AI’s Impact On Care, Pace of Innovation w/ Anish Patankar

Erik Sunset: [00:00:00] All right. Hello and welcome back. I’m Erik sunset, your host of the DocBuddy journal here at DocBuddy. We deliver healthcare solutions that take the pain and costs out of broken workflows. From the ASC to the clinic to on call at the hospital, DocBuddy helps riders access, create, and move data seamlessly, all from the point of care.

Erik Sunset: You can learn more about us and our solutions at docbuddy. com. And today we’re joined by Anish Patankar, who serves as SVP and GM of Oncology Informatics Software with Elekta. Elekta is a global Swedish company that develops and produces radiation therapy and radiosurgery related equipment and clinical management for the treatment of cancer and brain disorders.

Erik Sunset: Anish, thanks for joining us.

Anish Patankar: Thank you, Erik, for having me. Happy to be here and look forward to the discussion.

Erik Sunset: So are we, and thanks so much for making a little room in your schedule to, to cover some hot topics in healthcare today, centering around AI, AI and oncology, but before we, uh, sort of dive in, what else should [00:01:00] listeners know about you?

Anish Patankar: I think one thing I have, uh, I have a long background in healthcare, of course, uh, and I’m very passionate about delivering innovation. In healthcare technologies, of course, benefiting patients. And I think one thing I would like to maybe state out loud is that you do see a growing disparity in sort of how healthcare innovation is now reaching different parts of the globe on being part of a global organization.

Anish Patankar: One thing I really strive to do is to make sure that we can deliver this innovation to all of our clinical customers globally, not only where it’s easy to deliver. And that means it’s customers who adopt these technologies on the cloud, customers that deploy these technologies on the premises and so on.

Anish Patankar: So having that consistency in sort of benefiting patients across the globe is very important to me personally. And that is what I really aim to do.

Erik Sunset: Well, I, I know with your, with your background, that’s, uh, that’s no small task, but it seems like you’re the right man for the job. That that [00:02:00] flexibility of either on in the cloud or on premises for the solutions that you deploy, that sort of gives you a leg up here because health care is rushing to adopt artificial intelligence and the related solutions faster than it has ever sought to adopt a new technology.

Erik Sunset: Do you have any thoughts on why that is?

Anish Patankar: I think a couple of factors, right? If you look at the Specific impact on imaging technologies in healthcare, right? The impact of AI, the pace of innovation and the way you can deploy that innovation in the cloud or on the premises has been greatly accelerated by the advancement of the GPU tech, right?

Anish Patankar: And I guess we all see that a lot. So you see, you started to see that explosion starting from radiology where, you know, you could sort of contour organs in imaging. Right. CT images, MRI, PET scans, different kinds of images, and more and more now that expanding from radiology into oncology, neurology, [00:03:00] cardiology into different sort of disease areas.

Anish Patankar: And I think that the, uh, the Enablement of deploying GPUs in the cloud and the readiness of the models to be deployed anywhere where the physicians are to get access to those results in, in almost near instantaneous speed. That is causing one aspect of explosion of AI adoption in, in, uh, In our space, right?

Anish Patankar: The other aspect I would say is sort of the power to harness data and then you see this in sort of You have so much data in healthcare, which is so much siloed and I think so much has been spoken about this topic uh, there’s Just having uh, sort of the conversation between the doctor and the patient the amount of literature that’s out there the amount of innovation that’s happening uh in different areas that impact healthcare and then the um Taking all of these parameters as input into a large language model and then giving the right context to a physician who’s at [00:04:00] at the bedside with the patient in in the in the hospital room or in the clinic, giving something more actionable to the physician rather than them relying only on their experience or expertise.

Anish Patankar: I think that’s the other source of explosion of AI in healthcare, right? So making sense of all this data previously, which was so siloed, putting it together, getting sort of results out of it. We are seeing the adoption just really picking up.

Erik Sunset: When you hit the nail on the head, we, we kind of laughed before we started to record that prior guests of the DocBuddy journal, you know, highly accomplished individuals, great surgeons, leaders in, in US healthcare, you know, they end up with sort of a best guess around what AI will really mean for a provider and to an extent the patient as well.

Erik Sunset: And you see some of the wilder takes out there that AI will never work. We’ll never want it. I don’t think that’s true. But then the other end of that spectrum is we’re not going to need a provider anymore. We’re not going to need a doctor anymore. And I don’t, I don’t think that’s true either. So somewhere landing in the middle where [00:05:00] AI or predictive analytics or machine learning, or by whatever name that you want to call it, is really being built meaningfully to augment that provider’s workflow. And one of the things that you just mentioned, I want to key in on is sort of around that operational efficiency that, that an LLM, and that’s a large language model, something like a chat GPT for our audience, you may not be familiar, uh, the efficiency that something like that can provide. So can we get a little bit deeper there on what that may end up meaning in the future or maybe even what it means right now?

Erik Sunset: I

Anish Patankar: No, absolutely right. I think the reality is always in the middle. The providers are not going away and AI will definitely be adopted in the workflow, operationally, clinically and every which way. If you look at the impact that LLMs have brought us here, Erik, right? I mean, one thing is. There’s so much data to make sense out of in healthcare and I’m trying to sort of do that in a manual way or sort of relying on sort of just relying on the [00:06:00] physicians to be up to date with every new regulation on the compliance side, for example, uh, every new clinical, uh, innovation that’s happened in terms of outcomes.

Anish Patankar: It’s almost impossible for everybody to keep up with all of that. And on top of that, there are guidelines to be followed. These guidelines can come from the institution where a physician is working. They can come because of the payers that, uh, may be associated with that particular hospital or clinic, or could be any other.

Anish Patankar: regional or governmental factors because sometimes in socialized medicine in the NHS, in the UK, certain provinces in Canada, or in the U. S. also in depending on what sort of format you’re practicing medicine in, there are all these things that come into play. And when you talk about operational efficiencies, the ability of one person to rely only on their experience and expertise, to translate and digitize all of that.

Anish Patankar: It’s just not humanly possible. And that’s where the power of [00:07:00] a large language model to take all of these parameters into consideration. And then say, how does all of this apply to the patient who’s sitting in front of you? So the biggest impact I see in terms of operationalization is Helping the physician determine a treatment pathway in terms of operational workflow to say what’s going to be the most relevant for the patient case that you’re looking at right now.

Anish Patankar: And I think that’s where we see the biggest impact going forward in terms of efficiencies for the, for the clinical team.

Erik Sunset: couldn’t agree with you more there and I want to look backwards if we can for just a second that in the early days of EHR going back to the early, or I guess rather the late 2000s, early 2010s, this was when Fitbits were starting to be popular and things like the Apple Watch were on the horizon and things like the Whoop tracking band that created a tremendous amount of data. Going back 15 or so years, you’d ask a cardiologist. You know, doc, I wear a Fitbit. Can I [00:08:00] give you my data? And they would say, absolutely not. That’s too much liability exposure. I can’t handle that much information. I can’t do anything with it. And if I miss something critical to you, then it’s my fault. So keep it to yourself, not interested in it. Um, but with remote patient monitoring, with all of the patient care coordinators at work to be sure that sort of these patients in a value based care, uh, model are doing everything that they need to, I would imagine. That an LLM or something like it will be able to ingest all of this information and then turn it into data, I guess, to be, uh, very specific about the terminology, are you seeing anything along those lines?

Anish Patankar: No, absolutely. And I’ll sort of give you two data points on that. And just last week, actually, And I forget if this was in the Wall Street Journal or some other online publication, where I said, where I read that it was most likely the Wall Street Journal, where You know, they had a very nice article where actually physicians are now asking their patients to use Apple watches and make sure that they get access to that data because they [00:09:00] believe it’s so valuable in them understanding how patients are living their lives in addition to the medication that they’re on, right?

Anish Patankar: So, so I think it’s not that doctors are not pushing back. They’re actually requesting the patients to sort of use these devices and get them the data because I think now they have the adequate tools and resources, both human resources as well as technology resources at their disposal to make sense out of this data to mine this data and know that when an intervention is needed or not.

Anish Patankar: So that that’s one. And the coming back specifically to the area. Now I work in it. Electa, right? We also have, uh digital patient monitoring tool or remote patient monitoring tool, uh, which is baked into our EHR. And we, patients actually report their symptoms when they’re on treatment. And a lot of this data comes back.

Anish Patankar: It’s also accessible to the patient via their patient portal and so on. But what’s most important now is what we see is that We can now trigger early interventions from the clinical [00:10:00] side to say, okay, if the patient is on a particular radiation treatment or a medical oncology treatment, what is a serious side effect versus what’s a expected routine side effect when an intervention should be, uh, should be happening earlier versus what can wait until the next patient appointment or the next patient treatment in the workflow.

Anish Patankar: I think all of these factors with the sort of the ability to get this data in. instantaneously. And then the application of A. I. On top of it to say, you mind this data intelligently. This is a classic example of converting real world data into real world evidence, triggering early interventions and making sure that the quality of life that the patient has.

Anish Patankar: During some of these treatments, which can be pretty toxic in terms of those delivered to the patient or in terms of the toxicity in terms of, you know, chemotherapy and so on, can be managed much more tightly. So I really feel that this is already making such a [00:11:00] positive impact for patients who are on these kind of treatments.

Erik Sunset: Yeah. And this is a little bit out of my wheelhouse, but for, uh, the patients that are receiving that radiation therapy, whatever, whatever it may be, that’s, that’s a bad circumstance to be in, or not one that anybody would choose to be in, and then that’s a difficult thing Set of medicines to ingest, whether intravenously or otherwise, I would imagine anything that you can do for your providers to improve that quality of life, perhaps improve adherence to a degree to if a patient isn’t dreading their next appointment, that’s got to be Huge in terms of outcomes.

Anish Patankar: Absolutely. And when they see that the clinical team is engaged between treatment cycles between two patient appointments, they feel that they’re being, you know, looked after, they’re being monitored. They can report their symptoms, of course, but their symptoms are being managed. I think the adherence is much better.

Anish Patankar: The compliance of the patients is much higher. And this is extremely important for patients who are dealing with, [00:12:00] you know, chronic conditions as well as, you know, uh, certain types of, uh, treatments when they have cancer. So, so super important.

Erik Sunset: And this that kind of scratches at the surface of a value based care and generating a positive outcome. So we’re taking another kind of step off our main conversation topic, though. But there’s a large degree of individual responsibility that has to take place for any value based care panel to be effective.

Erik Sunset: You can’t you know, abuse your body or do the wrong thing and expect a positive, um, a positive health outcome. Um, so just again, to come, to come back to that, that main track with LLMs able to both mine data and perhaps even take a load off of that clinician, off of that provider or off of that mid level in terms of a piece of standard correspondence, perhaps from a patient portal, something like that.

Erik Sunset: Are you seeing any great use cases there?

Anish Patankar: Now we do, in fact, we, we, we also see that it’s useful for the, for the clinical team monitoring the patients, [00:13:00] but ultimately keep in mind, as we generate a lot more of this data, this, this classic example of what we call real world evidence, it is also super useful. For the drug companies, for the manufacturers of radiation therapy, for the people doing research in these areas as to what’s actually happening, they have, they have access to all this data now to say, how can we make, how can we control symptoms better?

Anish Patankar: How can we make the, uh, how can we manage toxicity yet being? medically effective in sort of treating the disease at hand. So I think these are the implications that we don’t, uh, we, we have not foreseen, but we can easily see that this is something that will impact the development of these new technologies, uh, going forward.

Erik Sunset: And that, that’s such a good thing. I didn’t quite finish my thought on the value based care, uh, individual responsibility, but still the care team needs to be able to intervene and say, Hey, I understand this is difficult, but we need you to do it. We have to have you do this. Here’s what we’re going to do based on your care. [00:14:00] And to go, uh, I guess to go a step back from that treatment of the patient and being able to ingest all the data that they’re providing to that care team and then relay it on down the line to those that matter. When you talk about AI and clinical decision making, especially with oncology, Uh, that has been a recurring response on this show of where do you think A.

Erik Sunset: I. and health care goes? And that’s to physicians, that’s to surgery center administrators, that’s to a wide variety of talking heads in this space. And that’s always the first answer. Diagnostic utilization or diagnostic use of A. I. That seems to be right in your wheelhouse. Give us the, give us the lowdown there.

Anish Patankar: No, I think, I think a few factors there, Erik, right? One, when you say, you know, uh, use and diagnostics, right? I mean, for me, when I look at sort of the workflow, diagnosis leads to treatment leads to in case of oncology, also survivorship, managing quality of life and so on. It’s end to end, [00:15:00] but it starts with the diagnosis of the patient, uh, and then deciding what needs to happen after that.

Anish Patankar: Right. And that, I think, um, In the past, right? For sure. It was completely based either on the physician’s expertise and sort of their own experiences more and more over the years. You saw a lot of, uh, how do I say this? Uh, a lot of guidelines being imposed either by intake in case of oncology, by bodies like the National Cancer Institute, where they have the NCC and guidelines, and that’s just stands for standard care pathways, uh, and so on.

Anish Patankar: And then, uh, a lot more coming from payers, like I said before, and also a lot more coming from, uh, specific provincial governments and then so on. Right. So there’s so many layers of information that have to be processed by a physician to say, now that I have this diagnosis, what’s best for this patient.

Anish Patankar: Now, what now there’s [00:16:00] another angle, which is becoming super relevant today, which is the patient’s own preferences. Now, what are the social determinants of health for this patient? Is this patient coming to your clinic from the neighboring city, if you are based in a suburb? Or is this patient going to drive 150 miles for every patient appointment because where they stay in a rural area, you are the closest clinic for this patient.

Anish Patankar: So what are these patients preferences? How much toxicity, uh, can this patient tolerate? Uh, what and what are their preferences, right? Some patients have a higher willingness and some patients have a lower willingness. Uh, what are the other aspects of caregiving that this patient then has to deal with based on the side effects and, you know, even longer, uh, not just symptoms, but side effects that they may have to live with after the treatment is over.

Anish Patankar: So all of these key Criteria, like everything I said in terms of the pair, pair guidelines, push pairs, push compliance, push the doctor’s expertise and so on. And then the patient [00:17:00] preferences, social determinants of health, what the patient has to then live with based on what is the expected outcome.

Anish Patankar: There’s no magic bullet to say, if, if a doctor chooses pathway A, that’s the outcome. And then so on. And if they don’t, You choose pathway B, that’s the other outcome and then which one is better. So I think this is where LLMs can be really useful in sort of taking all of these parameters into account and then recommending and helping the physician to say likelihood of occurrence of something which can be something to be discussed with the patient that, you know, This is what we are looking at the scenarios scenario analysis with the patient now I don’t think we are there yet in terms of uh, these kind of solutions which are very nuanced and rich But we are seeing steps in that direction.

Anish Patankar: We are seeing steps in terms of uh, Solutions coming out which are dealing with specific aspects of this problem And what I really foresee is that you will now start to see solutions that will take all of these things into account [00:18:00] Holistically and helping the physicians Take, uh, take really personalized treatment outcomes, decisions for the patient.

Anish Patankar: And I think that’s really what we mean by personalized medicine. I mean, personalized medicine is a very cliche term in our space, as you know, uh, but LLMs taking into account, I mean, they can take into account billions of parameters to generate a model, but taking into account data from so many sources, I think we are really looking at a future where personalized medicine is becoming more and more relevant yet managing all these other competing.

Anish Patankar: Not competing, but other factors that a physician has to deal

Erik Sunset: We’re our conversation is branching because I’m having two thoughts here. Number one is around LLMs. I want to pause for just a second for our listeners who are

Erik Sunset: maybe, you know, maybe just a casual tech enthusiast. They hear LLM. We’re not talking about chat GPT. We’re talking about a tailor made large language model for a specific purpose. There are more than one. type of LLM out there. [00:19:00] Some are custom, some are commercially available. So just want to be totally clear there. Hopefully I’m doing that

Anish Patankar: Absolutely.

Erik Sunset: and then secondarily personalized medicine. You know what we’ve talked about? Value based care. There’s a ton of different definitions, AI.

Erik Sunset: There’s a lot of different definitions, but personalized medicine is really interesting. When you talk about the sequencing of a genome, that’s a ton of information. You just mentioned billions of parameters to create a. to ingest and then spit out as, as part of a model. Are you seeing anything around, uh, human genome sequencing for an individual into an LLM to provide the very best possible outcome or maybe the most effective drugs in a given scenario?

Anish Patankar: Uh, no, absolutely. And I, I’ll try to combine my answer to address both these issues that you mentioned, right? Now, when you talk of LLMs, these are definitely custom LLMs. This is not just taking off the shelf, uh, LLMs that are, they’re more general purpose, they’re mining information that’s available on the internet.

Anish Patankar: What we are looking at is models that are developed specifically taking into [00:20:00] account, uh, healthcare information. And this healthcare information for as input to the large language model can be all the parameters that that are taken into account by taking the input from a human genome sequence. It can be taken, uh, as input from medical images.

Anish Patankar: Uh, they can be just EHR data. They can be lab results. So there are so many parameters that you can train a specific model in for a targeted outcome. And then you say, okay, what? therapies you want to, uh, look at and sort of what are the outcomes, what has happened over the past years. And like everything else in healthcare.

Anish Patankar: It’s not always technology that is the challenge. The challenge is getting the relevant data to train your models on that will continue to remain a challenge because it’s not easy to sort of structure this data. And that’s where the beauty of the LLMs come in because they can take an unstructured data to generate sort of, you know, patterns that, you know, is otherwise very difficult to do in a non LLM world, right?

Anish Patankar: So, so that’s sort of to answer your first question or for your first [00:21:00] point and responding to your second point, right? Uh, I have an example, maybe not so much from the, uh, human genome sequence, but if I look at imaging, for example, we can already see there are companies that are coming up who are taking into, uh, you know, They are taking input from medical images and then determining that, you know, there are very costly oncology treatments.

Anish Patankar: And they are saying they study the data, of course, mainly imaging, but also other data and possibly human genome. Uh, uh, the sequencing of the genome can be a factor in that also as an input, uh, parameter to say how likely is the patient going to be successful if they undergo this multimillion immunotherapy or certain other therapy, right?

Anish Patankar: And this is of interest to the payers because they want to know for sure before you sort of go down that route, uh, because even the patients have to pay, even if it is a covered treatments, very expensive treatments. So everybody has to be, what’s the likelihood of [00:22:00] success. to for the clinical team to say, okay, no, should we go down this route?

Anish Patankar: Because it’s an opportunity cost. You cannot simultaneously put patients on five different treatments. You have to say, what’s the best for this patient. And third for the patient to say, okay, apart from the financial aspect, uh, the clinical relevancy of, uh, the outcomes of after going through certain treatments.

Anish Patankar: So we’re already seeing. Uh, when does getting to use LLMs taking into account these kind of input from, from, uh, imaging parameters from human genome sequences to say how likely is the success of a certain treatment going to be? Should you even go down that path or should you explore other parts?

Anish Patankar: Because otherwise you are left at the, as a patient, you’re left Sort of, you know, just having a discussion with the clinician and then figuring out what’s best, right? It’s not that data driven. And I think what LLM’s help here is how LLM’s help here is to bring the data into the discussion, uh, and sort of forecasting sort [00:23:00] of the likelihood of success.

Anish Patankar: And I think that’s very, very important.

Erik Sunset: Well, with what you just said, you know, in a, in a patient speaking to their physician scenario only where they don’t have the full brunt of all this data to bear, you know, all you can do at that point is get a second opinion. And maybe that

Erik Sunset: provider agrees, doesn’t agree, isn’t using data, is using data. Um, this kind of ties into another part of your example here, that if you’re, like, meaningfully aligning the patient, the provider, and the payer, that seems like a big step towards Fixing at least some amount of what’s so broken with U. S. healthcare.

Anish Patankar: absolutely. It also is going to help us reduce the friction in the US healthcare system between precisely the parties that you mentioned here, right? By, by having, by bringing relevancy in addition to experience and expertise. And that’s how I always say the Holy grail towards personalized medicine is to bring relevancy in addition to the experience and expertise of the clinical team.

Anish Patankar: And that [00:24:00] relevancy comes in terms of bringing data into the discussion and the data being used to sort of. Uh, the models and the data helping you predict what outcomes are going to be more successful for this patient. And that is what will lead to best personalized medicine. This will take time. I think everything in healthcare, it’s not that it’s going to happen next year immediately.

Anish Patankar: This will take time. The level of adoption will vary based on the criticality of sort of, you know, is it sort of something you, you take up in a chronic care case or an episodic care case and so on. And so you will see the level of adoption. happening at different speeds in different therapeutic areas for sure.

Erik Sunset: And I want to go back to something you said to lead off the show. The thing that’s the most important to you, and I’m paraphrasing, is to make a positive impact on as many patients as possible. And this is a genuine point of curiosity that you have the perfect patient that has a lot of data. They’re affluent.

Erik Sunset: They have access to physicians. They see specialists, and they have a [00:25:00] a fairly complete personal health record. Longitudinal for, for their time. You have another patient on the other hand, who maybe doesn’t have access to all the same care, but is in a situation where they really need help. Uh, whatever that scenario may be. Uh, what, to use your phrase, either something episodic or something more long, with the advent of all of these LLMs and all of this data and all of this processing power that maybe isn’t personalized strictly to that patient, but are we able to provide a better outcome to that maybe historically underserved patient or that population based on everybody else’s data?

Anish Patankar: No, absolutely. I think that is the, that is the goal we are after, right? If you, if, if the patient is in a place where they do bring all the data to the table, because they have the means to undergo all the diagnostic tests, collect all the data, and then we can use the data to, of course, you know, get the best treatment for that patient.

Anish Patankar: But in another scenario, like you aptly described, a lot of [00:26:00] that data may be missing for that patient. Because of either the lack of not having the right diagnostics available, not everywhere in the world, you can get your genome sequenced, right? I mean, that’s just not possible. Not everywhere in the world, you will have the highest quality imaging equipment available to get the best imaging, but you may have a fairly standard set of imaging available.

Anish Patankar: You may have other lab results being available. So where can technology help you is to say, can we make certain I would not like to use the word assumptions, but can we make certain models where they can learn from whatever data they have as to how that can sort of map into other information that’s available elsewhere.

Anish Patankar: For training purposes, deploy that model in another scenario and still recommend the best treatment possible. Now, like I said before, this is the scenario we want to go after. This will take time. But this is really what and how we can help deliver the [00:27:00] best treatment globally to patients everywhere, even when there are limited means to collect all the data needed for that one patient case,

Erik Sunset: And I think, I’m looking for your clarification here, Aneesh, that even in that first patient example where they’ve got a lot of data for themselves and the population which, in which they’re surrounded also has a lot of data and you can fairly, with as much certainty as is possible, provide a strong prediction versus that second patient where we don’t only have your data but we have your population’s data and we can make implications or make assumptions about it, like you said. No matter what.

Anish Patankar: right? And I know. Sorry, just to add one point that you know, the good thing here is that there are many, many governments across the world. Like if you, if you look at the UK, I mean, the NHS is certainly under stress at this time, but they’re running a huge initiative called the UK biobank, where they’re trying to collect information across the country and then allowing, of course, they anonymize all this information and then they’re allowing researchers access to [00:28:00] this information.

Anish Patankar: And then you have like the initiative of the CRSEER data for oncology outcomes in the U S where this data is publicly available for researchers to use. So access to this kind of data. Deploying models then to learn from this data and then using that during a particular patient’s case. This is what I mean.

Anish Patankar: You can bring relevancy into context, of course, for the patient when you have the data for that patient available. But even if you don’t have all of the data available for that patient using all these generic widely, you know, wide abroad data sets available will be certainly of health

Erik Sunset: Way better than nothing and

Anish Patankar: way better than, of course, absolutely.

Erik Sunset: the clarification I want your expertise on here is that whether you have the complete personal data set in addition to a really strong like cohort or population data set or whether you don’t, we’re not talking about definite outcomes. We’re talking about a probability of an outcome across a range of [00:29:00] outcomes, right?

Anish Patankar: Correct. Absolutely. We’re not talking about definite outcome. It’s going to be extremely difficult to have definite outcomes, but I think for a long time, and there’s no magic at the science itself is so dependent on how the body reacts to treatment. So, but it’s sort of the range of scenarios and what will be better for, like I said, for that patient, given what conditions they have to face after the treatment is over as well.

Anish Patankar: So all of these things have to be factored out.

Erik Sunset: So here’s the, here’s the question. I’m not trying to trip you up with it, but, you know, I, I work in health, it, I’m fairly well versed in, in tech. God forbid, there’s some big health issue that I have. I not only want the best provider, just, just like we all would, we want the best doctor, but I

Erik Sunset: also want that doctor to be using tools that have, you know, if not a, a, a billion, a trillion data points, which it’s seen and it’s fairly consistent in the, the probabilities of outcomes that it provides. And I want this doctor to be in a group where all the other doctors are using these same [00:30:00] tools. I just, I want all of the information possible along with the human expertise. So I don’t personally. See a downside here. Is there one?

Anish Patankar: So just to understand your question, of course, I think as a patient, your desire is that you, you go to a physician group where all of the information is accessible. Maybe you can clarify that a little bit for me, Erik. What, what is

Erik Sunset: Sure. I was, I was painting it

Erik Sunset: a big, a big picture there, a

Erik Sunset: big picture. So

Anish Patankar: yeah,

Erik Sunset: if I’m a patient, I want to go to the best doctor and I want that doctor to be using the best software that has the biggest, uh, data set possible, biggest accurate data set possible so that not only do I have. The guidance of an expert physician, but he’s also using the right tools or she’s using the right tools. So I don’t see any downside to bringing AI into healthcare, bringing LLMs responsibly, you know, we’re talking, there’s a limit here, but the part that may isn’t meant to be the gotcha question, but kind of is. What’s the [00:31:00] downside? Why wouldn’t somebody want to use these tools?

Anish Patankar: Oh yeah. It’s, you know, it’s, it’s, it’s a very, uh, interesting question. Let’s put it that way. Uh, in an ideal world, absolutely. I think the best models should be something that every. Physician clinical team has access to and then they can decide what’s best. Now, also we live in a world where this information is procured.

Anish Patankar: Uh, it’s not easy to procure this information, to train your models on and, and sort of vendors who will have access to this information, they will deploy their own solutions. Now, I think the question to rather to ask is what are the incentives that we have in our system? sure that this information is shared.

Anish Patankar: If you, if I take the regulatory aspect of this, right, or the compliance aspect of this, the, uh, and the U. S. specifically, there have been very good initiatives in the last few years, uh, such as the 21st century cures act. If, if the viewers are familiar with it, it, it [00:32:00] sort of mandates that, uh, vendors and healthcare systems.

Anish Patankar: Share this information when patients want it or somebody who’s acting on behalf of the patient is requesting this information, right? So I think these are steps in the right direction. Now, keep in mind with every compliance requirement, there’s of course, the cost of compliance also goes up. So there are these trade offs that have to be managed in the system.

Anish Patankar: And there are people who are, this is beyond my expertise, of course, but who have to manage sort of what’s the right balance between. imposing the interoperability needs versus the cost of making, uh, these improper of meeting these interrupt needs by the providers and the hospitals and so on. But I think these are steps that will really help to answer your question as to say, the more this information is out there, the more we will have standardization in terms of making sure that everybody gets, you know, As [00:33:00] equal care as possible, regardless of where they go to.

Anish Patankar: Are they going to the best academic center in their area? Are they going to a community hospital somewhere else? And so on, right? So evening the level play or leveling the playing field as much as possible by making sure that the data is publicly available that these solutions and that’s where, you know, even the cloud comes in.

Anish Patankar: In the beginning, I made a comment that we tried to deploy all of this innovation, no matter how we deploy our technology, we In the cloud and so on. But if these technologies are being deployed in the cloud, then everybody has access to to these technologies, no matter where you’re based. So adoption of cloud pushing the right level of regulation in terms of interoperability, even if the interop is not off all the data itself, but off sort of what needs to be shared between health care systems to deliver the right Quality the right treatment for the patients.

Anish Patankar: These kind of steps will help. And then this will take time. But I [00:34:00] think when it comes to sort of the vendors who are developing these technologies, they definitely have the incentive that they want to deliver these innovation in their solution, right? So that’s sort of the balance we will have to look at.

Anish Patankar: For sure.

Erik Sunset: That’s interesting. I, um, I thought for maybe a second you might go, go down the path of, um, maintaining the confidentiality of PHI and other PII. So it sounds like it’s more of a roadblock to data right now, which we’ve had in the U S for the last 15 years, you know, all these disparate EHRs.

Anish Patankar: Absolutely. But I think we have crossed the hurdle of the PHI and PII because all of the models can be developed on anonymized data itself, right? And that has been happening for a while. Uh, getting the constant from the patient, of course, is absolutely critical. Uh, and then safeguarding that information is absolutely critical.

Anish Patankar: Making sure that Information does not go into the wrong hands and [00:35:00] so on. And I think a lot of safeguards by the vendors, by the security players in the industry, as well as safeguards from the authorities, I think those are already in place. The challenge is making sure the innovation that comes out of it, how do you make sure that innovation is shared?

Anish Patankar: How, how do we, uh, how does one kind of innovation interoperate with the other kind of innovation, for example, right? Those I think are the hurdles to be looked at going forward, more so in my mind.

Erik Sunset: Yeah. And to be clear on the, on the vendor side, it’s, it’s table stakes now, of course you need

Erik Sunset: that security.

Anish Patankar: absolutely.

Erik Sunset: Went more towards the, uh, the patient side, you know, that consent, I don’t want my information in the cloud.

Anish Patankar: Got it. Got it.

Erik Sunset: 2024,

Anish Patankar: Yeah.

Erik Sunset: most people in the U S at least have a smartphone. You don’t have privacy.

Erik Sunset: It’s. It’s already done, we’ve passed that, we’ve crossed the Rubicon

Anish Patankar: Yes. Yes. Yes.

Erik Sunset: Well then looking, looking ahead, we’ve, we’ve covered a lot of sort of the future of, uh, how AI and how LLMs [00:36:00] can help provide better outcomes for patients. Is there a, to your knowledge, is there a disease that’s in the crosshairs now, that if we have better, uh, management of huge datasets, we can eradicate this particular disease?

Erik Sunset: Does anything come to mind?

Anish Patankar: Well, I mean, working in oncology, I mean, if you ask me of the moonshot, I mean, the cancer moonshot for me is, is, is really the one that, uh, always comes to mind and is always top of my mind, right? Eradicating cancer. I don’t know if that’s possible, but certainly managing, controlling it. Uh, and early detection, early treatment and relevant treatment.

Anish Patankar: I think with the explosion of data, putting LLMs on it, I think we can really, really handle this challenge. And then unfortunately it is a challenge which is only growing as, as people live longer as our lifestyle change, as we have an explosion of so many other factors in our environment. Uh, the uptick of cancer cases is real.

Anish Patankar: And I think [00:37:00] for me, the Holy grail is. using technologies like large language models, harnessing the data, not only in sort of the treatment of the disease, but hopefully, and this is sort of beyond my expertise now, but hopefully also understanding why our environmental factors are causing an increase in certain diseases, right?

Anish Patankar: Managing and managing those factors. too, right? I mean, it’s, this is not just healthcare, then it’s also sort of managing our environment, managing the carcinogens that are seeping into our environment and so on. So I am also very optimistic that we can gain insight into those factors as well using large language models.

Erik Sunset: You took the words right out of my mouth because managing the environmental factors, uh, managing the personal choices that are made. I mean, yeah, I hate to frame it this way, but if you’re still smoking cigarettes in 2024, you got a pretty good idea of the risks. Um, but

Erik Sunset: you know,

Anish Patankar: Absolutely.

Erik Sunset: As LLMs become more and more powerful.

Erik Sunset: And I imagine we’re at the beginning of that uptick in the [00:38:00] J curve, the hockey stick, where we’re going to have essentially limitless. And that’s easy for me to say, being that I’m, uh, not on the, uh, the tech or the development side, but essentially limitless computing power. I would imagine the prevention of some of these cancers is become, will become more straightforward.

Erik Sunset: I don’t want to say easier, but if you’re able to look at a medical record and then along with. Uh, environmental factors and the personal choices, if they’re being logged correctly, I’d imagine prevention, uh, will be more possible.

Anish Patankar: No, absolutely. I mean, it’s one thing to have a conversation from, from a physician to a patient to say, Hey, if you’re smoking when you’re 25, And you continue doing that until you’re 40, 50, you are going to, you know, unfortunately you’re at a high risk of having lung cancer, right? But another discussion to say, if you are of this genotype phenotype and you’re still smoking two packs a day from 25 to age 40, the likelihood that you will have lung cancer is 95%, that’s a whole different conversation of saying, you better stop now, you’re being stupid, [00:39:00] right?

Anish Patankar: So I think we, we are sort of making that prevention case. To be made so much stronger. Uh, and I, I feel that should have an impact. And, and, you know, it’s, it’s easy to pick on smoking because the correlation has been proven. Uh, but there are so many other things that we don’t even know about, and I think that’s the hope that we can also look at those.

Erik Sunset: Yeah. Smoking’s the low man on the totem pole in terms of, uh,

Anish Patankar: Correct. Absolutely. Yeah.

Erik Sunset: So Anish, as we kind of get to the tail end of our discussion, what else would you want listeners to know? Whether you were screaming from a mountaintop to every physician organization in the U S or the world, or to patients across the world, what would you want them to know?

Anish Patankar: I think maybe two things and to, to sort of recap what you said at the beginning, Erik here, is to say, you know. AI is not going to replace providers. There’s no way that will happen. At the same time, let’s not push back on it. I think we need to sort of keep an open mind, look at opportunities where [00:40:00] the AI technologies will actually help us make patient lives better.

Anish Patankar: So let’s embrace those. Of course, keeping in mind that, uh, with You know, with every such adoption of a new technology, there’s a risk and we have to manage the risk and we have to manage it intelligently. So that’s the one thing I will continue to scream from the mountaintop.

Erik Sunset: Well put. And I shared this before we started to record all of my, um, Non AI, non technical guests. They’re clinical, but maybe not technical. I asked them, what are their thoughts about AI and healthcare? We’ve, we’ve just gotten your thoughts about AI and healthcare. So the one I really want to ask you, how far away from a generalized artificial intelligence do you think we are, if it’s even possible?

Anish Patankar: Okay. Maybe you have to, uh, clarify the question a little bit better, Erik, for me to understand,

Erik Sunset: Sure. Sure. So I was just leading in that listeners of the show

Erik Sunset: will know. I generally conclude with what do you think will be the future of AI and healthcare? We’ve, we’ve gotten your thoughts. We’ve gotten the state of AI and healthcare and where we’re going. [00:41:00] So for you, I think the more interesting question will be around general artificial intelligence.

Erik Sunset: So I don’t want to put the thoughts of Terminator and Skynet in anybody’s head, but is that

Erik Sunset: possible? And if it is, what’s it going to take to get to a real general AI?

Anish Patankar: I think it’s gonna happen. I have no doubt about it. It’s sort of, you know, I think the Terminator or the Matrix kind of analogies, of course, are scary, right? We don’t have to be scared by those kind of things, but if you even look at, uh, I mean, what I’m very optimistic about is that these technologies are gonna help us In terms of the profile of the population pyramid that we’re looking at, people are living longer, people are having fewer children, and I’m getting into sort of an aspect as a nation.

Anish Patankar: We have to deal with these aspects. We are going to have a more aging population. Going into the next century, [00:42:00] even into the end of the century, and technologies like AI are going to help us adapt to these changes much better. We have fewer younger people working, more aged population. The needs that we see in our society today are going to be radically different 50 years from today than they are, than, than what we see today.

Anish Patankar: Right. So when you look at the general AI, right. Uh, you already see the sort of the Alexa’s and so on helping with home automation, but imagine so much more automation coming into play with AI to deal with sort of the society that we would be staring at down the line. And I think that’s where generally AI will be of tremendous help to us.

Erik Sunset: Well said, really well put. So on our, on our way out here, Anish, where can listeners connect with you? Are you big on LinkedIn or

Anish Patankar: LinkedIn, LinkedIn is best to connect with me. Please reach out. Happy to connect with folks, uh, for anything.

Erik Sunset: I’ll be sure to get a link to your profile in the show notes. Anything else you want to [00:43:00] add?

Anish Patankar: No, that’s good. Erik, it was great talking to you.

Erik Sunset: Oh, it was a pleasure to have you on a niche and on behalf of the entire DocBuddy team, we want to thank you for listening. Be sure you’re subscribed on Apple podcasts, Spotify, and YouTube. So you always get the newest episodes of the show and until next time, I’m your host, Erik. Talk soon.