This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

Today on Keynote

(Intro)   I think we have to hold ourselves to a higher standard in healthcare. So we use words like responsible and trustworthy, those are really important words, but they're also really hard to do.

  My name is Bill Russell. I'm a former CIO for a 16 hospital system and creator of This Week Health, where we are dedicated to transforming healthcare one connection at a time. Our keynote show is designed to share conference level value with you every week.

Today's episode is sponsored by Quantum Health, Gordian, Doctor First, Gozio Health, Artisight, Zscaler, Nuance, CDW, and Airwaves

Now, let's jump right into the episode.

(Main)  All right. It's Keynote and today we're joined by Dr. Michael Pfeffer, CIO at Stanford Healthcare. And Mike, welcome to the

show. Thanks, Bill. Always a pleasure to talk with

you. I'm looking forward to it as well.

I just saw that you were on stage talking about, I assume you were talking about data and analytics at the Health Catalyst Conference.

Yes, that was great. The talk was about artificial intelligence, the possibilities, like we haven't heard about artificial intelligence before data analytics but really about, the potential.

And that's what I think is really exciting now in health IT. The potential is vast, really vast.

we are definitely going to talk about that some today. it would be crazy if we didn't talk about it some. I know some people think hey, the topic's over talked about, the potential is so vast.

That we have to be talking about it, I would think.

Agreed. And, there's so many unknowns that if we don't talk about it, we won't figure them out. I think we have to hold ourselves to a higher standard in healthcare. So we use words like responsible and trustworthy, and those are really important words, but they're also really hard to do.

And so we have to talk about it, right? And we have to get it right.

We'll definitely go deep into that. I want to start though, in the beginning we had our first conversation when you were at UCLA, I'd love for you to share a little bit about your journey into the health tech world and what inspired you to take the role at Stanford.

Yeah. I'm a hospitalist physician. It's really my, true love seeing patients, and I really thought that we could do better than paper back in the day, and that's what really drove me to health information technology and being an informaticist was how do I take The knowledge I have in being a physician and my background in engineering, process design, things like that, and add technology to that to try to make things more efficient, better, safer in healthcare.

And that's what really inspired me to go into IT and informatics. And I believe we've made a lot of progress, but we have a lot more to do. And, I love my time at UCLA. It's an incredible organization. It gave me a lot of opportunities. Incredible people there. And I've been there for 18 years and personal issues and an opportunity to be at another amazing academic medical center like Stanford right in Silicon Valley, who would have known that ChatGPT would come out and I'd be here.

It was a little bit fortuitous. Yeah,

excited that you're there because, you're right, it's right in the heartbeat of all the stuff that's going on. Plus, it's Stanford, right? So even the university itself, it's just a is a place that incubates such great thought and ideas and technology for that matter.

Yeah, absolutely. Every time I listen to a faculty member talk about the work they're doing, whether it's AI or not, I'm just blown away. one of the really cool things we do when we have a leadership retreat is we, our dean invites some of our junior faculty. To give like a little pitch on what they're doing and it's just mind blowing.

To hear some of the things that people are doing it's really incredible. Yeah, it's a great place to be.

Yeah let's see, from your vantage point, though, I'd love to understand what are the most significant tech advancements in healthcare today And how is Stanford positioned in that landscape?

obviously, AI is the talk of the town now. And, I have this way of thinking about it in terms of our evolution here. And, you can agree or disagree, but this is how I think about it. We spent probably the last 20 years in health IT digitizing the analog world.

And, We had to do that. We wasn't really transformative. We took things that were done on paper processes, and we translated that over, so we digitized them. And there of course were some transformational things, but overall we really started to hit a point where. We did all we could in digitization.

And so out springs digital. And digital's been thrown around forever. In fact, it replaced a lot of CIO titles and became chief digital officers. But I think now we have a foundational technology that can make healthcare digital. And just a perfect example is the ambient voice notes. We've been trying to do this forever.

You'd have, people would listen. People have, glasses on videotaping things. There'd be people transcribing things in the background. But now we're at the point where, it can, technology can listen to a conversation between a patient and a physician and generate a note. That's amazing.

With no human intervention that's digital. That's truly transformative. And those are the things that I'm really excited about.

So data becomes the building blocks of that. What data sets Are we going to have to tap into to make don't remember the exact word you used, reliable or trustworthy, something to that effect.

What data sets are we going to have to tap into and are we going to look at the national players to do this? Are we going to have to train our own models locally on our data sets? Are there data sets readily available to train these models?

All of the above. And I think that's what's really exciting here.

I think there are different models for different purposes. There are smaller models, large, obviously, GPT 4, GPT 5, I'm sure coming, that are just enormous. And what we've Steam, just internally, is the models, the very large models, really all perform pretty much the same.

It's really dependent on how you prompt them and how you integrate them into workflow. I think you're going to see all different kinds of models in this space. But, to really think about more the downstream problem, which is, okay, let's say you can predict something with a model. What do you do with that prediction?

And all of that is equally, if not more important than the actual model itself. Generative AI is really interesting because it generates things, text, mostly. So it doesn't generate the same thing twice. And so how you ensure that it is indeed reliable and trustworthy is an interesting problem.

But I think it's going to require all different kinds of data sets. If you want a multimodal model, you're going to have to learn from. Multimodal data.

it's interesting. We're advancing on this co pilot model and not to use somebody, somebody obviously branded the term, but that's the model that seems to be the most applicable in healthcare.

It's the AI comes alongside. analyzes something, gives information. There's a big push towards transparency at this point. with the physicians and the clinicians, are they looking for just I don't know, like a button they can click to say, hey, this was generated via AI, here's the model.

is that how we're going to provide the transparency for them? The patient's a whole other thing, but for the clinician how do they know how the model came up with the information?

So we've been doing AI for a long time. Drug interactions are a great example, right?

Yes, they're rules based artificial intelligence, but nonetheless, they do work that we cannot possibly do when you're looking at all the medications and determining if there's a drug interaction. And so we've learned a lot from them. They don't perform very well and they're overwritten very often.

So, I think it's going to be model specific, prediction specific, so to speak in terms of what it is we display. And that's where, implementation science and informaticists are really going to shine in this role. I don't think it's going to be a one size fits all. But let's take an automation example, right?

The revenue cycle is something that we should be able to automate in its entirety. Do we need to really show physicians all the billing codes that we submit if The AI is just as good? Probably not, you know, but if we are going to augment their workflows and combine very sophisticated data from the patient to deliver prediction, then we're going to have to show where that prediction came from, what to do about it.

We can't just display that. So I think it's going to depend on the use case in every case. That's really the basis of a responsible AI lifecycle. We built one here at Stanford. It's to really understand all of those pieces. And then what do you do once the model goes live? How do you monitor

it?

let's go through a couple of the use cases. You mentioned the ambient intelligence, ambient clinical listening being probably one of the most prevalent use cases. And I think the thing that's interesting in that space, a renaissance that's going on because this has been around for a little while and now it's not just one company doing it or two companies, there seems to be a bunch of companies doing it.

So there's pressure on the pricing and it seems to be much more prevalent now. People are rolling it out to more parts of the organization. It seems to be more trustworthy. I capturing? Why we're seeing a renaissance of this

area? Yeah, I think so. think, it's never been scalable in the way it was before.

And this technology is now more scalable. It's still in the beginning. The models are gonna have to get better. They, some specialties works better than other specialties. You could think about all the different kinds of healthcare workers that could potentially use this, from social workers to physical therapists.

All of that is gonna take time for these systems to generate notes that are worthy of what you want. What we've seen is it's across the board how people take to these. Some people don't like it at all. Some clinicians don't like them to use it.

Some love it. Some have time savings. Some don't. Doesn't matter necessarily. People who don't have time savings may still love it because it reduces cognitive burden. It's all across the board. But I do think they're going to get better quicker and learn. And, what I'm really excited about, obviously, the idea of, reduced time is important.

But also think about better capture of Data from the patients, better histories of present illnesses which I think will be really interesting we'll get better small data using these kinds of technologies. So I'm really excited about that but, at scale, they don't perform yet as well as we'd like them to.

a former CIO, I can say this. I don't want to get you in trouble for saying it, but the consistency and the quality of data with ambient clinical listening will be probably greater just from across the board because, clinicians didn't Go to med school to become clerks and capture all the information perfectly and those kinds of things.

They capture the information they need to perfectly, but that whole clinical history area is really interesting to me to get a consistent documentation of the

history. Yeah. I always believe that the notes were there to help you think If you go way back to medical school for me, we wrote.

to think about differential diagnoses and plans kind of put all that together. And then it moved really into the billing space, which, you need this many points for this and this many points for that. And luckily we've moved past that mostly, although cultures are to change. So most people write the notes in the same way to now a point where the act of typing and transcribing and kind of.

Using the subjective data and objective data with this technology allows you now to really focus on the assessment and plan part of it, or the impression and plan. What's wrong? As opposed to capturing the data. I agree with you. I think the capture of this subjective and data will be much more accurate and rich, which we will then be able to use.

So I'm going to ask you about AI in the administrative area and even the IT area, but I want to stay in the clinical area. Where else are we seeing Artificial intelligence make inroads into the clinical side?

On the generative AI space, I think you're seeing automation tasks happen. So in basket message draft creation, for example we talked about the ambient notes potentially generating HPI questions and having patients answer questions to generate HPIs, but kind of think about.

Automation of it doing tasks that, clinicians can do, but, AI could do it faster, better, easier, things like that. So I think you're going to see that more than you're going to see augmentation. This is where, the AI is making more clinical decisions for you. You'll predictive predictions we're already seeing, and I think that's going to grow.

And I think generative AI making medical decisions is a little ways off.

So we talked about this on a webinar last year. I think you came on one of our webinars and we were talking you, I, Chris Longhurst, and I think Brent Lamb with North Carolina, and we were talking about the the adoption of this.

And I started playing around with the AI generating drafts for me to respond to emails. I wasn't all that comfortable, like I was reading them and I'm going I just wasn't comfortable. And maybe that's just a training model and I have to work with it more. What are you finding with the

physicians?

Very similar. The models aren't trained to sound like you. It's not you. It's a draft. And. The idea is you make the draft your own. Some things won't require, changing because it's a pretty straightforward answer. But, you don't, I think for the most part, clinicians want it to sound like who they are.

And I think that will get there. I really do. And you can imagine as you learn how each clinician wants to what's the clinician's voice, then the drafts are going to be much more relatable.

  In the ever evolving world of health IT, staying updated isn't just an option. It's essential. Welcome to This Week Health, your daily dose of news, podcasts, and expert commentary.

Designed specifically for healthcare professionals like yourself. Discover the future of health IT news with This Week Health. Our new news aggregation process brings you the most relevant, hand picked stories from the world of health IT. Curated by experts, summarized for clarity, and delivered directly to you.

No more sifting through irrelevant news, just pure, focused content to keep you informed and ahead. Don't be left behind. Start your day with insight at the intersection of technology and healthcare. This Week Health. Where information inspires innovation. 📍 Increase 

I'm curious, computer vision, I'd be remiss if I didn't ask somebody from Stanford about computer vision.

Will we see that in healthcare and where will we see

it? Oh yeah, funny you mentioned that. We just had a press release one cardiologists, electrophysiologists, was the first to use an Apple Vision Pro spatial computing in the OR. And it was pretty cool. A big shout out to our biomedical engineering team that pulled that off.

But it was a proof of concept, so it was still very early of the idea that, you can use spatial computing devices to display information in a way that is accessible to surgeons in the OR in a different kind of way, you can move things around, you can grow them, without touching anything.

So you can really customize how you want it to be. So it's really exciting. So I think that is in its infancy right now in terms of what. What we can do. I think the technology is Gen 1 and it's not cheap. How we use it, I think we're still trying to learn.

But I do see the OR as a, initial use case.

you put one on yet? The Vision Pro?

I have.

Yeah, it's interesting. It has six cameras on the outside. Now, when you think of through what it can do with those cameras in the OR, it can really zoom in on things and bring that right to your vision.

I think you hit on one of the big things when we were doing pilots in our OR the surgeons would always say to us, it's I can't touch the keyboard. I can't like, And so we were using things like the original PlayStation had this visual thing. We were playing around with that. We, voice, we use voice extensively.

it's interesting you brought up that case. That'll be I think, it will be interesting if they can, look at the record, look at the vitals, look at all that stuff without losing their focus on the patient. That's a big win. Yeah,

it's very exciting.

I again enormous potential. We just have to figure out, the right use cases where it does provide value.

The administrative side and IT side, are we seeing start to penetrate, obviously on the revenue cycle side, I think it's been there for a while and the automation is there as well.

Are there other areas we're starting to see it?

I think. The rev cycle, for sure, is where I'm seeing a lot of it happen. I think that, we'll start to see more and more in this space around probably scheduling. That's always, if we could automate that even more than we can today, I think that'd be very useful.

So we're looking at use cases on how to do that. It's really, can you go from referral to the right physician without a human needing to touch it and then have a patient schedule the appointment online? And, that's a lot of work because you really have to understand physician preferences, which Right patient, right time, all of those things to get it right.

So we're doing a lot of work on that to see, how we can use the technology to do that. So I think those are going to be some big areas. Robotic process automation, RPA, we've been talking about that for a while, and it really hasn't lived up to its potential.

There's a lot of reasons, but pair that with AI, and I think you're starting to see the next gen. RPA, which is holding more promise, for sure. Again, like I think the transformation to digital is possible and automating the things that physicians, clinicians should be doing.

Yeah, it's it's interesting over here.

We've automated a bunch of stuff, but when generative AI and the open AI API, sort of say became available, it really took all of our automations to the next level. Because it was like putting a thinking, reasoning entity in the automation, and that's usually where it broke down. It was like, oh, it, this moved, or it didn't understand this, and then all of a sudden, boom, it broke something.

And when a computer breaks something, it breaks something.

A great example is, chart summarization is a task that physicians can do, right? But wouldn't it be nice if we could automate great. Imagine it got 99 percent of it right and switched left for right or left out the word no in front of fever.

That's a huge change, right? It may be statistically a very tiny change, but clinically a huge change. So we have a lot to learn in this space as we move forward with this. Human in the loop AI, that's where we're going to be for a while, for sure, on things that relate to clinical decision making.

the clinical summaries while possible are still fraught with challenges. The most recent API, you can now feed it large amounts of data through the API, so we might even be able to get a full medical record in there. The problem is. We still have to check everything that's coming out on the other side.

So it's amazing. We have a lot to learn, but I think, the potential is there for sure. We couldn't do this in the past. Yeah, I think it's a really exciting, but we just need to understand that these are real risks. You cannot just summarize a chart by throwing it into GPT, saying summarize, and then walk away.

Because Again, it only has to switch one or two keywords and, or leave them out and it's a whole different clinical picture. So how do we understand that I think is going to be a challenge we need to solve for sure. One of the things we did here is we basically packaged GPT in our own wrapper so that it's secure.

So it's, safe for PHI and open it up to all of Stanford. Healthcare and the School of Medicine to try things out. Now, it's not for clinical decision making, but it's a sandbox and it's secure. And it's really exciting what people are doing. And testing and learning and coming up with ideas.

And I think That's going to be really important.

want to close out. I want to, two more topics I want to hit with you. One is leadership and then the second is data. Since you did speak at the Health Catalyst Conference, I want to talk a little bit more about data. And I always love those conversations with you.

But I want to start with leadership. You left a really great team at UCLA. And then I got at UGM last year, I got to meet. a significant amount of your team up in Madison. And love to understand your approach to leadership. I think one of the things that's distinct we've talked about is you still practice.

You still take a couple weeks. I don't know how often you practice, but you're like gone. It's Hey, I'm not here. You guys run the thing. And if you absolutely need me, contact me. I think that's. That speaks to a leadership philosophy or approach. I'd love for you to talk about that a little bit.

It probably runs

better.

Oh,

man. I'll paint a very simple picture. Imagine a jigsaw puzzle. We're all just puzzle pieces. My puzzle piece is no larger, better, bigger, shinier than any other puzzle piece. Every puzzle piece is needed to make the puzzle complete. If you think of it like that, I have a role to play in the organization.

To complete that puzzle. But so does everybody else and everybody else has those unique skills that, that do that. So if you build a team thinking about how do I complete the puzzle and each one of us has that unique skillset, the diversity of ideas then the puzzle's not going to fall apart if one puzzle piece is missing.

It's not, complete, but that's kind of my philosophy on this. I have incredible teams. Because they're all really talented in, in their area and their ability to expand beyond their areas. And then you layer that philosophy with fun. This is fun. This is a lot of fun.

And it's a privilege to work in healthcare. So it's just acknowledging that. It's acknowledging like what, how lucky we are to get to do this. To, support missions of research, education, and patient care. If one of you, if

one of your senior positions became available, would you be looking for?

Is there certain characteristics, qualities you're looking for?

Leadership, leading people, managing and leading people is the number one most important thing. Can that has to be your priority and then being part of a team, a high functioning team. This isn't about you, it's about the team, it's about the organization and those are the primary things.

I think the easier part is kind of the skill set, the technical per se. But it's really being that team player and being willing to be on a high performing team, which means there's going to be conflict. How do we do conflict resolution? How do we be transparent?

how do we think about the needs of others? Sometimes before our needs, how do we think about working in an interdependent kind of environment instead of an independent or dependent environment?

It's interesting listening to you. We were talking earlier about speaking at Health Catalyst and we'll get to that right now.

And you're like, I can't believe I'm on the stage with these people. And granted, there's Nobel Prize people and that kind of stuff. But I think one of the things that makes you distinct as a leader is you're the CIO at Stanford, and you're an accomplished physician, and essentially your ego quotient's pretty low for someone in that role, and you see the value of the whole of the team.

anyway, I'm not going to ask you to respond to that cause it puts you in a weird spot, but I will ask you, what did you say? You got on stage, you're talking to a bunch of clinical informaticists and data stewards and data analysts and those kinds of people chief data officers.

what does the CIO for Stanford talk to them about?

It's a great question. Um, I would say to summarize it is there's incredible potential. We have to work together. This isn't about who goes first necessarily, or, but this is, Such a huge opportunity for healthcare in this country that we need to work together and it's gonna be data scientists, it's gonna be data analysts, it's gonna be architects, it's gonna be technologists, informaticists that really, I think, drive The strategy of healthcare in the future because of all the potential we talked about.

That's, I think, in a nutshell, what the conversation was about.

It's interesting when we talk about generative AI, one of the things we digitized everything, right? So everything's digitized. But we put it in closets and squirreled it away over here and over here.

One of the things that generative AI can do is it can suck all that in and make meaning of it. Now, we've talked several times about the concern about making meaning of it, but just finding it all and pulling it all in. has value to start with.

And, I think it's really determining where you want to go.

And because this stuff can also get very expensive by the way, right? It's a lot of compute to cycle through all this stuff. It's really saying what are the problems we want to solve? And I know CIOs say that all the time. It's what problem are you trying to solve?

But the point is. There's so much we could do. We don't have enough time or money to do everything that everybody wants us to do. So you kind of need to prioritize based on value and what are the needs of the organization. And each organization is going to be unique in some ways and similar in other ways.

And so is important to really say what is the value of what we're going to be doing. Should we be doing it in the first place? ,

📍 So let's close this out. Let's, as we come to the end of this love for you to sort of summarize, key message, call to action you'd like to share with our audience, predominantly healthcare providers, primarily in the IT space some clinicians and others listening to this regarding the future of healthcare.

and technology and what's going on and where we can expect it to go.

That's a big question, Bill.

you're in the seat. You're in the seat where people are expecting, we're going to see the innovations come from the academic medical centers. We're going to see it come from the large IDNs because, you're in Silicon Valley.

I

think what's really exciting about the technology is It's more accessible than ever, right? And I think innovation here is going to come from everywhere. And that's, I think the message it's get involved, play with this stuff, learn about what these technologies can do and partner with informaticists and technologists to bring this stuff to life.

And there's going to be bumps along the way. And so feedback is going to be really important. And it's going to be humbling sometimes. And I think if we really do embrace this change. And these technologies, we can really make a difference in many of the aspects of healthcare that we've been trying to do in health IT, such as reduction of administrative tasks and burden and reducing the cost of healthcare, those important topics.

So I'm really, excited about, and I would encourage everybody else to be excited, cautious.

Mike, always great to catch up with you and appreciate you taking time to come on the show

Thanks, Bill. Anytime.

Thanks for listening to this week's keynote. If you found value, share it with a peer. It's a great chance to discuss and in some cases start a mentoring relationship. One way you can support the show is to subscribe and leave us a rating. it if you could do that.

Big Thanks to our keynote partners, Quantum Health, Gordian, Doctor First, Gozio Health, Artisight, Zscaler, Nuance, CDW, and Airwaves

you can learn more about them by visiting thisweekhealth. com slash partners. Thanks for listening. That's all for now..