HCA’s AI-Driven Patient Safety Evolution, Part 2

Summary

---

Advancing Health – HCA’s AI-Driven Patient Safety Evolution, Part 2

Podcast: Advancing Health
Host: Dr. Chris Tirienzo (AHA Chief Physician Executive)
Guest: Dr. Randy Fagan (Chief Quality Officer, HCA Healthcare)
Date: September 8, 2025
Theme: HCA Healthcare’s comprehensive integration of AI to advance patient safety, with a focus on robust governance, frontline engagement, and actionable lessons for health systems.

---

Episode Overview

This episode delves into HCA Healthcare’s journey with AI-driven patient safety, offering a practical look at governance structures, variance reduction, the importance of frontline staff involvement, and real-world examples of AI-enabled decision support. Dr. Randy Fagan shares insights and concrete advice for hospital leaders navigating AI adoption.

---

Key Discussion Points & Insights

1. The Necessity of Robust AI Governance

(01:33 - 02:57)

• Comprehensive Stakeholder Involvement:
  Dr. Fagan describes HCA’s multi-disciplinary governance framework, integrating not just clinical leadership but operations, finance, marketing, supply chain, and more.

  “We've put into place a pretty robust governance structure that goes beyond just our clinical leaders... It's important as we look at use cases that we're looking at each use case through all lenses.” — Dr. Fagan [01:33]

• Objective Prioritization of Use Cases:
  Uses quantitative risk/opportunity scoring to identify high-impact AI use cases, with a primary objective to reduce variance in clinical and operational practice.

2. Variance Reduction: The Core Value of AI

(02:57 - 03:36)

• Impact Beyond Averaging Up:
  Fagan emphasizes that AI’s most promising contribution is reducing variability—by clinician experience, patient presentation, or geography—thereby delivering more consistent and safer care.
• Relatable Analogy:
  Dr. Tirienzo offers a “batting average vs. slugging percentage” analogy to distinguish between broad vs. targeted improvements.

  “Is this a batting average problem or is this a slugging percentage problem?... It really does branch into very different pathways.” — Dr. Tirienzo [02:57]

3. The Critical Role of Frontline Involvement

(03:36 - 04:45)

• Early Involvement for Relevance and Buy-in:
  Involving frontline staff from the onset ensures solutions address real needs and promotes adoption.

  “It is critical that we get the people closest to the work involved early in the process to help to shape that work.” — Dr. Fagan [03:36]

• Lessons from Early Failures:
  Dr. Tirienzo reflects on prior missteps, reinforcing this point:

  “We showed it to the people who are going to use it and they said, ‘What's this? We don't need this...’ That is a lesson that you learn exactly once in your career.” — Dr. Tirienzo [04:06]

4. Augmentation, not Automation: Preserving Human Judgment

(04:45 - 05:44)

• AI as a Decision Enhancer:
  Dr. Fagan clarifies that the intent is to augment, not replace, clinician judgment except where safe.

  “How do we enhance human decision making, not replace it, especially in the clinical space?” — Dr. Fagan [04:45]

5. Real-World Applications: Staffing & Documentation

(05:44 - 09:52)

• Staffing Optimization:
  AI helps standardize nurse staffing, freeing leaders to focus on patient care while optimizing team composition for both patient needs and nurse preferences.

  “…offloading that burden from them, it allows them to actually lead rather than spend hours of their day managing a schedule.” — Dr. Fagan [05:44] “This is exactly the kind of problem that AI is built to solve because AI finds patterns and helps develop the solution.” — Dr. Tirienzo [06:21]

• Skill Mix and Clinical Considerations:
  AI factors in disease states and nurse experience, aiming for balanced units rather than homogeneity (all new or all veteran nurses).

  “How do we create the right dynamic of skill sets and experiences to meet the needs of the patients?... There’s a clinical value in the work.” — Dr. Fagan [07:09]

• Ambient AI for Documentation:
  HCA pilots AI-driven documentation that reduces physician/admin workload and improves record completeness and continuity.

  “It can remove literally hours’ worth of work... Also, when you think about having a more accurate, complete medical record...it's just better for patients.” — Dr. Fagan [07:09]

• Benefit to Workforce Well-being:

  “This ambient technology is giving it [face-to-face patient time] back... improve the experience of our workforce... while we are in the midst of a workforce crisis that we will never be able to recruit our way out of.” — Dr. Tirienzo [09:52]

6. Key Lessons Learned for Health Systems

(11:08 - 12:27)

• Multi-disciplinary Teams:
  Include voices beyond clinical—finance, operations, etc.—to comprehensively address AI challenges.
• Start with a Safely-Tested, High-Impact Problem:
  Choose a problem where failed attempts carry low risk but successful solutions will have significant positive effects.
• Avoid “Solution Looking for a Problem”:
  Focus on identifying the core clinical or operational need before engaging vendors or purchasing solutions.

  “Don’t start with vendor selection. That becomes a solution looking for a problem, you need to first identify the problem you’re trying to solve and then identify the solution...” — Dr. Fagan [11:37]

---

Notable Quotes & Memorable Moments

• On Governance:
  “We've included those folks to be a part of the table because it's important as we look at use cases that we're looking at each use case through all lenses.” — Dr. Fagan [01:33]
• On Learning from the Front Line:
  “It validates the problem, it ensures the relevance and buy in.” — Dr. Fagan [03:36]
• On AI’s Role:
  “How do we enhance human decision making, not replace it, especially in the clinical space?” — Dr. Fagan [04:45]
• On the Future of Health Workforce:
  “You have picked the best possible time to go into medicine or nursing...because...our entire world got electronified.... this ambient technology is giving [face time] back.” — Dr. Tirienzo [09:52]
• On Implementation Wisdom:
  “If you fail, you’re going to fail safely in this. And…don’t start with vendor selection.” — Dr. Fagan [11:37]

---

Important Segment Timestamps

• [01:33] – Building AI governance across functional areas
• [02:57] – Reducing variance as a methodology for improvement
• [03:36] – Importance of engaging frontline staff early
• [04:45] – Augmenting, not replacing clinicians with AI
• [05:44] – Case study: AI in nurse staffing
• [07:09] – Clinical documentation and administrative burden reduction
• [09:52] – Workforce crisis and technology’s humanizing return
• [11:37] – Lessons learned: Team composition, focusing on problems first

---

Summary

This episode provides a playbook for scalable, AI-enabled patient safety initiatives. HCA’s experience underlines the necessity for inclusive governance, a variance-reduction mindset, early and deep engagement with frontline users, and disciplined implementation focused on tangible, tested problems. The practical wisdom offered is relevant for organizations of all sizes—in Chris Tirienzo’s words, “just as easily in a critical access hospital in Oregon as...a multi-state system like HCA.”

Transcript

[A]

Welcome to Advancing Health. The impact of artificial intelligence is showing across nearly every aspect of healthcare delivery today and continues to grow in reach and importance. In the second of this two part podcast on patient safety, we hear from a major health system about its steady integration of AI, how to govern it, and lessons learned from putting AI to work across many of its systems.

[B]

Foreign welcome back. This is part two of a terrific conversation with Dr. Randy Fagan from HCA. Again, I'm Dr. Chris Tirienzo, the Chief Physician Executive for the American Hospital Association. Dr. Fagan serves as the Chief Quality Officer of HCA Healthcare. And part one of this podcast went really deep on patient safety and how HCA is both thinking about and acting on patient safety at a scale that is incredibly challenging to rival within the American healthcare ecosystem. We got into a conversation around AI enabled technologies on patient safety and it made us wonder. There's a deeper conversation we should probably have around AI and Randy, I think that starts with government because I've got to imagine when you're talking about something like 190 hospitals across 2000 ish other rooftops that you have no shortage of people wanting to utilize AI in their workflows. And that's got to require a pretty significant governance arm.

[C]

You're 100% correct. And you know, we've put into place a pretty robust governance structure that goes beyond just our clinical leaders. It involves all of our functional areas from operations, finance, marketing, development, supply chain, you name it. We've included those folks to be a part of the table because it's important as we look at use cases that we're looking at each use case through all lenses. As we prioritize use cases, we look at them quantitatively based off of the risk and the opportunity in each one. It's interesting as we've looked at the opportunities, it's not always about average performance or going from this average to this average or one of the greatest opportunities that AI offers us the opportunity. And it doesn't matter which lane we're in, whether it's operational efficiencies or clinical performance. It's about variance reduction. And, you know, being able to reduce variance is one of the, I think, most powerful things that AI can do for us inside the clinical space. When we look at areas to prioritize, we look at where we have areas of high variance, whether it's variance by clinician experience, variance by patient presentation, variance by geography. How can we reduce variance to try to improve the consistency with which we deliver care to our patients?

[B]

I love the focus on Variance. One mentor of mine framed it this way, that when you're looking at a challenge, you have to ask yourself, is this a batting average problem or is this a slugging percentage problem? Because sometimes you're right, I've got to move a whole big boat from below the Mendoza Line, which is 200 in baseball for those who are baseball fans, to 300. But sometimes it's I just need seven more home runs or I need to not strike out 17 times. And it sounds like you're, you're taking a version of that approach that is not baseball related to, to healthcare. And I love that thinking because it really does branch into very different pathways.

[C]

100% agreed. And as we do that, one of the critical things that is kind of the next layer beyond governance is who you involve. And it's, it's critical that we involve frontline staff early, the people who are closest to the work being done. It validates the problem, it ensures the relevance and buy in. And as smart as the executive team may be, it is critical that we get the people closest to the work involved early in the process to help to shape that work.

[B]

For those who are listening only and not seeing me nod my head as vigorously as a head can be nodded right now, I just want to narrate that for you because I remember like 10 years ago, we had developed this terrific machine learning model to predict readmissions in the health system. I was working at the time. We spent nine months in development, like the numbers on the model were terrific. And then we showed it to the people who are going to use it and they said, what's this? We don't need this. And by the way, I don't want to use that. It took us a whole another nine months to walk through the people side. So that is a lesson that you learn exactly once in your career. And it sounds like it's being put to use by your group at hca.

[C]

Agreed. And as we bring those folks forward, one of the things that we try to make sure we're focusing on is augmentation, not necessarily automation. How do we enhance human decision making, not replace it, especially in the clinical space? We think that's a very important decision point.

[B]

Yeah, there are, there's a great article that was in Axios a few months ago that the two editors of Axios wrote together about the impact of AI on their own professional infrastructure. They recommended all leaders ask themselves the question, what are the things that I and my team need to do to be incredibly successful? And then ask the follow up Question. How can AI either automate some portion of that so I can focus my human time on a different part? That's going to make a big difference. Do you have some specific examples, Randy, about how you're seeing that put to use and how perhaps those of ideas have flown through the governance process and then into action?

[C]

Yeah, there's a few areas that we've addressed I mentioned in the last episode. Working towards being able to reduce variance in the way that we staff our nursing units. And it sounds kind of banal, but it's remarkable when you think about the variance in how we staff those based off the individual who's staffing it and the time it takes them to do that. It is incredible. And by, by offloading that burden from them, it allows them to actually lead rather than spend hours of their day managing a schedule.

[B]

The staffing example is, is fascinating because I interviewed Dr. Schlosser two years ago and Mike Schlosser, for those who didn't listen to episode one, is one of the HCA leaders who helped drive this technological transformation across the enterprise. And I remember him describing that nursing model in some detail because the number and DE of variables about weekend option versus not weekend option, Tuesdays and not Fridays. And for a human to try to manage your nursing unit can have hundreds of nurses who you're trying to mix and match in a way that meets personal lives as best as possible with patient and clinical needs. This is exactly the kind of problem that AI is built to solve because AI finds patterns and helps develop the solution. Right.

[C]

And on the clinical side too, of that. Chris, you know, this isn't just an exercise in how do you staff a nursing unit with the right number of humans for the number of patients that are there. There's the ability as you, as you head further down that road to say what kind of patients are on that unit today on this shift, at this moment, disease states, resource intensity, and then try to match that up with a dynamic of nurses who meet those needs. You don't want to staff an entire unit with all new grads, nor do you necessarily want an entire unit of all 20 year experienced nurses. How do we create the right dynamic of skill sets and experiences to meet the needs of the patients? There's a clinical value in the work that's done beyond just the offloading of that administrative burden from folks on the offloading administrative burden. I know a lot of folks have been using AI as a vehicle to assist with documentation where physicians, nurses, PAs, whoever can just talk and patients can just Talk. And that information is then aggregated and then put into the medical record in a way that it is consumable and it understands nuance, it understands context. And it plugs things in, regardless of the order in which you ask the questions, it plugs it into your documentation. And the idea of creating a greater level of completeness of our electronic health record has an incredible value to us. I mean, one on the, on the administrative burden side, it can remove literally hours worth of work of our physicians who need to be entering the information, editing the information, signing off on the information, all of that stuff. That's just an administrative burden. It literally can offload hours of it. Also, when you think about having a more accurate, complete medical record, the ability to transfer knowledge from shift to shift, and the ability to make clinical decisions based off the complexities of the individual, it's just better for patients. And I really see a value in that space. And that's another one that we're exploring and pursuing is we've got an entire cohort of physicians, both ER physicians and hospitalists. We're exploring the cardiac space as well. In Texas, that was our pilot area in the Dallas area that have been utilizing this and helping us to learn how do we allow this to give time back to the doctors, more time at the bedside, less time at the computer, at the same time, enhance the information that is able to be transferred, shift to shift, and the information that's utilized by each shift for clinical decision making. It's a really exciting space.

[B]

Ambient AI, the technology that you're referencing, and there are a variety of companies who are in that space, is one reason that when I talk to trainees, so folks in medical school or residency or nursing school, I say, folks, you have picked the best possible time to go into medicine or nursing or healthcare in general. Because you and I trained on paper and we were walking around units with giant charts and writing orders in triplicate. And then our entire world got electronified. And in doing so, we pulled people away from FaceTime with patients. And I see the value in all of that structured data. My background is a CQO as well, on a much smaller scale than you. So I love the fact that we now, thanks to the electronification of healthcare, have all of the structured data and the metrics. But my gosh, we transferred such face to face to face to screen time that this ambient technology is giving it back. It really is a way not only to, as you indicated, improve documentation, which has never been anyone's favorite part of their job, but also improve the experience of our workforce. It is such a crucial thing to focus on while we are in the midst of a workforce crisis that we will never be able to recruit our way out of.

[C]

Completely agree with you, Chris.

[B]

I think we've got time for one last question and I'm just wondering, obviously you all have had a number of successful implementations. You've probably had some that tried and were cut off. What other, you know, single biggest lesson would you have steering the AI ship at, you know, the, the 190 plus hospital entity that is HCA Healthcare are folks listening in who are perhaps trying to do the same thing within their own communities?

[C]

I would tell you one of the most important things that folks can do. One. Well, there's a couple of things you asked for. One I'm going to give you more than one. Well, one is, you know, make sure that, you know, if you're the clinical leader trying to use AI to advance care in your space, bring together a team that isn't just clinical, you have to have all different lines of sight to really solve these problems. Second, you want to define a single high impact problem that has a low risk if you don't get it right. You want to make sure that if you fail, you're going to fail safely in this. And then the last thing, don't start with vendor selection. You know, that becomes a solution. Looking for a problem, you need to first identify the problem you're trying to solve and then identify the solution that best allows you to solve for that.

[B]

Wise words and ones that I think could be applied just as easily in a critical access hospital in Oregon as they can to a multi state system like HCA. Dr. Fagan, it has been a privilege and again, my guess is we'll be asking you back again sometime in the future. Thanks for spending time with us today Chris.

[C]

An absolute privilege anytime.

[B]

Thank you. Take care everyone.

[A]

Thanks for listening to Advancing Health. Please subscribe and rate us 5 stars on Apple Podcasts, Spotify or wherever you get your podcasts.