A prescription leaves the exam room and starts moving electronically. AI wrote up the visit that produced it. At the pharmacy benefit manager, AI decides whether the plan will cover it. Somewhere in between, two machines argue over the price of a drug a patient is waiting on.
Will this AI-dominated future state of healthcare benefit the patient and address rising health spending? The Hopkins Business of Health Initiative assembled three expert panelists to grapple with this question during its September Conversations on the Business of Health webinar.
Frank Harvey, CEO of Surescripts, is a pharmacist by training with 42 years in healthcare, and runs the network that carries prescriptions and clinical data for nearly every physician, pharmacy, and health system in the country. Bob Kocher, MD, is a partner at Venrock and worked on the Affordable Care Act as special assistant to the president for healthcare and economic policy. Caroline Pearson is executive director of the Peterson Center on Healthcare and built the Peterson Health Technology Institute, which evaluates what digital health tools actually deliver. HBHI Director Daniel Polsky, PhD, moderated.
The central lesson from the conversation was that AI’s effect on healthcare spending won’t be determined by the technology alone. AI can lower the cost of doing things, but whether those efficiencies lower total spending or what patients pay depends on the incentives surrounding their use. In fee-for-service, greater productivity can mean more visits, tests, procedures, and coding. Under models that reward lower total cost and better outcomes, the same technology can make prevention, substitution, and more efficient care financially valuable.
A condition we know how to treat, and a prior authorization that clears in 18 seconds
Polsky asked each panelist where AI is already changing their work. Pearson picked a condition medicine has understood for decades.
“There are 120 million people in the country with hypertension, and only one in five of them is in blood pressure control today. That’s a bit remarkable and really quite sad, given that we have very effective and low-cost drugs that are quite good at helping people control their blood pressure.” - Caroline Pearson
Most of those patients see a doctor regularly. Blood pressure just loses out to whatever else is in the visit, and the follow-up slips. In the programs Pearson’s institute has studied, the patient gets a cuff at home and a clinician who does nothing but adjust the medication. Control arrives in three or four months. In primary care it takes about a year. The narrow job is also the automatable one.
“Enter AI. It turns out that those human virtual prescribers can be largely AI-enabled. A lot of medication prescribing and titration for hypertension could be done with AI, and that is the thing that would enable us to get that four out of five people into blood pressure control without dramatically expanding healthcare spending at the same time.” - Caroline Pearson
Kocher, who was chief medical officer at the Medicare Advantage insurer Devoted Health, called blood pressure control the most important public health work AI can help with. He blamed the calendar, not the medicine.
“The reason why it takes a year is that doctors are used to a tempo of seeing you maybe once every three or four months. You could change a medication once every three or four days. So you should be able to get nearly anybody with any disease at the right therapeutic dose in a matter of weeks, and then know if it’s going to work or not, and then keep them there.” - Bob Kocher
Harvey’s example came from the paperwork around that same prescription. A prior authorization can bounce between a pharmacy, a physician’s office, and a pharmacy benefit manager for days, and plenty of patients give up on the drug before it clears. Surescripts now handles some of those requests without anyone placing a call.
“With our touchless prior authorization, as soon as we see the doctor write it, our technology pulls all the data back, presents it to the PBM, the PBM checks it against their requirements for that product, and it comes back to us in 18 seconds. The physician’s office doesn’t have to worry about it, the pharmacy doesn’t have to worry about it, and the PBM doesn’t get thousands or millions of calls a year.” - Frank Harvey
Thirty thousand dollars in three days, and models that attack better than they defend
Harvey’s network handles 30.5 billion transactions a year and reaches 99% of the U.S. population. He told the panel not to lose sight of what the technology costs to run. His example was from that morning.
“I just reviewed one of my data scientists. In three days, he went through $30,000 worth of tokens running a query that I’m sure was very valuable.” - Frank Harvey
Security costs more. Surescripts is one of two healthcare companies in Anthropic’s Project Glasswing, which points a frontier model at a company’s code to hunt for vulnerabilities. Harvey said his information security budget will grow by about half next year. The models on the market, in his view, are lopsided.
“So far, the models that are out there are much better at attacking than protecting. There’s a whole group of the industry that’s now working on protecting against the attackers, so that will improve. But right now we’re better at attacking than protecting data.” - Frank Harvey
Kocher wanted to know why, asking whether the models were being trained “on hacking and not protecting.” The money showed up on the attacking side first, Harvey said, and defense is catching up. Pearson made a related point later in the hour, when the panel turned to how Medicare should pay for AI. Marginal costs fall over time. They do not reach zero.
What happens to the savings on their way to the patient
Those are real gains. Who ends up with them is a separate question, and Kocher put that decision with the payment model.
“AI has the ability to make the cost curve go in either direction, depending on the payment model. The more we get America into value-based care models, the more AI will actually help make those models work better and help people save more money. Those in fee-for-service will use AI to create more access, more appointments, more tests, more procedures, and better revenue cycle management, which will make healthcare more expensive.” - Bob Kocher
Hospitals, health plans, and physician practices are thin-margin businesses. A few points matter to them, and Kocher expects AI to deliver a few points.
“AI will add several points of margin to many low-margin companies. The question is, how does that get translated? It probably gets retained as earnings, not as lower prices.” - Bob Kocher
Pearson has been making the same argument about incentives that predate AI by decades. The market pays better for raising prices than for running a tighter operation.
“Right now it is easier to raise prices in healthcare than to operate more efficiently. And anyone who does operate more efficiently or deliver better care isn’t rewarded with the volume increase that would be the natural market force that would motivate that kind of change.” - Caroline Pearson
Ambient scribes are the test case, since they are the one clinical AI tool health systems have bought at scale. Pearson has been watching the billing.
“Ambient scribes are translating into immediately higher billing, because they’re more fully capturing the exact same clinical encounter that they were billing less for yesterday. They can now bill more for today.” - Caroline Pearson
Scribe vendors also sell the time savings, Kocher said, and the pitch to a health system is two more patients a clinic day. More money per visit, and more visits. He stopped short of calling it fraud. The systems were “giving away stuff for free before,” he said, and may now be billing properly. The trouble is where the billing starts from. The country already pays a lot. Harvey watches the other half of the transaction cross his network, and payers have bought AI too.
“Now we’ve got AI and ambient listening allowing doctors to code up to capture things that they weren’t coding for before. The health plans on the other side have AI that is looking at what’s being submitted, and now you’ve got AI versus AI saying, well, no, you really shouldn’t charge for this. It’ll be really interesting, the battle of the bots.” - Frank Harvey
Pearson put the escalation down to the codes themselves. Fee-for-service pays for time and effort, and AI’s marginal cost for both is close to nothing, which she said is how spending explodes inside such a system. Cheaper visits do not solve it either. A startup had pitched Kocher that morning on $19 visits with an AI doctor.
“Simplistically, that is cheaper than going to a doctor. The hitch is that when people go to the doctor, the doctors do a bunch of stuff. We do tests on people, we find things, then we track them, we initiate medications. We may improve their health, but we also spend more money on them. And usually these aren’t substitutes for a visit.” - Bob Kocher
The specialist inside the prompt, and the work worth taking off a physician’s plate
Kocher does see the arithmetic working in one place, along the line between generalists and specialists. His practices feed a patient’s chart to a model, ask it what a subspecialist would recommend, and show the answer to the primary care doctor before a referral goes out.
“For many straightforward things, it is incredibly predictive of what the specialist would say. Then we show the primary care doctor, here’s what the Hopkins doctor is going to tell you. Would you still like to send them? With a payment model that rewards lower total cost of care, you could probably avoid 30% of referrals to specialty care by having AI augment the generalist.” - Bob Kocher
In much of rural America, there is no endocrinologist to refer to, and audience questions kept coming back to year-long waits for specialty appointments. Not everyone welcomes the idea. Polsky asked whether the Johns Hopkins School of Medicine had invited Kocher to present the paper he wrote about it. They invited people to rebut it, Kocher said. Pearson set a condition of her own: incumbents have to stop bolting AI onto processes that need rebuilding.
“I am not a person who thinks that we are going to get to where we need to be by just handing every human an assistive AI. We have to take things off people’s plates. We have to let things go, we have to embrace autonomy, and we have to do that in a way that is thoughtful and patient-centered, because that’s how we scale affordability and access.” - Caroline Pearson
Higher wages for the doctors who stay
Attendees kept asking what all of this does to physicians’ work and pay. Kocher’s answer went the other way.
“I think wages go up. In value-based care, AI is going to help clinicians save more money, and that creates more shared savings, and that will partly go to the doctors. In fee-for-service, they’ll be able to see more patients. I think we will hire fewer doctors. If there’s job losses, and I think there will be, it’ll be in administrative labor.” - Bob Kocher
One attendee wrote in that this did not sound like the kind of medicine they wanted to practice, or the kind patients want. Harvey said the research at Surescripts keeps landing on the same variable.
“The key to that is having a human in the loop if needed. If it’s just AI, and the patient has no other option but the AI, they can’t escalate it to a human if they have questions. That’s when it falls down. As long as there’s a physician available, we find it’s an excellent experience.” - Frank Harvey
A look into the future
Polsky asked for 10-year forecasts. Pearson went first, and started with where the country is headed now.
“I want to be clear, we’re on the wrong track right now. We are on track to massively increase the cost curve. But my hope, and the whole reason that we keep coming to work every day, is that everybody on this call is going to be serious about taking the steps we need. The technology has the potential to deliver the change that we want to see. Now we have to make the system ready to embrace it.” - Caroline Pearson
Harvey aimed lower. Healthcare spending has been climbing eight to 10% a year.
“I want to be optimistic, but I think at best we’ll see a flattening of that curve. I don’t expect to see a downward trend anytime in my lifetime, but I certainly think we can stop seeing eight, nine, 10% increases a year.” - Frank Harvey
Kocher’s forecast took five words. “I agree with Frank and Caroline.”
The three took different routes to the same conclusion. AI will make clinicians more productive and patients healthier. Whether any of that reaches what Americans pay depends on decisions the healthcare system has so far declined to make about how it pays for care and how long it keeps the people it covers.
Watch the full conversation on YouTube.
The conversation was moderated by Dan Polsky, PhD, Bloomberg Distinguished Professor of Health Economics and Policy at the Johns Hopkins Bloomberg School of Public Health and the Carey Business School, and director of the Hopkins Business of Health Initiative.
“Conversations on the Business of Health” is a series of one-hour webinars hosted by HBHI that engages leaders from business and academia on the cutting edge of improving American healthcare.
Moderated by faculty members and jointly hosted by the Johns Hopkins Bloomberg School of Public Health, the Carey Business School, the School of Nursing, and the School of Medicine, the series is open to all. Learn more here.
