← Becker’s Healthcare Podcast
Becker’s Healthcare Podcast
Trust by Design: How Voluntary Accreditation is Shaping Responsible Healthcare AI
00:18:18
Loading summary
Transcript17 lines
- [00:00]
A
Hi, everyone. This is Lucas Voss with Becker's Healthcare. Thanks so much for tuning in to the Becker's Healthcare podcast series. It's fantastic to have you. Today we're talking about how voluntary accreditation is shaping responsible healthcare AI. And I'm so excited to welcome Dr. Shawn Griffin. He's the CEO of URAC. His career spans clinical practice, medical group leadership and innovation, and care delivery. After years of practice, Dr. Griffin stepped into that CEO role to broaden his impact, bringing his experience beyond the exam room to help advance quality, accountability and innovation across across the country and the world. At New York, he focuses on strengthening and evolving accreditation programs, influencing policy in and building a culture that supports both high standards and meaningful growth for his team. Dr. Griffin, I'm so excited to have you. Welcome to the podcast.
- [00:48]
B
Thank you very much, Lucas. I'm looking forward to this discussion.
- [00:50]
A
Yes, we have lots to talk about. AI is being adopted very rapidly across many different disciplines. Clinical, administrative, patient facing workflows. But a lot of leaders, even though it's being adopted, are saying they're unsure how to evaluate risk, transparency, accountability, a lot of different factors. What gap did you see in the market and how does the healthcare AI accreditation address that specific need?
- [01:18]
B
Well, I'm fortunate in that I was the Chief Medical Information Officer for a couple decades back when I was in clinical practice and working as part of health systems. And when you're a cmio, you get used to technology rolling in and you have rollout plans and training plans and all those sort of things. And AI really is similar to some of those things, but it's different also. And so what we were looking at is we're very involved in the policy level in Washington, D.C. and in guiding policy. You know, we've been around for three and a half decades. We do all this work in quality. And what I was seeing was I was seeing this come in into healthcare. And actually I spoke on AI in healthcare back in 2018 to a bunch of chief medical information officers who were early on in their careers. And back in 2018, it was mainly being done with PowerPoint slides because there really wasn't much meat underneath those PowerPoint slides. It was being used for funding and talk about, you know, sort of robot maids and flying cars and AI and healthcare. But over the past few years, what we saw was we saw that there was more of this rolling in and it was being, being sold. And if you went to a big technology and healthcare sort of meeting, the vendors were switching over from being EMR vendors to being AI vendors. And AI Wasn't everything. It was the sprinkles on top of every piece of technology. They're talking more and more about AI. And what I was hearing from my friends who are still in practice in the organizations that we work was, we need some guardrails on this, that this is moving faster than regulation is keeping up, and we can't tell the good guys from the bad guys. And accreditation has a history of helping you tell the good guys from the bad guys. I describe accreditation as like being a quality auditor. We step in and we say, we're going to check behind the scenes and we're going to tell you that they're doing things right. And not just the, oh, they say they're doing it right, but we're actually going to make sure that they're doing it right. And so we, we went to our board, and our board is a. A mix of stakeholders, everything from doctors to nurses to health plans to pharmacies to all these different groups. And we said, we're very concerned about this technology outrunning the rules and putting both patients and providers in a dangerous place where they don't know what to trust. They're not trained on it, they don't understand the risk. And honestly, we saw some bad actors who were coming to the table with tools that weren't fully baked and weren't trustworthy. And so we said, we think this needs some independent oversight. We looked at what was going on, and honestly, there's 200 best practices out there and best practice organizations and recommendations and all these sort of things. And oddly enough, they don't all agree on anything. But we said, is best practices. Saying you're following best practices is like saying you're gonna make a New Year's resolution. It's like, we're gonna do our best. And it's like in healthcare, just doing our best has never been good enough. So we said, we're willing to take our reputation and our integrity that built over decades and step into this space. So we built a program, we got stakeholders, an open call for stakeholders to come together. And we had some great minds around the table contributing to this. We had academic medical centers, we had practicing physicians, we had lawyers, we had ethicists, we had pharmaceutical companies. And when we got them together, we were planning on doing a program around users of AI, so an accreditation program to protect the users of AI. And when we got that group together, they said, well, yes, we need that. We desperately need that. There are no regulations right now guiding this on a national basis. But we also need one for the developers, because we're looking at the developers and we can't tell whether they know what they're doing or not before they start trying to sell to us. And so they were sort of looking for, can somebody check behind the scenes before it even shows up at our door to make sure that the vendors are doing it right. So that's what we stepped into. We developed that program. We rolled it out last year. It's the first AI and healthcare accreditation program. It's designed one module for users, one module for developers, and it's been very positively responded to. It took off faster than any program we've launched in about the seven years I've been here. And we've got good feedback on it. We've got regulators who want to know more about it because they're looking at how they can approach it. We've got big healthcare organizations who are looking at it because they're like, we're tired of just saying we're following best practices. We think this needs to be verified and trustworthy. And we're hearing from users who are like, you know, all of a sudden I've got this AI tool which has been stuffed into my emr. I had no training on it. I don't know about the data privacy. We hear all these horror stories about risk and bias and hallucinations and those sort of things. And so we just, we wanted to put some guardrails up, and that's what our program is.
- [06:03]
A
How do those standards then. If you want to talk a little bit about some of those core pillars. Right. Of the program in itself. And then you, you just mentioned some of those standards involve just an individual looking at o, how am I actually using AI on a daily basis? But how do those standards then translate into those safeguards in practice, both clinically and operationally?
- [06:22]
B
Okay, well, in any program that we do, and we have over 40 different programs that we do in healthcare, there are some things we call the foundational focus areas. Foundational focus areas are sort of those core quality structures that any organization should have if it's going to be in healthcare. It's things like regulatory compliance, regulatory monitoring. When the laws change, are you going to catch those things? It's about statements of work, contracting, data protection, privacy protection. We actually. Scalability. You know, I'd hate to deploy a tool into my system that couldn't keep up with growth. You know, I don't want to have a server crash and all of a sudden my doctors can't get recommendations for what they should be doing those sort of things. So we have those sort of foundational focus areas. Are the people who are doing it qualified to do it? Things that just make common sense, but we actually check those things because we say, you can't assume those things are being checked. A code of ethical conduct for both the developers and the users. Have they actually looked at this from an ethical standpoint? Have they looked at it from a technical standpoint? Have they looked at it from a clinical standpoint? Sadly enough, there are some AI tools which are being rolled out which really don't have clinicians involved in their development. So I've heard of physicians who say, I just got given this new tool to use for patient email responses, and it doesn't have medical dictionary to make sure that the words are the right words that are supposed to be used. It has a technical dictionary, but not a medical dictionary. And that's just a case of, you know, where things are may work great over in finance, but when you're going to bring them into the clinical world, we actually have a higher standard that we expect people to follow. So we have all those foundational focus areas which go into either program, whether a user or a developer. And be honest, there's some organizations that are doing both. I mean, I've heard of big hospital systems where they say they have over a thousand AI tools that they're deploying. And if we said they have a thousand drugs that they're developing and using on patients, we'd go, maybe somebody outside the organization needs to be looking at these things. But in AI, you can have six guys in a garage write a bunch of code, and all of a sudden you have a tool that'll be showing up at your next vendor fair.
- [08:30]
A
Yeah, absolutely. And again, we know that part of the issue, as you've mentioned before, there's not a lot of regulation around it. No, the garage guys can develop their tool, bring it to market, and then effectively sell it to healthcare. A part of the process right now. How does then voluntary accreditation differ from regulation in the context of healthcare AI? And why is that distinction so important, especially now when we have the guys in the garage developing these AI tools?
- [09:01]
B
Yeah, so. So AI under the previous administration, AI was. Had some constraints on it. There were. There were some regulatory frameworks that were being thrown around at the federal level. And when we looked at that, we're like, okay, there are some guardrails that are being appl. And with. With the most recent change in the White House, they said, we're concerned about holding AI back too Much. And so they. They suspended some of those executive orders and some of those rules. And whether those rules were great rules or terrible rules, they suspended those rules. And so now you've got sort of this vacuum at the federal level as to what are the rules going to be. You have states where they have. You're developing a patchwork of states where maybe Colorado is concerned about this and California is concerned about this. And my old home state of Iowa may not have any rules whatsoever, but if you're a developer, how do you sell into those markets when not every facility respects state lines? When I used to practice, I was in St. Joseph, Missouri, and in St. Joe, I had patients who were from Kansas, from Missouri, from Iowa, from Nebraska. And it's really strange to think, well, wait a minute, they're from Iowa, so I can use this email tool, but their cousin is from Missouri, and so for their chemotherapy, I can't use this tool. And that's not how medicine has tended to work. So we looked at the regulations. I mean, you're still supposed to follow HIPAA. HIPAA's been around for a while, but the thing about AI is that AI changes. And so there are some particulars that can change when you're talking about an AI deployment. And what we say is, it's really sad to think about it, but right now there's actually more regulatory oversight for the hospital cafeteria than there is for the hospital AI program.
- [10:43]
A
Absolutely. And again, I think it's partially also because the word regulation comes with a lot of stigma around it, specifically for folks that want to move fast and that we have to be cutting edge. We have to drive innovation across the enterprise. And then you say, oh, wait, regulation. And it's a worry for a lot of folks, again, that it slows innovation and adds friction in the AI process within an organization. And from what you're seeing, though, how can those responsible AI frameworks actually accelerate this process and help organizations then scale AI safely?
- [11:22]
B
Well, I would say if you look at F1 drivers doing over 200 miles an hour in a car, they do wear a seat belt. You actually need some controls if you're going to go fast. And when I think about regulation, the nice thing about accreditation is accreditation has been around for decades, and accreditation actually has the ability to change and move along with a rapidly changing care environment. So whereas I'll tell you right now in Washington, D.C. it's difficult to pass any laws. There's just so much, so many, so much tug of war going on in D.C. but accreditation being voluntary right now, accreditation is being approached by people who say, we recognize there needs to be some other oversight on this, and we're going to go for that. For my generation to be the Good Housekeeping seal of approval, you know, for younger generations going to go for those Yelp reviews or whatever it is. But some. Somebody needs to be reviewing the safety and the efficacy of what's going on in healthcare. AI in particular, we're not putting out an accreditation program for your bank's AI. We're saying healthcare is a different, different beast. And it needs, it needs a little bit of a collar at least, so that people feel they can come and pet it.
- [12:36]
A
And it touches so many different people. We all are involved in healthcare. That's part of our lives. We have to be able to say, okay, there is regulation, there are some guardrails, there is that collar, as you've mentioned, to be able to control some of this. I do want to come back to the guys in the garage a little bit here because, again, I think it's an important part of the conversation. Looking ahead as we head into the rest of 2026 and certainly beyond, how do you expect accreditation to influence that vendor selection and also the competitive positioning and just in general, the overall trust
- [13:11]
B
in healthcare AI, I think there's a couple different places where this can come in. I will tell you. I've talked with hundreds of organizations, and some of them have a very good governance and adoption framework that they use for their organization for any technology coming in. I actually used to chair the technology adoption committee at my hospital, so system. And, and so I'm familiar with that. And, and if, if you're going to buy a new CT scanner, one of the things that you do is you look at, you know, what's it to be used for, you know, who developed it, are they going to be in business next year? And then how do we bring it in? How do we train on it? Do we have the electrical hookups? Are our systems ready for it? And, and, and what's it going to do for us? And so I, I think that what we're hearing is that people who are looking at vendors coming in from the outside have said, we really think that we should start writing in this idea of an accreditation for. Consider them as being part of the process. You know, it's, it's sort of like a, you know, if it's four guys in a garage, they could go out of business next week, or mom could kick them out of the garage, and then they suddenly don't even have an office. And so when it comes to AI tools, especially anything that influences patient care, patient care has always been a special trusted area. You know, we talk about evidence based medicine and, and what's the research show and those sort of things. And AI is breaking some of those rules. And not only is it breaking some of those rules when it shows up, it's breaking some of those rules a week later when it might have changed a little bit. So, for example, when we talk about AI programs, we say, before you adopt an AI program, you should analyze it from a risk standpoint. What is the risk of the usage of this tool? If it's high risk, you have to keep a closer eye on it. If it's low risk, you know, you can, you can check it every month or so, something like that, if it's not affecting patient care. Just to illustrate that point, one of the analogies that I use is think of anything you're asking AI to do in healthcare and imagine it's a person doing the same thing. Okay, so if you went into the doctor's office today and they said, Larry is going to come into the office, into the exam room with you, and he's going to write down everything that we say so that I don't have to write it all down, are you okay with that? Well, it's a very natural question to say, well, who's Larry? You know, what's Larry's background? Was Larry, you know, picking up the recycling earlier today and he just happened to wander by? But he's got really nice handwriting, so this is fine. So, so there's a credentialing analogy here for when you bring something into your system. Is it trained, is it qualified, is it doing the job that you expect and who's going to keep an eye on it until you trust it? And we're saying the same things for AI. Our accreditation standards are actually out there on the web. You could look them up. You could go to our website, urac.org you could see what our standards are, what they talk about. And that gives you a really nice framework for what you should look for, both from your developers and also in your deployment. Now the next step is, okay, we say we're going to follow those things, but patients don't know to trust us, providers don't know to trust us. They don't want to be hung out to dry with liability from some tool that's not all the way baked. Or is the four guys in the garage who developed it. The other thing is that, are we training as we roll these things out. I know that some of the AI scribes have been rolled out to hundreds of thousands of users, and I'm not familiar with how much training they're getting on those tools. Some organizations are doing a great job with it. Others are sort of like, oh, we put a new AI scribe into the EMR today. We want you to take a look at the notes and use those for your notes going forward. And that's really not how we have done things when it comes to this. And there's risks that pop up. There's horror stories of just terrible things. But there's also, you know, I'm hearing from doctors, these tools have been rolled out. And, yes, they're letting me pay attention to the patient in the exam room. But then that means after the exam room, I have to go read through all of the notes and make sure they got it right. Because a mistake in the record, it can be damaging to the patient in ways you don't expect. Yeah.
- [17:19]
A
Well, Dr. Griffin, I feel like we could talk for another hour. There's so much, as you've mentioned, there's so much to talk about and so many different aspects to this conversation. I want to thank you for your time and for your insights. This was fantastic. Thanks for being here.
- [17:32]
B
Well, I appreciate the opportunity to talk a little bit about this, but, you know, take a look at our website. There'll be other AI programs that'll probably come out, and we'll update this as need be, but it's a. It's a very sensitive area, and we're just offering to. To help organizations to do it right, to protect their patients and their providers. So thanks very much, Lucas.
- [17:50]
A
Absolutely. It's great to have you at least talk to the guys in the garage. Don't just buy their product.
- [17:56]
B
Yes, the guys in the garage may not be that much smarter, and you're letting risk into your organization if you don't check them out.
- [18:04]
A
Exactly. Well, Dr. Griffin, thanks again for your time. We also want to thank our podcast sponsor, Urac. You can tune into more podcasts from Becker's Healthcare visiting our podcast page@beckershospitalreview.com.