Health systems have spent the past few years building the case for AI one use case at a time. Less clear is what AI will do to the nation’s healthcare bill.
The question came up during the opening keynote at Becker’s 11th Annual IT + Revenue Cycle Conference in Chicago, where Bobbie Byrne, MD, executive vice president and CIO of Charlotte, N.C.-based Advocate Health, gave an answer that rarely shows up in an AI vendor’s pitch deck.
“I do think … AI in general is going to raise the cost of healthcare, not for our organization, but in total,” Dr. Byrne said at the September event. “We’ve talked about how we’re using AI to improve care. Well, care is what costs.”
The research cuts both ways. A 2023 National Bureau of Economic Research working paper estimated that wider AI adoption could save 5% to 10% of U.S. healthcare spending. A September Blue Cross Blue Shield Association white paper pegged the cost of rising hospital coding intensity at about $942 million over two years and said it coincided with the spread of AI-assisted billing tools, findings the American Hospital Association challenged in October. Health system CIOs are also working to rein in usage-based AI token costs that grow as adoption does.
Philadelphia-based Penn Medicine has budgeted for $105 million in benefits from AI by fiscal 2028. But Mitchell Schnall, MD, PhD, Penn’s senior vice president for data and technology solutions, who shared the keynote stage with Dr. Byrne, said much of that value comes from better performance rather than savings, starting with the revenue cycle. AI has made Penn’s pharmaceutical prior authorization work 30% more efficient, for example, allowing it to handle more prior authorizations and capture more pharmacy revenue.
That revenue, he said, helps pay for AI tools that make life easier for patients and clinicians but don’t directly bring in dollars.
“I think you’re seeing the first use cases raising the cost of care, and then potentially you’re seeing investment in use cases downstream that can potentially reduce care going forward, reduce the cost,” Dr. Schnall said.
Nabile Safdar, MD, chief AI officer of Atlanta-based Emory Healthcare, told Becker’s he expects AI to be roughly cost-neutral or to raise total U.S. spending slightly over the next five years, with stronger returns further out. Licenses, infrastructure and training costs arrive on Day 1, he said, while savings depend on changing how work gets done.
“Electric motors are a good parallel,” Dr. Safdar said. “Factories that simply swapped them in for steam engines saw little gain for decades, and productivity jumped only when they redesigned the factory floors in the 1920s. Most health systems are still in the swap-in phase.”
Jane Moran, chief information and digital officer of Somerville, Mass.-based Mass General Brigham, also distinguished between making care cheaper to deliver and lowering total spending. If AI helps find disease earlier, close care gaps and reach patients who faced access barriers, she said, health systems may appropriately deliver more care.
“The opportunity is to use AI to change the trajectory of healthcare costs by delivering better care more efficiently, not simply less care,” Ms. Moran told Becker’s.
Even savings inside a single organization can be slow to materialize. Jeff Gautney, senior vice president and CIO of Chicago-based Rush University System for Health, said during a Nov. 2 panel at Becker’s 14th Annual CEO + CFO Roundtable in Chicago that IT carries the full cost of AI tools, while the departments expected to reduce or redeploy staff get them for free.
“You have to put in place this ability to track and make sure that you follow that value chain all the way through to its end,” he said.
The revenue cycle is where the cost debate gets most heated, since revenue one organization captures shows up as spending for another. Rush’s quickest AI payback has come from a prebill coding review tool that has found about $5 million in gross revenue since going live in March, or roughly $3.3 million after paying the vendor, Mr. Gautney said.
The tool has also added work for Rush’s revenue cycle staff, he said, because it triggers a second review of charts that have already been coded. “Now, the payback is worth it, but it’s an unintended consequence that we originally didn’t plan for,” Mr. Gautney said.
At Boston Medical Center Health System, evaluation and management coding at levels 4 and 5 rose from about 61% to 71% after the safety-net system rolled out an ambient AI documentation tool, according to Tinu Tadese, MD, who until recently was its vice president and chief medical information officer. The organization has not yet tied that shift to revenue, she said at the CEO + CFO Roundtable session.
Dr. Safdar said payers and providers are both partly right, since much of what documentation tools capture is care that was delivered but never recorded.
“But when providers buy AI to document and payers buy AI to deny, we end up with two algorithms arguing over the same claim, and patients get nothing from that fight,” he said.
Ms. Moran said it’s important to separate more accurate documentation from inappropriate increases in coding intensity, with accuracy, transparency and clinical oversight remaining essential.
Judging whether the added spending pays off may take new yardsticks. Dr. Byrne suggested an adjustment such as quality-adjusted life years so the industry can make apples-to-apples comparisons.
“Measure the workflow, not the tool,” Dr. Safdar said. “As AI gets embedded in the EHR and other tools we already own, isolating its precise ROI will get harder. It’s still worth estimating with well-founded assumptions, or we can’t tell good investments from expensive habits.”
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.