Intro: The following SHSMD Podcast is a production of DoctorPodcasting.com.

Bill Klaproth (Host): On this edition of the SHSMD Podcast, we talk about AI in Healthcare as we talk with Dr. Juan C. Rojas. He is the author of the article in Future Scan 2023, The Evolving Role of AI in Enhancing Clinical Care and Business Operation. Everybody's talking about AI right now. It's the perfect time to talk with Dr. Rojas about AI in Healthcare. So let us get to it, right now. This is the SHSMD podcast, rapid insights for healthcare strategy professionals in planning, business development, marketing, communications, and public relations. I'm your host Bill Klaproth. In this episode, we talk with Dr. Juan C. Rojas, Assistant Professor Department of Internal Medicine, Division of Pulmonary Critical Care and Sleep Medicine at Rush university in Chicago; about the evolving role of AI in enhancing clinical care and business operation. You can read more about it in Future Scan, 2023. Dr. Rojas, great to talk with you and thanks for being here.

Juan C. Rojas, MD: Thank you, Bill for having me and I look forward to talking a little bit more about AI and medicine today.

Host: Yeah, this is, the topic du jour, if you will. That seems like all we're hearing about is AI technology and it's going to be interesting to hear you talk about the role in healthcare. So let us jump in then. How is AI being integrated into healthcare systems now?

Juan C. Rojas, MD: The answer Bill is really, spreads the gamut of different things that healthcare systems are trying to do to hopefully improve patient care and hopefully improve operational efficiency for how they're delivering care. So I can give you some examples for my own work and I think are representative of what I think others are doing in the field.

Starting for I would say one of the biggest areas of interest for health systems and actually doctors and nurses taking care of patients is how can AI feed clinical decision supports tools that we use when we're taking care of patients in the hospital or in the clinic, that can help augment our decision making, not replace it. I think that's the important word, not replace it but to help really enhance or augment the way we think about certain diagnoses. So things that people are doing in the hospital that I've sort of lived and seen in my work as a critical care doctor when I take care of patients in the intensive care unit; one of the things that we worry about is a life-threatening dysregulation to infection that we call sepsis. And so a lot of health systems are trying to find those patients before they get sick enough to go to the intensive care unit and see if you can use AI and the signals of what patients developed this syndrome called sepsis and intervene earlier such that potentially we can get them the right resources at the right time, early in the work, earlier in the workflow than what you could have done with just routine clinical judgment and how hospitals work today.

And so that's one kind of example that potentially if done well can really identify patients earlier and maybe get them some lifesaving antibiotics and therapies earlier and may prevent them from getting as sick and having any morbidity or mortality from say, sepsis. That's like one big area of interest for me as a critical care doctor, but also I would say very on the top of mind for many health systems both because it obviously is a big driver of inpatient mortality for people who go to the hospital, these really bad infections that sort of get dysregulated. And also because health systems are judged against each other with they do for sepsis patients, and so it's one way of seeing well you're doing and if you can improve care, obviously, you might be doing it better than your peers down the street.

Host: So you said you use it to enhance or augment clinical decisions. Do you think that's how most healthcare organizations are using this now?

Juan C. Rojas, MD: Yeah, I would say that's a general thought on how we want it to be used. I think we're still trying to figure it out in medicine, how best to use it. But I think the general rules are, you want to be able to sort of identify patients who are at risk for certain things in the hospital or certain things on the outpatient side, earlier than you would've thought of it yourself as the doctor or nurse taking care of them, or at least equally as well. Right? So I think the key is trying to develop tools that really speak to the needs of people that are taking care of 20 patients a day in the hospital or taking care of 20, 30 patients on the outpatient side a day and making sure the things that you may miss in routine care, might nudge you in the right direction.

Say, oh, by the way, have you thought about this? Or this particular person seems to be at risk for Y. And then that may make you stop your work and say, Hey, that actually makes a lot of sense. Let me actually think about that and go down that path. And not necessarily at this point in time telling us specifically what to do as a doctor or a nurse taking care of a patient.

I think that might be AI 2.0 in the future. But right now, I would say AI 1.0 is really helping us figure out how can we better care and identify patients for risk of a variety of things in healthcare earlier in the trajectory of their illness such that hopefully we can maybe make outcomes better for those patients.

I think that's the aspirational goal. I think we have a ways to get there, in using these tools and figure out ways where both the prediction is accurate that the clinicians who are using the predictions actually buy into using it. And that there's transparency in that system such that the patients understand that there's these new models predicting a variety of things in the healthcare system where everyone's sort of on board with it and making sure that there's no distrust or fear in the system around how these tools are being used.

Host: So if used right, this can be a very valuable tool. So in the article in Future Scan, which you authored, you also talked about the role of AI in operations and patient safety. Can you touch on each of those?

Juan C. Rojas, MD: Operations, I think is another hot topic, especially for health systems that are increasingly in a current healthcare environment coming out of C VIDO with a lot of staffing issues, specifically around nursing and other areas, and really trying to find ways to increase access to care for patients that they care for in their community.

But also find ways to maybe automate certain things that maybe could save money for the health system. And so I think one example that's being used, I would say a fair amount and has been used already with varying levels of success, is trying to think about for patients who are admitted to the hospital for a variety of reasons, how long that patient may stay in the hospital, such that we can have we'll call it a marker if someone has an expected time in the hospital of seven days, are we sort of meeting that expectation?

Or is someone staying for nine days because things are running slow in our system? And so, that's what we're calling length of stay prediction. So can you think about kind of ways where when people are medically ready and they're better to leave the hospital to go to their next care setting, whether that be home, a rehab facility, that we as health system operational people can really make sure that the system is delivering care well and safely. That's the most importantly, but also efficiently such that things that need to get done for patients aren't taking an extra day or two because of a weekend or because that resource is only available one day a week.

And figuring out if you can use AI to help enhance that operational efficiency is definitely something I've seen in the health systems I've worked for, but also with my peers doing this work across the country.

Host: So it sounds like adoption is happening across the country. So what are the possibilities for the expansion of AI for patient safety and outcomes in the future?

Juan C. Rojas, MD: Right now I think we're in the phase now where we're trying to find a signal of the right people for the right diagnosis and that right time flow for their trajectory of whatever's going on with their health. But I think eventually as you've of seen in the press right now, in the last many weeks, or not months even large language models like ChatGPT, Bard from Google, others are really resonating with people in a way that potentially some of the other AI tools that have been used in the last five, 10 years by other industries and other companies maybe in a more relatable way because it's language, right? And so in medicine, not surprisingly, for those of you guys listening to the podcast that aren't in clinical medicine, the vast majority of data that we have on patients is still words.

And it's actually not an excel spreadsheet of different things that are going on with the patient, but really just how we're talking about that particular patient's clinical condition in words and, the words we use every day. And so I think in the future, what I expect to see is really an increased use of that big pool of data that we call unstructured healthcare data.

It's not in a spreadsheet, it's not something we can grab very easily. It's very raw. And using that to hopefully, use that unstructured data to sort of unlock potential patterns such in this case, for example, for patient safety; right now, for example, the only way I know a patient has an allergy is if someone took the time to use their electronic health record and mark it, you know, that patient has an allergy to penicillin. I shouldn't order it. Now my modern day electronic health records will tell me, Hey, by the way, Dr. Rojas, you're ordering penicillin. The computer tells me this patient's actually allergic to penicillin. Maybe you shouldn't do that, but what happens if someone wrote that down somewhere, buried a Word document, if you will, but it's not, hasn't really made it into that right registry that we have in the chart.

Then potentially I could order that penicillin not knowing I was doing any harm. And then that patient could have an allergic reaction, right? So that's one example where if we're able to look for signals of, say, for example, for an allergy with words as opposed to having to rely on a human to go put it into the right place in the chart every single time, could we prevent errors like allergies or anything else that might be a boon for patient safety, allowing us to not have to rely on the right nurse, the right medical assistant, the right doctor to put the thing where it needs to be every single time, because we do know that things fall through the cracks, right?

And so, even if people have the best intentions and people get busy, and they may have forgotten to put that allergy, that first, first time they heard about it. So can we use a computer that's been trained on language to associate penicillin with the word allergy, and then query me when I'm putting that order in saying, Hmm, it looks like I saw the word allergy next to penicillin two years ago. You sure you want to do this? And then maybe points me into that, to that documentation with a really easy to use hyperlink. That's something that in my mind would be really helpful, but just doesn't exist in today's modern day electronic health records. And I'm hoping as this technology evolves, both on the end user side, myself as a clinician, kind of gaining more trust to use it on my patients and with the vendors providing a lot of the tools we use to do the work we do as far as charting in the hospital really trying to see ways that they can innovate in this space to enhance patient safety as this space matures over the next couple years andkind of the next 10, 15 years, we have a different feeling EMR that we really feel like is patient safety first, clinician first, and really is helping us have less burnout in, in the clinical space, if you're not spending all your time charting and number two, you feel like it's actually helping you take care of patients better.

Host: Yeah, that is really interesting to hear you talk about this and what potentially could happen in the future and ways that AI could be used. So, I know you talked about operations a little bit, in Future Scan, you also talk about, how it could potentially optimize business practices. Can you talk about that a little bit?

Juan C. Rojas, MD: I think there are a lot of different use cases that come to mind. But, what I think about the most as far as business arguments or business use cases is a couple things. The biggest one is oftentimes in healthcare systems there are a variety of things that are machines that maybe are costly, that need to be running most of the time from a profitability point of view, and also just like a throughput point of view. One example of that is an MRI machine. Often that's a really high cost instrument to help get better pictures to make diagnoses for patients. But often maybe one or two of those for a five or 600 bed hospital. How do you think about schedule scheduling and prioritizing who gets to use that machine as fast as possible for the right sort of acuity of why you're ordering the test?

And right now, really the only way you can do it in healthcare is a combination of the doctor saying, Hey, this test is what we call stat in medical parlance, but really it just means it's really important, like we need it now versus sort of a routine test that can be done, you know, say tomorrow or the day after.

And then on top of that, as a selfish doctor myself, sometimes I'll be like my patient, Mrs. Smith really needs it and can you get her down as soon as possible. So I might, actually call two or three people on the phone, try to see where my particular patient is in the queue for that machine.

But boy, wouldn't it be nice if we actually had a more thoughtful way of scheduling than it was just first come, first serve, but really could use AI and, machine learning to think about who is most likely to benefit from getting the fastest result from an MRI like right now, and then making a work order queue be based on that.

Or maybe even including things like where the patient actually is in the hospital. How fast can we get that patient down there to the machine such that maybe you also think about geography, right? In a huge hospital that patients are closer or farther away and those sort of things a computer's really good at.

But sometimes it's harder for us as human brains to sort of put all those disparate points together into a summary benefit score for getting a test, right? So I think we'll start seeing more and more of those type of things rolled out or tested just to see can we make the burden of scheduling, and obviously there's someone, there's a job somewhere where someone actually has to schedule those things and boy, wouldn't it be great if that person felt like they had an assistant, as opposed to having to do it themselves hundred percent of the time. And so I think that is I'm hoping one of many operational use cases that I think people are trying to work on today and will continue to grow as this technology sort of evolves in healthcare.

Host: It's amazing how it can touch virtually all aspects of the hospital or healthcare system. So, Dr. Rojas, what about the risks of AI in its use for healthcare?

Juan C. Rojas, MD: That's a great point. I think that's my probably biggest area of interest in this space is to make sure that while I would say we're still on the early stages of how healthcare is going to adopt AI in its daily practice; I do want to make sure you know healthcare currently, as many of us know, both in healthcare and outside of healthcare, has inequities for a variety of reasons that sort of are outside the scope of what we can, I think this podcast today. But certainly patients who come from certain racial, ethnic backgrounds, from certain parts of the city just have different abilities to access care and therefore their health is impacted. And so, one thing I worry about is like any machine learning tool or AI tool that relies on data, and that data often has baked in bias into it. And so I think it's important as we train and learn from the data that we've developed over 10, 15, 20 years of having more we'll call electronic based medical records, that we acknowledge that that data has some baked in biases in it, and that some of the patterns that these models will learn, may actually not necessarily be reflective of biology or health, but really just someone's ability to get to the doctor, for example, on a routine basis.

And I think accepting that, that's number one. Number two I think is really the harder one is, what are you going to do with these powerful tools potentially? And making sure that the way I think about it is from like an equity point of view, is you want to make sure that in, in the current healthcare system, we know healthcare inequities exist. And what I don't want is someone who's really passionate about this is for the use of these algorithms, these models to potentially exacerbate or make those worse. And I'm hoping we can actually have AI and ML actually, maybe shorten that gap from patient A to patient B as far as health inequities go.

But you have to be conscious as you develop these tools, both from a development point of view, but also from a rollout point of view and what you're going to do with them locally in your own healthcare environment to make sure that you're able to potentially bridge that gap as opposed to make that gap larger. And that's not easy work, cause every use case is different. Every health system is different. But I think the key there is to have a local team of people really thinking thoughtfully about the local governance of these type of tools, right, and saying, okay, we're maybe have a team of a clinician both maybe doctors, nurses, and other allied health professionals, a data science person, an ethicist, even potentially, a diversity equity inclusion leader. Really having people that have different viewpoints of the world come together and say, Hey, we need these two or three AI tools that might really impact healthcare locally for us. How do we find those tools, source those tools, build those tools and then potentially roll out those tools such that all of us feel good about it?

Host: That is the concern. So thank you for sharing that. And what about physician and staff buy-in? Are people reticent to think about AI or is there pushback, or is there acceptance? What's your thoughts on that?

Juan C. Rojas, MD: Yeah, that's a great point. I would say it's a little bit of a mixed bag. I think those of us who are working in this space are really excited about its potential, but we right now acknowledge that the keyword word there is potential, right? I think now the onus is for us who work in this space in healthcare to really show that proof's in the pudding, right? We want to actually turn these on, turn an algorithm that's says for a specific diagnosis, say for sepsis, for example, and really prove that we identified patients earlier than we would have in a counterfactual world. And those patients did better.

There's starting to be some more literature in the academic journals around this, around making sure we have some rigor in these studies evaluating these algorithms just like we would've a drug or a device. And so I'm really hopeful that we'll continue to build that body of evidence that there's a signal of health or improving health as opposed to harm.

And then the second thing. So I think that's going to kind of get buy in from the kind of, we'll call it the general lay clinician staff, if you'll, but also I think the key is for the communication to be really important, right? Like I think the teams that are doing this work at each individual hospital or healthcare system have to be transparent with the people who are going to be using it.

And I think that's where making sure that people understand what went into some of these tools, like how they were trained. What data that was. Why the model, if it's a model or an algorithm is flagging your patient as being high risk for something, what we call machine learning and explainability, I think is going to be important.

So people can really feel that they understand the ins and outs of what's going on. Maybe not so much the math in the background, but more the like the Gestalt, does this kind of jive with how I think about my current patient that i'm caring for and we want to make sure that's done a way that sort of user friendly where they can just hover over their patient, they see a score oh, that's interesting. Let me take a look at that and dive deeper. And they say, oh yeah, that actually makes sense. That's actually value added as opposed to just being another click in the chart that they're trying to ignore. And I think as we think about how we can do that, both on the health system level and then more on the national level with this type of technology, I think we'll start seeing less concerns about is this AI going to replace me or, replace my clinical decision making?

I think the answer is no. Clearly it's going to, I think, hopefully enhance your decision making, but not replace you as the trained doctor, trained nurse, trained pharmacist, et cetera. And so I think that key is it hopefully going to help you do your work better, but not replace you. I think that's the first key messaging.

And then the second thing is really showing them, Hey we turned this on and then actually did something. Getting away from the hype and the buzzwords in AI and saying, Hey, like meaningful things got better when we turned this on because we studied it with some rigor. And then I think if you do those two things, that combined with transparency of with both patients and with the clinicians using it as the end users; I think that's what I hope is a roadmap to success where we can look back 10, 15 years from now saying this was the beginning of a really great time in healthcare where we were able to see improvements in, in relevant patient-centered outcomes and operational efficiencies that we just didn't have because we didn't have the same level of technology not that long ago.

Host: Yeah. Well you seem like an early adopter, so for you, this has really got to be an interesting tool, an interesting development to help you with your patients and enhance better outcomes. Is that right?

Juan C. Rojas, MD: Yeah, I mean, I, would definitely say given my interest in the space and what I do in my day-to-day, that I'm definitely fall onto the early adopter category. This has been what I've been thinking about for the last five or 10 years, really with a lot of the work I've been doing outside of my clinical time in the intensive care unit.

But that being said, I think this is the first time where I'm starting to see a signal of hope in other areas. Right. I had a patient I saw a few weeks ago in my ambulatory clinic where I see patients with lung problems as a pulmonologist. And I had a patient actually for the first time ask me if I'm going to use that new ChatGPT thing on them.

In that moment, it was a moment of brevity and laughter. But then I thought about it after I got home that night and I said like, I think we're not so far away from that. Like I think there are people already thinking how can we use something like ChatGPT to sort of augment the workflow for, patient to doctor communication. Right. Like, if I could open up a message from my patient and then I have a potentially a prompt to say, my patient wrote me a message about this question and then I can get like a two or three sentences drafted from me and I can vett them.

That allows me to not have to spend five, 10 minutes drafting a response to every message and makes my quality of life better and probably gets the answer to the patient that they want faster and right. So I think we are going to start seeing some of those things that have been pain points for either clinicians, patients communicating with their doctor. Can we use some of this technology to decrease the barrier to, something as simple as like an email message or a MyChart message if people who are using the Epic electronic health record, that's sort of a portal that I use in my daily practice. And then also things like communications about, you know, I have, as a lung doctor, I also take care of patients who often get, maybe I prescribe a medication and then the insurance company tells me, you know what that's not the right one. We want a different one. Or you have to tell me why that one's approved what we call a prior authorization. And so that's often me just sitting at my computer after I get home from the office and saying, I'm going to write a two or three paragraph novel on why I think it's really important for my patient to have that specific medication.

I would've loved to have a way to make that work easier for me and hopefully maybe even more compelling cause maybe my writing isn't as convincing as something that maybe I can tell ChatGPT

Bill Klaproth (Host): Right. Or

Juan C. Rojas, MD: to write.

Host: Or concise right. It, it can help you in all of that.

Juan C. Rojas, MD: So I think there are definitely ways things like something like a prior authorization that maybe doesn't seem like it's the top of mind for every doctor in the trenches and every patient taking care or being seen in the clinic or in the hospital. But those are things that are happening every day in healthcare that right now for often are happening on fax. If you can believe it or not. And it would be great to sort of move that needle a little bit faster than right now what was me printing off a letter and sending it via fax?

Host: Yeah. And that's a great story about your patient. Hey, are you going to use this ChatGPT thing? I love that. So you're right. There is an added side benefit to this, too. It can definitely be a time saver to help you get emails out quicker to your patients or to pharmaceutical companies, or when seeking prior authorizations, it can help you do your job quicker and faster, which allows you to concentrate more on the health of your patients.

Juan C. Rojas, MD: Totally agree. I think that's the hope. I think we have to make sure that we do it in a safe and an effective way. But I think, one of the biggest things that we're facing in healthcare is a crisis of burnout. And a lot of that for better or for worse is due to the burden of just the work needed to do what you think is right by your patients in the current system we have.

 And a lot of that is just mundane paperwork, to be honest. If we can make that paperwork just a little bit less painful for clinicians who are really trying their best to serve their patients, then that's something that I think it's definitely a value added and really important if we can find ways to do it.

Host: So we talked about how we're currently using it in healthcare, the potential down the road, the risks. So can you sum this up for us, Dr. Rojas, what are your key takeaways of AI in healthcare?

Juan C. Rojas, MD: I think the key takeaways is I think we are in the early sort of stages of how best to use some of this technology, to really, hopefully now for the first time, meaningfully change the trajectory of someone's health outcomes, which I think we're getting there and I think we're starting to see early signals of this technology really trying to move the needle in the right direction for patients.

And I think in parallel, I still think the key takeaways are for me as someone who's been working in this space now, is we just have to be cognizant to do it right the first time, and then not to have some stops and starts. And what I mean by that is really I don't want to see a year from now some terrible thing that happened either with patient health data or bad outcome because the model was wrong.

I think you have to remember, this is just like anything else we've had before. Before we had things like lab tests, a new CT scan that was new when we only had x-rays. This is just like another new thing in healthcare that we're going to have to find a wayto build into our current workflows but it shouldn't replace what we're already doing, which is just communication with your patients, making decisions with a multidisciplinary approach. Your patient is one of many things and a score or algorithm is just one part of that story. But there are some other things as a clinician you're going to take a look at to sort of make a decision. And i really want to emphasize that this is hopefully augmenting or improving your decision making, but it isn't going to be replacing decision making for people out there.

Host: Yeah, well I think you said it really well, this is for you using it to enhance or augment your knowledge or decision making, if you will. So I think that's probably the better way to do it. And as you said, you're vetting this process now, for its uses and how we can use it properly now and in the future.

So I think, that's probably the right way to look at it. So, great thoughts, Dr. Rojas, and I want to thank you for your time. Last question. Any final thoughts you want to add as we talk about AI in Healthcare?

Juan C. Rojas, MD: No, I'm just excited. I thank you Bill for having me today. I think, it's one of those things that we're like I said, a couple times already, we're in the early stages of figuring out the AI and healthcare journey, but, I'm really looking forward to being one of many people on that bus or train, if you'll and I look forward to really having more patients eventually that instead of being that ChatGPT thing, they really are seeing the benefits of how their communication with their doctor is, or an actual meaningful outcome that actually got better for them. And I, I so look forward to having a story where I can say, you know, Mr. Smith, I actually made a decision today in part, because I learned something from an algorithm that I didnt know before that is going to make you feel better in the future and I just look forward to having more of those feelings or more of those sort of insights where I can really feel like it's sort, making my job easier and safer for patients. It won't be an overnight journey, but I think there's enough of us out there that are really excited about its potential and now it's really transiting that potential into action.

Host: And what a powerful tool this is, and down the road to be able to say yeah I saved your life with an assist from ChatGPT. I mean that's pretty cool.

Juan C. Rojas, MD: I think we're on that journey now, just making that a reality.

Host: Yeah. Well this is really fascinating. Dr. Rojas. Thank you for your effort and time and the authorship in Future Scan 2023. We really appreciate it and in all your thoughts on AI in Healthcare. Thank you again.

Juan C. Rojas, MD: Thank you so much.

Host: And once again, that is Dr. Juan C. Rojas. He is featured in Future Scan 2023 with his authorship of the Evolving Role of AI in Enhancing Clinical Care and Business Operations, and you can purchase your copy at shsmd.org/resources/futurescan-2023

 And if you've found this podcast helpful, and of course, how could you not, please share it on all of your social channels and please hit the subscribe or follow button to get every episode. This has been a production of DoctorPodcasting. I'm Bill Klaproth. See ya.