DHiC 26 – Physician Perspectives on AI

This transcript was AI-generated and human-corrected. It may contain minor errors.

“When I saw my first AI-generated clinical note, it was as if someone took off a backpack full of rocks off my shoulders.”—Dr. Samuel Gareau-Lajoie

Katie Bryski: Hello, and welcome to Digital Health in Canada, the Digital Health Canada Podcast. I’m Katie Bryski.

Shelagh Maloney: I’m Shelagh Maloney.

Katie Bryski: AI is opening new doors in care delivery, but its true impact depends on how people experience and shape it. Today, we invite two leaders to explore how clinicians build confidence in new tools, why some solutions thrive where others stall, and what health leaders can do to bridge the gap.

And we are so thrilled to welcome to our virtual studio Dr. Samuel Garreau-Lajoie, family physician, co-lead at Care for Tech, and clinical lecturer, AI/QI lead in family medicine at the Université de Montréal, and Dr. Payal Agarwal, family physician, chief medical information and innovation officer at the Waterloo Regional Health Network, and provincial clinical lead, digital health, at Ontario Health.

Welcome to you both. Thank you for joining us.

Dr. Samuel Gareau-Lajoie: Thank you, Katie. Happy to be here.

Dr. Payal Agarwal: Thanks so much. Yes, happy to be here as well.

Shelagh Maloney: Let’s kick off. We have one question that we start all of our podcasts with, and people are really interested in leaders such as yourselves and what your career journey is. So Samuel, why don’t we start with you?

Tell us how you got to where you are today.

Dr. Samuel Gareau-Lajoie: So, uh, I’m a family physician based in Montreal, Quebec. Somewhere along the way of my, uh, 10 years career at this point, I became very interested in why so much of my work days were spent doing stuff that has nothing to do with my patients. So since I practice family medicine in a busy practice group, interdisciplinary team, uh, continuity patient, and chronic condition, uh, with all the complexity of primary care, I’m under a lot of admin burden, and that’s a bit what of led me to get interested in, uh, technology in the first place and how it can help me and my colleagues.

So, uh, that frustration came to its apex after the pandemic. Everyone was tired. I had a newborn child, and I was under crushing admin burden, and, uh, I took more patients, and I couldn’t keep going the way I was going ’cause I, I was burning out. First thing I did, I tried to master all the features in my EMR.

That helped, but it wasn’t quite enough. So late 2022, uh, the world changed. Gen AI, became available to the public, so, generative AI, like ChatGPT, and at that point, I was like, “Oh, this thing, it can make text based on instructions, and it could help me probably with my clinical notes.” So that’s how I got interested in AI scribes, and I had the pleasure to participate in co-developing Carrousel, which is the first AI scribe based in Quebec, so adapted to the particular bilingual context we have in Quebec.

After that, like, I participated in some research project, a teaching role in Université de Montréal, and, uh, that all started from admin burden, so thanks, admin burden.

Katie Bryski: I feel like that drive to turn frustration into improvement is a theme we’ve seen, uh, with other guests as well. Payal, how about you?

Dr. Payal Agarwal: Yeah, so I’ve had, um, a bit of a varied journey through digital health.

I actually, so even before medical school, I started at engineering. I did systems engineering, which, for those that might not know, but is very grounded in user-centered design. So I spent five years very, very grounded in user-centered design, and that’s something I’ve carried with me throughout the years.

Then went to med school, kind of forgot about engineering. Honestly thought I’d never use it again. And then kind of, you know, similar to Samuel, as I went out into practice, I started to realize, like, hey, there’s a lot of things here that I feel like could be done better, and the, you know, the engineering brain in me started to be like, along with seeing patients, I really want to start, you know, making the system overall function better.

And have since then have just been sort of finding different ways to hit that same goal. So for a while I worked for an EMR company and, um, sort of led product and built an EMR that’s now actually reasonably well used across the country. I have done research in primary care and digital health here in Toronto for the last almost 15 years now, and that’s been really exciting, ’cause that’s allowed me to really see a lot of different big projects, government projects, but also did some of the early evaluation on things like AI Scribe, virtual care over the pandemic.

So that’s given me a really nice sort of scientific structured lens on digital health. And then, um, increasingly moved towards operations. ‘Cause I realized, you know what, we can build the best technology, we can evaluate them, but are, a lot of our challenges just lie in getting them out into practice, getting them scaled, getting them funded, getting the policy levers moving.

I work, um, at the Waterloo Regional Health Network. I’m their CMIO, so I, I do a lot of day-to-day operations, how to leverage digital health to run a, a large community hospital. And then more recently, in the last, I think, about eight months now, have taken on this role at Ontario Health, which is very much now thinking across sectors, thinking, you know, a lot about, like, policy levers.

So it’s been an, an exciting journey, and I think it’s always, digital health is large and complex. You kind of need to, to sometimes work from a policy point of view, sometimes need to work from an actually building technology point of view, and that’s sort of how I’ve, I think, ended up where I have.

Katie Bryski: I’m curious, like Samuel, you mentioned this a little bit, like that encountering ChatGPT, you know, when it became publicly available.

I’m wondering for both of you if you have that moment when you first encountered AI in like a care delivery setting, and A, what that was like, and also like how it might have shaped your expectations and hopes for what it can do.

Dr. Samuel Gareau-Lajoie: When I saw my first AI-generated clinical note, it was as if someonetook off a backpack full of rocks off my shoulders.

That was back when I was still 100% clinical, so I was like seeing patients all day, all week long, uh, from morning to night. And I saw that, oh, I maybe won’t have to do like long clinical notes, uh, every day for every patient for the rest of my life. That was such a relief. AI scribes were not very widespread at that point, and I saw the potential and I was, “I need to tell everyone about that.

All my colleagues, they also need to have this relief.” I had a lot of hope, but in the usage in the few, few days after that, well, I saw the limits because there’s, there’s quite a few. There is like factual error, hallucination, the lack of context. So AI scribe, they’re not perfect, and generative AI, it is not a, a perfect technology, and by design, it is producing variable outputs.

And we have, as clinician, uh, we have to, uh, keep an eye on what it does. And so that’s kind of what I’m trying to teach now, both to resident and to a clinician that are in practice, that AI, it, it does help. It can, uh, accelerate, it can augment you, but the physician always have to, uh, keep in mind that it can be wrong, it can be misleading, and it can lead you in, in the wrong way, and that is also true for patient.

So I’m having more and more patient that come and see me and have, uh, have talked to Dr. ChatGPT before seeing me. So that, that is a, a whole new challenge.

Shelagh Maloney: Interesting that Dr. ChatGPT has replaced Dr. Google. You talked about it was an epiphany for you and life-changing and, and we hear a lot of those kinds of examples.

But, you know, there’s also issues and, you know, you said there’s hallucinations, there’s lack of context. And so same question to you, Payal, and really around what changed it for you? And I suspect and maybe going into sort of the next question is, some physicians don’t adopt it at all. Uh, some physicians may adopt it, and the first time something gets it wrong, they’ll completely dismiss it.

And so how do you build trust among those clinicians who are still very skeptical, and, and rightfully so, potentially?

Dr. Payal Agarwal: Yeah. Great questions. And I’ll just say, so for me, the aha moment, kind of similar experience to what Samuel was saying. For me, it was really when I did my research. ‘Cause, you know, I’m a, I’m a techie, like, I like cool technology, so I thought, hey, maybe this is just me.

But when we started, we, we got it in the hands of some doctors, and we specifically looked not just for, like, the typical early adopter profiles. The stories I heard coming out of it, right? Like, “This has changed my life. I am now home to see my kids and home for dinner in a way I never have been. It was the first time I took two weeks of vacation and holiday without ever opening my laptop.”

Like, in the research that I have done for 15 years, I said, on digital health, we have never gotten anywhere close to such a strong value proposition for such an early technology. And of course, it’s not perfect, and of course, it still need, you know, we need to change it, we need to think through policy and privacy and all of those pieces.

But to have such an early signal for a value proposition for your end users, you know, I had never seen before in all my, my digital health work. So that’s when I was really like, there is really something here and, you know, that’s really special and, and in a way that it hasn’t been before I think as you looked for adoption, and I would say AI scribes again more than typical, I saw the early adopters not like your typical early adopters of those people who like technology.

It was, you know, the busy parents, like Samuel said, just trying to, you know, get through their day and get home, and are actually very skeptical of technology and would only adopt it if it truly provided value. That’s who I actually saw a lot of the initial kind of uptake with, which was quite exciting and I think spoke to the strong value proposition.

You know, how to build trust. We did a lot of work to really understand people’s concerns and barriers, and a lot of it was like, my patients trust me, and their trust in the tool relate to the trust in me. And I think a lot of physicians took that to heart. So how do we help them, you know, maintain trust with their patients was really important.

So in Ontario, we did a lot of work to try and make sure we had a, you know, vendor of record that really resolved some of the issues around privacy, security, making sure we’ve managed where the data goes. That’s very hard for physicians to do independently, so we, we tried to do some central support to resolve those issues.

And then just, you know, traditional making sure the champions were heard. We had some really, really great early champions, you know, who could go out across a community and share their experience. But I think starting with listening to the issues and, and, you know, we had to get the CPSO and the privacy commissioner, and we had to get many people around a table to try and resolve some of the privacy concerns early.

And I think taking that kind of bold approach was really helpful instead of just sort of leaving it to physicians to try to figure out these privacy issues by themselves.

Shelagh Maloney: Yeah, and I think that’s where it’s, that leadership really comes in when you’ve got like every individual physician trying to address technology and privacy questions, and that would probably save them a lot of time and energy.

But, but, you know, I was thinking, Payal, about when we looked at EMR adoption, that took a long time, and like, like it was years, and then we, the rate at which AI scribes were adopted was like crazy fast compared to other technologies in a clinical setting. So that has to have some kind of power But, you know, it’s kind of interesting.

We had in one of our episodes Onil Bhattacharya and Simon Hagens, who you both know, who did those evaluations. And one of the questions now is, was that AI scribe adoption, it was sort of that Gartner curve. Are we at here? Are we at the depths of despair? Are we gonna see a plateau? What, what are your, what’s your sense about this whole AI adoption, and maybe not even just AI scribes, but AI adoption with clinicians?

Any thoughts around that?

Dr. Samuel Gareau-Lajoie: The curve, it, it, I think it is maybe, uh, shrinking or there is some, uh, displacement in it, uh, at the moment. So we have, yes, the early adopters, like probably me and Payal are very excited when new tech comes out and we wanna try it. But there’s also other people that are usually not so into tech.

And o- one example I, I, I love to speak about is one of my colleagues, which has practiced for more than 40 years and is in the end of his career, is almost 80 years old, and he hates keyboards. He really hates keyboards. And AI scribes was the perfect solution for him. He loved medicine. He loved practicing with his patient and seeing them, but he hated clinical notes.

And the shift from paper to EMRs, uh, in Quebec it wa- it was forced. So it was mandatory. You had to have an EMR. That made his day much less enjoyable. So when he saw my clinical notes and came to me and said, “Hey, you changed something,” I told him about it, and he was the second doctor in my clinic to adopt it.

We have this misconception and maybe, uh, uh, older folk are, uh, gonna, not gonna be, uh, early adopters of technology. But with AI, it’s kind of changing that, and that’s something I lo- I love to highlight to my colleagues. It is easy to use. It’s also easy to use it wrong, but it is much easier, uh, from my perspective, to use generative AI for multiple purposes in medicine, be it AI scribes or looking for clinical information or looking at clinical studies, than navigating some EHRs.

That’s something that’s, that’s gonna help adoption for sure. And we’re seeing that, uh, it, the adoption was very quick of this technology, and that was new. What is also new is that doctors took out their own personal credit card to buy technology. That didn’t happen before, so yes, clinics and hospitals, they’re gonna pay for, uh, EHR, EMRs.

But for individual doctor to see value in a technology and to take out their own credit card and say, “I need that so bad that I’m gonna pay for it, I don’t care,” that is unseen, and I think that’s, uh, that speaks loudly about AI and how it can help.

Katie Bryski: Super interesting. We had a question about, you know, where has AI solved a clinical pain point.

But, like, what I’m hearing from both of you, it’s really the value seems to be more on, like, the human side, right? It, it’s not forcing you to fit to a new workflow the way, Samuel, your, your colleague may have felt with EMR implementation. It’s something that makes the rest of your life that is not work, or not directly to work, easier.

Dr. Payal Agarwal: Absolutely, and I think that’s the power of AI in gen- like, generative AI in particular, because it sort of is more human and, and meets us where we’re at. And I think it allows us to, if we use it right, be more human in our healthcare. So the other big thing when we did the studies on AI scribes is people said, “It lowers my cognitive burden as a clinician.”

And the patient said, “I can see that I have much more eye contact and direct face time with my clinician. My clinician is listening to me, and they’re paying attention to me.” And that’s what we all want. That’s what we want as clinicians. That’s what our patients want. It, it stops you from having to be so clicky and, and face the computer, and go back to kind of why we went into healthcare.

So yeah, I think that’s absolutely the power that gen AI offers if we use it right, is to actually bring more humanity back into how we deliver healthcare.

Dr. Samuel Gareau-Lajoie: I’m definitely guilty of typing and, uh, speaking and looking at my computer for a whole five, six years of my practice because I didn’t want to finish my notes at home or on the weekend, so I was typing as I was speaking with my patients, and that costs a lot of energy.

So y- you spoke about the mental burden, the mental drain of having to type a clinical note, have your clinical reasoning going on, and maintain the relationship with the patient, and that led me to an interesting thought that people, they prepared, they use, uh, gen AI to prepare their appointment. They speak with ChatGPT.

ChatGPT is very, uh, all generative AI are very sycophantic, so they are telling you what you want to hear, and they are super nice about it. AI creates this relationship between patients and AI that they feel very heard and seen, but it also enables a clinician to be more attentive to them. So it, it kind of works on both hands, and we, we have to leverage that to be, uh, able to match Dr.ChatGPT’s very empathetic responses.

Katie Bryski: So that’s an interesting question then. Does AI, like, almost become the third person in the room during an appointment if they’ve got a relationship with the ChatGPT? I don’t know if you have a relationship with your AI scribe. Like, what’s that dynamic like?

Dr. Samuel Gareau-Lajoie: Yeah. The… It’s, it’s like a third person in the room, and patients, they come in and they are convinced they have a diagnosis, X or Y, X, Y, or Z, and they, they spoke with their AI chatbot, and now both the chatbot and the patient, they are convinced they have lupus. But AI, it can tell everything and the opposite.

So I have patient take out their phone, open their, their app, and just ask the chatbot, “Why wouldn’t it be lupus?” And AI is great at saying anything you ask it to say, and the response often helps me to deconstruct maybe s- some, uh, uh, some thoughts that have been, uh, wrong with- between the patient and the AI.

So that, that, that’s a kind of a life hack that I’m trying to, uh, teach my colleagues. You can use AI to reverse the effect of AI.

Shelagh Maloney: One of the things that I’m thinking about is we talked about an AI scribe is by far the biggest example that people use when talking about AI in clinical settings. Are there any other applications that you’ve seen that work very well and/or applications of AI that have not worked so well?

Dr. Payal Agarwal: I think where AI works well is, you know, administrative burden, right? You know, low risk, repetitive tasks that nobody wants to do. You know, I think that’s where it, it’s such low-hanging fruit. We see things around, like inbox management, you know, managing faxes and, and making that easier. In hospitals, you’re seeing a lot to support just the administrative tasks, writing minutes, things with finance.

Like, those are all very, very strong use cases. You are seeing a lot of clinician uptake on, um, I think as Samuel mentioned as well, like helping me search through evidence better, right? Like, we, we had our tools before. They were a bit clunky. To find a piece of evidence to treat a patient might have taken me five minutes before, it takes me 30 seconds now, right?

So I think that is where you’re seeing a lot of uptake. I think the place where, I wouldn’t say it’s not working well, but will require a lot more thinking, is the clinical decision support, right? The alerts, the, the sepsis tools. Um, you know, my hospital, we did some work on discharge kind of planning and discharge prediction.

And those pieces for two: one, there’s more risk, right? You are starting to predict clinical outcomes and, and there’s a lot more risk associated. But importantly, there is a lot more process and people change that has to happen, and that’s the part, you know, it’s all the stuff we know about digital health but sometimes forget.

And I think in those cases where, you know, around clinical decision support, we’ve got to make sure we’re pairing the great technology with managing the risk with the really in-depth people and process redesign. ‘Cause, you know, you’re not gonna realize a clinical outcome from a prediction a- alone. That prediction has to drive a change in how we care for patients and, and that’s not easy, so we have to always remember to put our effort in there.

Katie Bryski: You’ve both mentioned a few times, like, the importance of using AI right, and how easy it is to use it wrong, and we know there’s all the other change management pieces that come with any digital health technology. So as you think about implementing AI in a sustainable way, like, what do you think is really important for clinicians to keep in mind?

Dr. Samuel Gareau-Lajoie: It is to preserve their patients’ privacy and to keep their data safe. It has to be taught. Like, there are lawyers that got fooled by AI and came to court with fabricated judgment from AI. So it’s not because people are not intelligent, it’s because they did not learn, and they don’t know it, and it’s f- it, it’s very fresh and very new.

And I’m happy to hear that CPSO or, uh, the College des Medecins du Quebec are trying to create more of technological literacy and AI literacy, because that is something that has to be done. Someone has to do it, so, uh, the, the, the regulatory bodies can, uh, take position. But I don’t think that in the upcoming years, uh, a physician can say, “No, I’m not learning anything about AI.

I’m not using that.” That maybe can work today, but in five years, in 10 years, definitely not. So people have to have a minimal understanding of how AI works, what are the caveats, what are the limits, to be able to, well, use it to help them, to help their patient, but not to, uh, uh, cause, uh, any problem.

Dr. Payal Agarwal: I would just agree with that.

If I would… Like, I think we’re all gonna have a professional responsibility to learn about, you know, AI and have that competency, just like we do of the medications we prescribe or anything else. So I would just very much agree with that.

Katie Bryski: As we wind down the conversation, we have a new ending question for this next season of Digital Health in Canada that we’re very excited about If you were to look ahead three to five years into the future, what’s one hope that you would have for digital health?

And Payal, maybe we’ll start with you.

Dr. Payal Agarwal: I think I’ve said this before, like 10 years ago, and maybe now it’ll finally come true. I kind of hope we no longer call it digital health. It’s just kind of part of health. It’s another tool. It’s another way we do healthcare, just like anything else, instead of treating it so different and separate.

And I, you know, I do think AI gives us that opportunity, back to what we were saying. It, it can actually really meet us as clinicians and patients where we’re at. So I think that we’re no longer seeing digital health as still such a separate function in organizations and in bureaucracy, and we just kind of, it’s much more embedded in how we think about healthcare delivery.

Dr. Samuel Gareau-Lajoie: Yeah, I, I agree with that. So, uh, that, uh, clinical AI right now, it, it is its own conversation. There’s, like, champions and there’s committees, and, uh, like we’re, we’re speaking about AI in healthcare, and, like, we’re not in an echo chamber, but it, it feels like we’re a bit not part of medicine as a whole. Uh, so, uh, for the integration of both to happen in the next few years, uh, would be great.

The other thing I would hope for is that we are able to preserve, uh, humanity and the, the relationship between patients and, uh, physicians and carers. It’s easy to say, “Oh, we’re gonna send pre-questionnaires to the patient, gather all the data, and then the doctor says three words and prescribes something, and that’s over.”

But that’s not how it works. That’s not how medicine work. We, we need to have a relationship. We need to build trust. That is not possible if there is a super high interference of, uh, of technology between the caregiver and, uh, and the patient. My hope is that primary care is measurably less exhausting in five years

That’s something I, I definitely hope for, and that the tools like A scribes have been great at, but they keep removing friction and, uh, uh, addressing pain points and, uh, still are be- are properly eva- evaluated to make sure that they are, uh, safe to use. But, uh, uh, the, the relief, um, if it ke- if it can keep coming, we, we’re, we’ll take it.

Shelagh Maloney: Retaining that human side. I like that. That’s wonderful, and, and it’s so important. Healthcare, at the end of the day, technology can advance us and take us so far, but it’s that, as you said, it’s the personal relationship. When people come to hospitals and come to seek care, they are in a position of not being well.

And, and so to have that empathy and have that human side of healthcare is so important.

Katie Bryski: Thank you for showing us the humanity of AI today, and thanks again for your time. It was a great conversation

Shelagh Maloney: Well, that was a, a really important conversation to have with clinicians who are on the front line. It’s great to see those folks who have experimented with it. It’s changed their lives, and now they’re sort of become these evangelists for it, and really at a more systems level are sharing their experiences and expertise.

That’s, that’s really helpful.

Katie Bryski: I mean, I confess, I went into this, I think, with a bit of a clinical lens, no pun intended. But yeah, very focused on, you know, what are the clinical pain points, like that more slightly removed perspective. And I was really surprised and delighted by, yeah, how much of that humanity and emphasis on relationships really came through.

In hindsight, I, I suppose it makes sense. You know, presumably people become physicians because they want to connect and they want to help.

Shelagh Maloney: One of the things that I really was interested in seeing is how Payel in particular, but Samuel too in his teaching roles, have- Understood what the implications are around privacy and data quality and keeping the data secure and safe.

And those are the things that, you know, the average clinician I’m sure has no time to think about. Uh, but to have larger organizations taking on that role and that giving them advice around those kinds of things, I s- expect is very, very helpful.

Katie Bryski: I was reflecting, I’m sure most clinicians did not go to law school, right?

I’m sure some did. I’m sure some, uh, did both law school and medical school. But I think most clinicians became clinicians to deliver care, right? Like, I think in all of our jobs, everyone needs, uh, some working knowledge of data safety and data governance and cybersecurity. Like, I think it’s just the price of admission working with computers.

But yeah, I think it’s also a good example of how it takes many different perspectives and expertise and people around the table to make a solution work at scale.

Shelagh Maloney: You know, I listened to a presentation not too long ago, and the gentleman who was speaking told a story about, it was a, an older woman in, I think she was in Spain or Portugal, and she had a chatbot that she was chatting with and didn’t realize it was AI.

But you know, she at one point went into the clinic and, and she said, “You know, before I leave, may I see Evelyn? She’s been such a joy. She s- answers all my questions, and she asks me how I am.” And she had developed a real friendship and, and you know, “She has all the time in the world for me, and she’s interested in my grandchildren and asking me questions.”

And so she had really developed this relationship with this chatbot, who, chatbot who from her was, was Evelyn. And you know, when they explained to her that it was just AI and a chatbot, she wasn’t upset. She was just, again, she just had a great experience and, you know, she was surprised that it wasn’t a human.

But it, it just, but it is gets to that point, right, is that chat can be empathetic and can be those things that-

Katie Bryski: Or it can appear empathetic …

Shelagh Maloney: Yeah. But – Everyone has their off days, and chat can be wrong or can be empathetic. And, and I don’t know, would I be, would I be surprised if I, or disappointed if I found out that this person who I thought was a person is, turns out to be a, a, an AI?

I wonder. I’m not sure I, I would or wouldn’t be.

Katie Bryski: Depends on the relationship, ’cause I think many of us are hardwired to, when we’re talking to, you know, the chat support for, you know, a company, to say, like, “Are you a human or are you an AI?” Like, I think we, we expect it in certain situations now. But I think that story also just really illustrates the human drive to connect.

As humans, like, we just are wired to form relationships with things, whether they’re, I don’t know, animals or chatbots, and I think that’s also why a lot of people go into our line of work, so it all comes around.

Shelagh Maloney: Yeah, I was just gonna say, that’s probably the resounding theme of all the podcasts we’ve done, right?

It’s all about relationships and communicating and, and building that trust and all of those things. And, and it’s easier to do it with a human, I am sure. You know, can you tell now the difference between a human empathy and an AI-generated empathy? Likely, in many scenarios, you can. Uh, but maybe not always.

Katie Bryski: This is where we go into the Twilight Zone. But before we do that, we will be back next month with another podcast, not necessarily going too far into the realm of sci-fi. Until then, as always, you can check out lots of resources about AI and other things on Digital Health Canada’s website. Thanks for listening, and we’ll see you next month, right here on Digital Health in Canada, the Digital Health Canada podcast.

Thank you for listening to today’s episode. Be sure to subscribe to the podcast to get new episodes as soon as they’re available, and tell a friend if you like the show. We’ll see you next month. Stay connected, get inspired, and be empowered.