Building Trust in Clinical AI with Yaw Fellin of Wolters Kluwer Health

Summary

---

Podcast Summary: Building Trust in Clinical AI

Podcast: Becker’s Healthcare Podcast
Episode: Building Trust in Clinical AI with Yaw Fellin of Wolters Kluwer Health
Date: October 20, 2025
Host: Erica Spicer Mason
Guest: Yaw Fellin, Senior Vice President and General Manager of Clinical Decision Support and Provider Solutions, Wolters Kluwer Health

---

Episode Overview

This episode explores how clinical intelligence and artificial intelligence (AI) are shaping the future of healthcare, with a focus on building trust and driving adoption. Yaw Fellin discusses the evolving role of AI and clinical decision support tools in tackling major challenges facing hospitals and health systems—such as clinician burnout, rising costs, and growing care complexity. The conversation emphasizes practical approaches to integrating trustworthy AI, the necessity of transparency, expert involvement, and strong governance, as well as strategies for workflow adaptation and results-focused experimentation.

---

Key Discussion Points & Insights

1. Yaw Fellin’s Role & Wolters Kluwer Overview ([00:33]–[02:07])

• Yaw’s Background: Over 20 years in healthcare, with leadership roles at organizations like Optum and The Advisory Board Company.
• Current Focus: Now leads all aspects of clinical decision support at Wolters Kluwer, especially the UpToDate product.
• Company’s Mission: Delivering evidence-based clinical intelligence for providers, patients, and nurses, integrated directly into clinical workflows.

2. The Value of Clinical Intelligence in Addressing Industry Challenges ([02:07]–[04:40])

• Ongoing Challenges: Cost pressures, workforce strain, and care complexity remain acute.
• Value Proposition of Clinical Intelligence:
  • "I think of this as having a clinical thought partner...an expert peer clinician built into systems and workflows." — Yaw Fellin ([03:23])
  • Saves time and helps clinicians collaborate for better care decisions.
  • Potential for improved outcomes through more informed, reference-based decision-making.

3. Building Trust in Clinical AI: Guiding Principles ([04:40]–[10:06])

• Skepticism is Declining:
  • Growing patient comfort with AI (e.g., using ChatGPT to research health issues).
  • Survey findings: 40% of providers are ready to use AI-supported clinical decision support; two-thirds recently changed their views towards embracing AI solutions.
• Three Key Principles for Trustworthy AI:
  1. Transparency:
     • Not just showing sources, but also revealing the system’s assumptions and reasoning.
     • "In the work that we’re doing, one of the key guiding principles is...we actually want to show you the assumptions or the clinical reasoning that the AI system took to arrive at the answer." — Yaw Fellin ([06:45])
  2. Expert in the Loop:
     • Not just “a human,” but specifically trained clinical leaders and specialists validating the outputs.
     • "I would actually be a horrible human to be in the loop...for me to be in the loop...doesn't improve the safety." — Yaw Fellin ([07:57])
  3. Governance:
     • Addressing risks from “ungoverned” AI use at the end-user level.
     • Strong, thoughtful governance frameworks needed for responsible, safe adoption at scale.

4. Case Study: St. Luke's and Wolters Kluwer’s Expert AI ([10:54]–[12:57])

• Real-World Example:
  • St. Luke’s University Health Network implemented Wolters Kluwer’s new generative AI-powered clinical decision support.
  • St. Luke’s highlighted as a model for transparency and robust governance in AI adoption.
  • Rolled out incrementally, aligned with their care delivery goals.
  • "...the focus that they've put on governance...the evaluation and the phasing and rollout..." — Yaw Fellin ([12:27])

5. Workflow Integration Strategies ([12:57]–[15:56])

• Integration is Essential:
  • EMRs (Electronic Medical Records) are central but often slow to innovate.
  • New technologies, particularly “ambient” systems, are creating fresh possibilities for integrating decision support into workflows in more seamless, less disruptive ways.
  • Anticipation that both EMR and ambient solutions will continue to evolve and unlock better patient outcomes.

6. Priorities for Innovation: Experimentation & Outcome Alignment ([15:56]–[18:12])

• Top Immediate Priority:
  • Organizations must prioritize "intentional experimentation"—testing, learning, and iterating with new technologies.
  • "The ability to experiment, to pivot when those experiments aren't working, that would be number one." — Yaw Fellin ([16:31])
• Results Must Align with Outcomes:
  • Experiments should be directly tied to improving organizational goals and patient outcomes.
  • "I think the most important result that we can all be aiming towards is improving outcomes..." — Yaw Fellin ([17:43])

7. Final Thoughts: Need for Better AI Validation ([18:33]–[20:07])

• Validation Gaps:
  • Current public benchmarks (e.g., USMLE-based testing) are insufficient; they don’t reflect the complexity of real-world, point-of-care decisions.
  • Many private benchmarks show only 50–60% AI accuracy, which would be unacceptable for clinical practice.
  • "If I told you your doctor was going to be 50–60% [accurate]...good question as to if that would be an acceptable threshold." — Yaw Fellin ([19:37])
• Call to Action:
  • More rigorous validation standards are urgently needed for clinical AI. Wolters Kluwer is publishing research on this soon.

---

Notable Quotes & Memorable Moments

• Clinical Intelligence as a 'Thought Partner':
  "I think of this as having a clinical thought partner...an expert peer clinician that is built into hospitals’ workflows." — Yaw Fellin ([03:23])

• On the Need for True Expertise in AI Loops:
  "You don’t just want any human in the loop...you absolutely need the right experts—really the clinical leaders and the specialists—to validate these systems." — Yaw Fellin ([07:31])

• On Governance:
  "We just see a lot of ungoverned use of AI systems...and on the flip side, tremendous hospital peers that are really being thoughtful about the governance process." — Yaw Fellin ([09:15])

• On Experimentation:
  "The ability to experiment, to pivot when those experiments aren't working, that would be number one." — Yaw Fellin ([16:31])

• On the Importance of Improved Validation:
  "Most public benchmarks...aren't very representative of what happens at the point of care...validation is probably a topic that's not getting quite enough attention." — Yaw Fellin ([19:05])

---

Important Timestamps

• Yaw’s Introduction & Background: [00:33]–[02:07]
• Defining Clinical Intelligence & Its Impact: [02:46]–[04:40]
• Guiding Principles for Trustworthy AI: [05:25]–[10:06]
• Case Study — St. Luke’s University Health Network: [10:54]–[12:57]
• Workflow Integration Strategies: [13:26]–[15:56]
• Prioritizing Experimentation & Alignment with Outcomes: [15:56]–[18:12]
• Call for Improved Validation in Clinical AI: [18:33]–[20:07]

---

Conclusion

Yaw Fellin underscores that trust in clinical AI requires transparency, deep clinical expertise, and rigorous governance—especially as skepticism gives way to cautious optimism. Successful organizations are those that experiment intentionally, prioritize seamless workflow integration, and relentlessly focus on patient outcomes, all while demanding high standards of validation from their AI systems. As AI’s influence grows, the industry must keep its standards high to harness the promise of clinical intelligence without sacrificing safety or efficacy.

Transcript

[A]

Hello everyone. This is Erica Spicer Mason with Becker's Healthcare. Thank you so much for tuning into the Becker's Healthcare podcast series. Today we're going to talk about how clinical intelligence and AI are shaping the future of healthcare. And joining me for this conversation is YA Phelan, the senior Vice President and General Manager of clinical decision support and Provider Solutions at Wolters Kluwer Health. Ya. So great to have you back with Beckers today. Welcome to the podcast and thanks for making the time.

[B]

Thank you Erica. It is great to be back and reconnecting with you.

[A]

Likewise. And I know some of our listeners are likely already familiar with you and Wolters Kluwer. But for those who may not be clued in, can you share just a little bit more about yourself and your work in healthcare?

[B]

Yeah, no, I'm happy to. It's actually an update since I was here last time. When we last spoke just a couple months back, I was leading the product organization for the clinical effectiveness business at Wolters Kluwer. That's basically the up to date portfolio. And as you just noted in the intro, I'm now leading the business holistically. So everything that we're doing around clinical decision support at the point of care, that is still the up to date product portfolio. And then for those that aren't as familiar, we are very focused on how do you take evidence based knowledge or clinical intelligence and provide that to hospitals, health systems, you know, providers, patients, nurses as clinical decision support within their workflows. So that's kind of the focus of the business. I've been in health care for over 20 years. At this point I've had roles at Optum, the advisory board company. I've worked across SaaS based technologies as well as on the ground with hospitals and health systems as a consultant. So healthcare is kind of, well, certainly my passion and you know, Certainly the last 20 plus years what I think about on a day to day basis.

[A]

So great to hear that about you. Ya. And appreciate you also sharing with us some of the changes in your own role and your work. I'm just really glad to have your expertise on the line again because what hasn't changed since we spoke last is some of these challenges or I should say the magnitude of the challenges that hospitals and health systems are navigating right now. So we're still seeing rising costs, strain on the clinical workforce and growing care complexity as well. So from where you sit right now, what is the role of clinical intelligence in helping leaders approach these challenges with greater clarity? And also confidence.

[B]

Yeah, it's such a great question. And I think your lead in was perfect. Right. These are not necessarily new challenges for hospitals and health systems, but I do think there are a set of challenges that are exacerbating. We did a study not too long ago, basically showed 76% of the organizations that we spoke to viewed reducing clinician burnout as a top priority. And I think what you're seeing in the here and now around clinical intelligence is that in an optimistic way, the technology is getting much, much better to start to address some of these long standing issues within the industry. So for me, clinical intelligence, it's pretty straightforward. It's almost as it sounds, but maybe a little bit more of a practical way to describe it. I think of this as having a clinical thought point partner. So, you know, think of an expert peer clinician that is basically built into the systems and workflows hospitals are deploying as, as they provide care. And, you know, that role, I think can drive a whole host of value propositions. One, it can save organizations time. Two, it can help improve the clinical decisions that they're making. So think of two clinical peers debating what's the best way to treat or diagnose a specific patient and ultimately coming up with a better decision or an answer. And therefore, three, it has just tremendous potential to improve outcomes for patients and the care that they're receiving within these organizations.

[A]

I appreciate how closely you tied those outcomes to clinical intelligence. Yaw. And also I think it's a great way to think of this type of tool or this technology as having a clinical thought partner. A really helpful way to make this a bit more tangible for our listeners. And you said something about the way that technology is advancing, acknowledging that it has certainly improved recently and over time, but we still see AI generating optimism and that healthy skepticism not just among providers, but patients as well. So what guiding principles do you think are essential to ensure that AI is applied in ways that are both trustworthy and can also actually improve care delivery?

[B]

I think it's important to start with I do see the skepticism, or we do see the skepticism reducing. So on the patient side, I was spending time with a number of clinical leaders last week. We had a forum around clinical decision support. And certainly what those leaders were sharing were anecdotes and stories of patients becoming very familiar with ChatGPT or generative AI capabilities and tools, using those tools to try and understand questions around their clinical care, and then bringing those questions directly into interactions with providers. So lots of data points to tell us that patient skepticism is trending down. A moment ago I referenced the survey that we did. If you go a bit deeper into that survey, 40% of providers are basically ready to use some of these new technologies and AI supported clinical decision support at the point of care. And over two thirds of providers had recently changed their views on using these types of systems. So we are certainly seeing an improvement in skepticism from a guiding principle perspective. I'm pretty notorious for always thinking in threes, so I'll probably give you three on this one as well. Erica. First is transparency. I think this has been one of the primary guiding principles that organizations or folks have taken to try and put AI to use in a practical way. On the transparency side, I would say this is tended to start with just being transparent about the source or the reference within an AI answer. But I would push that a bit further. In the work that we're doing, one of the key guiding principles is not just that we're going to be transparent with the source, but we actually want to show you you the assumptions or the clinical reasoning that the AI system took to arrive at the answer or the output that you're seeing. So that's, that's definitely number one. The second for me is this concept of having an expert in the loop. And I actually think the, the industry bar right now is, is too low in terms of how we think about who and what do we want to have in the loop on these AI systems. Generally people perceive that it's safer when you have a human in the loop of the AI systems is the term that you hear bandied about. And I think if you stand back and think about it, you don't just want any human in the loop of these systems. I would actually be a horrible human to be in the loop of a clinical decision support system. Not because I'm not well intentioned and hard working, but I've never actually trained as a practicing clinician. So for me to be in the loop of a system, mediating what is a good or a bad answer around clinical decision support doesn't improve the safety. So the guiding principle that we've taken here within Uptodate is you absolutely need the right experts in the loop, really the clinical leaders and the absolute specialists in their areas to help make sure that the outputs and the validation of these systems is, is up to the appropriate standard. And then the third kind of key guiding principle or area that, that I think is important to focus on is just around governance. Candidly we do see, and I think this is occurring in Large part because a lot of AI systems are being adopted more at an end user level right now versus an enterprise level. But we do see a lot of ungoverned use of AI systems just as we scan what's out there in the market. And then on the flip side of that, we just, we've got some tremendous hospital and health system peers and partners that are really being thoughtful around, you know, what is the governance process as we put these AI tools into the hands of our clinicians? You know, how do we make sure that we've got all of the right checks and balances so that this is safe to deploy at scale? And I'm sure I'm missing other things, but if you ask me for the top three, I think those are the top three guiding principles.

[A]

Yeah, yeah. I appreciate the way that you think in this series of three, it's easy for me to understand and follow and I'm sure for our listeners too. And these are great, I think, very practical tips for hospitals and health systems that are looking to really try to alleviate some of this healthy skepticism that they're seeing. So that was just to sum up again, transparency, having an expert in the loop and governance. So in your experience working with partners, I know Wolters Kluwer partners with a lot of organizations across the country. Can you reflect back on maybe a specific example of where you've seen an organization do those three things or maybe even one of those things really well and maybe some of the tangible changes that you saw ensue as a result?

[B]

Yeah, we've got a number of organizations that are excelling across all of those dimensions. But for a specific example, you know, it's just a couple weeks back, on September 24th, we announced the work that we're doing around our expert AI solution. We've basically created a new generative AI powered clinical decision support capability built on top of the, the up to date product and the expert community that curates that. And our first, you know, key partner to roll that out is St. Luke's University Health Network. They were participated with us as a part of this press release. And you know, I think they're just a prime example of the exact type of partner that you want. You know, when you're working in this type of space. Right. That I know that they view this, you know, not just as an upgrade to what they did and up to date, but as a way to really put clinical intelligence to work across their health system and to support their care delivery goals. So I think the mission piece of how they focus on their work and where they see the potential is huge. But I think critically as important, and I know we're just talking about all the guiding principles, we just see them adopt and adhere to those in spades. In particular, I would say, you know, obviously choosing a system that has the right type of transparency and assumptions built into the system in and of itself, but most importantly, just the focus that they've put on governance. Right. How do they, you know, really be thoughtful about both the evaluation and the phasing and rollout of these types of systems to their clinicians? That's probably the best example. You know, that's top of mind right now.

[A]

Yeah, it's a great example. So thank you, thank you so much for sharing. Ya. And I want to just take one more second to talk a little bit more about the operational piece of leveraging clinical decision support. So we know that the impact of this technology, it really does hinge on how well it can integrate into real workflows. So what strategies have you seen help organizations adapt and evolve those workflows to better serve clinicians and patients?

[B]

Yeah, workflow is the key. And if you stand back and think about hospital and health system workflows, the primary or dominant workflow in that area is the electronic medical record or EMR space. Generally speaking, I would say that workflow, you know, innovation in the EMR space had been a little slow. You know, particularly when you're talking about putting clinical decision support in the heart of those systems. We've been working at that for quite some time. And you know, while we've made meaningful progress with a number of partners and health systems on that front, it's just a heavy set of work. Right. Tweaking content to fit with inside of almost a customized proprietary EMR for a hospital or health system where we've seen tremendous evolution is on the ambient side. And I think a big part of that is ambient is almost reopening what's possible from a workflow perspective within hospitals and health systems. So, you know, I think that's a prime example of the positive impact of technology and just some of the big shifts that have taken place within the industry. And in terms of getting back to what's ultimately most important, which is can you leverage those workflows to improve the outcomes of patients. And we actually expect that both channels are going to continue to evolve in a much more rapid way. I think even on the EMR side, the building of ambient systems into EMRs and the deployment of those types of capabilities is really starting to unlock a lot of different workflow potential then obviously ambient adoption, while very popular, is still in the early, you know, early innings, if you will. So I think we also expect, you know, just to see that as greater ambient adoption takes place, that the opportunity to, you know, improve clinical decision support in the workflow, you know, kind of rides along.

[A]

Yeah, you've touched on some exciting developments that are on the horizon or already in motion. So as you look ahead, what's one priority that you would recommend organizations focus on in the near term to make meaningful progress either in clinical decision support or just innovation more broadly? Any final thoughts there?

[B]

One priority is so hard. You know, I like three.

[A]

True.

[B]

So I think it's important that organizations continue to experiment. And so if I had to just. And maybe this is a little bit of a, you know, maybe it's cheating in terms of how I'm answering the question, but I think the top priority should still be experimentation. There's so much change and the speed at which we're able to move with these new technologies has gotten dramatically better. Erica, I think in previous iterations you might have asked me a question of what's the priority for the next 12 months or what's the next priority for the next two years? So even the way you ask the question I think indicates kind of the pace that the industry is moving at. And yeah, I just think the ability to experiment, to pivot when those experiments aren't working, that would be number one. And then just because I can't help myself for a second, I won't go all the way to three, but for a second. I also think it's just critical to make sure that the yield or the results of those experiments are aligned with the ultimate goals of the organization. So certainly I think the most important result that we can all be aiming towards is improving outcomes. And you never get all the way there if you do your experiments in a bite sized enough way. But if you're doing, setting up a system to do experiments and then really grounding those experiments in terms of can we show a meaningful improvement in outcomes? I think any organization or set of organizations that are following that pattern are really in a prime position to thrive.

[A]

Well, yeah, your thought leadership is so appreciated and I really appreciate the sentiment as well about organizations continuing to experiment. How much of a necessity that is. But that experimentation should be intentional and aligned with the organization, the mission. It's been great having you with us again today. Ya. Is there anything we didn't cover or any final notes you wanted to share.

[B]

With Listeners first just want to say thank you. It is always great to join you and have these discussions. If I had one final thought, we still do see an opportunity to improve the validation of the AI systems that are out there. And specifically when you're focused on clinical intelligence, if you just look at some of the public benchmarks that are out there right now, most people are grounding around the usmle. And when you talk to clinicians, everybody knows that that's not very representative of what happens at the point of care. So that's kind of the public benchmark that's been out there in private. What you hear is that different validation benchmarks that, you know, companies are experimenting with, maybe have, are showing 50 to 60% accuracy of these systems. And you know, if I told you your doctor was going to be 50 to 60%. Right. You know, good, good question as to if, if that would be an acceptable threshold. So, you know, validation, I think is probably a topic. It just as a final thought, that's not getting quite enough attention or enough rigor. We're actually going to be publishing on that topic pretty soon. And you know, clearly within the work that we're doing around clinical intelligence, you know, we're going to have a much higher standard than what's being out there. So I just think that's a good place for organizations to, you know, to increase their focus on.

[A]

It's a great note. Ya. And we'll keep an eye out for that upcoming research from the Wolters Kluwer team. So thank you again for sharing all of this with us and for sharing your passion and expertise for this area of the field. Great having you with us again.

[B]

Thank you. I really appreciate it.

[A]

And we'd also like to thank our podcast sponsor for today, Wolters Kluwer listeners. Be sure to tune into more podcasts from Becker's Healthcare by visiting our podcast [e-mail].