The Future of SMI Treatment with Dr. Scott Feers

welcome to Mental Health Evolution, a podcast about what's changing in mental health and why it matters. I'm your host, Rachel Harrison, inviting you into honest conversations with people from all perspectives in the field. Clinicians, tech founders, investors, insurance companies, and all the folks in between. Let's explore what's working, what's not, and what's next. Welcome back everyone to the Mental Health Evolution Podcast where we are talking about how the landscape is evolving in the mental health industry. And today we are joined by Dr. Scott Fiers of AMAI Health, a company that is integrating technology and data to improve care for people living with severe mental illness. In this episode, you might hear us refer to severe mental illness as smi. That's kind of a common way that it's referred to in the industry, so I just want to make sure everybody listening kind of has that little tidbit. Dr. Fears is a board certified psychiatrist and neuroscientist and serves as the Chief Medical Officer at AMI Health. He is also a professor of psychiatry at UCLA and has spent over 20 years dedicated to advancing care research and education specifically for individuals facing severe mental illness and addiction. At Imai Health, Dr. Fiers and his team focus on combining face to face clinical care with innovative AI tools and wearable technologies to monitor and support patients aiming to improve outcomes and reduce re hospitalization. AMI Health has partnered with major hospital systems such as Cedars Sinai, New York Presbyterian and Stanford Health to expand the reach of their model. As always, I want to kind of give you a high level information our listeners a high level here about some articles related to the conversation. So we all kind of have a baseline for where we're coming from. The first one I want to talk about is from CBS News and it's an article called for profit companies open Psychiatric Hospitals in areas Clamoring for care. And this article kind of talks about a trend in which for profit companies are coming into the market for psychiatric hospital care. And this kind of after many nonprofit and government run organizations have had to close their doors. I know we've seen a lot of that in the mental health treatment world where we've got hospitals and residential care closing their doors, unable to fund care any longer. So this article talks about some of the competition, the issues with Medicaid payment as well as staffing challenges that are not solved by new businesses moving into the space. So that's a good kind of place to jump from. Another article here is about private Equity among US Psychiatric hospitals. And this one is from the JAMA Psychiatry Journal and it talks about trends in private ownership of psychiatric hospitals and evaluated outcomes of care, ultimately showing that there are no indicators that private equity owned facilities decline in quality of care. This is one of the first research articles I've seen kind of talking about this quality of care piece and they're showing some signs that these facilities are actually improving care. So I thought that was also an interesting tidbit for our conversation. But lastly, I want to talk about this article specifically about Dr. Scott's company called AmaiHealth Announces 25 Million Series B to advance toward a cure for severe mental illness. And that was from the business Wyat. And this article talks about the plan from Amaya Health to change the way that smi, or serious mental illness is treated, specifically applying these inpatient clinics. There are three technologies that his company is looking to integrate with human clinical care for optimizing treatment results. And these are going to be kind of the foundation of our conversation. So I want to spend a minute outlining them as I understand them and then we will dig in further and I'm sure Dr. Scott will give us some more insights here. But one of them is called Patient at a Glance. And this is the idea that there's a dashboard for clinicians thinking about like a dashboard like you would see maybe on your phone or even if you want to think about your car, like something that shows you the different information at a glance at a quick way to see data about a patient. So that's one of the tools that they are using. Another one is called a point of care assistant and this is an AI powered clinical assistant that monitors the status flags, changes, offers treatment recommendations. They're clear that providers remain completely in control and adjusting this information with their clinical judgment. And last but not least, symptomatology, phenotyping. And essentially my understanding of this is looking at symptom clustering based on longitudinal patient data, which allows them to create digital phenotypes that then providers can use to deliver highly effective, precise care based on a unique set of symptoms. So with that we will dive in with Dr. Scott. So I want to start understanding a little bit about amai. They represent present so many changes happening in the industry applying to inpatient care and serious mental illness. But I'm curious how you decided to get involved with this organization and what some of the evolution of this treatment integration with technology has been like.

Well, yeah, so great to be on the show. Great to meet you and be a Part of this conversation. But yeah, so I've, you know, been a psychiatrist for about 20 or so years, and really there's been kind of two threads to my career. And Amai brings both of those threads together. One thread is the clinical thread. I've always been interested in more complicated psychiatric cases, you know, kind of what we call severe mental illness, which includes schizophrenia, bipolar. I actually think of borderline personality disorder when it's severe. I include that in the category of smi.

Okay.

Some people don't, but I think some folks with severe borderline personality suffer tremendously and they have a very heavy burden of symptoms and they can really cycle in and out of inpatient care and just not have a great place to land in the outpatient to really get stable and kind of start working on recovery in a more focused way. So, again, SMI has always been an interest of mine clinically. My other career thread has been more of a basic science thread. I actually started out molecular genetics of cancer, and then during my medical school training, became fascinated with mental health and psychiatry, and in particular the fact that many of the molecular genetic tools that had been developed for cancer were moving into mental health. This was in about 2000. So really kind of interested in the research transferred from cancer into mental health with the focus on gene mapping for psychiatric disorders. But we had a particular focus, which is using quantitative, what we call intermediate phenotypes to map for genes. So instead of saying, I'm going to map genes for bipolar disorder, what we would do is we would collect a lot of objective data from people with bipolar disorder, including, we would have them wear fitness trackers, and we would be able to pull out lots of circadian rhythms, sleep phenotypes, that sort of thing. From these, the Fitbits or fitness trackers, we would also do cognitive tests, temperament tests, and we also did quite a bit of neuroimaging and then would be able to measure kind of specific brain structural phenotypes and then use those to map genes. Now, there's been a tremendous amount of work in this field. For example, pretty much every psychiatric population. If you dig into the literature, there's data out there showing what the Fitbit profile for people with bipolar disorder, schizophrenia, adhd, and showing that these fitness tracker profiles are quite different in these populations. As powerful as a scientific tool that these have been, they've never really been effectively integrated into the clinic. So when I had the opportunity to join mi, it was really with the goal of combining the good old conventional human driven care therapy groups, individual that sort of thing, and also to integrate in all of these tools, which again, have scientifically been used to study the psychiatric populations, but have really not been integrated into the care delivery. So I met the founders of the company, Estaf Sokolin and Sonia Garcia about four years ago. Now, at that time, I was the medical director of what the UCLA Depression Grand Challenge, which is still going on. And UCLA's depression grand challenge had the goal of building scalable models of treatment for mild and moderate anxiety and depression. And we had been working on that. I'm still kind of associated with that group now. They're continuing to do very good work. About four years ago, Stas and Sonja found me through UCLA and said, listen, we want to do something similar, but for serious mental illness. And so at that time we began conversation. We opened our first clinic in Los Angeles about three years ago. We now have five clinics across California, New York, North Carolina, with plans to open another even half dozen by the end of next year. And the clinics are really combining again, conventional human focused care, brick and mortar, in person care, which is the foundation of all mental health treatment, to my mind, is always going to be human based. So I want to say that from the outset is AI is not going to replace human care. I think it has an important role for augmenting, making certain aspects of treatment more efficient and really kind, kind of again augmenting what humans do. But I have a very strong belief that humans and connections to humans is the core of mental health treatment. So just want to be clear up front, that's the core, but that we can then bring in all of this technology to augment that and make it more efficient and again, kind of integrate the technology into conventional care. I also would like to add one important theme for MI is we have a very strong focus on physical. I think mental health often focuses on therapy and meds. Often as psychiatrists, we're prescribing meds to folks and then as they walk out the door, we're saying something like, oh, by the way, make sure you exercise and eat well. Which is an extraordinarily difficult thing to do. I mean, it's very hard to change habits of exercise and diet. And yet we know how important it is. And yet it's not really a central focus of mental health treatment. So one of the things we do at MI is about half of our treatment program is focused on changing diet, being able to exercise, move, kind of change those habits, and also deliver primary care. So we have primary care providers in our clinics. We know that folks with serious mental illness often have reduced lifespans. You know, some folks, Even lifespans are 20 years less than someone without a serious mental illness. And a lot of the causes of mortality are basic things like cardiovascular health. So included in our treatment plan is a very intentional focus on physical health.

Wow, I really love that. So that kind of leads me to this first piece about the wearables, which I think is so fascinating that you even have some of the phenotypes with these wearables. So what does that look like as an example? I mean, obviously you're tracking steps potentially, right? Sleep cycles, things like that. But how is that used then in the clinic when somebody comes in, Right?

So I think these wearables, I mean, generate an incredible amount of useful data. For example, in mental health, we know how important sleep is, right? And when folks come in, it's probably one of the most common questions, how's your sleep doing? You know? And we know self reports of sleep are notoriously difficult to really get, right? I mean, some folks, they feel like they're awake all, all night, and the reality is they're up for an hour or so, but they do fall asleep. And then they're kind of in maybe a restless state of sleep, but they're actually sleeping. And they'll come in and say, I didn't sleep at all last night. And having the objective data regarding sleep just adds a lot of depth to the conversation, because then we can start to look and say, well, it looks like you were actually in bed for seven or eight hours. How was that? When we initiate a treatment intervention like a medication or CBT for sleep and CBT for insomnia, really helpful to be able to track, to say, hey, it looks like we're on the right track. Your Fitbit is telling us your sleep looks like you're sleeping longer. Or we start a medication and say, actually, it looks like we're going the opposite direction. It looks like you're getting higher rates of sleep fragmentation. You're not falling asleep early. And this is something we need to know and be on top of it. So right now in the clinic, we really are focusing on those basic measures that you mentioned. Daily activity. If you're walking 2000 steps a day and that goes up or down, we see it. We can also target that as something that we want to change and say, hey, you're at 2,000 steps a day now. Let's see if we can get to 2,500 next week. And we have something to track and follow that. We also look at the heart rate variability, which is really emerging as an interesting measure of stress, that sort of thing. And again, I mentioned the sleep, where we can kind of track sleep. Now, those are the basic measures. Anyone with a Fitbit app or an Apple watch or an OURA ring, you can open your app and see nice displays of all of those measures, including things like readiness scores and that sort of thing. So us using that data is not that innovative, or I should say it this way, the innovative part about using that data is we're taking this data that's available to customers everywhere with any of these devices, and we're integrating it into a routine part of the assessment and treatment planning. Now, that's the basic summary data. These things also collect a tremendous amount of data, even below the level of the summary data. So the summary data might say, hey, you took 3,000 steps today. But these things actually contain pretty granular data. And they can look at things like fidgety nests, restlessness. We can combine this with phone data and say, you had 3,000 steps today. Was that using GPS data? We can say, was that going around the block or was that pacing around your living room? Because the individual that's taking 3,000 steps and a walk around the block, that we want to encourage the individual that's pacing in circles around their living room, we might want to say, listen, is this a problem? Are you kind of restless? What's going on there? So where we're putting effort now is to collect all this data and really start to build the algorithms that can dig much deeper. One of the first things I want to develop is a relapse detection system, such that we're pulling in all of this tracking data plus other data. We can talk about the other sources of data. We can work with and use algorithms on the back end to really identify anomalies that might represent early, early evidence of symptom relapse. That then allows us to reach out to that individual and say, hey, it looks like something might be going on. And what do you think now? They may say, nope, everything's good. You know, I've been traveling, or I went to a wedding and so my routine's been off, but I'm doing great. Or they may say, things are starting to slip a little bit. I think I'm doing okay, but things are slipping. But that allows us to have the opportunity to come in and with a much lower risk intervention, really help someone before they fully decompensate and need to go back into inpatient. Inpatient care can be tremendously helpful. But the reality is a lot of times that's a week of being in the hospital loaded up with medications. Medications can be very helpful, but they also carry risk. So our ability to identify a relapse very early, intervene again, hopefully that's a lower risk intervention where we can say, hey, your sleep is starting to slip. Let's work on some mindfulness exercises at night to see if we can right the ship before things fall apart. Right. So it allows that early intervention that can then hopefully prevent the full decompensation. Save a lot of suffering, saves a lot of money, resources, but again, the cost of a full decompensation. There's a lot of human suffering associated with that. So those are the sort of things we want to catch them early, see if we can do low risk, simpler intervention to prevent the more complicated treatment interventions.

Is there also kind of that biofeedback component utilized where patients are actually also self monitoring, maybe even having some kind of a look, you met your goal today, reward system kind of piece?

Yes, correct. So part of it too is the Fitbit. It does help engage people, right. A lot of times folks come in, we give them some therapy, we give them some meds and they walk out. And then they're kind of outside of the treatment environment as opposed to if we give them a fitness tracker and say, okay, here's our goals. This is what we're aiming for. We're aiming for eight hours of sleep at night and let's track it that helps engage folks. Not everybody. I mean, there's a lot of people that have Fitbits that never open that app or never open that app after the first week. But I do think when we work with folks, we really are trying to use it to engage in their care. Now, in terms of the feedback, where we're at for now is we very much want to have this to be a human interaction. So if we identify something that we feel might be an anomaly or represent a miss, we want humans care trained clinicians looking at that and reaching out to the individual to have a conversation. I think where it goes over time. And we're working on this, but we're still keeping the human clinicians as a very central piece of it. But where it goes with time is we start to automate some of the feedback. So if there's the algorithm or identifies an abnormality, we can send a simple ping to say, hey, how are you doing? Looks like there might be an anomaly. Let us know how you're doing. Right. And then so we get that feedback, which by itself is an intervention, kind of ping someone and say, hey, it looks like you're kind of stressed, your heart rate variability is getting constricted. Do you notice anything that by itself is an intervention because that person may not be aware of it. And just that simple alerting might be enough for them to say, yeah, I am experiencing a little bit of stress and let me use some of my tools. Right. And then over time I think we can develop more complicated algorithms that are actually starting to do some clinical decision support. Now. Again, need to be careful here. I think the analogy of the autonomously driving cars is a good example. These tools are really developed with a lot of human involvement at first. And then over time, as the algorithms get better and better at what they do, we can kind of turn more and more over to them. So I do think there will come a time in the future where these systems, these algorithms, these technologies are kind of, of monitoring moment to moment, day to day sort of things and can start to make some clinical decisions, you know, and make recommendations. We're not there now, right. We're, we're at the stage where we've got the, the human driver is still sitting in the autonomous driving car, really doing a lot of the work, allowing the, the, the AI algorithms to, to really develop and become more mature and that sort of thing. But eventually it will go in other parts of medicine, for example, radiology, image analysis, AI is doing tremendous amount of work. Most of that, as I understand it still has a radiologist involved to make sure there's good quality on the reads. But that's an example of something where computers are good at image recognition or these pattern recognition in images. And so you can see there as a place where the algorithms, the AI is starting to, I wouldn't say take over radiology, but starting to do a lot of the kind of the base work and create a scalable system where these images can be read more efficiently.

Yeah, yeah. And that kind of leads into this AI piece that you have, this point of care assistant. So it seems to me that that is an AI generating potentially patterns, ideas, thoughts for a clinician to review and look at. Am I getting that right?

Right, right. Right now it's actually a little more straightforward than that. It's not really recognizing complex patterns that the clinician couldn't recognize. What it's doing is really augmenting the clinician. So for example, we can do something like we'll develop a treatment plan and with the individual patient, we call patients members at mi. So with an individual member, we'll sit down and say, hey, let's make a set of treatment goals here. Some that are gonna have to do with physical health, mental health, school and work, kind of vocational activities, social activities, family relationships, right? So we can build a set of goals. Now we do this all the time in mental health care, but the reality is once you're into therapy or treatment over the long term, it's very easy to lose track of the fact that you have this kind of whole range of goals that you're working and often you kind of end up focusing on one or two those things get better and you kind of forget about the other treatment goals. AI and these large language models are great for giving it the treatment goals and saying, here AI, here's 12 treatment goals. Look at our sessions over time or on a session by session basis and tell me what treatment goals we've been working on. On, right? So the idea is, as our providers open up a note, there's actually a little dashboard that says here's the treatment goals, here's ones that you've worked on and are successfully kind of moving towards completion, and here's the ones we haven't even started talking about yet, or we started talking about them three months ago, they fell off the kind of list. But we need to keep our eye on these goals or at some point return to kind of talking about these goals. Right? So that's one kind of very basic function that again, that doesn't really change. I mean we've always had treatment goals, we've always known it's important to have these multi domain treatment goals, but they're very hard to track. And so here's kind of a simple way to have the AI in the background going hey, by the way, checkbox on these three goals. But we're still in an incomplete stage of these other goals, right? So that's one, one kind of issue that I think AI can be quite helpful with. There's quite a few others. We can pull meds from outside records. We can make sure the physician or the prescriber is seeing all the meds that are on and able to kind of in a, in a straightforward interface kind of do a medication reconciliation to make sure everybody's kind of understanding what meds are being taken or what, what meds aren't. We can introduce the, the fitness tracker goals to say here's what their average sleep looks like over past week, over the past month. Is it going up, is it going down? Right. So this very basic set of data that Again, conventionally, this is not adding anything new in terms of how we think about treatment. But the new thing it's adding is the ability to kind of pull all this data together and present it in a way that as the provider, you know, whether it be a therapist, prescriber, is, is able to look in a very succinct way to say, okay, here's where we're at.

This may not quite be your purview, but I'm curious from a cost perspective, like, do you think there is a, I'm hearing you say there's a benefit to patients. Certainly. But I'm also wondering, with things like insurance and all those things, is there a cost savings associated with this? Is there something that Amaya is looking at or not?

Yes, we are. I will say right now we're not getting reimbursed directly for this care. At the same time, I believe the cost savings, to go back to the example I gave before, if we can detect a decompensation early, intervene with a low risk intervention that prevents a hospitalization, that's tremendous cost savings. Right there it is. I think the other, Yeah, I mean, again, hospitalizations in addition to the cost of human suffering are quite expensive. Right. And so you know that, that is one of the costs that we're directly kind of targeting in terms of savings. I think the, the other cost savings are one the, you know, making care more efficient. So for example, you know, AI scribes, I think those are becoming quite popular. We've been using them for a year plus now. I feel like they do. And they, my notes now, you know, we transcribe session, my note pops up right after the session in two or three minutes I can review it. Most notes are pretty good. Do find some, you know, kind of errors or you know, kind of miss misquotes that I can correct. But even when I have to correct a note, my notes have gone. An intake note that, that might take 20 or 30 minutes to write now. Five to eight minutes. Right. And a regular follow up used to probably take me, I don't know, 8 or to write that follow up note now really within two or three minutes. Even with some edits I can kind of read through, have a good note sign. So the cost savings in terms of efficiency, those, I can see those now, those are actually making my job as a clinician much easier. So those cost savings are there. I do think a lot of the cost savings are going to be kind of spread across decades, which makes it a little hard to argue to reimburse it but in terms of individual's physical health and kind of long term general physical health, I think there's going to be a lot of cost savings kind of spread over there. The one other place where I think this is going to be more of a challenge, but I think the one other place where there can be cost savings is that same system I'm talking about as a relapse detection system where once someone's kind of in a healthy balance state, you know, that system is going to detect anomalies in a way that allows us to intervene, you know, kind of step in and intervene. That same system, if we put that in place as we initiate treatment so, for example, start a medication, that same system can detect changes that indicate either, you know, we should make an adjustment to the medications or we're on the right track. And I think there's cost savings there, again, both in dollars, but also human suffering in the sense that that will help us move more quickly to getting someone stable. Right. So if we can start a medication right now, often we have to wait a week or two or sometimes, you know, in the case of antidepressants, sometimes we're waiting four, five, six weeks to say, how are we doing? Is this working? Ad this one's not working. It's been six weeks. Let's get another round. Let's try a different medication. And if you're waiting another four, five, six weeks to kind of decide what to do, if that one's not working, you're now going on two, three months worth of failed treatments. If there was a way for us to use this data to say, hey, it looks like your sleep's lining up, you're more active and to know that within two or three days and say we're on the right track, let's just kind of do what we're doing or hey, we're not seeing much change that we would expect. If this medication is gonna work, it's been two or three and we're able to in two or three days make an adjustment and be able to do that in lieu of that four to six week wait. Now we're moving that, that treatment trajectory, we're tightening it and we're really allowing that thing, you know, and the ideal to say, hey, let's what would have taken two or three months to kind of find a good balanced place? We're finding that in two or three weeks, right? Or at least knowing in two or three weeks, weeks we're on the right track. We're pretty confident that we've Got the right medications, the right treatment, and may take longer than two or three weeks to really get better. But if at two or three weeks we know we're on the right track, we can be pretty confident. I think that saves both time and money and useless switching of meds and that sort of thing. So I think from that perspective, it makes the initiation, an early phase of treatment more efficient.

Yeah. Lastly, I'm curious about the system clustering and the digital phenotypes that you're using. Can you talk a little bit more about that?

Yeah, yeah. That's the more basic science side of things. And so I think, you know, I come from the world of psychiatric genetics, and we, the geneticists have known for a long time what everyone now pretty much knows. But when we put someone into a diagnostic bucket, schizophrenia, bipolar, depression, and I think there's value to it. But in general, a diagnosis by itself isn't, especially in mental health, it's not the same thing as saying you have type 1 diabetes, and that gives us a clear path to treatment. Right. Depression is probably one of the most heterogeneous disorders there is. And by that I mean there's many, many, many pathways to depression. Similarly, there's many, many pathways to schizophrenia. Right. Schizophrenia or disorder and psychosis in general. Thought disorder, that's a very complicated thing. And our brains, our minds have tremendously complicated interacting circuits. And there's many different ways to kind of perturb that, such that you end up with a thought disorder. And yet we put someone into the same bucket. Oh, you have a thought disorder. You're schizophrenic. Okay. You meet these criteria from the dsm, you fall into the depression bucket. And if you fall into that bucket, here's what we do. We start with Lexapro for you, or, you know, the individual with a thought disorder. We're going to start with an antipsychotic, Right. So we drop them into this bucket that we know is highly heterogeneous. It has a very complex mix of different pathways to that disorder. And yet we give that person the one size fits all treatment. We wait until it doesn't work and then we try something else. Right. And that sort of thing. And so what we would like to do is to be able to use all this data to more clearly define groups that people fall into that do predict different interventions or do point us in the direction of more specific treatments. Right. So sure, you have a diagnosis of schizophrenia and you have a thought disorder, but using all of this data, we can cluster you in A way to say, hey, you seem to fall over here within this group. And we know from past that this group tends to respond better to such and such a medication or such and such an intervention. So, again, similar to what I was talking about in terms of using this technology to tell whether we're on the right track or the wrong track, we can use this technology to really define the starting point in a way that we have better ideas is where to start for your particular group. And then again, use the technology to say, hey, right track, wrong track. Is this working out or not? But that ability to more finely divide people into groups that we think have meaningful clinical relevance will accelerate the treatment process even more.

Yeah. How do you find clinicians? Think about this. Like, as you're working with clinicians to integrate this technology and use these tools, what are you seeing there? Are there difficulties to bringing this on? Yeah, okay.

It's a great question. It's a great question. And I would say it's a complex reaction. I think everybody really recognize the value. Again, Fitbit, reporting your sleep. It may not be perfect, but clinicians know that's a pretty good measure. Right. And I'm not gonna bank everything on that one measure, but it's helpful. Right. Especially over time to track it similarly activity, that sort of thing. So all clinicians know how important that is. And so I think a first conversation with clinician, they're going to say, this is great. This is exciting. We need to do more of this. The problem is, you know, we're trained, you know, in certain, you know, techniques and methods and approaches, and, you know, especially. Let me speak to physicians directly. Physicians are trained never to order a test unless they have a strong clinical reason to order that test. For example, you would never just order a chest X ray on someone just to get a chest X ray, because if you do that across thousands of folks, you're gonna identify lots of false positives and you don't know what to do about it. You know, someone has a lump on their chest, it could be completely benign. But at that point, do you have to go do a biopsy to tell us it's benign? And if you do a bunch of biopsies on benign masses, you know there's going to be risks associated with that. Similarly, physicians are trained, do not order lab tests unless you have a reason and you're looking for something. So I say all that to say, as we get all of this data, we're giving it to clinicians, and they're looking at this going, I didn't order this. I'm not trained to read this. Some of the data is quite complicated. Right. Heart rate variability. How do you interpret a heart rate variability? And there's, you know, some straightforward ways to interpret it, but if you start interpreting it over dozens or hundreds of folks, you really get into a gray zone. So I think clinicians rightly are conservative about introducing all this data that they haven't been trained to interpret. And frankly, it's not totally clear what you do about a lot of these things. Again, if you see an anomalous heart rate variability measure, what do you do about that? Right. And there's certainly plenty to think about, but it's not a cookie cutter question. So I would say clinicians are intrigued, fascinated, enthusiastic about the availability of this data, but integrating it into their actual practice is challenging. And I would say they are appropriately kind of conservative about that part.

Yeah, yeah, that makes sense. We are about out of time. But I'd like to ask you one last question. And it's a little more bigger picture about what's happening. There are all these advances and changes in our industry, and if you could sort of say what you see on the horizon or give a piece of advice or thought that you have overarching all of this change, what would that be?

I don't think this is quite the answer you're looking for. But we need, we need payer reform. Right. We need to pay for the right things. We're stuck in a fee for service. Right? Code your cpt. We only code. There's only code for a, a psychiatric visit, you could for a therapy visit. But the reality is we need to pay for peers, we need to pay for ancillary services, we need to pay for physical health. We need to do a better job reimbursing a broader range of services. Again, our big focus is on physical health, which is absolutely critical to mental health. And yet mental health providers, it's a struggle to get those things paid for. So again, we've had a great conversation about kind of these really cool, innovative technologies that I think are gonna add a lot to mental health. And I am excited about their potential for making treatment more efficient. And, you know, these, in the future, these clinical decision support systems that I think will allow us to scale mental health treatment out to populations that otherwise don't have much access to it. So I'm very excited about that. That's going to happen, right? That's happening. We and others are really pushing that forward. So that's gonna happen. But underneath, we need payment reform. We need out how to recognize and properly pay for the services that really get people better and then keep them better over time.