Healthcare AI Update: General Discussion

Good morning, everyone, and welcome to Signals and Symptoms PASC podcast. I believe Dr. Manering is going to join us pretty soon. It is just the two of us, Harvey Castro and me today. And then the core goal is that we're going to learn from Harvey as we did a few weeks ago, understand, uh, the core features and the core capabilities that we need to build in the age of AI, how it is going to impact patients, physicians, providers, payers, etc., the whole stack, the 5 what we call Ps of healthcare stakeholders. Again, reminding everyone that please do follow Harvey Castro on LinkedIn. Also his latest LinkedIn post of the Dr. GPT as well. He is 5 times TED Talk speaker, not to mention an amazing businessman. And of course, you know, talking about, you know, phantom space as well. So without further ado, uh, Harvey, thank you so much again for educating us. I'm extremely grateful to you always and to learn from you. So the floor is yours today, Harvey.

All right, buddy. Well, let me get these slides up. Uh, I know we got some people helping us. If you could put the slides up because I don't have the power. Okay, they're already there.

So I am presenting for a group of gentlemen this month Harvey, for the focus, I'm gonna get off the screen and then of course I'm gonna get in when you need me to. Sorry, just wanna—

Sure. So first of all, the reason I'm dressed up today is because I got really blessed today. My undergrad— not my undergrad, my graduate degree for medical school is through UTMB and they are coming to my home and taking pictures. I'll be on the fall edition of the alumni going over AI in healthcare and Kind of stuff like that. So I'm excited. I'm gonna be presenting later to a group of men called Metal, and basically it's gonna be the future of health. And my goal is it'll be a monthly presentation that I'm doing. So I wanted to share it with you guys because this particular talk really correlates with what we're talking about. And it'll be short; it'll be about 15 minutes. And then from there, we can have a great discussion. And my business partner here, Janade, and I will talk about. Several things that he has. So number one, when we talk about AI in healthcare, this is taken from the World Health Organization. We look at 6 pillars. Number one, trust and safety. And I know we talk about that a lot and everybody talks about it, but can we really rely on it? Is this something simple where maybe it's just helping me make my appointment, or is this my surgery? Is this neurosurgery that we're talking about? Which one? Second, governance and accountability. And we hear that a lot, but at the end of the day, it just boils down to something simple. Who's responsible? Is Dr. Janai and I responsible since we're using that AI, or is it the AI company?

Who is it?

And in the state of Texas, and for now, it'll be us physicians because we do have the license. Third, and I really think this one's the rate-limiting step, culture and workflow. How are people using it? Just because we have AI, does that mean we use it? Or just because we want to use it, does our institute have what we want, or are we using our own tools that are not technically allowed? Which one is it? And that really boils down to the culture and workflow. The next one is the privacy and security. Who sees my data? If I'm sharing everything with you, if I created my digital twin on my phone, does everybody have access to digital Harvey? Or is it set up correctly? The next one is equity and access. Man, this one's huge. I'll be in New York City and I'll be talking about AI and bias, and this will be one of the topics. Is this equitable? Who can use it and who can afford it? As we know, there's some treatments out there that unless you have the resources, you'll be able to take care of those. If you don't, is that equal or will everybody have access? And that's more ethical, but it's It is definitely an issue. And the last, outcome and value. What actually improves? And I have some cool slides to that. Recently, the FDA just approved the ability to check your glucose and ketones in one wearable. Now, for an ER doctor in me, I'm like, oh my gosh, this is amazing, because as you know, there's physicians out there that know this, If I have someone with diabetic ketoacidosis, I'm looking for those ketones. So if the sugar's high and the ketone's high, then I know that person has it. And just for layman purpose, why does it matter? If your sugar in your diabetic gets to a certain degree, it could be so high that you become really acidotic, and it gets so acidotic, believe it or not, you can die. The ability to have a wearable at home to tell you that it's catching. Oh my God, this is crazy. So big picture, what it boils down for this device, if a regular consumer that has diabetes is using it, is the company that launched this having a 24/7 doctor ready to know what to do, or is the patient supposed to know what to do? So it makes it interesting because if you think about it, going back to those pillars, we have the technology, But is the culture, is the system, is the workflow ready for that new technology? Quite interesting. Another one was really interesting, clinical AI and the evaluation of does AI really help in the medical record? And this was a study in Kenya, and basically the skinny is this. The individuals—

So going back to Castro, the last slide, the question I had is essentially not just for medical, right, or healthcare. I mean, the broader question is AI adoption. And of all the chat talk and the chatter we see online, the noise essentially is all about essentially saying that, you know, AI adoption. But again, we, as you said, from a human perspective, is we are talking about change control and change management. And whenever there is nothing new from, you know, telephone switchboards to everything, you're looking at this. What do you see specifically in general AI adoption challenges and then in healthcare? And then is the pace really the one that is going to make things significantly harder or is it something else?

I think the rate-limiting step is culture and trust, those 2. We can have all the gadgets you and I decide to make, but at the end of the day, is the institute ready? Is the culture there? Is the workflow— are we creating it so that we're not disturbing the workflow, where it just becomes seamless, where you don't even realize that the AI is actually running?

And going back to this thing that you said in your prior podcast, the culture, and you had a very interesting answer for it, and it stuck in my head. And, uh, that was, uh, that the culture is going to change when you're going to get sued for And which is really interesting because let me give you another example, right? We have the breast mass detection, people are starting to use it, et cetera. The pushback is coming from director of radiology and says, well, when the AI predicts it and radiology does not agree with it, who's medical legally liable? Then of course the problem is that if at some point in time, is it something that's going to happen that why didn't you use AI? That's a medical legal flaw that you should have used AI in the first place. So comment on how to, because cultural change is extremely hard. Like, you know, I know, I mean, we have gone through digital transformation, we have gone through, you know, these EHRs. How do you see things moving in this direction given, of course, your—

Good question. Big picture, I see the following. I see the dichotomy, and the only way of getting to that point for that radiologist is it's kind of the Wild Wild West. What do we do? So we need to establish the rules, we have to establish the culture, and as education plays into it is the big thing. If we can educate our physicians, our patients will be able to advance this further. Think about everyone right now. Take a second. When you first heard about AI, what were your thoughts? How did you see it? And now wait and think of all the ways you do things differently today compared to that day. That's a good example of how education changed it. Now, to answer this legal question, that's where it's kind of crazy. Europe's doing their own legal stuff. California is doing another, and then now all the states are kind of doing their own, and then the national is kind of trying to keep it all under one to make sure that we can continue to advance. As we move forward, the governance will be more and more important because we got to make sure we get this right, because so many things, if we don't get it right, so many things can fail. So bottom line, to answer Jeanette's question, I foresee the legal system really having a hand in this, and I wasn't kidding when I said Legal will really change the culture because from the underlying— let's just— my example back then is there's 2 hospitals, one using AI, one not. This is hypothetical. One is consistently creating better patient outcome to the point where it should be the standard of care. And technically for that region it is. And hospital across the street decides not to get it and the patient has a bad outcome, a lawsuit will happen and eventually that will change it all. So that's how I kind of foresee that playing out. On the international experience, it's a different culture. When I speak in Dubai, when I speak in Singapore, a lot of it is from top down. This is what we're going to do. This is the AI. This is the electronic medical record. These are the genomes, and everything will be done as such. Whereas here in the United States, that, that's not going to happen with our culture and our laws. So going back to this slide, big picture, it basically said Kenya, had this study where they used AI documentation and they thought, oh wow, this should help. And it helped the physician be more efficient with their documentation, but they didn't show any benefit for that patient, which was interesting. The next one is this eye therapy data. It's just phase 1. All they're doing is looking to see if this is safe. The only reason I mention it is because we're getting a lot of noise and signals. We get all this stuff like, oh my God, there's this thing coming out and At the end of the day, this might be overhyping that. All it's saying is we're just looking for safety. They may be hoping to get something amazing, but at this point, what they can prove, or what they're trying to prove, is that this particular item is just safe. Now, when it becomes what is the better information that can help a decision, that's where you're coming to these podcasts. That's where your experience, if you're a physician, healthcare, how can we make sure that these items that we're learning can actually improve our health? Just because we get a special number, does that mean, let's say assuming a better score, does that technically give me better health? And this just gets going into the weeds, but if you're looking at clinical trials and they give you a number from 10,000 feet high, make sure that when you're looking at it, and you can always ask AI to help you, is that clinical study valid? In this particular example, we're just showing you an odds ratio, and that particular number also includes no difference. So just because as you look at that image, the majority could in that experiment help, but there's a part that in that same experiment it actually makes no difference, and in some cases can make it worse. I'm just going to go quickly just through some of these slides. So a possible service progression, let's look at that. Separate measures and contacts, alert each by phone, and verify response if the service delivers. So In this example, I was just looking at that example of that wearable. How can we make that work? Is it the patient that's just getting the alert? How is that workflow? And we really haven't looked at that. But at the end of the day, who's responsible for that? That's my biggest worry. If that patient with DKA has that diagnosis at home, what do we do next? Is it the patient's responsibility? Is it the caregiver's? Is it the healthcare team? Who's going to respond on a Sunday at 2 in the morning? And that's quite interesting because we got to make sure we do that. Questions that we must take actions— I get a lot of CEOs asking me like, Harvey, how do I implement this at my hospital? And these are the things that I say. I said, number one, look out for the outcome. Was it actually what was trying to be measured? 2, who's responsible and when? Again, that same example— is the AI responsible for this? Is it the doctor? And when? And the third, what would justify continuing this experiment or this AI widget or just stopping it? Now, 60%— this is just my personal forecast in the future. I believe that with 60% confidence, by this time next year, we will have an AI that's going to be used at the hospital system or clinical records that will show that the morbidity and mortality will be improved by using this particular AI as opposed to the standard. And then I just want to leave you with this. At the end of the day, for everybody out there, what would make you try to use these type of service? What would make you wait? And what evidence would change your view and your change of mind? And that's basically the skinnies. It's supposed to be just a short, just to kind of get us going in the conversation and started today. So, Jeanette, you come back and let's put the slides back down and answer questions.

Insane amount of knowledge packed in in 15 minutes. The Harvey Castro method of distilling knowledge in easy-to-stay. Uh, we're gonna put this, uh, I'm gonna actually make sure that you guys do subscribe to Harvey Castro's Substack because, um, he is being, uh, bringing in a lot of knowledge, uh, condensed. And I, I'm assuming he's gonna Add this into his Substack as well. Yeah, so, um, Hari, let's unpack one by one. And then, uh, I love predictions because, you know, it makes things amazing. Um, I, I think hospital systems are gonna be forced into it, and that's been the case, right? I mean, if you really look at it, they have been essentially very hard to to accommodate new technology. As a matter of fact, if you look at healthcare in general, interestingly, it has been essentially been slow to adopt any new technology. I mean, just to let you know, I mean, we still had faxes, for God's sakes. We still have pagers in some hospitals. Like, oh my God, like, you know, it's been 20 years since the iPhone came out. Are you not freaking insane? And there are issues, right? I mean, um, and just to let you know, one of my friends actually does telecom. And, uh, I walked around, uh, they walked around one of the hospitals and they realized, how do you guys function with this bad of a Wi-Fi? Interesting. And then the, the priorities are completely off, right? Off in the sense that in the waiting areas, the Wi-Fi is amazing because, you know, the patients will complain and they can, you know, go through, scroll Facebook and everything. But in the basement, in the CAT scanner, when I have to contact the neurointerventionist so that the patient can get thrombectomy, they're not going to provide the goddamn service. So I don't know. Hospitals, I believe, are still behind, but I feel like we are doing right. I mean, Save Life AI is going into imaging centers. It is going into RCM solutions for hundreds of clinics now. And we've seen that everyone else is actually starting to pick up. I mean, true implementation is going on in front office automation, back office automations, RCM automations, et cetera. But what I do see still is this, where the clinical perspective and clinical documentation and everything, and there are some lacking issues with workflows. So let's unpack my questions one by one. Number 1, redefine hospitals by the end next year. Do you still think that? Number 2, we talk about healthcare in 2 essences. One is, of course, back office operations, and then of course clinician-facing operations and patient-facing operations. So let's just talk about these 3 different ways of AI and AI adoption curve that you see. Of course, I have another one when you were presenting, Dr.

GBD, digital twin and all of that, and we'll talk about it. I see the following with our healthcare system. We obviously, we have early adopters, late adopters, and some of the early adopters with AI have gotten burned in that if their hospital system is using it wrong, there's some lawsuits in place and they're getting burned. Simultaneously, we're seeing other hospitals cautiously just playing right behind the early adopters, seeing where they messed up, what mistakes they did, and how they're moving forward. And I think this plays perfectly into my talk earlier in MarTech's Law. And just a quick review, MarTech's Law is AI is exponential, our organizations, our individuality is logarithmic. Now, MarTech Law predicts that when the gap between technology and the individual or organization gets bigger and bigger, the disruption happens. And therefore, AI obviously is disrupting a lot of these organizations, but here's the magic. Here's to this point. This is why I made this point. The different organizations that are leveraging AI are disrupting the ones that are not. And let's just add to it for the governance and everything, the ones that are using it correctly are making it more efficient and are going to outlive the others. And so now that goes to the prediction, and I hate using that as a word, but now that we start seeing hospital systems, we're going to start seeing hospital systems taking over the other hospital systems and being more efficient. with these tools and being able to leverage them for competition.

So true. And I think as much as we try to believe that there is competition, interestingly, it's a regional mafia of massive regional healthcare systems that are controlling the market. But what we're seeing is this evolution is really changing how we think. Another question that came up is very interesting. Again, when you're talking about the 6 big ones that United Nations is also adopting. And of course, your whole process is AI governance. And guys, uh, we're going to be launching something, uh, very soon with the help of Dr. GPT, uh, and everything. Uh, it is not going to be truly inside Save Life AI because of the, you know, ensuring that sanctity of data and everything, but we are coming close to it, uh, with our decision engine and context engine. We talked about it the weeks ago. or we're going to operationalize it differently to make sure that Harvey's vision of governance is ensued everywhere. How do you measure responsible AI for clinical-grade use? Which is basically very interesting. And this is like, as a startup, right, Harvey, you ask— as an investor, you ask me, what is your KPI? And that's like key performance indicator. And this is like an interesting, very hard question to answer. that fidelity, precision, accuracy. FDA is still trying to figure this out. FDA is coming up with, you know, new methods to even, and they haven't figured it out just to let you know. And they're very open about it. And that's one of the things I loved about the new FDA that they're like, dude, we don't know, we need help. So what are your thoughts, Hari?

Well, great question. Thank you, Maryam, for asking that. And I'm gonna punt this back to you and then I'll go after you 'cause you're the one that's in the front lines right now with your product And how are you addressing it? I know you're using AI agents. How are you— are you using that to help leverage this question?

Yeah, so first of all, you're absolutely right that we need essentially a measurement. And then every AI that we build is clinical grade. And let me step back and let you understand this. Even with clinical trials, it is not 100%. Right? I mean, we do thousands of patients, then we do post-market surveillance and we figure out, oh my God, we have to take this drug off the market. So first understand the whole expectation of this, because people for some reason demand a lot higher from AI as compared to what we have been doing. Because if you really look at it, which we have been involved, both of us, in clinical trials, you have no freaking idea how This shit happens sometimes. More importantly, some of the clinical trials that failed, that never published, is interesting to see because we are on the back end of it. Then we have been doing this, and then we don't even get sometimes as a site on the clinical trial why the clinical trial stopped because some site messed up. Some quote-unquote event happened, which is data safety event or patient safety event. So that's first of all understanding backwards and then understand the expectations. So clinical grade comes from that, right? The only difference is that you have done clinical trials and you're thinking, oh well, what is going on? So that's number one, to reset the expectations. Then the number 2 is every single AI that we build has a different, uh, KPI attached to it. So for example, if you're building a vision model— vision model is, for example, in the CT abdomen I'm going to say segment out the liver or segment out the liver mass. Now those are essentially areas in which they have to segment. For that, we have to use Dice score. For our medical coding, which is essentially being able to understand that the prediction that the AI produced, how much it is, then we use F1 score. If you are actually going through, you know, these back office operations, etc., and for example, there's a prior authorization call that my voice agent does, then for that is task completion rate, not full task completion rate, not the small bits of— because every task is every job, so job completion rate, and then underneath is what we call task completion rate. So both of them needs to be managed. So for every AI, there is a KPI attached to it as far as As far as that is concerned. So long story short, what the way that Harvey, what we are doing, because we are a global healthcare AI factory, most companies, startups don't go through this whole massive process anyways. We have 4 FDAs, HITRUST, we're doing RCM and everything, but because we believe that the real value comes in when the integration of the system, there's a vertical stack that helps the physician and the patient. So there are so many different areas in which we have to focus and then make sure that this happens. And then of course the technology changes. We just released our own decision engine. I think next week I'm going to show and tell inside Save Life AI that how it is working and functioning and everything. But what you're going to see is that, that the technology, when it changes, we actually have to ensure. So one of the things that we have done consistently in our Well, I don't know, is that we have what we call a golden dataset. That golden dataset essentially is our hard-coded physician-verified, clinician, multiple clinician verified. And we can quickly check that, okay, this is the ground truth. And then once you have the ground truth for different applications, for different KPIs, then you assess that back and say, okay, the new model or the new variation of the new model or a completely new goddamn engine like decision engine comes in, then you adjust, uh, the, the, you know, the F1 scores. And then depending on if the results are good, then you update it. Some of this we can update, some of this unfortunately has to go through FDA directly, which is still a long process. And again, um, I think FDA is working on improving that because It takes 3 months to show them that, okay, our model is new, and by the time they approve it, we have another version that is actually improvement. Mary Ann, thank you so much for this amazing question. Anyone who has questions, please do comment us so we will reply that. I think today we're going to take one last question. One of the things that we're going to talk to Dr. GPD unless you guys come on. is this. Let's just make it a little fun. Is privacy dead in the age of AI? And then the question really is that going back to this, that okay, somehow there will be private servers and your MacBook is going to run your own private AI and personal AI. However, how would it look like? Because we we already can't solve the goddamn data silo issues with all these issues. How the hell am I gonna share my digital twin to a person who's sitting in, I don't know, over here in town, like Baylor, McKinney, you know, all of these, for the physician to be able to use it now with their AI? So I know, Harvey, hard question, but I just love it that you have created it. And then I'm so— I'm gonna learn it from you more of creating my own digital twin. I'm working hard on it, just to let you know, collecting data and everything, but now connecting my EHR records as well. But go ahead.

No, I think the age of privacy, I think my problem in general, I don't even think it's a problem. I think in general, I see everything positive. Even things that are negative, I see as positive. I just think it's just a matter of time. With that said, what about the future of leveraging edge technology where all my medical records on my phone, leveraging blockchain technology where I can change it and transfer it to whom I want and have the keys and leverage, you know, like this out of space, leveraging the technology in space in the cloud to now allow the processing power that maybe my iPhone doesn't have, but the collective cloud in outer space does. And that's what Phantom Cloud is going to be doing, working on its own cloud in space. And I think that could be a great position for privacy.

I love it. I'm actually gonna, you know, restate that. So There, there are 3 layers that he defined, and I'm going to tell you from a more technical perspective what the data layers look like. One data layer is my layer, which is I own, which first of all, that's a big problem, which is going to solve. We're going to go to Vermont system as far as the state. Basically, at some point in time, the— these baby boomers are going to go away and the millennials are going to come and they're going to demand data sovereignty on their own body. And it's fair. We are gonna see that coming because I, it doesn't matter. No hospital doesn't own it. Physicians doesn't own it. Basically, the patient will own their own records. It is coming down the line. It's gonna take 2 years, 3 years, 5 years. I don't know. I think that is coming. And at that point in time, I think when physi— when patients have access, they will be using Edge, as we— as he talked about, for simple stuff. And for complex queries, it's going to go to cloud, and most likely it's just going to go in some version of— by the way, did you know that Tesla just— sorry, SpaceX just bought a spectrum of mobile spectrum?

Oh yeah.

So yeah, so you can see where the shit is going, right, at this point in time, right? I mean, this is like, oh my God, what the hell did he do? Anyways, he's just crazy at some point in time. So he basically is building A direct sell to satellite, sell to all services, by the way. So you will essentially— are looking at a future where you're not— your health records are directly connected to your cell phone service, which also adds on to a Grok in space and Phantom space that he's working on directly and built directly for it. So we are definitely looking at that future coming forward. in my opinion. So that's the second layer. And then the, the third layer is PHI scrubbing, which is again, we are doing so efficiently, consistently with AI that even if you share that layer through blockchain technology, your PHI can be scrubbed. And then of course, the overall blockchain will actually ensure a level of security that is highest encrypted, and that encryption can be, okay, well, I want to share my records for research, so it's PHI scrubbed, but I want to share my personal records with Junaid Kalia name on it with my physician. I believe Harvey is exactly right that we are personally looking into it in Save Life AI, that do we need to build a blockchain ourselves or not? I just don't want to complicate the technology, but with Harvey's guidance as our investor, and, uh, we will look into this technology to build in SaveBlack AI at some point in time too. So again, grateful, Harvey, for that question. I think I'm gonna let you close for today, and then next week we have some, uh, exciting, uh, guests are coming. Otherwise, uh, I'm gonna get into the routine and I'm gonna go through all the updates, uh, one by one on this system. And then again, there are so many updates and I wanted to be fair and show a balance and not just Save Live AI. But Harvey said, man, you need to just summarize it, show it on Save Live AI and move on. So I'm going to start doing that as a great advisor he is. So again, important that you always have great advisors as well.

Yeah, I'm just going to put it out there. I'll be in New York this month, Atlanta, New Orleans, and then next month I'll be in, uh, Orlando. That's just the next 4 weeks. So if you are in any of those areas, feel free to reach out. Love to meet you and have conversations with you. And more importantly than that— oh, I forgot, this just got released. I wanted to share that with everybody.

Oh.

So Wiley sent me the book and it's, uh, AI Basically in Emergency Medicine. So I'm really, really stoked. This is everywhere. Uh, but more important than all of that At the end of the day, guys, have fun with AI and see how you can use it. It's all about human plus AI. What is your individual characteristics, your personality, your clinical gestalt, or your secret sauce? And then add that to AI, and you'll see that both will give you an amazing output. Thanks for having us.

Thank you so much, guys. See you next week.