Kathleen Wessel (Host): As health systems look for innovative ways to integrate AI into clinical workflows, one principle stands out: technology should strengthen not dilute the meaning behind clinical data.

Welcome to AHA Associates Bringing Value, a podcast from the American Hospital Association. In this series of podcasts, we speak with AHA Associate Program business partners, check in on their efforts and learn how they support AHA hospital and health system members. I'm Kathleen Wessel, Vice President of Business Management and Operations at the AHA. And today, I am joined by Dr. Steven Rube, Chief Medical Officer at IMO Health. Together, we'll explore how AI can improve documentation, reduce clinician burnout, and support better decisions by aligning with clinical intent. We will also learn what responsible and scalable AI integration looks like at the point of care. Dr. Rube, welcome to the podcast.

Steven Rube, MD: Thank you very much for having me.

Host: You know what, I think we're going to jump in with the first question. As AHA members turn to AI to improve care, how can AI solutions ensure clinical intent is preserved while improving accuracy and usability of health data?

Steven Rube, MD: So, that's an excellent question because clinical intent is obviously one of the key issues that we always struggle with when we move away from pure dialogue. I think one of the first and most important elements to the entire conversation is that we have the right people involved. It's going to be very important that we not only have data scientists who create these unbelievable solutions, but we also have clinicians and not just clinician informaticists, actual, you know, clinicians who still work day to day at the point of care.

That being said, I think one of the key things to remember is that an AI system-- or an ambient AI system probably is more relevant to this situation-- needs to be more than a fancy dictation system. In other words, we've had the ability to listen and record and voice to text for quite a while. So, why is this different? We're going to need these systems to not only listen to what we're saying, but to really understand what we're saying.

And I'll give you an example. If a patient came in to see me, I might say, "How is your diabetes?" What I'm not going to say is, "How is your type 2 diabetes with stage IV chronic kidney disease while you're using insulin?" However, that is what I mean. So when is an exact match not an exact match? Well, it's in that situation. So, these ambient AI solutions are really going to have to understand the patient in total. And that's just one example of preserving that clinical intent. We have to be careful not to dilute these clinical terms to the abstract so that all diabetics just become diabetics, and we really still maintain the richness of these diagnostic terms. That is the trick, and that's going to be very important to, as you said, maintaining that clinical intent.

Host: Thank you for that example. I thought my impression was you were going to go with more kind of nonverbal signals. But no, you've actually introduced kind of some concepts that I hadn't even really considered that need to be very specific. So, I appreciate diving in deeper there.

So, I do want to back up just a touch. Before we get into some of the details, could you maybe share your professional journey? What brought you to IMO Health?

Steven Rube, MD: Sure. My professional journey is a circuitous one. I graduated medical school and went and did a residency in family medicine. I was one of the last few people to set up a private practice in a major metropolitan area. I had a private practice with a friend of mine in the city of Chicago in the urban area. We were private.

The point you're going to see as I go through this journey is that I've touched in many different arenas that are relevant to many of the people that we end up interacting with. So, I have experience in private practice. We were eventually purchased by a large hospital system, so I have that experience. I was faculty at two medical schools. So, I understand academic medicine and teaching as well. I like to joke, I was sentenced to be a CMIO at a hospital and got paroled for good behavior after meaningful use ended. All kidding aside, that was incredibly useful experience. It was back in the days when CMIO was a very new position, most people didn't even know what it stood for, yet meaningful use, whether you look at it as a carrot or a stick or both really was effective, but it really was important. So, I learned a lot during that. After that I went to work for an actual EHR company as their chief medical officer that focused in emergency medicine, and I was working in emergency rooms at that time. That's the end of chapter one or book one.

And then, I came to IMO almost by accident. Very interesting. I've been at IMO for about 13 years, IMO Health. And I came in originally writing content, so I like to joke I started in the mail room. But it gave me a very deep understanding of clinical content, going back to our first question and understand that the 18,000 terms of ICD are not enough richness to describe medicine. And in my first early days at IMO Health, I was, "Don't we have all the terms already?" And we're approaching in the millions and you really see that the descriptive medicine that we need to capture is really quite extensive.

 And over time, I've moved through IMO all of its corridors, and now I lead a team of incredibly talented clinicians who share stories like mine who are both an internal soundboard as well as an external-facing group, nurses, doctors, clinical coders who could walk into a room and say, "We understand your problems and we're going to make sure that ours work for you." So, that was a bit of a longer answer probably than you were expecting. But it does kind of paint that picture of who I am.

Host: Yeah. Well, and to your point, I mean, just that all of the different inputs and all of the different experiences have shaped how you work with this information now. So, I think it's brilliant. Thank you. So, back to the subject matter and pulling all of these pieces together, what does successful AI integration into clinical workflows look like? And how can hospital leaders avoid disruption to care delivery?

Steven Rube, MD: Well, that is one of the million dollar questions, although there are several million dollar questions. The moniker that doctors are averse to technological changes I take offense to, I don't think that's true. But doctors ask two very simple things: make me better and make me more efficient. And part of the make me more efficient is make these solutions into it. That's very, very important.

The actual method of the doctor-patient relationship is very, very old, and it is very effective. And what we have to try to do with these solutions is to enable that interaction to be better. And I think we've kind of gotten away from it a little bit. And I'm sure we'll speak about more of doctors as data entry people, but the idea now that the simple idea that doctors have their back to patients while they're typing into a computer, that they're no longer making eye contact, that the patient feels like I'm not being listened to, that clinical notes become very generic and very copy and pasted, if you will. All of these things are taking us away from that original two people having a discussion and understanding what they're saying.

So, these solutions, they need to work. They need to be incorporated into a workflow that clinicians-- and I use the word clinicians and not just doctors because it's really anybody who interacts with the patient. So, nurses, therapists, dieticians, anyone who's going to have that interaction with the patient needs to be able to maintain that while these systems integrate almost seamlessly. Someone once asked me what's the best way to have one of these systems appear? And I said, it should appear to be invisible. It should operate cleanly and in the background and only come out of that background when asked to, kind of like our parents used to say to us when we were children. So if a system can do that, I think, it'll really be effective in getting us back to that doctor-patient relationship that really makes a patient feel comfortable in that situation.

Host: That's wonderful. So, what are some effective ways to reduce clinician burnout while driving trust and adoption of AI across care teams?

Steven Rube, MD: Right. Another very pressing question that comes up at almost every discussion around hospital staff. Clinicians and, in this situation, I will say the word doctor because it's what I have the most experience with. We not only don't want to be data entry people. We're very, very bad at it. Even the ones who tell you they're good at it, they're bad at it. We have too many other things going on. We're thinking about too many other things. Most of us did not take typing classes. We speak in shorthand to each other or to ourselves. So, we are not good at data entry.

And I'm not going to say that data entry and the advent of the electronic health record is the only contributor to physician burnout. Clearly, there are many things from social, political, economic, regional issues that are weighing on the healthcare system in general, and physicians in specific. But certainly, myself ending a day knowing that I have 14 notes to write on patients that I may not remember exactly everything that I did. And on my two finger typing style, I'm going to be here until 10 o'clock at night and not see my children at dinner. That is a major contributor. And I can say from personal experience as a patient, that going into a doctor who's even testing a new ambient AI system, looked at me with a smile from ear to ear and said, "At the end of the day, I have no notes to write."

Now, the caveat to that must be attached, that the notes need to be quality and go back to the first question you asked me, they must preserve the clinical intent. It cannot be just a generic list of stuff, and it can't be an incredibly dense dictation-- or transcript is a better word-- transcript of our conversation. "How are the dogs? How was your trip to Europe?" And so, it must be a note. A note that is good and useful and accurate and specific. That goes back to your clinical intent question. I think we're approaching that, and it will be something that I think will help relieve some of the pressure in the system from a clinical sense.

Host: So, thinking about metrics, which metrics should hospital leaders prioritize to evaluate AI's impact on data quality, clinical outcomes, and operational efficiency?

Steven Rube, MD: Sure. I think the metrics have to be many. And I think you can divide them really into two categories, the subjective and the objective. And can we turn subjective feelings into objective data? And we just touched on one, are physicians happier with it? Are they able to, you know, finish their day? Are you able to recruit a higher level of employee because your systems themselves are not obtrusive and driving people away? Can you draw people in by saying, "We have this system and it will make your life better"? So, that is a metric that is a bit gray. It's hard to capture that, but you can. You still can understand physicians' attitudes towards their workplace environment. So, I think that's one.

I think another main one is, again, as we touched on earlier, there are metrics in medicine that were created for different reasons when you get into value-based care. We talk about describing patients accurately and specifically, and one of the things that I've seen done is-- without going too deep down that rabbit hole of HCC and RAF scoring-- can we look at how sick patients are being documented and are we capturing how sick they are compared to how we were in the past?

Believe it or not, that can be done with metrics. As I said, there's something called a RAF score, which you could attach that is meant assign multiplier to how sick a patient is. Well, if that same patient six months after installing an ambient AI solution looks much less sick, they either got dramatically better or we're not capturing the full extent of their diagnosis. These things need to be tracked. And then, I think you just need people looking at these notes and saying, "Are we doing a better job than we were back in the day of dictation or back in the days of handwriting?" Maybe nobody could read my handwriting and maybe the chart couldn't be anywhere at once, but I knew what I was trying to say and I captured it very clearly. And we should be able to replicate that with these systems.

Host: I think adoption, implementation and just the improvement across the board, it's really at the early stages. So, what excites you about this work?, The future of AI and driving innovation and improving point of care?

Steven Rube, MD: I think what really excites me is I think we're at an inflection point that is similar to when we went from paper to electronics. And what I mean by that is, while that was a dramatic step and the improvements are absolutely documentable, error rates, warnings, charts being able to be in an infinite number of places at one time. How many times did I round and the chart was in radiology and I wasn't able to see it? That was a massive change.

I think where we are now is at a point where we have to stop thinking as the electronic medical record is just the paper record, but on a computer. And what I mean by that is I think that these AI systems will be able to-- without getting too geeky informationally-- they will phenotypically or a better word is to accurately describe a patient in real time to the doctor and present that information to them at real time when it's needed.

So right now, we live in a world where it has to be in the doctor's note. If it's not, I have to search through even the electronic record to find it. It shouldn't be that way. It should be presented to me and understand that if I am a cardiologist versus an orthopedic surgeon, this is the information that describes this patient this minute right now. Whether it's in the electronic medical record or not, it could go anywhere, it could go to the pharmacy, it could go to previous notes, it could go to previous dictations, other visits. And it can be presented to me in real time. So, I have the best information to take care of that patient.

I think we have the technology to do that. I think there are people thinking about the interaction of the doctor, the AI system, and the patient that way. And I think that's really what excites me to go from not only generative ai, in other words, generative AI gathering information, but agentic AI where now I have an AI agent, If you will, that's helping me, that's creating differential diagnosis, that's going through the medical literature and letting me know what is the most up-to-date version of treatment for any specific disease and being able to apply that quickly and efficiently.

I think we're on the precipice of being able to do that, and I think it will really unlock a different age of medicine for us. I know that sounds, you know, a little bit greeting card. But I think it is true. I think we were on our way to seeing that become a reality.

Host: Absolutely. I'm right there with you. Dr. Steven Rube, thank you so much for joining me on the podcast today and sharing your takeaways with AHA members. I really do appreciate it.

Steven Rube, MD: it was absolutely my pleasure. Thank you for having me.

Host: For listeners, if you'd like to learn more about IMO Health or anything that you've heard on this podcast today, please visit us at sponsor.aha.org.

This has been an AHA Associates Bringing Value Podcast, brought to you by the American Hospital Association. Thanks for listening.