Speaker 1 00:00:00 By automating those difficult, time consuming tasks. We're giving that time back to the physician, whether they need that just to recharge their batteries, or they want to spend that time with the patients.

Speaker 2 00:00:20 Welcome to Off the Chart, a business and medicine podcast featuring lively and informative conversations with healthcare experts, opinion leaders and practicing physicians about the challenges facing doctors and medical practices. I'm your host, Austin Littrell. This episode features a conversation between Medical Economics managing editor Todd Shryock and Mark Daly, chief technology officer of Digital Diagnostics. They're talking about artificial intelligence in medicine and why diagnostic tools might be something you want to consider.

Speaker 3 00:00:54 I'm here with Mark Daly, the chief technology officer of digital Diagnostics, the first company to receive FDA clearance on an autonomous AI diagnostic tool to discuss AI and medicine. Mark, thanks for joining me.

Speaker 1 00:01:08 Thanks, Todd. Excited to be here.

Speaker 3 00:01:11 So more AI is often seen on the administrative side of medicine with note taking or other kind of back office functions. But how is AI transforming the diagnostic side of medicine?

Speaker 1 00:01:23 Well, there's a lot of really cool things happening in the space right now, particularly in terms of image interpretation.

Speaker 1 00:01:29 So there's a lot of tasks that physicians do. I think folks probably most commonly think of a radiologist interpreting an x ray. and depending on what those diagnostic tasks are, they can potentially have a huge impact on the patient's outcome and their pathway and journey. But getting that initial interpretation, whether it's a diagnostic screening or some kind of other test, can really have a huge delay on getting the patient into the right place. So we've seen a lot of great opportunities in diagnostic image interpretation, both in the radiology space and where we are in ophthalmology, where there's a lot of screening tests and high volume procedures, where there's not enough, physicians available to be able to interpret the images in a quick enough time. And so there's a huge opportunity just from a sort of initial screening and placement for different types of diagnostic tests. and I think that that's really where there's a huge opportunity to help have a huge impact on patient outcomes. Of course, the challenge is that interpreting medical images is a lot higher risk, higher stakes process than doing a dictation or report that's going to be over read or reviewed by someone.

Speaker 1 00:02:32 So you have to have a lot more comprehensive, thoughtful development process. It's a lot more work to do, but you also get a much higher impact and potentially on the patients and the outcomes.

Speaker 3 00:02:43 Many doctors worry about AI replacing human judgment. How should they think about these AI diagnostic tools?

Speaker 1 00:02:51 That's a great question. I think that a lot of folks, I started my career in the radiology space, and I know that, you know, years ago there have been several famous predictions from folks in different industries about radiologists won't have a job in so many years, or there will be, you know. But I still have friends who drive their Teslas into work to interpret x rays. Right? So there's a lot of potential there. But I think that, the reality is, just like any tool, AI is something that can be used in a lot of different ways. And ultimately it is just a tool, though, in the toolkit to help people. And I think the way we're seeing it apply in medicine, like we're doing for diagnostic screening and digital diagnostics or other folks are doing a different radiology workflows, there's a lot of different ways to do it.

Speaker 1 00:03:33 And really what you're doing is instead of having physicians or other folks wasting their time on, activities that could be automated, we're having physicians do that, right. So that's where the opportunity is, is letting folks really work at the top of their license or spend the most time with patients in meaningful encounters rather than going through big basically sorting exercises, which is sort of how these AIS interpret images, right? You're looking at different patterns. Eventually you're going to make a decision based on more information than just that image, but that image is the key to unlock your pathway.

Speaker 3 00:04:04 What regulatory hurdles do AI driven diagnostics face, and how might that affect adoption and clinical practice?

Speaker 1 00:04:13 I think there's a lot of interesting challenges in the regulatory space, particularly for folks who don't have a lot of experience building medical devices or are coming from other industries where there's really rapid advances happening in AI. One of the key things that regulators do, for instance, in the US, the FDA, they have a mandate to not only make sure medical devices are safe, but they're effective and they're secure.

Speaker 1 00:04:36 And being able to prove that requires a lot of different technical documentation and process adherence and compliance, that folks who don't come from the medical device world are not familiar with all those challenges. Now, having said that, there's again, a lot of great tools to help with that, right? Even writing the software, you have GitHub Copilot. You have a lot of great AI tools embedded in quality management systems. So there's a lot of ways to sort of mitigate the challenges and the the large amount of work that regulatory has on building a medical device safely and effectively, but ultimately following those processes and going through not just doing sort of the bare minimum that you need to, but also thinking about the ethical considerations and the bias and that you might be introducing with training data, like in an AI system, there's unique risks particular to AI that are different than other medical devices. So the way I kind of think of it when I explain it to new hires is we've got all this work we need to do to make sure that we're treating patients safely from sort of our regulatory standpoint, plus all of these the new challenges and things that we're learning in the AI space.

Speaker 1 00:05:37 And so we almost have like a double dose of sort of a regulatory overhead, but ultimately by following the, you know, an ethical, significantly, you know, thoughtful process, we end up with a really safe and effective device. And that's one of the challenges to getting these devices sort of delivered at scale is to go through that whole process takes a lot of work, takes a lot of time and expertise and a lot of cross-functional stakeholders. This isn't just an engineering exercise. You really need to have folks who have deep experience and quality and regulatory and all these different areas sort of coming together to put together a really comprehensive view and then be able to deliver that. So that's what's one of the things I love about digital diagnostics is we've put together a really incredible team of experts in all these different areas that we get to work with and help bring us, help us bring these products to market.

Speaker 3 00:06:25 How can I driven diagnostics be designed to ensure transparency and trust among physicians?

Speaker 1 00:06:32 Yeah, that's a great question.

Speaker 1 00:06:33 And I think it really it's very connected to the regulatory piece. Right. Today there is not an extensive amount of hard guidance on what you need to do from how you build an AI safely and ethically. However, there is a lot of sort of pending or draft guidance from both the FDA and industry groups about best procedures and best practices. So I think one of the things that we found when we talk to folks, there's a couple different elements of the conversation you need to go through, right? One is just that first idea we talked about earlier in the conversation. We're not here to take your job. We're here to help you spend your time with patients in the best way possible. Right. So that's the first you're sort of getting on the same page of what are we here to help you do? We're here to help you take care of more patients more effectively. the second piece is the sort of that data trust efficacy performance piece. Right. So we want to look at sensitivity and specificity.

Speaker 1 00:07:23 We want to look at diverse populations of folks that we're validating these systems against, where we make sure that we don't just perform in a particular subgroup or a particular geography or a particular, subtype of disease. We need to cover all those different pieces. And so then we have a lot of great clinical evidence and papers that we work on to sort of show how the systems work. And then the third piece is really just that ongoing integration and workflow and, kind of the connection to making sure that we're not just, another tool in your toolbox, but we really are part of the patient journey. So when we talk to our customers, this isn't just something to help check a box or improve quality score. We really want to integrate into your workflow. We want to understand if you have this type of patient population, what are your targets? How many folks do you want to be seeing with this? How can we help you achieve that? There's a lot more to a successful adoption of an AI system than just the little AI piece inside.

Speaker 1 00:08:16 That's the hardest, most difficult part. But that's also the part that, that doesn't stand alone. It needs all those other pieces. And that's really what we found makes a successful system is looking at the whole, not just the AI part. The AI part has to be great, but that's, that's that's assumed. We assume the device is going to work. When you turn it on. There's all the other pieces that's really key.

Speaker 3 00:08:38 What about on the patient side? Studies show they prefer humans in the decision making process. Do patients need to be convinced of the value of these AI tools, or do you think that will kind of come naturally with the popularity of ChatGPT and some of the other tools that are out there?

Speaker 1 00:08:54 There's some really interesting new papers coming out about different ways. different folks are using those AI, lm generative systems, right? There's a lot of really cool potential and opportunities. But at the same time, there have also been some pretty terrible mistakes and ethical lapses that folks have done in the industry, too.

Speaker 1 00:09:12 Right? There's sort of a broad spectrum. So if you're a patient and you're not, you're just reading the news. It can be kind of hard to tell what kind of system am I getting? What's really going on is my doctor just using ChatGPT on their phone, or is this really part of an integrated, safe medical device? So one of the things that we found that really has a big impact on how patients respond to our system is that our results are delivered right at the point of care, right after the test. So when someone gets their diabetic eye exam within 60 to 90s, they're getting a printout that says, hey, here's a picture of your eyes, and here's what we saw, and here's what you need to do about that. And what we've seen is actually that immediate impact of, you don't have to wait a couple days later for the physician to call you back with results, or email you, or get in contact with your physician. That immediate point of care impact has a huge transformation on people's perception and their behaviour and then subsequently the outcomes.

Speaker 1 00:10:04 We actually have a really cool white paper about a patient who had been not really doing a good job controlling their agency, and they didn't because they didn't understand the impact. And then they got this diabetic eye exam and all of a sudden boom, like the the light bulb went off and they realized, hey, I can start to change and impact my health. I'm going to get on top of that. And so stories like that give me a lot of hope that there is a potential when you're using the tool in the right way. But that's because we've gone through and thought about the workflow and how we integrate and include. When does the patient get informed? Who's telling them? What are they going to know? Like, how does that work? We've put a ton of work into thinking about that. And I think a lot of times when people think about like AI models in a radiology space, it's cool to be able to go, you know, measure the shape or the size of something or see how it transforms over time.

Speaker 1 00:10:47 But how do you integrate that into the workflow? So the physicians are using it and changing the outcomes. That's what we think that there's there's a lot of opportunity there, but there's also a lot of work. And I think that's where the building trust with patients is really that's the opportunities to show them, hey, look at we can give you this really important information at the right time. We're not we're not proposing, anything that is going to disrupt the provider relationship with the patient. We just want to help augment it and make the provider have more time to spend with the patient, rather than looking at images, they can just talk about the results.

Speaker 4 00:11:23 Say, Keith, this is all well and good, but what if someone is looking for more clinical information? Oh.

Speaker 5 00:11:29 Then they want to check out our sister site, Patient Care Online, the leading clinical resource for primary care physicians. Again, that's patient care online. Com.

Speaker 3 00:11:43 With clinical workflows and the potential impact these AI diagnostic tools could have, do you see it as helping reduce physician burnout?

Speaker 1 00:11:53 Absolutely.

Speaker 1 00:11:53 There's some really great literature and publications about, being physicians, being able to use AI tools to expand their impact and their ability to see more patients because it can accelerate what they're doing. And I think that, from our from what we've seen, particularly with the folks we're talking to with these diagnostic screenings, a lot of times the workflow today in like telehealth settings is, physician will have a big stack of a couple hundred exams to read at the end of their day, after they've done everything else, they've gone through their epic inbox and reply to all their messages. Now they've got another 300, 500 exams. They got to read through. That's where they start to get that burnout is, man, I just put in 10 or 12 hours. Now I've got to go read another 500 of these where they're mostly going to be negatives, except for a handful of positives. And I'm going to have to go through this complicated physician outreach. And by automating those difficult, time consuming tasks, we're giving that time back to the physician, whether they need that just to recharge their batteries or they want to spend that time with the patients.

Speaker 1 00:12:53 That's where I think the opportunity is to reduce the impact and reduce the burnout and improve the impact. And we've seen that in some early publications. And I think that there's I suspect as folks get better at the workflow piece of it, we'll continue to see those those stacking benefits.

Speaker 3 00:13:10 Which medical specialties are likely to see the most rapid adoption of these AI driven diagnostic tools?

Speaker 1 00:13:17 I think the quickest place you'll see adoption is sort of the space we're in, which is, low cost screenings that have a really big impact. So there's a lot of different diseases that you need to screen for. It's sort of at a pop health level or the or to manage or monitor at that level, and the diagnostics to do so are just not present. and whether that's because the you need some special type of equipment or you need a physician, typically I think that's where the big opportunity is. The other reason that that's nice, from a sort of regulatory pathway perspective, is a lot of those screening things are lower risk than something like a pacemaker or something that's delivering drugs, like in an injectable pump where you have to go through much more rigorous, complex, in-depth validation process.

Speaker 1 00:14:04 So that's the other opportunities that because you have a sort of lighter regulatory overhead on a lower risk test, you can also likely have a faster timeline to drive adoption. So I think that that's really the opportunity is being able to take these sort of low cost diagnostics and scale them out, to be able to get people on the right pathway sooner for our for our particular focus. Diabetic retinopathy. It's the leading cause of preventable blindness. and so being able to get folks on a track to keep their vision saves, I don't know the specific figure, but there's a huge amount of money just from like an insurance perspective, that keeping someone in sight intact has huge stacking benefits for not only for the health care system costs, but obviously for their life and quality of life too. So I think that's where the big opportunity is, and that's going to be where a lot of focus is. I think the more complex things you see in radiology, there's a huge, huge value there. But even there I think that where we're seeing progress and clearances are on more of the screening tests, things like a chest X-ray or mammography, rather than the more complex things that need a lot more diagnostic interpretation and are more, much more open ended.

Speaker 3 00:15:14 A small rural clinic doesn't have the same resources as a large urban hospital. How do we make sure that all patients can benefit from these advancements in AI?

Speaker 1 00:15:24 That's a huge, huge concern for our group. In fact, we're actually we have publications we're both working on. And I think that we've already published in the past that show deploying these systems actually has a huge impact on underserved populations or folks that historically haven't had the same level of access. So that's a big driver of our mission, is understanding that in places where you don't have a lot of physicians or you have folks that don't get to be able to go see specialty care or don't have access to nearby specialty care because our our focus is in the primary care market, because you're just going to your regular doctor. This is something that you do when you're going in for your regular physical anyway. It's not an extra trip you have to do. It's not an extra special place or special physician you need to go see. And so because of that, we see that that really is expanding it.

Speaker 1 00:16:12 But that's because of the model that we chose to do. Right? I think that, you know, going back to the radiology example, if you have to go down to a university specialty hospital in an urban area, that's going to be a lot to take advantage of the system. That's a lot harder to do than these sort of our system can work out just, you know, anywhere that you have access to the internet and you can deploy the camera correctly. You can you can use our system. You don't need to have special equipment or a huge budget, right? We're not putting giant servers. Like I remember back when I started in the radiology days, they would install a new 3D interpretation system. You'd have to get a whole rack in the data center with all these equipment and GPUs and stuff. And now we're using cloud computing. So we have a lot of folks that even are using systems without great internet connections. But you just need that little 3G, 4G modem, get it plugged up to the internet, and you can still get results back in a reasonable amount of time and that ability to be flexible and to.

Speaker 1 00:17:02 We have a couple of folks who who have mobile clinics even. Right. And so there's a lot of opportunities and flexibility that because we're not having on premise big hardware that that that unlocks to. So I think for folks that, don't have access, that really I is a huge potential to increase access to care if it's thoughtfully deployed. And I think that that's, for me, really, one of the major things that's super exciting is that there's so many. If you look outside the US, like in, in the Middle East, there's diabetes is a huge growing problem and there's just are not enough physicians there. This is the only way, I think, that we're going to be able to solve that in a reasonable time and have a big impact, is using AI systems.

Speaker 3 00:17:43 AI tools are great, but they aren't free. With healthcare costs already soaring. How is society going to pay for all these new tools that are coming online?

Speaker 1 00:17:53 That's a really good question. I think that there's a lot of interesting work that's happening in terms of folks figuring out how can we build and train these complex systems in a safe and ethical and fair way? So one of the things I've seen across the health, the health tech industry is there's a lot of interesting consortiums and alliances and groups forming that usually have a specialty area or focus, whether that's a disease or an imaging modality or some specialty.

Speaker 1 00:18:21 And then those groups will get together and they'll start sharing data and collaborating as a community. And that starts to create kind of an open standard or way of doing things. And I think that having these big open data sets that people can use really starts to level the playing field for those pieces. I think that there's a lot of opportunity also on the trust building side, in terms of having validation data sets that people can say, hey, this is sort of like this is just like your MCAT or your Usmle or these other licensing exams sort of having these are the gold standards and I can perform as well or better than a physician. I think that's another opportunity to help build trust there. So in terms of how all that gets funded, I think one of the most interesting pieces right now. I'm sure you've heard the phrase data is the new oil. There's a lot of folks that think there's a lot of value in data. And while that's true, what we've experienced in the field is there's a huge amount of variability there.

Speaker 1 00:19:14 Just because you have a lot of data, is it labeled? Is it high quality? Do you have the rights to use the data? Right. There's a lot of questions that folks have to figure out that really has a modifier on the value of that data. And so one of the things we've been doing a lot is we talk to people in industries. We're trying to figure out what how do we do this in a fair and equitable way. We have a group in our company that helps us make sure when we do work on data sets, that we're doing it in a fair and ethical and sort of thought, thoughtful way in terms of being fair about the pricing and the value of the data and looking at market comparisons and other public information. So I think there's still a lot of interesting work to do, but having that sort of data collaboration is one of the opportunities. I think that will help kind of offset some of those cost pieces because ultimately everyone is generating. I mean, we're generating huge amounts of data every day.

Speaker 1 00:20:04 The question is just how can we capitalize on that data in a way that's fair and equitable for everyone involved, whether that's the patients or the physicians or the folks that are hosting the data? there's there's definitely interesting models to do it. And we're exploring some of that right now.

Speaker 5 00:20:21 Oh, you say you're a practice leader or administrator. We've got just the our sister site Physicians Practice. Com your one stop shop for all the expert tips and tricks that will get your practice really humming again. That's physicians practice.

Speaker 3 00:20:37 Looking ahead, 5 to 10 years. What's your vision for AI and medicine, particularly on the diagnostic side? And what challenges still need to be addressed?

Speaker 1 00:20:47 I think that the the really big opportunities in the diagnostic side are these sort of low cost, high value tests like we're doing for our diabetic eye exam. I think there's a ton of, for these some of these screening tests, right? I think everyone is familiar with the sad story of Theranos and their vision of sort of one drop of blood.

Speaker 1 00:21:04 We can do all these tests from it. Well, their sort of implementation was fraudulent. The vision of having these low cost, high value tests. I think that's really interesting to me. And from what I've seen in sort of both the radiology space and what we're doing, there's a lot of potential there in terms of things that you can learn from a simple image that the eye can interpret or pull inferences from, that humans don't necessarily even see from the same imaging modality. So there's a lot of really cool potential for other diseases that you can see from the eye. The eye is sort of the window into the brain and the heart and a lot of other complex systems. And so we think that there's a ton of opportunity there to keep adding additional diagnoses on top of these sort of one shot tests that you're not just getting one result, but three, 4 or 5 depending on what the right indications are for, for the patient. So that's one cool possibility. I think the other piece that's really exciting that we see a lot of potential for is helping find unique or small patient populations for rare diseases.

Speaker 1 00:22:04 Right. So you have these really, really big data sets, and you have these really rare diseases using AI to pound through these huge data sets and find what are the interesting correlations. What are the threads that a person can't get or statistically doesn't understand? How can we pull on those threads and learn new things? There's a lot of cool projects that I've seen people publishing on and that we're working on with partners. I think there's a huge amount of opportunity in terms of using AI to just crunch through these giant data sets and pull out really unique inferences that a normal person or physician wouldn't. But somewhere in there, there is a statistical connection that the that the model can find. And I think that's there's going to be a lot of really interesting insights that get tapped out of that, assuming people can continue to do it in a thoughtful, ethical way, because one of the one concern I have of what might put all that at risk is just taking lots of data and not thinking about the provenance of the data or the labeling or all those pieces, you know, you can start to run into, you know, it's the opposite of the flywheel where you just have bad inputs, creating worse outputs over time.

Speaker 1 00:23:04 And so I think that we need to be careful about that. But I think there's huge potential, in those spaces to kind of find new things and then also deliver lots of, lots of high value insights at a low cost.

Speaker 3 00:23:17 Is there anything else that you would like to mention for our physician audience, that you think they need to know that we haven't talked about?

Speaker 1 00:23:25 I think that one of the one of the things I'm really excited about is just the idea that AI systems can really help improve patient outcomes, at a pretty cost effective way, because when when you look at how expensive it is to have a physician interpret an image versus how much an AI can scale and how it can go, you know, the AI isn't going to get tired. The AI can work across time zones. There's a lot of really cool opportunities. I think, that making sure we can really just get those results delivered to patients faster? Like, it just keeps changing the outcome. And so what I'm excited about is as we as we build new devices and get into new, new specialties and new areas, seeing how we can keep having that patient impact those stories of people who are changing their behavior when they see the results.

Speaker 1 00:24:16 That's what I want to. That's what I want to see more of. And I think that we have a huge opportunity with with the great stuff that we're seeing with the data, and the research that folks are doing. So we're really excited about the potential to keep delivering more innovations like we already have into into lots of new areas in the future.

Speaker 3 00:24:32 Very good. Mark, thanks for joining me today.

Speaker 1 00:24:35 Awesome. Thanks so much, Todd. Really a pleasure.

Speaker 2 00:24:49 Once again, that was a conversation between Medical Economics managing editor Todd Shryock and Mark Daly, chief technology officer of Digital Diagnostics, on behalf of the whole medical economics and physicians practice teams. I'd like to thank you for listening to the show, and as you please, subscribe on Apple Podcasts, Spotify, or wherever you get your podcasts so you don't miss the next episode. Also, if you'd like the best stories that Medical Economics and physicians practice publish delivered straight to your email six days of the week, subscribe to our newsletters at MedicalEconomics.com and PhysiciansPractice.com

Speaker 2 00:25:16 Oh, and be sure to check out Medical Economics Pulse, the quick hitting news podcast that offers concise updates on the most important developments affecting your practice, your bottom line, and the broader health care landscape delivered by the editorial team at Medical Economics. Off the chart, A Business of Medicine podcast is executive produced by Chris Mazzolini and Keith Reynolds and produced by Austin Littrell. Medical Economics, Physicians Practice and Patient Care Online are all members of the MJH Life Sciences family. Thank you.