Q&A With HHS's Health IT Czar

 Thomas Keane, MD, MBA, began his tenure as national coordinator for health information technology at HHS in June 2025. He sat down at his office with Washington Editor Joyce Frieden to discuss how he hopes his office will help ease the prior authorization paperwork burden and what artificial intelligence (AI) may portend for physicians -- especially radiologists.

The interview, which was conducted with a public relations person present, has been edited for length and clarity.

Your office recently put out a request for information related to AI in healthcare. What are you hoping to learn from it?

Keane: AI is so new, and it's progressing so quickly that we're trying to, across HHS, get AI regulation correct. We don't want to put too many regulations in place that can strangle the field and damage the progression of this technology that holds for such promise. On the other hand, we understand that there has to be some regulation in order to assure trust and reliability and to make sure that patients are safe. We wanted to tackle a particular problem, namely the lack of adoption of existing AI.

There are currently 1,300 FDA-approved AI medical devices. But if you actually look at how often these are used or how widely deployed they are, the adoption has actually been very low. There was an article in the New England Journal of Medicine about 2 years ago that looked at billing [records] as a means of seeing how widely adopted these were. And it turns out that 75% of all AI apps that were billed for involved a single application called HeartFlow. So we wanted to know, why are existing AI apps really not being adopted? What's missing?

We have a couple of theories. Probably the biggest one is that they don't offer a return on investment for the people using them. They say, "This is a nice tool, but it really doesn't make my life easier. It doesn't get me extra pay. It doesn't necessarily make me more accurate in my work." Another barrier to adoption could be that it's not being reimbursed properly. But we're only speculating. There's also the question of liability. How much do liability considerations impact the way physicians use it, or impact the sort of AI that's being produced?

So I wanted to ask, why are some doctors using it, and some doctors not? That's really the purpose of the request for information, and we'd like to see if there's anything government can do to help.

What about clinician concerns that AI might eventually replace them? As a practicing interventional radiologist, you must certainly have come across this. Could that also be a barrier?

Keane: I have been using AI in mammography since 2005; you could actually use it to overread mammograms for you. It was called computer-aided diagnosis, and you could press a button on the mammo machine, and it would put little dots, circles, and stars on the mammogram to say, "Hey, the computer thinks this might be a suspicious finding," and you would re-review those. Typically, after you read the mammogram, you'd press the button and you'd see if there was anything that you may have missed. That's been around for 20 years, but I think a lot of radiologists felt the systems were only mildly helpful or not greatly helpful. They adopted them because we got payment for them.

That was 20 years ago, and now I routinely use, when I'm doing diagnostic imaging, a program called Viz.ai that helps detect stroke. If I'm going to be honest -- and I don't want to get sued for saying this -- but it works some of the time, and it's overall helpful. It can be helpful to the emergency room doctors who are trying to triage patients, but it's oftentimes wrong. On balance, though, it's helpful because I can catch it when it's wrong, and it's right enough of the time that it helps us triage patients faster.

My personal view is that within 5 to 10 years, the machines will probably be as good as the diagnostic imagers. I actually think they'll be as good as human radiologists. And then the question becomes, what is the role of the human radiologist? Do we use the AI to triage cases? Do we use the AI to help make us more efficient, or does the AI replace us? I don't really know, but I think at some point, particularly in the field of diagnostic imaging, it could possibly replace us. It's a possibility.

As part of the 21st Century Cures Act, Congress barred electronic health record (EHR) vendors from engaging in "information blocking" and required that patients be given full access to their medical records. How have those requirements been working out?

Keane: Well, let's be honest. Up until now, they haven't been effective. The rules themselves are powerful. The reason they haven't been effective is we haven't prioritized enforcement.

In 1996, HIPAA [the Health Insurance Portability and Accountability Act] gave patients the individual right of access to their medical records, and then the Cures Act said they should be able to get access to these electronically, without special effort and without fees. And it gave the responsibility for enforcement to us and to the Office of Inspector General, and it created pretty stiff penalties -- it's $1 million dollars per instance.

Up until now, the department has not chosen to enforce the rules, but [HHS] Secretary [Robert F.] Kennedy [Jr.] announced about 6 months ago that he was going to prioritize enforcement of these rules. He wants information to flow. His vision is every patient having their data in their hand, on their phone, or in a form that they can use to empower them to take control of their own health. They have hired attorneys, they have allocated a budget, and they are now investigating the 1,500 information-blocking complaints that we've gotten through our portal.

Speaking of patient records, one solution that has long been on the table to simplify patient records has been a universal patient identifier that would link to a single medical record that all clinicians could access, as is done in many other countries. Current federal law prohibits such an identifier, but do you think it would solve the problem?

Keane: I'm a person who plays with what he is dealt. Wishing doesn't give birth to horses, and there's a statutory prohibition against a universal patient identifier. So I have to make information exchange happen in the world we live in, not in some other world. That being said, we're a much bigger country than many of the European countries that have a single medical record, and we have a federated health system -- every doctor or nurse practitioner or healthcare provider creates their own medical record for a patient, and these belong to them by law. There's really no way to create a universal medical record in our system where all of these records dovetail together.

Now with AI, you can pull all of this data, you can query all of it, bring it into a single repository and turn it into a comprehensive medical record, and that's what health information exchange allows. But I don't think there's any way in our system to have a single medical record for the patient that everybody works off of.

The question about the universal patient identifier, people have held this out as a way to know who each individual patient is. But the reality is, it would not be a magic bullet. We currently have Social Security numbers, which are supposed to be unique and protected and are used for government benefits determination. But the Social Security number, by itself, is not even a particularly powerful identifier for patients, and the question is, why? Some places don't collect it, and some places mis-enter it. I've mis-entered my own Social Security number. It typically requires a lot of information about a patient -- about eight demographic variables -- to uniquely identify that patient, and having a single number sounds like it would be a magic bullet, but it probably wouldn't be.

There has been a lot of talk on Capitol Hill recently about trying to improve the prior authorization process. What role does your office have in that?

Keane: On July 31, we implemented our HTI-4 regulation, which addresses prior authorization and real-time pharmacy benefit determination. We regulate the electronic health records, and we mandated that the electronic health records have to implement standards to allow for real-time electronic prior authorization. Insurance companies tell us about 80% of prior authorizations can be adjudicated in real time, while for about 20% it's a little bit more complicated.

We've told EHR [vendors] that by the end of 2027 they have to adopt these standards that will allow them to communicate with the insurance companies in real time, so that when an MRI is ordered for you, instead of having to go through a long prior authorization process, the data to support the MRI is pulled off of the electronic health record, piped to the insurance company, and the insurance company computer makes a determination and sends it back.

In that same rule, we also required them to adopt real-time pharmacy benefit functionality, which allows patients and providers to see at the point of care how much a drug that the [provider] wants to use will cost the patient, and also what therapeutic alternatives would cost. So if one drug is going to cost the patient $100 and another drug is going to cost the patient $25 and they're therapeutically equivalent, they can choose the $25 drug. We found that in instances where physicians get presented with this information and they make a switch, it saves about $70 on average for routine prescriptions, around $130 for diabetes medicines, and around $800 for specialty medicines.

What other initiatives is your office pursuing?

Keane: The 2010 High Tech Act mandated the adoption of electronic health records and provided tens of billions of dollars to support the adoption of electronic health records. But behavioral health providers were not involved in that, and so they've been left out for the most part of electronic health records and the information exchange that goes with them. So we just announced, along with our colleagues over at SAMHSA, the Substance Abuse and Mental Health Services Administration, that we are giving $5 million in grants to nine different states to allow for data interchange between 45 different providers. We essentially specify data elements that are necessary for the care of behavioral health and substance use disorder patients, and we want them to be exchanged between providers and states so that the care of the patient can become more efficient.

For example, the grant that we gave to the state of Florida will support data interchange between behavioral health providers, vocational service providers, and housing providers. Many people with behavioral health and substance use disorder challenges require housing, require employment, and as part of their recovery, getting housing and getting employment can be essential, and the ability to exchange data with social services providers and healthcare providers will allow for better care of these patients.

https://www.medpagetoday.com/practicemanagement/informationtechnology/119838