May 8, 2023: Dennis Joseph, Sr. Director - Product Management & Healthcare Practice Lead at Digital Scientists joins Bill for the news. What are the latest technologies that Digital Scientist is implementing in healthcare, especially as it relates to telehealth, AI, and getting devices connected? What are the current challenges in using AI models in healthcare due to the quality and availability of data, as well as potential biases in the models? What are the challenges in building a comprehensive patient profile, including genomic and ancestral history, and how can this be addressed in the future? What are the potential use cases of ChatGPT in medicine and wellness, and what are the challenges associated with them? Can ChatGPT be used to provide real-time evidence-based recommendations to healthcare providers to improve patient outcomes, and what are the potential risks and benefits of doing so? What are the limitations of ChatGPT in medical record keeping, and how can they be addressed to ensure patient safety?
Alex’s Lemonade Stand: Foundation for Childhood Cancer Donate
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 This Week Health.
I'm one of the few people who still has hope in value-based care. Yeah. I'm one of the few people who still has hope in accountable care organizations because it's in that type of care setup. It is in the vested interest of that ACO to manage the patient holistically. Yep. And the only way to manage that patient holistically is to have a view of their patient record, irrespective of the care setting. (Intro)
Welcome to Newsday A this week Health Newsroom Show. My name is Bill Russell. I'm a former C I O for a 16 hospital system and creator of this week health, A set of channels dedicated to keeping health IT staff current and engaged. For five years we've been making podcasts that amplify great thinking to propel healthcare forward.
Special thanks to our Newsday show partners and we have a lot of 'em this year, which I am really excited about. Cedar Sinai Accelerator. Clear sense crowd strike. Digital scientists, optimum Healthcare it, pure Storage Shore Test, Tao Site, Lumion and VMware. We appreciate them investing in our mission to develop the next generation of health leaders.
Now onto the show.
It's Newsday. And today we are joined by Dennis Joseph, senior Director of Healthcare for Digital Scientist. Dennis, welcome to the show.
Thank you very much, bill. Glad to be here.
So Senior Director of Healthcare Digital Scientists.
Talk to me about some of the things you're doing in healthcare.
Yeah, so we are basically a technology strategy and development firm based in Alpharetta, Georgia. So we essentially work with healthcare clients across acute and post-acute care spaces. We basically work as an innovation and sometimes a digital health transformation partner all the way from strategy to design, to development to transfer.
And so, a lot of the latest technologies that's coming into play as it relates to telehealth as it relates to. AI ML as it relates to, getting devices connected. We play in those spaces and yeah, excited to be here.
Cool. Well, we've got we've got a fair number of stories.
We've been talking a lot about chat, G P T, there's a couple of generative AI stories and those kind of things. There's cures, act some new things around that. There, there's plenty to talk about. Let's start with how health tech leaders compete in the Battle of Healthcare ai.
This is, David Chow wrote this. He's been on the show. It's been a while, but he's, he was on the show three or four times early on in our Life, and he writes about how tech leaders can Evaluate the technology, implement the technology. He has some quotes in here, but for the most part, he narrows it down into three things, integration and customization, performance and vendor support, and customer success.
You had a chance to look at this article. What are your thoughts on the article itself?
You know, we can talk endlessly about this topic. There is a lot of I would say Frivolous development going on in ai and healthcare obviously is a very serious topic. But the marriage of the two is really interesting and it leads to some very interesting propositions.
I think for me the true north is what are the outcomes that we are looking to drive, right? There are some serious issues going on with healthcare as it relates to health equity, as it relates to, managing patients outside of the acute care setting. And AI can really help. As you think about the quality of the data that's coming out and the availability of the data, there are some real issues there that's proven
Let's talk about the quality of the data.
Yeah. Right. I mean, essentially we've digitized all this stuff. And now we're laying on, we had clinical decision support and some really well written algorithms that we were putting on it to try to do some things. Now we're actually dropping AI on top of this, some narrow ai. Around specific use cases, maybe sepsis, maybe digital imaging, maybe.
So we have some narrow ai and then we have people talking about applying these broad general models like chat, g p T to healthcare. But first talk about the data. Are we at a point now where we feel confident in the data that we're going to be able to feed it into an AI model and get good high quality information and insights from the data.
I personally don't think we are there yet. I think we are making progress, right? But as you think about availability of real world data that is quality data coming out of our systems until we get to that point. All of these models are basically just a theoretical model, right? You wanna make sure that these models, when they are getting created, they have good testing that goes on real world data.
There's a good. Amount of sensitivity, specificity and we are not there yet, right? A lot of these models are not getting trained enough for us to get to a point where we are comfortable in terms of the recommendations coming out of it. So I think there's this whole bucket of issues as it relates to the availability of the data.
The other piece is the actual model itself. As you think about the bias that it develops over a period of time, a model that you develop on, let's say a population set that's based in Boston. And you lift that up and shift it to Florida, it doesn't mean that it's gonna work. And so as you think about responsible AI development, what are the companies doing to be honest with themselves in terms of making sure that these models meet a certain threshold?
I think there's a lot of a lot of distance to be covered to get to that point.
It's interesting because when I was cio back in 2012 we introduced this concept of the whole patient profile. And, as we were sort of walking down this path, there was one aspect which was getting all the longitudinal patient record, right?
So all the things we had about the patient into one location, which was. Herculean effort in and of itself so that we wouldn't have to ask 'em the same question over and over again so that our call centers would know who they were and those kinds of things when they called in. So we had that aspect of it.
But one of the things that came up pretty quickly as we were talking to clinicians and we were building out this whole patient profile is how complex the person is and how little information we really do have about the person. Like in a lot of cases we didn't have their genome, and in a lot of cases we didn't have their ancestral history.
In a lot of cases we didn't even know where they lived. We didn't know if they lived in a food desert. We didn't know. In some cases we didn't know their education. How educated are they? What's their job? All these things that contribute to the health and wellbeing of an individual we didn't have.
And so we stepped back and said, can we really call this a whole patient profile. We're making decisions on this, and we only have this much information about about dentists, and yet we're making these pretty dramatic things now around healthcare. We have enough information around healthcare, hopefully, that we're making decisions.
We're gathering all the variables we need in order to make a diagnosis and those kind of things, but it's interesting as these AI models are getting going. I'm wondering if we can paint enough of the picture that if they're summarizing my medical record, they're going to do it effectively or it's gonna do it effectively.
I'm, as I'm ascribing it, a human characteristic. But you know, it's, no, I agree.
I think, a classic example is, think of a patient who has congestive heart failure. This patient gets stabilized out of the ICU and is eventually discharged and comes home. Well, home is a very dynamic setting.
You could have one Thanksgiving dinner and one too many chips and your sodium level goes up. It starts retaining water, and suddenly it's hard for you to breathe. And now you're very close to getting, getting back to the hospital. And so as you think about that holistic view of the patient, which not just includes the C H F diagnosis, but it also includes what are the other comorbidities that this patient has?
What are all the medications and are there any drug interactions? What are they taking on a day-to-day basis? Right? And so just staying on top of that, And getting almost, one step ahead in terms of, Hey, is this patient showing signs of exacerbation so somebody can intervene? I think that's where it becomes really powerful, right?
And we are not there yet. We are starting to get to a point where we are piecing those data points together, and the latest Cures Act helps with that, which is, Hey, can we have one single view of the patient, irrespective of the setting, right? That takes one step forward. But that just gives you insights.
But how do you collate that? How do put all of that together that actually leads somebody to actually make a decision on something where a decision needs to be made? And that's where I think, I'm hopeful, I'm truly hopeful that large language models multimodal learning, right?
These are all very interesting topics for me, cuz it has the potential to sort of summarize all, these huge. Varying pieces of information variables and really boiling it down to, okay, what does this patient need now? Yeah, is this patient stable and can we keep this patient stable?
So let me give you some of the onc ONC proposes new Rules for Cures implementation certification, so forth.
Here. Here's the highlights, making the electronic health record reporting program a new condition of certification for developers of certified health. It. Number two, expanding and modifying exceptions to information blocking regulations to support wider information sharing. So again, they're continuing to push this, get the information out there, get it in the hands of the patients, allow them the opportunity to utilize their information and empower the patient.
Next thing, revising some certification criteria, including rules around clinical decision, sport, patient demographics and observations. Electronic case reporting and APIs for patient and population services. Clearly that's a little vague I'd want to dig into that a little bit more. Next one, adopting the version three of U S C D I.
And I think that's important. They just version one had a handful of things and they just keep expanding that that core data set. And then finally updating some standards and implementation specs adopted under the certification program to advance interoperability, support, enhanced health IT functionality, and reduced.
Burden and cost. Let me ask you this, as a developer as somebody who's ideating new systems and whatnot, how much of a challenge is it to get to the data, to get to the information for the patient and utilize that information? Today, like right now, March of 2023.
I'll tell you that entire journey is riddled with challenges, right?
And when you think about the influx of wearables, that's, if it could be a blood pressure cuff, it could be a pulse oximeter, it could be something that's tracking your heart rate, your O two sat, there's incredible amount of data getting generated. But who's there to receive it on the other end?
And last of all, who's gonna make sense out of it? And what are the clinicians gonna react to? And so in that kind of a day and age there, there's very little being done. The other is all of the critical pieces of information that is in an EMR or a case management system. It's very hard to get to that data.
And so as you think about piecing all of those information variables together and to be able to have a coherent story of who this patient is, it's incredibly hard at this point. And so I feel that there's a lot of industry players who's solving it in pieces and parts, but there's really nobody who has a cohesive picture of who the patient is.
And I think they've got to make some significant steps. And hopefully this cures act, or the latest version of it, helps us to get close to it. But I think we're still pretty far away.
It's it's interesting to me how things don't tend to happen unless we have carrots and sticks because there's really no financial incentives for the health system.
To share the information. We were relying on the altruism of the health system to share the data. Because at the end of the day, in 49 states, the health system owns the medical record, right? Everyone likes to say, oh, the patient owns the medical record. Legally they don't because they didn't create it.
And yes, even though the information's about them, they didn't create it. Therefore, they don't own it. And since they don't own it, we're relying on the health system to share that data. Well, the only one who can make them share it. Is the federal government, right? So they're the only ones that can step in and say, okay, enough, these, this information's about them and it can be used for good if you put it in their hands.
And we still haven't created that economy. Where you and I are like begging for our health information, like we're pounding on the door and saying, give me my data as they say in the industry. Because we the economy hasn't grown around it. Like nobody, I'm not like signing up for a service over here where they're saying, Hey, give me your medical data and we will summarize it for you.
We will do these things for you. We'll do that. It's sort of a closed system. It's like, Doctors sharing my medical record with another doctor who sh and whatever. But my thought is if we get enough individuals with their medical record that there will be a, an economy that sort of grows outta that, where people go, Hey, do you wanna make some money by participating in this study just by giving us your data?
And I might sit there and go, eh, you know what? I care about heart disease. I have history of heart disease in my family. I'll give you my data for money, but that doesn't exist today. And therefore, Health systems make money from our data and they utilize it for studies and they launch companies like Tru Veta and whatnot, which are at billion dollar valuations.
But the individual still is not empowered with it. I'm sorry. That's my. That's my rant for the day on.
I keep ranting cause cause you're onto something. So I'm one of the few people who still has hope in value-based care. Yeah. I'm one of the few people who still has hope in accountable care organizations because it's in that type of care setup.
It is in the vested interest of that ACO to manage the patient holistically. Yep. And the only way to manage that patient holistically is to have a view of their patient record, irrespective of the care setting. And once we are able to figure that out, I think there, I think there's some hope there. So I know there are some ACOs who are doing a pretty good job with it.
There are a lot of a ACOs who are still struggling with it, but I'm still hopeful that once a c realizes its true potential of what it was originally intended for, I think we can hope, we can cut through some of that, some of those challenges that you just talked about. Yeah.
Alex's lemonade Stand was started by my daughter Alex, in her front yard. It By the time she was four, she knew there was more that could be done. And she told us she was gonna have a lemonade stand and she wanted to give the money to her doctor so they could help kids like her.
It was cute. Right? She's gonna cure cancer with a lemonade stand like only a four year old would.
But from day one, it just exceeded anything we could have imagined because people responded so generously to her.
We are working to give back and are excited to partner with Alex's Lemonade stand this year. Having a child with cancer is one of the most painful and difficult situations a family can face at Alex's Lemonade Stand Foundation, they understand the personal side of the diagnosis, the resources needed, and the impact that funded research can have for better treatments and more cures.
You can get more information about them at alex's lemonade.org.
We are asking you to join us. You can hit our website. There's a banner at the top and it says, Alex's lemonade stand there. You can click on that. And give money directly to the lemonade stand itself
now, back to the show.
At the risk of people getting really sick of this conversation, I'm gonna bring it back to chat, G P T.
And the reason I'm going to, cause I, there's an article out here that I feel is fairly reckless. In its assertations and I, I wanna run it by with somebody and it has the top 14 use cases of chat, G p T in medicine and wellness. And I don't want you to comment on the potential because I think each of these 14, you read it and you go, yeah, as a clinician we could use help in this area.
It would save me some time. It's pretty interesting. And whatnot. So I think each of 'em, from that perspective will be presented in a way that people go, oh wow, there's a lot of potential here. And we all agree there's a lot of potential. I wanna talk about the challenges with such things.
So ChatGPT, LLM,, large language model. Generalist for the most part. And it it hasn't been trained on specific healthcare data and whatnot. And these are the things that this person a search that it can do. We'll start with the first one. Virtual assistant for telemedicine chat. G PT can be used to develop a virtual assistant to help patients schedule appointments.
Absolutely. It's an administrative task. Receive treatment. And manage their health information. Maybe with the rise of telemedicine, many patients, this is how they interact. We're gonna go through a couple of these. What are your thoughts on that one?
I mean, the administrative tasks I get, because if they make a mistake, ah, the calendar appointment's wrong, it's a pain, but it's not life and death, but in healthcare, Sometimes these things have to be a hundred percent accurate. So, so what are your thoughts on that one?
Yeah, a lot of noise there. I think we are several years away from actually reliably using a technology like this in healthcare, especially when it comes to diagnosing and treating conditions. I think, we need to get to a point where there's some sort of control put in place where, We understand where the recommendations are coming from.
Is it a reliable, credible source? Is it research backed? Right? Is it evidence based?
Right? The transparency's not there. Like if it, yeah, let me go to the next one. Cause I think it's even more more pointed, which is clinical Decision. Sport chat, G P T can be used to provide real-time evidence-based recommendations to healthcare providers to improve patient outcomes.
Now the reason I think you get away with that statement is, It's providing it to the healthcare providers, not to the patients. So the healthcare providers can be that human check, if you will. A trained healthcare provider, right? It said, for example, ChatGPT can used to suggest appropriate treatment options for particular condition, flag potential drug interactions, and provide clinical guidelines for complex medical cases by providing quick and reliable support.
ChatGPT can help clinicians save time, reduce errors. And improve patient care, by the way? I don't I have no doubt that chat, G p t can do this, even do this today, but the error rate is 8.5%, or, error rate is 10%. Let's just say it's 10% is the error rate, and we have no idea what it did behind the scenes to give us the information.
As a patient are you okay with that?
Definitely not. And even if it goes through a provider, I think there is still a level of bias that could seep in over a period of time. And so I'm extremely wary at this point that any of that even suggests a recommendation.
And so again, as I said, it goes back to, it needs to evolve quite a bit. And I think it needs to be put through the same sort of rigor that we are putting AI ml Amen. And be able to kind of be confident that okay, the error rate, both on the positive side and on the negative side is up there in a way that we feel confident and it's not there yet.
Yeah, I, yeah, he goes on medical record keeping generates summaries and whatnot. It's important to note. That chat, G p t has no idea what it's writing like. It doesn't understand what it's writing. It links words together. And so if it's gonna start doing summaries, is it gonna capture the whole summary?
Is it gonna miss something? Is it, I mean, again, it's a really powerful model. Don't hear me. Saying anything, but I think there's a lot of potential here. But I think we're getting pretty close to dangerous recommendations here until we have that level of rigor, that level of testing. He has, but
I'll tell you, I'll give you a quick example.
One of the things that we are working on is, how do you bring some sanity as far as telehealth encounters are concerned, right? There's free flow conversation. It averages anywhere between 20 minutes to an hour. But when you think about that conversation and the amount of content that gets generated in that one discussion, right?
These providers, they're responsible for transcribing it and documenting it. And we've been working on a technology that does the speech to text transcription. It summarizes the telehealth encounter, and it also has a smart layer that basically figures out the coding based on the conversation.
Okay. That to me is. Somewhat of a safer area where you can apply these technologies, but as it relates to diagnosis and treatment, Way too early. Yeah,
I mean, he has medical translation, medication management, disease surveillance, medical writing and documentation, clinical trial recruitment creating symptom checkers, patient triage, drug information, medical education, mental health support remote patient monitoring.
Again I believe there's potential for chat G P T in all these areas. There's absolutely opportunities for generative ai. In all of these, and in fact, I could think of solutions that are being deployed that are narrow AI models that have been trained specifically on health information that can do these.
I think the next step that needs to happen is chat. G p t needs to go to med school essentially for and I don't know what that looks like now that. So OpenAI agreement with Microsoft. Microsoft has Azure for OpenAI, which is now getting integrated with Nuance, which will now. Be doing transcription which will be checked in multiple ways before it goes into the medical record.
And so that's a form of reinforced learning human based oversight and that kind of stuff. And it's gonna reinforce the model and make it smarter and smarter. Maybe that's the start of going to medical school. And just over time we're gonna see this thing get really good at understanding.
Medications, understanding, diagnosis, understanding images and imaging. I, again, we all see the potential and I don't want people to hear me as the, oh, don't use it. But we've gotta be really careful, I think, at this point in the in the game.
Yeah, no, I agree a hundred percent.
But I do think that true North needs to be the patient. As long as we can measure it from an outcomes perspective and constantly keep that as the barometer, I think we'll be headed in that, that, that direction. But to meet that barometer, I think it's gonna take quite a bit of time.
Fantastic. Let's close with this. I don't have another story. I'm just curious. The coding and development, one of the things chat g p t does is it documents code really well, I've heard companies tell me it's like, Hey we've already started to document our code using chat, G p t, cuz it can take spaghetti code and go.
Here's, here's what this code does and whatnot, and it does it very quickly. Are there other applications you're seeing in that area, or is it, are you still a little leery of utilizing it?
No, I think at least the feedback that, we are getting from the developer community is it does make them more efficient.
As it relates to checking the quality of the code as it relates to more efficient ways to code as it relates to areas of coding where they might not be aware of and they, it saves them time and effort to kind of research and cross that gap. And so from that perspective, I think from a coding perspective, it is.
Driving a lot of efficiency. Obviously you gotta be very careful in terms of the code that you're outputting. But I think so far I've heard positive things, obviously some watchouts there. But I think as you think about putting languages it's gonna blur the barriers in terms of, coders knowing specific languages.
And I think at some point you'll get to a coding community that's so versatile. In terms of the language and the platforms that they use
Are you worried at all, and this will be the closing question, you worried at all about your staff, like putting your company's code into this thing without like an agreement with open AI to, they're gonna be using your code so the next time they write code for me, they, your code might inform my code that's being written
Yeah, so we've got strict policies around leveraging these technologies from a coding standpoint. And so I think there's a high level of awareness in terms of the risks that comes with it, but is there a responsible way to understand certain aspects of coding that would otherwise take many hours, sometimes days to learn?
I think that's where there's there's some opportunities. So I think from a learning perspective, there's definitely potential. Obviously inputting anything that's proprietary is a big no-no.
Yeah. Dennis, I wanna thank you for your time and great discussion. I really appreciate it.
bill. Pleasure being here.
📍 And that is the news. If I were a CIO today, I think what I would do is I'd have every team member listening to a show just like this one, and trying to have conversations with them after the show about what they've learned.
and what we can apply to our health system. If you wanna support this week Health, one of the ways you can do that is you can recommend our channels to a peer or to one of your staff members. We have two channels this week, health Newsroom, and this week Health Conference. You can check them out anywhere you listen to podcasts, which is a lot of places apple, Google, , overcast, Spotify, you name it, you could find it there. You could also find us on. And of course you could go to our website this week, health.com, and we want to thank our new state partners again, a lot of 'em, and we appreciate their participation in this show.
Cedar Sinai Accelerator Clear Sense CrowdStrike, digital Scientists, optimum Pure Storage. Sure. Test Tao, site Lumion and VMware who have 📍 invested in our mission to develop the next generation of health leaders. Thanks for listening. That's all for now.