[00:00:00] Speaker A: Welcome to the heart rate variability podcast. Each week we talk about heart rate variability and how it can be used to improve your overall health and wellness. Please consider the information in this podcast. For your informational use and not medical advice, please see your medical provider to apply any of the strategies outlined in this episode. Heart rate variability. Podcast is a production of optimal LLC and optimal HRV. Check us
[email protected]. Please enjoy the show.
Welcome friends to the heart rate variability. Podcast. I am Matt Bennett. I'm back with a very special guest today who I'm really excited to talk about even our pre podcast. Before I hit record, we're having just some great conversations. Dr. David eddie. Dr. Eddie. Welcome to the to. I was in a rabbit hole with your BIOS online with all the great work you've done, looking at your articles, came across one of your publications in the AAPB Journal. But I'd love just to introduce yourself briefly to our audience, and I always love to start out with the question of how did you come to Heart Rate variability?
[00:01:18] Speaker B: Yeah, thanks for having me. It's a pleasure to be here. So I'm a clinical psychologist at Massachusetts General Hospital.
I'm also a researcher there at the Recovery Research Institute and center for Addiction Medicine. I'm also a researcher in the center for Digital Mental Health within Mass General Hospital and an assistant professor in the Department of Psychiatry at Harvard Medical School. So I wear a lot of hats. I'm a little clinical, a lot of research.
I've always been interested in heart rate variability as a biomarker of risk of pathology, and that really comes out of my training as a graduate student, which I'll talk to briefly as well. But as you know, I'm also really interested in heart rate variability by feedback, like, how can we leverage what we know about heart rate variability to affect real world changes in people's lives and to improve treatment outcomes?
I'm a little interested in antecedents, but I'm also interested in clinical interventions and outcomes as well. And I'm primarily focused on substance use disorder. I've done quite a bit of work on borderline personality disorder as a condition as well. Most of my work now is really focused on substance use disorder, on addiction, and really my training as a graduate student, as a PhD student, I was also quite focused on substance use disorder. I studied under Dr. Marcia Bates at the Ruckus Center for Alcohol Studies. I think it's now the center for Alcohol and Drug Studies.
And that was really a psychophysiologically oriented lab. We were doing HIV research.
She had some collaborators in that lab who were really notable figures in the field, including Evgeny Vashillo, who really well known Russian physiologist, really did some of the important foundational work in this area. So I was really privileged to study under Dr. Bates and Dr. Vasillo, as well as Dr. Jennifer Buckman there and other researchers, including Bronnie Vasillo.
So I had a really rich training experience, and it was really working at the center of Alcohol Studies that I got interested in this work. I had some training as an undergraduate student in neuroscience, and I had some lab experiences, but it was really when I got to grad school, I got indoctrinated into this important work. And there I also ran a few clinical trials as a graduate student on heart rate variability by feedback, which, of course, was developed by Fgeni Vascilo and Paulira At Of. Makes sense that this is where my career has gone, right?
[00:04:17] Speaker A: So just a question about that graduate. I mean, I'm assuming not because and I just keep my hair short so it doesn't turn gray as well. You're distinguished gray hair, but more your accomplishments is that you've been in the profession for a while. So when did you start to do this in grad school? What year are we talking about?
[00:04:39] Speaker B: Oh, yeah, sorry.
I started grad school in 2010.
[00:04:44] Speaker A: Okay.
[00:04:45] Speaker B: I finished grad school around 2015, 2016. I went and did my clinical residency at Mass General Hospital, stayed on to do a postdoctoral fellowship, and I've been on faculty since then.
[00:04:56] Speaker A: Awesome.
[00:04:57] Speaker B: Yeah, absolutely. I've been in the field for a while now, and that's really where I came from. That's how I got interested in this work. One of the things I've really tried to do as an independent investigator is take the work we were doing I was doing as a graduate student at Rutgers and really extend that. And one of the things we were doing a lot of in grad school was curious paradigms, and we were observing and studying how people would respond to stresses in the lab and how that might affect craving for alcohol and other drugs, for instance, and in terms of her risk for relapse. And we learn a lot in those curio activity paradigms. But something I've really tried to do as an independent investigator is test some of the assumptions that came out of that work in real world settings and under real world conditions. So I've done a lot more Ambulatory research in the last five or six years using ecological momentary assessment, which is where we prompt people with brief surveys in real time through their smartphones so we can assess their mood or their emotions in real time, craving, so on and so forth. And with that as well, I couple Ambulatory psychophysiological monitoring using Wearable biosensors. So I have folks, while they're doing, say, a week of EMA surveys, they're also wearing an ECG monitor and perhaps also a Smartwatch or a device like the Empatica E Floor, which is monitoring heart rate variability and skin conductance and body temperature through a watch.
[00:06:38] Speaker A: Awesome. So what are some of the insights that you have gained along the way? I know that's such a huge question for all the experiences, the research, the thinking you've done on this topic. But as you have worked with heart rate variability and got to work with some of the pioneers in the field, just sort of working with, we know, folks, substance use, personality disorders, highly dysregulated nervous systems in many ways.
What are some of the insights, lessons that you've learned being really deep into this research?
[00:07:23] Speaker B: Right, so one of the things we were talking about before we hit the record button here was how people with substance use disorders tend to respond differently to our curioactivity paradigms and healthy controls. And one of the inherent limitations I think we're seeing in the field is we were talking about stress detection algorithms. How can we leverage heart rate variability to identify stress that's perhaps occurring outside of conscious awareness using wearable biosensors like Smartwatches fitness trackers, for instance. That's an important area of my research right now. I'm trying to develop the algorithms that we could embed in Smartwatches to detect stress in real time. As a substance use disorder researcher, I'm particularly interested in this question because a lot of the risk that's conferred for relapse comes from factors that are outside of conscious awareness. So even though folks might identify in early substance use disorder recovery that they're experiencing aversive effective states, that they feel uncomfortable, that they're experiencing significant craving, they're not always very good at identifying the antecedents of that and acting effectively around that. So the idea is if we could detect stress in real time, we could prompt individuals using a just in time intervention through their smartphone, for instance, to act differently, to become more aware of their stresses and in turn, hopefully prevent relapse or at least prevent lapses substance use. And so development of stress detection algorithms is an important clinical goal. Now there's researchers in industry who have been working on this problem for years and there are stress detection algorithms embedded in some commercially available devices. The problem is we suspect that most of those algorithms were trained, their machine learning models were trained using folks without pathology. Yeah, the problem is it's an empirical question.
We need to test these algorithms to determine how well they're going to potentially support individuals who are experiencing some form of psychopathology. And so we have potentially some good stress detection models that are out there. Unfortunately, most of them have been developed by industry and are behind a firewall and we're not going to be able to access them anytime soon. That's one of the reasons the NIAA has funded my research so that we can develop publicly facing algorithms that perhaps work a little better with people who are experiencing psychopathology.
[00:10:21] Speaker A: Yeah, out of curiosity, because working and a lot of my lessons were with severely traumatized teenagers working in child welfare, substance use could be in there as well with the individuals I worked with. But almost that the folks I worked with, you could almost flip everything you know about the nervous system and stress and just turn it upside down. It's like the higher their heart rate, it seemed like they at least self reported more regulation and we would see it in their behavior and in their moods as well. And we're like, this makes no sense really. And cortisol used to be kind of more cortisol. You were always stressed out. But we were looking at research where actually there was a higher and lower cortisol could have lead to dysregulation and it's just like nothing got simpler, let's put it that way.
[00:11:17] Speaker B: No, there's tremendous heterogeneity. And in the study I just mentioned, we've got our phase one data and we're looking at it and we're starting to train an algorithm or to develop a model using our machine learning approaches. I'm collaborating with Dr. Rich Fletcher at MIT on this. And one thing we're finding is just tremendous heterogeneity in terms of how folks are responding. In this study, we did both a Q reactivity paradigm and a week of Ambulatory monitoring with the same participants. And we actually wanted to see whether the model for stress detection trained on the lab data performed better or worse potentially than the model trained on Ambulatory data. But we're seeing tremendous heterogeneity and that's sort of within our sample of individuals with substance use disorder. But we also know that folks with substance use disorder often respond quite differently to healthy controls in our curioactivity paradigms. So that in part could be a reflection of the pathological state they're experiencing as people experiencing substance use disorder.
It could also be a factor, an individual factor that's different about people with substance use disorder that's contributing to the problem, right? And so one of our hypotheses is that sure, substance use can lead to problems in self regulation. It can affect US neurobiologically psychophysiologically, it affects alcohol and other drugs, can affect the cardiovascular system in adverse ways. And so we see on average people with addiction have lower on average heart rate variability than healthy people in the population. That's also true for major depressive disorder, anxiety disorders, post traumatic stress disorder. Generally, as you know, and probably most of your listeners will know, low HIV tends to be associated with mental disorders and physically pathologically states as well, medical conditions.
So low HIV is probably a symptom of substance use disorder. But we also think it's a contributing factor to the maintenance of substance use disorder because once you have lower biobehavioral flexibility represented by lower heart rate variability, you probably have less capacity to dynamically respond to stresses or challenges in your environment in more effective ways. And of course, we know that people with substance use disorder become more and more reliant on substances to cope emotionally.
As you rely more and more on alcohol and other drugs to cope you become less able to harness other kinds of coping strategies. And of course the problem just compounds. It's a vicious cycle folks find.
[00:14:17] Speaker A: Absolutely. So how in the world with what you just said and the vast deviation, the population that you're working with doesn't really match up with other norms, but there is a lot of variation within the population as well. How the heck does one go trying to create an algorithm with all this variation?
I'd love to just get a little peek behind the curtain and when you're seeing this and you're looking for an algorithm that I would assume would go on this very diverse population's wrist measuring it, how do you go about even starting that process?
[00:15:07] Speaker B: Well, the first question is, can we do it?
Because stress detection has been achieved empirically by academic researchers and these results have been published. But again, typically it's in a population of folks who aren't experiencing mental disorders and are physically healthy. Right? So the first question is can we do this? And I think the answer to that will be yes, but it won't be easy. And so what we're really doing now, we've been able to identify, create a crude algorithm that does the text stress, but it's not at the level of specificity or accuracy that we want it to be at. So then the next thing we need to ask ourselves is are there individual factors essentially covariance that we could be entering into our machine learning model that could potentially dial up the accuracy of the model so that there are individual characteristics like age, sex, body mass index, race, substance use, history, primary substances used that might actually help increase the accuracy of our algorithm. And that's really what we're testing now is we're trying to understand if there's these individual factors that can increase specificity or alternatively, do we need to be stratifying individuals as high or low on certain measures. And of course, long term, if we ultimately have a device or an algorithm that we cook into an app that people can download from the App Store onto their smartwatches, perhaps that app, when you're setting it up as an individual end user, you might have a series of questions that that app asks you to collect the information that it will then use to sort of dial in the algorithm. So it might say how old are you? What's your sex, what's your height and weight?
Et cetera, et cetera. Obviously there's limits to that. We don't want to have to ask people too many things and we don't want to make it an onerous task. We really want something that just kind of works when they install it, but we might have to ask for some individual information to get it to really work.
[00:17:34] Speaker A: Well, as we yeah, I would assume just what I know and my frustrations with population norms is standard deviation is so big in the studies and the studies are honestly fairly limited in what you can find. But the standard deviation, all these are so huge and that's within gender and age demographics, even within those are so big. So kate, do you feel like because one of the things that I guess maybe a conclusion I've come through with, with my deep dive into the research is you're almost an in of one.
I hope we get better population norms and that kind of research gets funded. I don't know if standard deviations will get smaller. I assume maybe they could as our technology gets better. But I'm kind of wondering is to get an individual's to say, hey, you're in the red with your stress response, practice some resonance frequency breathing or mindfulness or whatever intervention that might be set up for them.
Do you need some previous heart rate variability biofeedback on them as an individual? Because it seems like with the populations that we both seem to have a passion for working with, there's probably a lot of dysregulation, I would imagine. And we've seen this is people have a lower overall than the population norms would be. So are you thinking about getting some baseline data to get those sort of alerts for them? Or how are you thinking about sort of the end of one in your thinking?
[00:19:23] Speaker B: Yeah, it's a perennial problem, isn't it? And there are other ways we can go about this, right? I mean, I've been talking about developing an algorithm that perhaps dials in using some individual information.
But another way you could approach this is by dialing in the algorithm, by training it in real time using that individual's data. So rather than using a training data set from a population that you presume represents broadly the individuals that are going to use this app. You could actually train the app in real time using that individual's data. So you could use self report. So you could, for instance, ask an individual to rate their stress several times a day over the course of a few days or a week. And the device could monitor their physiological arousal or state at those moments and could actually then be trained to identify that individual's pattern of physiological arousal and predict stress in that way. So there are other ways of doing this and that might end up being where we land.
If we can't do this, if we can't predict stress to the level of accuracy that we would like to because there is just too much heterogeneity in the populations we're interested in, then that's plan B. Yeah, that's a little more complicated and a little more onerous for the end user, isn't it? Because then you're actually asking them to do additional work. It would be much nicer if they could just download the app and it just works.
[00:21:15] Speaker A: Yeah.
Not to totally problem solve this off on the podcast, but I wonder too, because this has just kind of been my thinking about where this could go.
Could you get baseline? Because that to me, is what really starts to make the end of one a powerful tool when you look at just that individual is what is your baseline over a week and using that data, plus probably some survey data and demographic data as well, to really say, okay, we know what your baseline is over the week. Now, is that a good week or a bad week? There's still limitations there, but then when you see sort of outlying spikes in the data against someone's own baseline absolutely. Are we getting I think with what we do at optimal is like the morning readings, three minute morning readings, which is just a different way to measure we use RMSD. So when you see outliers in that to question, okay, what's going on? How to plan my day based on how that compares to my week or 30 day or all time.
[00:22:32] Speaker B: Absolutely.
[00:22:33] Speaker A: And you're collecting all that data 24/7, which is really exciting of what we could learn in the process.
[00:22:41] Speaker B: Absolutely. The most crude stress detection algorithm could simply identify an individual's heart rate baseline and look for deviations from that, of course, controlling for movement because that's going to impact heart rate. But yeah, you could potentially have a very crude stress detection algorithm based on heart rate alone. And that would certainly have some advantages because we know that risk worn devices are much better at identifying heart rate accurately than heart rate variability. Because HRV assessment is really obviously affected, as you know, by noise and artifact in a way that heart rate monitoring is not. You can have quite a lot of missing data and still assess heart rate pretty accurately. But heart rate variability indices are really affected by noise.
That's another thing we're working through as well.
The risk is possibly the worst place to be collecting HIV data and PPG sensors, which are the sensors that smartwatches contain that it allows them to monitor heart rate, are not as good as or as accurate as, say, gold standard ECG.
But again, we want a practical intervention here.
We don't expect that people are going to go out and buy ECG devices, much less wear them. But we do know people wear smartwatches and they use smartphones. And so we really have to work with the technology that already exists in people's lives as best we can, knowing that there are some limitations.
[00:24:23] Speaker A: Yeah. So I'm struck by a fascination with creating an algorithm. One of the things. And this will be a podcast I've recorded but won't be played till a week or two after we have this one live is sort of my surprise when I first got really interested in HRV. Is, I was stuck in this I think it was a 1997 Journal article that I'm sure you're familiar with over the European society of something or the other that continues to be sort of the Bible of HRV algorithms. Not to say that there's not been work done, but I couldn't name an algorithm that's been added since then. Is any good grad student. You always want to see something within the last couple of years, but I see that referred to. So as you look at creating a new algorithm, are you approaching that to say, okay, obviously you're building upon past learning, but we're creating it from scratch, or we're going to use low frequency and RMSD and very low frequency. How do you even go about what I think is incredibly exciting? Because there's got to be new algorithms out there, I would think we haven't found everything. So tell me about the thought process and then how to realize and even test a new algorithm.
[00:25:54] Speaker B: Yeah, I guess it's an important distinction to make here, right? Because you're talking about that 96 97 work group HIV workgroup paper that really articulated the main HIV variables that were in use at that time and define them. And that's really the sort of the paper people tend to cite or reviewers one sided if you don't, and that's fine, even though there's some information in that paper that's now outdated. So, for instance, they talked about the low frequency high frequency ratio as being possibly an indicant of sympathetic parasympathetic balance. And since then, sort of the idea that the low frequency is a measure of sympathetic control has been refuted, and that's no longer believed to be the case. That's not to say that there aren't some sympathetic components in the low frequency bandwidth of HIV, but we know that it's actually much more complicated than that. So the low frequency high again, nothing.
[00:27:01] Speaker A: Got easier on us.
[00:27:02] Speaker B: No people still, when I get papers to review, refer to the low frequency high frequency ratio as the balance of sympathetic and vagal or parasympathetic tone or output. And that's because they're reading these older papers which speak to this and highly cited and largely fairly trustworthy even now. And a lot of what they said absolutely stands to this day. But there's been some aspects that have been refuted. And when I'm talking about algorithm development for, say, stress detection, I'm not referring to we're not developing new HIV indices. What we're doing is we're taking existing HIV indices, either high frequency HIV or RMSD, if we're principally interested in vagal or parasympathetic control, which is usually what we're interested in, or heart rate even. And so we're using those sort of well articulated, well known HIV indices in our models. But one of the things we're testing presently is which HIV indices give us the best predictive ability in terms of stress prediction. Like RMSD and high frequency HIV are really highly correlated.
So if they produce similar results, then we're probably going to use RMSD because it's computationally less dense.
It's an easier thing to calculate for a device that uses less energy than high frequency because that requires a fast four year transformation, which is just a few extra steps. So both are valid, but we're going to take the less energy expensive measure if it gives us just as much value in our final stress detection algorithm. An important distinction there. We're not creating any new HIV indices.
We're really just utilizing the preexisting measures that we have in new ways.
[00:29:12] Speaker A: Are you one of the fascinations that I'm questioning?
The validity of my fascination. I think I need to drop it. But I got to ask you as you explore this, because I've become very excited, a little worried, but that's a whole different non HRV arena with artificial intelligence.
How when I feed Chat GBT RR intervals and what it can tell me back spectacular.
And the time frame that it does is amazing. So we're getting to the point and let's just pretend we have the perfect AI, which I know we don't right now, but we got really good stuff when it comes to heart rate variability, it's really good. I'll give it that from my testing. With it, we can ask questions of the algorithms in different ways.
And initially I got excited of, okay, we collect all this data during a 20 minutes biofeedback session.
What's? RMSD and low frequency together telling us or in a three minute morning reading, what does Sdn and Max Min tell us together? Are we measuring slightly different? And the conclusion I'm kind of coming to is it doesn't really matter. RMSD is going to give you 95% of anything you'd want to know, and everything else is just going to confuse 99% of the users out there that want to know if they're doing where their polyvagal system is doing today, kind of how their vagal nerve is operating. So I wonder if maybe you could spark hope in me again.
Do you see any arena and doing 24/7 kind of monitoring obviously is a little bit different of an approach than how I was thinking about it. Do you see any room there where the perfect AI could tell us, look at these algorithms and tell us slightly different things? Or is Low Frequency just going to give us probably the best that's out there, RMSD for other things? I'd love to get any thoughts or epiphanies you've had on this, a few thoughts about that.
[00:31:41] Speaker B: I think perhaps you're speaking to in your question, you're speaking to some HIV measures that are less well studied. And I think there's a lot of potential for interest in measures of complexity. SD one, SD two, which we don't really understand as much.
Chaos theory, right? And that's sort of like the vanguard, I think, of HIV research.
But I think we're a ways off really fully understanding those measures.
The sort of better studied RMSDs and Scnn and maybe it's a little plain vanilla, but we understand those measures a little better.
And can AI be leveraged to help us better understand some of the complex problems we're working through? Of course. But we also have to be really careful.
AI can represent a black box that we're feeding information into. We're not always sure how the information that's being output was achieved, calculated, derived. So we have to be really careful. And this is a problem I see in HIV post processing where researchers maybe don't have a background in HIV analysis and they rely heavily on an algorithm baked into a post processing software like Kubios.
And they don't really take the time to manually inspect the raw signal and kind of check for noise and artifact. And they just sort of trust that the baked in filters are going to do the job and just one needs to be really careful. Similarly, with a lot of output HIV statistics from wearables like Fitbit have their algorithm for calculating HIV statistics. Now they're using the same formula for RMSSD that everybody else is. But what's different is how they're managing noise and artifact in real time on their device. And that's obviously going to influence the output HIV statistics. And again, we've got a black box problem where we've got information coming into a system and we're getting perhaps the HIV statistics output. And sometimes researchers are using those indices in their studies, but they don't really know how those HIV statistics were calculated, how some of these trickier points were managed, like noise detection.
Windsorization so I think we have to be really careful.
[00:34:35] Speaker A: Yeah, I mean, you dive into that and it gets really we were talking a little bit ahead of the show and then asking and trying to find out even how you may have better luck than I do. Like how do you calculate a readiness score? Or one app had a productivity score, which from somebody, I would love to give you a productivity score, but it's proprietary data. So as a clinician, I can't run on hypotheticals, especially when we're talking about relapse suicide, potential suicide, thinking like those sort of things, I can't trust you that this is if you're not going to tell me how you're getting this data, how do I trust that data? A piece of this and this is kind of the commercialization of the field is a little scary to me because you can make these leaps and a lot of folks who haven't done the research that you've done and I've started to do, you might be given really bad data to folks. And it's tough enough when you're just, hey, here's your RMSSD.
How do you handle artifacts, how do you handle movements? Don't pay attention to your breathing, just breathe normal. But what is normal breathing? And if you want to change somebody's breathing, tell them to breathe normal. So all that stuff in and so somebody who's attempting, if I'm understanding correct, to really kind of get a continuous 24/7 monitoring. There seems to be a lot of challenges in that of if I stand up, my heart rate variability is going to change. If I sit down, if I take a drink of water, if I cough. So it seems like you're going to be working to handle and account for. A lot of artifact that hopefully if you're sitting down and trying to be halfway decent and quiet, taking a reading, you can account for some of that.
How do you approach that in getting accurate data when the world is happening around the individual?
[00:37:01] Speaker B: Yeah, it's a great question.
Humans are complex heterogeneous in their presentation, their physiology, their psychology.
It's hard, right? It's hard to develop kind of predictive models that can have real world clinical implications. But in a recent study, based on data from previous Ni study that we completed a few years ago, we recently published, and I think actually you might have referred to this at the beginning of the podcast, this particular paper, we're actually able to predict alcohol use in people in early recovery from alcohol use disorder based on their average Ambulatory HIV. And that was collected under real world conditions.
That data was collected in real time.
We weren't calculating their HIV based on a 24 hours recording. It was epochs of five minute epochs of HIV that were calculated related to the EMA surveys participants were completing in that study. We could have looked at 24 hours recordings, but that's another going to be a different paper.
[00:38:26] Speaker A: Right.
[00:38:27] Speaker B: But nevertheless, we were able to calculate average HIV for these individuals as they were going about their day, perhaps certainly under variable conditions, they might have been sitting, they might have been standing, they might have been walking during these epochs of recording that we pulled out of their longer 24 hours multi day recordings. And we were able to use those average HIV scores to predict how much somebody would drink. Subsequently, low HIV was associated with greater alcohol use. That's interesting. Now, what are the clinical implications for that?
Potentially, you could argue, you could say that individuals HIV statistics derived from Ambulatory monitoring could have some kind of predictive utility in this population of people. In this case, it was folks with alcohol use disorder in terms of predicting risk for lapse or subsequent alcohol use. Now, would you use that solely in isolation to drive all your clinical decision making? No, of course not. But it's an objective measure you could potentially leverage or utilize in your clinical decision making that could complement your self report measures. Remember, in psychiatry, clinical psychology, I mean, like 99% of what we do is self report. We ask people, how are you feeling? What do you think you're going to do? And that's all really valuable. We don't have a lot of bioassays, right? That's a real limitation of our field.
And so anytime we can have an objective measure of risk, I think there's potential value there. But of course that needs to be couched in the broader context of treatment planning and clinical considerations and other kinds of assessments. But HIV, could that complement clinical decision making?
[00:40:37] Speaker A: Absolutely.
[00:40:37] Speaker B: Knowing that of course there's going to be noise in that recording. There might be some accuracy issues. But if an individual is presenting with low heart rate variability, I want to know that as a clinician, because that suggested me that there's probably a logical vulnerability in addition to potential psychological vulnerabilities or social vulnerabilities that that individual is experiencing that could pretend risk for relapse. And that's just another piece of the diagnostic or clinical decision making pie.
[00:41:14] Speaker A: It's huge. And that was one of my real trying to fit the pieces together and I just couldn't with the existing how do we then get that data to the clinician in real time? So there is that way to reach out with a phone call. It say, hey, I've seen your scores have been trending, what's going on? And it may be, hey, they decided to train for a marathon. It could be something very good. But that one phone call. And I think both of our experiences, especially in substance use, that social support. And if you can break a pattern of behavior that might lead to a relapse, a phone call, a text message might I mean, not to be overdramatic save someone's life because relapses are such a dangerous thing because it's usually not at least my folks, it's not a mild use.
A relapse is a big use. And with Opiates or other potentially could be drugs.
[00:42:28] Speaker B: Absolutely.
[00:42:30] Speaker A: It's amazing what this could do.
[00:42:33] Speaker B: You're talking about real time interventions and I think there's potential utility, but I think there's also utility for introducing HIV measures into clinical practice. If we have a kind of a measure of folks HIV across days or maybe we even just do a quick five minute HIV assessment while people are sitting across from us in a therapy session, if we can use that information to complement the other sources of information we could have, that could be really valuable. And you asked how would we get this information? Well, it's pretty easy to measure HIV using little clip on wearable devices now that we know have in the clinic. In our also, you know, there are devices like the Vivi Health System, which I'm scientific advisor on the Vivi Health Advisory Board, and that's a really nice device that utilizes a relapse prevention smartphone app. But it also includes a wearable like a Fitbit that's connected to the app so that it is actually collecting heart rate and heart rate variability information in real time as people go about their days, their weeks. And that information is available to clinicians. And I think HS as well is another relapse prevention app out of the University of Wisconsin that I'm not sure if that integrates a wearable, but potentially it might.
But these apps are feeding information to the clinician in real time, but also so that it's available in session so that one can review the information with the patient and look at the week that was and what were the highs, what were lows, how was your sleep? What was your physiological arousal looking like?
So there are ways we can sort of begin to pull this information into our clinical practices. And I think we're going to see more and more of that as these devices and apps become more and more ubiquitous.
[00:44:36] Speaker A: Absolutely.
I want to ask you about HRV biofeedback, so let me throw that out there because I know I'm throwing out another big topic when we only got a few minutes left. But I got to ask you just sort of how as you think about the really exciting and I love that you're tackling these problems because my questions are not to challenge, but they're out of excitement is I haven't seen a way to get over these barriers, whether they're technical, whether they're just our biology, whether they're algorithms. It seems like we kind of could get stuck in these boxes. And what I love about your work is you're getting out of those boxes. So I'm really excited. So I'll just ask you a general question with hopes that you'll come back and record another episode with me down the road, just on this. But talk to me about how using HRV biofeedback, same science just as helping to heal, helping to regulate a supplement to treatment, how's that informed your work as well.
[00:45:52] Speaker B: That's a separate track of my research. Then I talked at the beginning about the clinical interventions. I'm interested in how we can leverage what we know about psychophysiology and how we can leverage that knowledge to better clinical interventions that are going to have real impact in people's lives and improve treatment outcomes.
One of the things I always felt was a limiting factor of heart rate variability biofeedback was that the technology we were utilizing in the early days was just it wasn't really accessible. It was never going to scale up. And for those who don't, in the early days, we would bring people into the lab and we would have these big pretty expensive setups and we would hook people up to an ECG monitor and a respiration sensor.
Some other devices, it was kind of complicated. There were some early devices, ambulatory devices for HIV biofeedback practice were okay, but they were kind of clunky. And frankly, our study participants never really like using them because it was like an additional device they had to carry.
So there were real limitations there. And I always worried that even though HIV biofeedback really seemed to help people and the data really showed that it improved, reduced negative affect and we were showing in folks with substance use disorder, it was significantly reducing craving. And it was like, this is great, has real potential to benefit people. But who's going to use it? Who can access it first? Like try finding a provider who is trained in HIV biofeedback and then has the equipment.
There's a tiny portion of the population that are going to be able to access this intervention. So I was really excited to see companies like optimal HIV leaf therapeutics as well, who have really sought to develop HIV biofeedback technology to make it more accessible, to make it scalable.
Like the leaf device, for instance, is a product people can order a subscription to online. Anybody can access it. It's a wearable biosensor. You don't need to find a clinician who's trained in HIV biofeedback. They have training apps and videos sorry, baked into their app and they have online coaches and it's accessible, it's scalable. And I got really excited when I saw that product launch because I really felt for the first time, it's like, perhaps HIV biofeedback can be an intervention that people can actually access and use in the real world, right?
[00:48:50] Speaker A: Yeah.
With our shared passion for the populations we work with was just like especially worth working with dr. Hassan. It's know the research around residents frequency, breathing and bringing know even if you're just using the pacer and you're not tracking, you have that additional coping skill and trying to do that in a trauma informed way. Knowing not. Everybody is going to respond well to the traditional mindfulness. Has been really a great thing to be able to bring that science into the affordable category with it because I found it so powerful in my own health and wellness.
Like I said, it's like mindfulness but on steroids. It's like, oh, this works.
[00:49:41] Speaker B: It's a powerful tool. Right. I mean, we can affect big changes in our physiological state, physiological arousal, by modifying our breathing. And this is widely recognized now, even in dialectical behavior therapy, which is a really well established treatment. Now, it was originally developed to treat borderline personality disorder, but it's now widely used for other problems. And it includes a module on pace breathing or a pace breathing skill and it's essentially six breaths per minute breathing.
It doesn't include the biofeedback and a device, but we're talking about pace breathing in other areas of clinical psychology as well. So I think that's exciting too, that this knowledge that's really come out of the study of HIV biofeedback really has trickled down into other know. It's not to know interventions like CBT weren't talking about know, they've been talking about it for a long time, but I think we're talking about it a little bit differently as a function of what we now know about HIV biofeedback and its confirmed benefits.
[00:50:59] Speaker A: Yeah. And I think we're evolving in and take a deep breath to a more specific protocol that helps that regulation. And that's the exciting thing, especially working with folks with trauma where you don't know what can be triggering like the interventions like breathing. I love the tapping stuff as well, is we're getting to these supplemental pieces to help folks integrate, to heal, to regulate.
I love being in psychology right now because my joke is when I was trained in the late 90s in grad school, it was kind of boring. You sat in a chair, I sat in a chair. We talked and it wasn't really boring. But now we got EMDR, we've got more and more integration of biofeedback, different breathing strategies, trauma informed mindfulness has really taken off and it's really exciting the tools that we can give individuals moving forward.
Which leads me back to my final question again. I hope to have you back because I'm going to keep definite eye on your research that we've talked about. But what do you see when you look 510 years into the future? And you may be writing a lot of this future with the work that we've been discussing in detail.
Where do you see with AI, with algorithms, with technology all advancing at pretty exponential rates?
Where do you see us 510 years from now from your experience and what you're working on currently?
[00:52:52] Speaker B: Yeah, I think we're going to see greater uptake of heart rate variability, biofeedback, because it has become more scalable. But I think also we are going to see greater integration of AI, which is really a very sophisticated form of machine learning. I mean, ultimately AI technology is based on algorithms that are created by people so they're not complete black boxes in that way.
But what I'm really excited about is how we're going to be able to use passive monitoring to improve or develop new or better just in time interventions. So I'm thinking about things like digital phenotyping and using pulling information off people's smartphones, utilizing that information as indicants of emotional state, of affect, of mood of risk for substance use relapse, but also perhaps suicide. So that we can be using that information in a complementary fashion to affect better just in time interventions where we're getting out in front of people's risk before they're even aware that risk. And I think in terms of the way we monitor HRV, I think we're going to see greater and greater passive HIV monitoring. Right now we're at a point in the development of technology where we've gone from these giant holder devices, right? It used to be like having a phone book around your neck and the holder devices got smaller and smaller and smaller. And now we've got these tiny ECG devices and we've got these smartwatches that people can wear. But I wonder in the future whether we'll even be monitoring HIV using wearable devices. For instance, at MIT some time ago, there's an app that I think is still available for download and the name of it's Escaping Me. But essentially it converts your smartphone camera into an HIV monitoring device simply by monitoring the capillary blood flow in your face. So you just hold your camera up and it can see the flushing, endometrial flushing with each heartbeat that occurs in our faces. It's not visible to the human eye and it can calculate heart rate and heart rate variability from that. So could we be wearing like meta are coming out with sunglasses and glasses that include cameras that track our eye movement?
Are they also going to monitor our skin and potentially be calculating HIV and HR statistics for us in real time that could then be fed into inform our just in time interventions? It sounds like maybe like that. It's a bit far fetched and a bit ways off, but not really. I mean, the technology is around, it just needs to be sort of managed in a way that makes it sort of scalable financially, sort of cost effective. But it seems like it's something that could be doable think about where we were ten years ago with Wearables and how far we've come in just ten years.
We didn't have smartphones in 2005 exactly that long ago.
[00:56:26] Speaker A: Well, the exciting thing is I'd sort of love to talk to these departments sometimes because of how they treat heart rate variability. But when you've got Apple Fitbit's owned by Google, now you got Mean, you've got the biggest, wealthiest companies in the history of the world coming to this science sometimes. Well, sometimes, like I said, I just love to sit in on team meeting and see how they talk about it, but only get better. And research like yours, I'm just so excited for what you're doing because we need to really get scientific about this and have it open to how are you figuring these stress scores and having that box not be shut behind AI and kicking out things that may be valid. May not be valid. So I just want to thank you Dr. Eddie, for your work.
We will put some information in the show notes at optimal Hrv.com you can find those. I really hope that this is a first of many conversations that we have because I'm definitely going to follow your research. Where you're going with this because like I said, you are addressing a lot of just walls that I've hit along the way and I didn't think we were going to probably get to the other side of them. So I'm just so excited about that you're working on that and to be a fan rooting you on in that journey, I'm excited to see where you go.
[00:58:09] Speaker B: Thanks so much man. It's been a pleasure being on the show. Thanks for having me.
[00:58:13] Speaker A: Thank you so much.
