing changes to the value equation, where the reimbursement is lower, and the value to the patient and payer are higher. When the printing press happened,” he noted, “we needed far more printing press operators than we needed scribes. So the scribes didn’t all lose their jobs. Will the same thing happen in medical imaging? Some are saying we’re already over-using imaging. You could imagine a world where AI is significantly helping to manage patients in a significant way—and where we start screening for additional types of cancer. Right now,” he said, “we screen only for breast, colon and lung cancer. We don’t screen for renal or pancreatic cancer, because the value isn’t there.”
Nicola went on to ask, “Is there a market where we might screen for other types of cancers? To understand what might happen to the workforce will take some time. Another element, per disruption, is, we’ve looked at the value equation and what happened with the printing press: technology always democratizes a skill set and makes it eligible for others to do.”
Still, he noted, every new technology introduces a variety of disruptions. “Per the modern-day equivalents of the printing press, we now have copy machines in our offices. For easy-to-do photocopies, we simply buy a photocopy machine and do the copying ourselves. We expect AI to eventually democratize the practice for radiology, for sure. At the same time,” he quickly added, “ we still have printing press operators 600 years later. I fully expect radiology to be pushed into the new world by technology,” he said. “It’s likely there will be a radiologist pushing processes through.”
Panner said to both Dr. Nicola and Dr. Golding, “I love the way you think about different technology paradigms. In the public policy realm, it feels like we’re a little stuck on some old paradigms. At 30,000 feet, is AI a way to change the paradigm, or will we be in the pre-Gutenberg stage forever?”
“I think that if something is completely disruptive, we’ll see policy change faster,” Nicola said. “But I don’t see policy budging at all for the moment, with people pulling their hair out, per radiology policy and AI.”
Panner then said, “Dr. Golding, what would I say to a policymaker about what they need to know about AI and policy and payment issues? Is AI going to end up being ubiquitous like CAD, for instance?” he asked, referring to computer-aided design.
“I do see a lot of people bringing up CAD,” Golding said. “I use CAD in my day-to-day life. It’s sometimes helpful, sometimes not. It’s a completely different category. To me, CAD is a feature on a PACS [picture archiving and communications system]. AI will do something that humans either can’t do or will do it infinitely better than humans. But it absolutely can’t increase the cost of HC. If AI algorithms will detect certain nodules but will cost more, they’ll never fly. When AI is used for things like figuring out who’s at risk for cardiac disease, then it will be implemented. It will have to be useful and not cost more.”
What’s more, Panner noted, “AI is accessible. If you have good ideas and see places in the workflow where you could make a difference, there are a lot of new ideas that will come up. How do you see physicians getting involved in AI development? We have a saying in medicine that whoever’s closest to a problem should get involved in solving it.”
“Impactful AI absolutely will require significant physician guidance and input,” Nicola said. And, by “impactful,” he said, it means the leveraging of AI that can actually help to change radiological outcomes. “And we’re just not seeing that. But when we do, I guarantee that radiologists will have been involved in the development. It takes significant clinical research. Real, impactful AI is several years away, and it takes multi-modality data, and data analysis, and lots of physicians will be involved, and that’s where we’ll see the market change.
Payment paradigms—will they shift?
“We talk about some of the challenges in paying for the current AI algorithms and the duplicative-ness. That’s been one of the challenges,” Golding said. “We hear from the vendors that AI can do it faster and better. But right now, Medicare doesn’t pay primary care physicians for seeing a patient faster. So the CPT system is a dead end for a lot of AI algorithms in their current form. That could change, but it hasn’t right now.”
“A lot of physicians are upset at the current payment paradigm, but I think that some things that could be paid for, it would be a mistake to pay for,” Nicola added.
With regard to the policy and payment discussion, Panner said, “I love the focus on outcomes. Somebody asked me—they said, culture is everything, what’s the culture of your company? I said, we aspire to be a part of the care team. To understand that everything is about the outcome at the end of the day, and somebody is asking for our help, and that’s the most impactful interaction that you can have. I do want to talk about a new paradigm that has promise but makes some people a little more nervous, and that’s risk-based payment. What do you think? It feels like a little bit of a new frontier.”
“You’re exactly right; value-based care is a completely different playing field” from making changes in the fee-for-service world, Golding said. “And none of these applications or vendor submissions; all that matters is whether you provide value to the payers and purchasers and consumers of care. AI can help you risk-stratify and help you with the actuarial aspects of your population, and help you to make some decisions. But we have to progress both the payment mechanisms and the AI tools, and eventually, that’s where healthcare’s going.”
“Dr. Golding and I published a paper recently saying that that’s exactly where healthcare will have to go,” Nicola said. “And it’s the most logical way for this technology to go. Now, the venture capitalists don’t like that; they want to create cash registers; but that’s where we’ll see it take off; it’s vital for things to be managed locally.”
Panner asked Golding and Nicola for their final thoughts. “This is a fluid topic, though I’ve been pretty firm with this opinion for about two years,” Nicola said. “But speaking to all my clinical colleagues, I’m part of a company that’s outside of radiology, and many people are alarmed by things. But disruption is actually wonderful, and the only way to survive is to innovate, to be part of the disruption. I don’t think any doctor should be afraid of it; we’ve got to dive in and guide it. And that’s true in any industry.”
“I’ll echo Greg’s comments that this is a fluid topic,” Golding said. “Every few months, there’s some new government rule or regulation that governs how all this plays out. I do think we’ll continue to see the challenges in the FFS system, so until there’s significant HC reform, there will still be significant barriers to reimbursement of AI algorithms. But there are significant opportunities beyond detection algorithms to show value. And products designed without radiologist input won’t be helpful to us or anyone else. So you should get involved and stay involved, and the process will evolve much better with us side by side.”