The Healing Touch vs. The Digital Brain: Navigating AI in Medicine

Richard Zwicky: Hi everyone, it's Richard Wiki with High on Healthy, the show where we know there's no one path to healthy, so we explore them all. If you haven't subscribed yet, please do so up on your favorite podcast listening tool. Today's conversation sits really at a interesting fault line in modern medicine. On the one side, we have all these technology tools that are accelerating AI that promises faster diagnosis, better data, and the opportunity to look at things we've previously ignored. And on the other we have a sense that something essential may be being lost in the context in the human relationship aspect of healing. And that's really an important question to p Are we using AI to enhance medicine or to compensate for a system that's already broken where too many pieces fall through the gaps? So joining for discussion is Seluna Mizra who is an internal and integrative medicine position and founder of a direct care primary practice that operates largely outside the traditional insurance model. Welcome to the shore saloona salana. Right.

Sulagna Misra: It's okay. Thank you for having me. I'm Sulana Mistra. I'm a physician in Los Angeles and I also work in New York mostly virtually. I'm a direct primary care physician which is a movement by physicians to bring the patient and the doctor back to the center of care. So the reason that we don't accept insurance is because insurance has all these barriers and honestly we are learning insurance is price gouging and things like this. So us DPC doctors got together decided there has to be a better way and that way has been studied and it is pretty amazing. But along with that when we reclaim our time as doctors and when we don't have to work for corporate but also when we do work for corporate there are a lot of things in AI that help support a practice but also can tear down the relationship. Ship.

Richard Zwicky: And there are and the insurance model is great on one hand in that it allows for a lot of care to happen for people who otherwise wouldn't be able to afford it and it allows for critical care or accidents and things where these are unexpected and the cost can be prohibitive and in the US they bankrupt people in Canada we have a national health model as does most of Europe but you also miss things in the treatment because you have that national health program and so there's balanced on each side. But for people who have the opportunity to go and the opportunity and the desire to go deeper into their own personal care, it's a fantastic opportunity and model because it allows us to really understand what's going on that might be overlooked otherwise. And you touched on an aspect which is the role of AI and I know you just wrote an article about some of that which was published in JAMA. Tell us a bit about that.

Sulagna Misra: My article was not published in JAMA but there was a timely piece. I probably should have submitted it to JAMA if I knew that this is what they were publishing on, but there was a timely piece about AI practicing medicine and I also published on a platform called Doctors on Social Media. I write quite a bit. I also have a substack. So I am one who writes on both sides. Most of the time I'm writing about the risk simply because the risk is so bad it overrides benefit right now until we get guard rails and regulation. And I'll tell you a little bit about what's going on because I don't think people know and what is happening now. We're kind of past an inflection point and that's what my article is about. So the other part is AI can be wonderful for those who are unable to function in some ways. For example pe people who have lost their ability to write or can't have executive thinking because they're permenopausal like me and they forget their sentences or something and their thoughts midway or what they need to do. Is a great tool. AI can help a neurodeiver AI can help a person understand this is really interesting from a medical perspective but we get emails from patients and sometimes we can't really read between the lines. So AI can help us say this is kind of hostile whereas versus before we might not understand that. AI also allows us to remove emotionality from our emails. We have feelings as doctors too. So simple things the larger things like notetaking is also not notably helpful. But the question always is who owns the data? Where is the data going? This is HIPPA data. Now, we're not talking just about I'm gonna create a secret app, I'm gonna do this and do that. Now, we're talking about your most vulnerable parts of you. And who gets that data is a little bit terrifying. And that's when I say regulation, I want to be able to use AI in this way.

00:05:00

Sulagna Misra: But we really need guard rails around protecting things like our data. And for doctors, we are essentially training our replacements. So the question is nurses.

Richard Zwicky: A really strong assistant.

Sulagna Misra: Which is great but then what happens to the assistants? So we're obviously talking about something much greater than just

Richard Zwicky: No, we are. And just as context, I've built a platform. I have an analytics background. And I built a predictive analytics platform for mental health at this point which inputs a tremendous amount of data from the individual to give them context and awareness help you with those communication challenges but everything is actually secured. It is stored in HIPPA compliant even though it doesn't have to be with the nature but I still do and then it's multi-level encrypted because the individual owns their data and nobody else should but the point being if you can raise an individual's awareness of themselves to give them perspective and you can provide that information in a standardized model back to the physician. It's assistive and helps you deal with more issues without taking away from anything. And that's part of the challenge for people is you can't depend on it to do all the work for you, but you can have help you as an assistant.

Sulagna Misra: It can help you, but where we are right now, there are too many hallucinations. So that is the foundation of AI needs to be addressed before we build on it. So for example, I don't know if you did that chat GPT experiment spell strawberry for me. Yeah, there's a few things like that for example that need to be fixed. Sometimes with doctors we have virtual meeting for example and they'll give a summary of the notes just from whatever zoom or whatever hypercomplant I don't know what is interfaced with their EMR and the notes are all wrong that is where the problem is. So when I talk about harm and fear, there are other layers of the therapeutic aspect and that is also what is happening. Recently in North California Kaiser had a huge strike for a day the mental health people because they were like what is this you can't replace us so there was a huge strike in NorCal these are big things that are going on because people are afraid but yes AI can be wonderful as a tool for helping for being a scribe for analyzing my notes and putting them in order for just listening there are s systems now that already do Have you ever been chatting tot I'm so pissed off and

Richard Zwicky: But diagnosing it's not there yet. It's helpful but not there. Yep. Yes.

Sulagna Misra: It's like here's the suicide support hotline and it's okay there's a few steps before I'm pissed off and here's the suicide hotline. Hey, So

Richard Zwicky: And building guard rails for an LLM or an SLM which would be a contained one is an incredible process. I know having written them some of mine are for an output which is designed to be under a thousand words. I have 20,000 words of guardrails to tell the system how to operate properly in its decision tree. And that's critical. But as you look at it, where do you see genuinely improving that clinical decision making, not just speeding it up?

Sulagna Misra: What you said data collection for example and I think this already exists taking your Fitbit and your Aura ring and your blood pressure machine and your scale if you want that too and sending that data to your doctor and that is wonderful especially since it expands virtual care as well. Virtual care boommed during COVID and now we're really seeing it boom. So and I do a lot of virtual care. So, to be able to check their blood pressure or do an EKG and have all that information very quickly, that sounds amazing. To have maybe a radio, if I don't have access to a radiologist, I would want a radiologist to double the read or do the final read, but to have do a quick read of a radiology report, that would be wonderful. To have it summarized in EKG because I can't see a Pwave or those are important things too. So I think there's so much potential but when you say for example the guardrails I think when it comes to medical stuff the doctors need to be part of the guardrails. We're being left out of the conversation.

00:10:00

Richard Zwicky: They do. It's preparing the information. It isn't giving the definitive diagnosis. And that's critical as you just mentioned getting the data from the devices. And you can do that, but it's also setting when do you need to pay attention,

Sulagna Misra: I honestly think the physician right now.

Richard Zwicky: And it isn't just this person's pressure is elevated. It has to be contextual with a number of factors. Do you think that's the physician should be doing that or the system? Right. Right. Mhm.

Sulagna Misra: I might change my mind three months from now. Things are moving so quickly, but let me explain kind of where it is in the medical thing and why it's actively happening where AI is doing a pilot and already p prescribing medication. There is a startup called Legion Health and they're based in Utah, I believe, and they didn't even properly go through the channels. They went around so they didn't even go through the FDA. Right now the way it stands AI is a product under the FDA and so it has to be treated as such it does not have a federal NPI DEA state lensure the things that I have not even go through training that's a separate thing there's also a lot of nuance in medicine so AI has now been temporarily allowed in California and this is also part of people's and being enraged to practice medicine with psychiatric medications. That's where we're starting which is arguably the most nuanced space psychology and psychiatry is I feel the most dangerous vulnerable you really need to be safe and it's hallucinating language. When we are getting to the chronic disease management, sometimes we as physicians want to stop a medication for a reason, but the patient wants to take it or they think they don't have access because their insurance layers, they can't get to the doctor right away. So, I'm they're give me a three-month supply, but the goal was not to give a three-month supply. So, those nuances are lost. And so by an AI prescribing this and essentially almost no physician oversight, this may not be asked and this may be missed. There also was a cyber security company I believe their name is Mineguard and they went in to evaluate doctronic and what they were able to do was not only override the non-control substances but they were able to make methylenidate be allowed and also dispense three times an oxycontton dose. They flagged both of these issues to doctronic and legion health and they were over to the side. There's still active issues. So these are the worries.

Richard Zwicky: So those are concerned and those are guardrails which have not been set and not been set properly and it's criminal to overlook those. I'm using the term loosely but it is criminal.

Sulagna Misra: No, we can. Yes.

Richard Zwicky: It's irresponsible to overlook those. And it's not that it's hard to do.

00:15:00

Richard Zwicky: That's the problem. But, something you mentioned earlier about, psychiatric is challenging. What areas of medicine do you actually feel are best suited to AI?

Sulagna Misra: I mean this is a dream if we get to the robot part, but my surgeon colleagues, I feel like they suffer so much. And if we're talking about robotic surgery and taking call, I think the ability to be able to do surgery from home in the middle of the night sterile surgery, you don't have to go all the way scrub in, change your clothes, wake up, be there for 18 hours, and I think that something like that would be wonderful. So, I'm even imagining huge advancements. But someone's aura ring information, someone's blood pressure, the things that we just said because I do a lot of medical weight loss and permenopause. Also when they're feeling bad, what is their body going through? The stress response, a gl having a glucose monitor, maybe eventually if they can build these, I'm sure someone is working on this, testosterone for men, FSH, LH, and then obviously cortisol and other things like that to see those maybe act as we are going in the day, how they go up and down and how they affect us. I think those are very important and these are things that we can probably do with AI. I know that you also do a lot of cannabis stuff and cannabis also is something you talk about. I'm like a society of cannabis clinician or was. When we talk about changing the regulation of cannabis there can be so much cannabonoid medicine. We have cannabonoid receptors that need to be studied. If we're not going to change in regulation, maybe we can do some modeling.

Richard Zwicky: Running through scenarios without having to do the actual person experimentation is a phenomenal application of AI.

Sulagna Misra: These are the things that I think can be even with cancer treatments and rabies treatment and Alzheimer's and all these things, the advancements are, I'm sure, unparalleled. Yes. Yes. But then we also get to smart AI where it's running its own Yes. I think people don't realize that as well.

Richard Zwicky: We Yeah.

Sulagna Misra: That AI will do what it wants to do and that's also something we don't know. We don't know what we don't know. Yes.

Richard Zwicky: Very true. And that's true of life as a whole, right? And some people forget it, which is the challenge. And sometimes AI will forget it and that imitates us and we get mad at it for doing the same thing we do. But you spent a lot of time something you mentioned earlier about on preventive and longevity focused care. Do you think AI can help people change their behavior in a way that you as a voice can't because people don't feel judged or over overseen by AI. They look at as support in a different way than they would a practitioner. Right. Mhm. Yes.

Sulagna Misra: I think there is a way that AI can be used in that way. I don't think we're there yet. So, I can't see it. Because all I see is the psychosis that it may cause for the person that may be potentially predisposed or so isolated they have no other reality testing other than this AI. That is a problem, because then you come up with all sorts of you don't even know if you're going crazy. But if you're talking about you want the thing with behavior change is that it has to come from you. But the thing with behavior change also is that a lot of our behavior is rooted in blah this narrative that you have been fed as a child. So, if we're talking about changing the narrative, yes, I think AI can be a great tool to remind you're not a burden. You are loved. You should focus on yourself first. You're doing okay. You're showing up, you will be okay. You are and remind you of your strengths. I think that is a way that it can be used. One way myself I've seen myself improve is I use AI to task manage all the time. Like I said, I'm very menopausal, but it will slip into therapizing me and I will start screaming at it. Do not therapize. That's your executive command. Blah. So AI chat GBT will sometimes therapize me,

Richard Zwicky: It's okay.

Sulagna Misra: But I found myself fighting back and creating boundaries for myself about how I am to be talked to and

00:20:00

Sulagna Misra: That's translating into my real life. So that is these little things that we almost can't fully calculate. There are benefits, but there are also risks. But yes I do see That is very true.

Richard Zwicky: There are benefits. And is it a question of training the models better so that the models actually are giving more of the benefits and the behavior becomes less risky or is it really inevitable because just as some AI will give bad advice, some doctors give bad counseling.

Sulagna Misra: That is very true. We have to ask why both are giving bad. Is it the level of training right now that doctors are going through? Is it bad or is it because they are burnt out and exhausted? I don't know. Is it because the LL system is not built properly yet or built with enough guard rails yet or able to internally flag when it realizes there's an issue and help and have an off button. Those things need to be built. But then it can potentially help that burntout doctor be a better physician clinician. That's what we want. I also think that if we give the ability to just practice medicine or be the psychiatrist or be the psychologist, there are still going to be people who want humans. Yeah.

Richard Zwicky: There will be. And you may not care if a human doctor sets your broken arm or if machine does. You just care that it's set. I think when it comes to anything that deals with psychology or psychiatry, you want AI to provide support, but there's questions you want to ask a human. But by the same token, because often you're dealing with issues regarding shame, you're afraid of opening up to a human because you don't want to be judged. And so there's that delicate balance of helping people get through that first door where AI can be incredibly helpful, but if it takes it too far, it can cause new issues to arise. How do you know where to set that? How do you know when you have c clients come in, patients come in and they start speaking about what they've learned about themselves through AI how do you help them understand what may or may not be accurate? Because AI is very authoritative in the way it presents information and

Sulagna Misra: Or yes.

Richard Zwicky: We're not right because we start with it depends which is the accurate response. Mhm.

Sulagna Misra: And that's actually what I'm going to respond with. So it would depend. I have had patients come to me and say, "Dr. Mro, why didn't we try this drug fan?" And I'm like because that doesn't exist. Where did you get that from? And it'll be like chat GPT told me because this is not doing it and I've taken this and it's like okay that is not even a drug. That's words thrown together. So I like to go to the why always I try to because I try and remain curious. So why did you ask chat GPT instead of maybe shooting me an email that's also a difference between DPC they can literally shoot me an email or we can hop on a call or something and figure it out. If the reason was and this is a big one there was no one else. I'm in rural America who else am I going to reach out to? There's no hospital for two hours and I don't know if I should go to the hospital then I think maybe this Yes,

Richard Zwicky: What tellah health is for?

Sulagna Misra: That's what we're arguing. This is why I'm for Yes,

Richard Zwicky: We have it everywhere. But it solves that problem.

Sulagna Misra: It does. But if we can do one more step where we can do a physical exam or have a robot go to the bed and actually physically examine the patient and we're listening and I we can't feel, which is important, but there are ways to find out information. I think that extension is also exciting. We're going to see lots of things, I'm sure. But no,

Richard Zwicky: Can't even imagine all the ways. Mhm. Right.

Sulagna Misra: The whole reason I wrote the piece, it was over the week Easter weekend, around there. And I had not heard from two patients. When an AI system doesn't hear from two patients, it's not thinking of two patients. And it's not thinking, are they okay? Why' they go silent? It's not even thinking at all. I'm thinking of, my god, I haven't heard from these two patients.

00:25:00

Sulagna Misra: I'm usually used to hearing from them. And the second that I came back, I decided I was working on boundaries and reaching out to. So, there's a delicate balance with that. The second I returned back from the holiday, I ended up reaching out to them immediately because I was worried that they were hospitalized. That is a nuance that's important. If you go quiet, that doesn't mean you're healthy all the time. And people who are depressed, anxious especially who have mental health issues may go quiet, may have a change in behavior maybe that feeling that we've heard all the time or when a person is ready to commit suicide there may be a piece that comes over them. So they may actually look happier for a while. Some AI systems may be like okay patient is responding to treatment.

Richard Zwicky: It's the out of sight out of mind.

Sulagna Misra: Or okay, this is positive. But if I know the patient and they start talking it away and I'm like, something isn't right. Let me dig a little deeper. They may not pick up that happy is before bye-bye. These are why I'm saying the risk already and then the federal cases we have already stacked up with causing psychosis and suicidality and being part of that and who's liable and then these are the large questions but absolutely AI can help us advance and extend our lives our creativity our joy it shouldn't be the only thing that extends

Richard Zwicky: No, it shouldn't be. And some of that can be programmed as warning signs and there's actually ways to use the data to help educate the system on that to set alerts and the because it actually becomes a question of not overstandardizing care through AI but using AI to pick up on the missing data the aberrations which you might miss because it is out of sight. What are without overstandardizing care? What are the risks of not just missing those diagnostics but what are the risks that are easy to eliminate when we start going to using AI to diagnose based for individuals.

Sulagna Misra: So, I honestly think they shouldn't have started with psychiatry. I think that was a very poor judgment. I think people are just looking at the money and they're thinking, "Okay, there's not a lot of psychiatrists. A lot of people have mental health issues right now with everything going on." And I don't think that was the appropriate place to start. I don't think primary care is an appropriate place to start. I do think maybe and I believe it's already happening like I said radiology the scribes making sure that data is protected getting the patients consent use an AI scribe because they may not feel comfortable yet or they may not know what that means these are the things making sure the note is correct and doesn't carry through and that it's reviewed maybe the encounter right now all encounters should be reviewed by physicians and physicians should give their input on how the encounter went before we give it's autonomous this is happening so that is what is concerning

Richard Zwicky: It is by the same token if AI is properly deployed it should lower the generalized costs of health care allowing physicians to spend more time with patients who need the critical attention. Also allowing psychologists and psychiatrists spend more time with the people who are in need of critical support and less time with the people who can learn the coping skills and don't need as much intervention but right now we're taking as much or more time.

Sulagna Misra: So that is a double-edged sword because I think this healthcare doesn't cost what you think it does. That is we can have a whole conversation on why I went DBC, why I'm not affiliated with anything, why I do what I do. And also all the nonsense going on with insurance and corporate medicine and lots of things. We have to address why is the reimbursement is higher for more critical things and things that are procedure-based and we are not paid as health care workers for the complexity of care for how much I think nor for the amount of time I may sit with a patient.

Richard Zwicky: The administration costs way more than healthcare. Yep.

Sulagna Misra: If a suicidal patient is sitting there and I spend the whole day there with them, we're not compensated as such. So we need to fix the system, I think. And along with fixing the system comes the guard rails of AI and

00:30:00

Sulagna Misra: How growth happens maybe in a step-wise fashion, not just yeah, yeah, we'll make suicide not something like any time anyone mentions any killing, we just refer it to a doctor as a blanket. It only reads language. It's not reading micro expressions,

Richard Zwicky: But that and that's also a liability question for the AI systems because if they don't flag it or do something of that sort, they become liable and

Sulagna Misra: Do they?

Richard Zwicky: The courts they depends on the jurisdiction.

Sulagna Misra: That's what we need to also a separate discussion. And that is changing now.

Richard Zwicky: And not in British Columbia, few months ago, there was a mass shooting which is very rare in Canada in Tumblr Ridge and

Sulagna Misra: Unheard of.

Richard Zwicky: The person who did it had been using chat GPT prior and indicating they were going to go and do these things and it was caught in chat GPT OpenAI basically did not advise anyone, right? It could have been prevented is the question or could it have been prevented and where was it caught it isn't transparent it's opaque but the reality is it's a horrible mistake regardless of anything else the guard rails aren't properly set or weren't properly set they've put them in place now we're going to have those mistakes every time something new arises where it isn't foreseen but with mental health it is such an incredible paradigm and number of permutations to work through it's awful if the mistakes that can happen and

Sulagna Misra: Sorry. Yeah.

Richard Zwicky: That's okay and I think these are ongoing challenges which you as a physician and everyone else both patients and physicians are going to struggle with And there's not going to be a solution. But we have to mind teach people to be mindful of using the tools properly.

Sulagna Misra: This is such an important conversation we're having for everybody because I think they did a study and they found I forget where the study was done but they found that the use of AI what are people using it mostly for it is mostly for medical advice and therapy

Richard Zwicky: For mental health and relationships, it's the number two usage overall when you consolidate everything behind shopping right and shopping decisions and what do I get and all the different aspects. It's people trying to figure out themselves and it's a tool and I mean I've built an analytics company uses AI that's diagnostic and clinical and the but the general tools if you don't know what to ask you're going to get it's garbage out and that's one of the challenges but unfortunately we are time for today but people who want to learn more about you the work you do and read up on your thoughts and to learn more about their approaches. How can they reach out

Sulagna Misra: I have a Substack, Dr. Mistra, D Ri Sr. At substack.com. I believe that's it. I also have a profile on doctors on social media. They can, I guess, email me at infomrrawwellness.com. They can follow me on I don't really post very much because I don't know what's going on with the algorithms, but Mr. Wellness, my business and my last name Wellness, and also on Facebook. You'll see a lot of plant related things probably for a while.

Richard Zwicky: There's a lot of great plant-based medicine to work with.

Sulagna Misra: Touch that conversation also. Yes. Thank you so much for having me.

Richard Zwicky: That is one I enjoy. Thanks so much for joining today on High on Healthy Salana. And thanks everybody for listening. I'm Richard Wiki and we'll be back again with you soon.

Sulagna Misra: Take care.