Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Autism with guest Matthew Goodwin
418 Plays11 months ago
This episode is dedicated to the memory of colleague and friend Grace Baron.
Joe and Rolf talk to Dr. Matthew Goodwin, a professor at Northeastern University whose research focus is on autism. His most recent paper used biosensing (heart rate, skin conductance, etc.) to predict aggressive behavior in profound autistic individuals up to three minutes before the event.
Recommended
Transcript
Introduction to Matt Goodwin
00:00:09
Speaker
Welcome to Cog Nation. I'm Joe Hardy. And I'm Rolf Nelson. Today we have a guest that we're really excited about, Matt Goodwin. Matt Goodwin is a founding and key faculty member of a new doctoral program in personal health informatics and the director of a computational behavioral science laboratory at Northeastern University. He was also a visiting associate professor in the Department of Biomedical Informatics at Harvard Medical School.
00:00:35
Speaker
the former director of clinical research at the MIT Media Lab, and an adjunct associate of research scientist at the Department of Psychiatry and Human Behavior at Brown University. So Matt has worked for a number of years on assistive technologies and biosensing in autistic populations.
00:00:53
Speaker
And that's what we're going to be talking about today.
Overview of New Research
00:00:56
Speaker
He has a new paper out called wearable biosensing to predict imminent aggressive behavior in psychiatric inpatient use with autism. So we're going to talk about that paper and also some other things that Matt's been up to. So Matt, welcome to the show. Thanks for being with us. Thank you so much for having me.
00:01:15
Speaker
So tell us a little bit about your interest in research in autism. What got you interested in this?
Journey into Autism Research
00:01:24
Speaker
Yeah, we're going to go back to 1994. So after high school, I was a psychology major. I enjoyed child development. I was also a baseball player and a ski racer. I thought I was going to go to school and play baseball. I crashed my junior year ski racing.
00:01:47
Speaker
fractured a kidney had to stop thinking about baseball scholarships and go to whatever school would take me at the end of my senior year.
00:01:58
Speaker
My college or high school guidance counselor and parents got a little bit more creative and said, well, let's look around at schools that might be less focused on GPAs and standardized tests and more interested in potential and experience. And one of these programs was St. Claire's.
00:02:20
Speaker
at Oxford University in the UK. And I was going to do a psychology and philosophy certificate for that year in 94.
00:02:32
Speaker
It so happened that the director of my program's wife ran a unit for children with autism in Oxfordshire. And I got talking to her one day and she was describing to me some of the characteristics of her students. And you have to remember, this is 94. So this is pre, like autism speaks, curious incident of the dog and the night. I had not had public exposure yet to autism.
00:02:57
Speaker
And I'll come to learn that she had a high percentage of children with islets of ability or savant-like abilities or prodigious abilities. There were children in there who were nonverbal but could sing in perfect pitch. Kids who could draw these extraordinary drawings well beyond their developmental age.
00:03:19
Speaker
kids who did have some language, they would ask a new person, what's your birthday? And then they tell you, you were born on a Wednesday. And, you know, my mind is just like expanded and confused at how do these kids do these things?
00:03:36
Speaker
while at the same time seeing that the kids, the things that typical kids do without learning effortfully, like orient to faces, look at people in their eye, use gestures and join attention to recruit people's interest. They didn't
Balancing Scientific and Personal Views
00:03:56
Speaker
do any of those things. I really got a feeling like there was a social
00:04:03
Speaker
awareness or willingness to engage that was quite different than typical children. So I started to read more about it and I started to spend more time kind of clinically observing and helping out with the teachers. And at the time it was, you know, it's in the UK, it was very much phenomenological sort of describing what the experience seems to be like. And then very much kind of coming out of the cognitive revolution
00:04:31
Speaker
Pre-Mac and Woodruff and thinking about theory of mind, this idea that people have thoughts, beliefs and desires that are different from each other and we can kind of simulate our perspective and theirs and meet in the middle that this might be a core deficit in autism. And then some key experiments that were done at the time were suggesting that theory of mind might be a global impairment in autism. And then you kind of calculate the consequences of that
00:04:59
Speaker
You wouldn't know that other people are social creatures. You may not have a rich philosophy of mind. You may not be attending to stimuli in the environment that helps you negotiate with other people. You'd be more drawn to objects and special interests and some of the characteristics of autism. But here's where things kind of started to get interesting. I'm reading in the literature what they're suggesting individuals with autism can't do.
00:05:26
Speaker
And I'm spending day to day with these kids and they're doing those things. After they familiarize themselves to me, after I let them approach instead of me going towards them, as I show up more consistently, I become a part of their routine.
00:05:47
Speaker
then they're looking at me and they're looking me in the eye and they're using gestures and they're referencing other things. And so I got really kind of second level interested in how does science see this and define this and how does experience see and define this when you spend more time in the natural environment.
00:06:10
Speaker
So there's like the how can you have ability and inability in the same brain? How can you come to understand something when you're with it? And then when you were supposed to take doing air quotes, an objective look at quantifying ability and inability through science, you get a different conclusion. And then over the course, I developed relationships with these kids. And there's one I want to tell you about specifically, but
00:06:36
Speaker
It was these relationships that I started to see development and progress. And then I got really interested in the power of clinical supports and interventions. And even though there may be developmental delay, there is a lifetime of cognitive behavioral affective social development that still occurs. And so those three
00:07:00
Speaker
I'd call those like the pillars of my passion and interest today.
Autism Diagnosis Rates
00:07:04
Speaker
But there was a young boy at the time who was completely nonverbal. I'd never heard his voice in all the time I ever spent with him. But very rich facial expressions. You could tell when he was surprised or angry or happy or confused. And I used to do a lot of narrating around him about, oh, you're looking at that.
00:07:28
Speaker
you seem to be enjoying this or, you know, this, um, this is something new. Uh, that's a surprise. Why don't you share it with me? Kind of speaking as though his receptive language, which seemed to be more intact certainly than his expressive language, but he would do this, um, it's kind of repetitive, uh, running around of the,
00:07:52
Speaker
playground, I would watch him sort of look at the side of a wall, look at the path, look at where some of the play equipment are, and pick a line. And he would run linear to these different angles. And I used to watch him do this over and over. And then so I started to play. I would stand where I knew he was going to come to a point that he hadn't approached yet. And he would run right up to me.
00:08:22
Speaker
And he would look up, he's three foot something, little guy, look up and he'd make a bunch of funny facial expressions kind of giggle, turn around and go back the other way. He wouldn't go around me. He wanted to complete his cycle without someone perturbing it. And so I kept doing this standing in different places. I was trying to make a game out of it.
00:08:46
Speaker
I got a minor cold. I was out the next week. And the staff told me that at all the places that I had stopped in that previous playground play, he stopped and was looking around for me. And to this very day, and I think it's where I, there's a lot of themes here that are reflected in this paper that we're gonna talk about.
00:09:15
Speaker
this designation of profound autism or more severe impact, the IQ at 50 or below, the no or minimal language, the increased rate of what we might call challenging behaviors or aggression to self other property.
00:09:39
Speaker
understudied and underserved. And we know now because we're getting designation of that profound label that the Center for Disease Control is including in their surveillance for prevalence, that this makes up roughly 27 to 30% of the entire autism spectrum. So one in 36 children by age eight are now being diagnosed.
00:10:05
Speaker
That's near 3% of the total population in the world. That's 78 million people worldwide who have an autism diagnosis right now. And 30% of them meet this profile that I'm describing. Because we have a difficult time, including them in research, we have very little evidence-based understanding about how to provide the best supports.
00:10:35
Speaker
And so this is where the public health part of me comes in. The people who need our support the most have the least amount of science to suggest what that support should be. And we have an economy of scale problem. We have more people who require support than we have people trained to provide it. And so this is where someone like me trained as a behavioral scientist and experimental psychologist gets
00:11:04
Speaker
clinical translational interest, but then starts hanging out with computer scientists and electrical engineers and machine learning people. I spent a decade working with clinicians and families, but the last 10 years has been working with people to help me figure out how do we instrument people in places and do real world data instead of relying on bringing individuals into the lab.
00:11:31
Speaker
which are strange places with strange people doing tasks you've never done before for ill-defined periods of time and hoping to keep it together and act natural.
00:11:39
Speaker
So you've been around from a time when autism was much less well understood, I think. And you've been in a position to observe a lot of research come out. Do you feel that we're getting it now, that we're approaching understanding autism? Or is it still a mystery? I mean, you talk about it also as something that's fascinating because you're trying to relate to another human mind and one that is different, that's organized differently and approaches the world differently.
00:12:08
Speaker
So, I mean, there's this personal understanding of it that I'm sure you've worked on this for a long time. About 30 years. It's a good set of questions you're asking. I think if I recall correctly, it was 1 in 10,000 being diagnosed in 1994, something around there. 1 in 36 now. So, diagnosis has gone way up as there's been more awareness. Do you think that's accurate or do you think that's over-diagnosis now and it's kind of
00:12:38
Speaker
Hard questions. There are several factors to think that there's an artificial increase. The diagnostic criteria changes. It brings
00:12:54
Speaker
Asperger's in and out, which is sort of the more cognitively and socially able, but still requires clinical support. So higher functioning individuals might have not been diagnosed in the 90s. They weren't. So that's going to inflate the numbers, you know, five years after you introduce that criteria that people can meet. We got smarter about as
Genetic and Environmental Factors in Autism
00:13:14
Speaker
a field about the earlier signs. So now people are starting to add, you know, American Academy of Pediatrics is asking
00:13:23
Speaker
pediatricians to do screening at 12 and 24 months. We didn't used to think of this until six years, eight years of age. So now you're gonna get a big swell in people who were being counted earlier in life that then were traditionally. We also at the same time that these numbers are really starting to increase a lot of the what were smaller single institution or sort of
00:13:48
Speaker
caregiver, family-run foundations are starting to turn into things like the National Autistic Society and the Autism Society of America and Autism Speaks, and now you're getting more resources and lobbying and surveillance and IDEA, new regulations that are passing about special education benefits in public settings. All of this is bringing more attention to it.
00:14:13
Speaker
So I think people are starting to now be more aware, people aren't slipping through the cracks maybe like they were before, or they're not just being called an intellectual disability, or a pervasive developmental disorder not otherwise specified.
00:14:29
Speaker
we're getting finer tuned in what differentiates autism from those other conditions or syndromes, that's going to inflate the number. And then we started to get much better insurance reimbursement for early intervention if you had an autism label that you don't get if you just have an intellectual disability. So there might be in a positive way, but it messes with the numbers.
00:14:53
Speaker
clinicians are giving that label because it's going to provide paid for better services and so diagnostic substitution. It's been a while since I've looked at this, but last I saw kind of big surveys, people trying to estimate and simulate what percentage of an increase would that account for those various factors. And even after they kind of estimate what they think the contribution is, the
00:15:23
Speaker
average statistic that was being talked about at that time was there was still, even after accounting for those factors, seemed to be a 725% increase in the rate of autism over the last decade. And that has continued up to present day. And it's interesting that the same
00:15:44
Speaker
there's a range around any central tendency around to mean there's going to be some places at higher and lower. We're looking in aggregate at the population. These numbers are the same in every continent that we do the evaluations in the same years. And it is irrespective of race, of sex, of social economic status, of ethnicity. It is
00:16:10
Speaker
I think it's happening more. I mean, I'll just say in the 30 years that I've been in the field, I cannot go anywhere and ask someone about autism and they don't either know somebody in their neighborhood or know somebody in their own extended family. And the more severe profile, the not subtle autism, the walking on your toes, engaging in repetitive motor movements, aversion to other people,
00:16:38
Speaker
I see that much more than what I used to see, even in schools only for children with autism. Is your suspicion that autism has always been around in a prevalence like this or that it's increased? I mean, I don't want to go into crazy theories like vaccines and stuff like that. Yeah, yeah, yeah, yeah. I think it's always been around. I think we've called it different things throughout history.
00:17:08
Speaker
It can look like schizophrenia. It can look like intellectual disability. It can look like pretty extreme social anxiety or pretty extreme obsessive compulsive disorder. At the same time, from the best I can deduce from consuming the science and talking to other scientists and public health people, and then just my own experience with service providers, families,
00:17:36
Speaker
the people I know, I think it really is happening more. I don't think it's just that we're better attuned to it. But you know, cancer is happening more, asthma is happening more, ADHD is happening more, obesity is happening more, epilepsy.
Complexity of Autism and Individuality
00:17:53
Speaker
This is not the only thing that is increasing. A lot of things are increasing, but it does seem to be happening at a higher rate than a lot of those other pediatric conditions. And so
00:18:05
Speaker
speculating on why. It's always dangerous to speculate too much, but it'd be interesting to hear anything you have to say about it. Best I can tell, it's a gene by environment interaction. You know, there's been a lot of genetic research looking across populations and genome-wide association scans. There's something impaired everywhere.
00:18:29
Speaker
when you look at the autism sample at that scale. The problem is the same thing doesn't seem to be reproducing for all cases or many cases that have autism. There are many different paths to have the kind of the same, what we'd call an endoclinical phenotype, the same presentation. And it doesn't look like it's directly heritable.
00:18:50
Speaker
There do seem to be increased relationships if you have a great uncle or great aunt who might be clinically impacted but not enough to reach diagnosis, what they call the broader autism phenotype. That seems to produce a child eventually
00:19:07
Speaker
more often than not having broader autism phenotype. We know from twin studies, monozygotic, dizygotic, that there's a differential rate between having the same DNA and not at the rate of autism. There's increased likelihood if they are monozygotic. We know that second-born children of a family that has a first-born with autism have a higher recurrence rate than those that don't.
00:19:33
Speaker
But these genes that look to be a lot of the big hits that show up in a lot of different cases with autism that don't show up in non-autistic cases, they're changing de novo. They're mutating after inheritance. So that means now what's going on in the environment. And that could be the environment within their bodies, their mother's bodies, their living environment, or
00:20:00
Speaker
nature, right, are built environments. And then there's a lot of data saying that pesticides, exposure to industrial chemicals, exposure to more pollution and particulate matter, those increase higher rates of autism, but not for everybody that lives there. So this is the G by E is
00:20:23
Speaker
There's probably a variety of different genetic risk factors, a variety of different environmental risk factors. And if certain individuals get the right combination, all the time I'm saying, right, it's wrong. I mean, it's producing something negative. Well, it may be the cascade that leads to autism.
00:20:45
Speaker
What's so tricky, guys, is we don't have a biomarker. It is all behaviorally defined. If you look in the DSM, it's all qualitative impairment in X, Y, and Z. It's social communicated and restricted and repetitive behavior.
00:21:04
Speaker
with secondary, sensory, and some anxiety. Hard stop. When you look at the DSM, there are several different items or qualitative impairments that if you have two or three out of the list of
00:21:20
Speaker
30 some odd, you meet criteria. Run those numbers. That means that just in the taxonomy itself, you have over, I don't know, 600 permutations of how you are the thing that we're saying that you are. And then there's many different candidate genes or biological profiles that put you at risk. And then a lot of things in the environment that could trigger that. So
00:21:46
Speaker
One of the incredible things about this condition and probably why I remain so kind of obsessed and passionate about it is it's like the ultimate scientific challenge. It is highly heterogeneous. Individual differences are the norm.
00:22:03
Speaker
We have a science that does a lot of lumping. Autism requires splitting. It changes how you think about what is a statistics and a p-value and generalizability. Are you talking about what's shared across people or unique across people?
Predicting Aggressive Behaviors with Biosensors
00:22:19
Speaker
And when we get into the clinical realm,
00:22:21
Speaker
who's in front of you, big data is an N of one, but maybe a million data points versus the big data approach that we take in mainstream medicine, which is usually one data point over a million people. It asks and answers different questions. You know, one of the things that I found really surprising about the paper that you wrote
00:22:44
Speaker
was actually just the clinical population itself, these inpatient individuals with really severe symptoms. And I guess I haven't encountered people with those severe symptoms in my own life. So I was surprised to hear there were so many folks in this situation. Maybe you could tell us a little bit more about this population and maybe a little bit about what's going on with them, but then also maybe about the prevalence.
00:23:12
Speaker
Yeah, yeah, you bet. So what I wanted to do was introduce new recording technologies in a natural environment, not in the lab, and observe behaviors that are naturally emitted, not experimentally evoked, and to see how well could the individuals with autism wear a biosensor and comply with it, and how well could
00:23:40
Speaker
clinical staff make behavioral annotations on the fly. Like this is the messy, dirty world and data collection was really wanting to evaluate feasibility and ecological validity. So this is why it's a, why we're in the inpatient setting. The behaviors that I was interested in better understanding and trying to support are
00:24:06
Speaker
aggressive behaviors. So aggression to other people, hitting, kicking, biting, scratching, poking. Self-injurious behavior, so hitting oneself, biting oneself, scratching oneself. Major meltdowns or tantrums, sort of falling to the ground, kicking, screaming, crying, yelling, but maybe not making physical contact with anyone. And then property destruction. This is what
00:24:34
Speaker
families who have a child with autism who engage in these behaviors will tell you is the primary threat to quality living. And it's because in many instances, and this is what will partly describe the population and who is there,
00:24:51
Speaker
We tend to see a higher rate of engagement in these behaviors with individuals who have no or limited language, have IQs at 50 or below, who seem to have a more difficult time with emotion regulation, and oftentimes who have what we would call medical complexity, so higher rates of seizure disorder, insomnia, gastrointestinal problems, other metabolic conditions.
00:25:21
Speaker
One of the primary factors for referral to inpatient stay is high rates of aggressive behavior. And what that means for the family is that child's aggression is not easily managed. And so they're a threat to themselves. They're hurting their siblings. They're injuring their parents. They're destroying the built environment. If they're in classroom settings, they're putting other students or staff at risk.
00:25:51
Speaker
You probably haven't seen many folks like this because the way families cope is they don't go out into society.
00:25:57
Speaker
They stay at home. For the reason, which is partly the approach to what's described in the paper, parents who know their children better than anybody will say these behaviors more times than not come out of the blue. We don't know when they're going to happen. We can't predict it. We're often caught off guard. And it's the fear of that happening
00:26:23
Speaker
in a restaurant, at church, in a shopping mall, at a playground, at school, that they voluntarily stay at home. And then this has a whole negative sequelae of parents and children not being seen by others and seeing that it's normalizing, that we can accept it. Parents are not getting support from others that, man, this looks hard. Is there something I can do to help you out?
00:26:51
Speaker
The children with autism are not seeing neurotypical kids interacting and getting the benefit of social learning and peer modeling. Neurotypical kids are not saying that these are not just violent, dangerous, angry kids. These kids laugh and play and have other things to offer it. That seclusion, I think, has a two-way street that is not making matters better for the family. And then what'll happen is these kids get older
00:27:19
Speaker
They go through puberty, they get stronger. And you can't just give them a hug and cool them down. And, you know, parents can't necessarily, the kids are getting stronger than them and are getting more willful. And then we see higher divorce rates in those families. We see more substance use. We see more family dynamics that are getting straining.
00:27:41
Speaker
The teachers are getting injured more in classrooms. Insurance claims are increasing for care providers. You start to get burnout and people leaving the profession. And so eventually what happens for a lot of these families, you know, medication works in some cases, but it doesn't, it'll stop working for some or the side effects are very significant. People gain tremendous amounts of weight, et cetera.
00:28:06
Speaker
Parents will find themselves in a situation where this child of theirs is violent and it's occurring frequently enough that they start calling 911. They start going to the ER. They start going to psychiatric inpatient hospitals for long-term stay to try to
00:28:30
Speaker
reduce this what I'll call dysregulated or disorganized aggressive state. And these are acute resources in society, police, ERs and inpatient. They were not meant for long-term utility and service. So they're very expensive to society. So this
00:28:54
Speaker
Project, I've been working on this for 20 years. I know that sounds maybe hard to believe. This observation that what if these are fight and flight responses? What if this is not about escape only? It's not about a void only. It's not necessarily forensic. It's not bad kids taking pleasure out of harming other things. If I can't predict somebody else's behavior, if I have
00:29:25
Speaker
sensory issues. If I have not slept well for seven days, if I have clinical or subclinical seizures, if I can't voice my internal state to somebody else, wouldn't the world be potentially more stressful? And might I have a harder time regulating my own internal state
00:29:50
Speaker
and the world keeps coming at me, we know all the way down to reptiles that if we are under threat or there's ambiguity around threat to confront or flee is adaptive. And so I've for a very long time wondered if some of these aggressive behaviors are not forensic, they're maladaptive stress responses.
00:30:16
Speaker
Second to that, if a child is going to stand up, walk across the room and start vigorously punching and kicking, their body has to engage in a set of preparatory responses to facilitate that increased metabolic demand. So the insight 20 years ago was what if physiology could be observed changing prior to the onset of those behaviors
00:30:46
Speaker
Could we then better understand cause-effect relationships between precursors, behaviors, and consequences?
Biosensing for Preemptive Interventions
00:30:55
Speaker
Could we take the out of the blue away from the caregiver? They know that something's about to happen so that they can get prepared. They can put their eyes on their child. They can rearrange the environment to keep it safe. They can get themselves ready to try to deescalate this child, meet whatever their
00:31:13
Speaker
stressor need is, try to facilitate coping so that the aggressive behavior is not the only means for them to get their need met. I mean, you have to remember, once you aggress, everybody goes away or you get secluded, which is taking you away from all the other environmental stimuli, or in some cases you get restrained, which somebody is providing over control over your situation. It probably also burns a lot of energy and you're tired after.
00:31:42
Speaker
and people sort of leave you, you know, they're reinforcing properties to even that aggressive behavior, but we know behaviorally, behavioral therapy.
00:31:53
Speaker
There are ways to get those same needs met, the same function of those behaviors without the negative cost of hurting yourself or somebody else. Well, the inpatient unit, that's why families are sending them there. It's staffed 24 seven. People are trained in data collection. They're trained to provide responses to aggressive behavior and
00:32:20
Speaker
we get long-term observation because they're there for anywhere from 9 to 100 days. So it seemed like a good safe place to do some experiments where we start to introduce new sensing technology within a clinical workflow. And then what you'll see reported in the paper is
00:32:41
Speaker
How well does that biosensor data, and we get volumes of it quickly, can machine learning help us ask a set of questions about how much data, physiological data from the past, is needed to make a prediction how far into the future of a behavioral event that hasn't yet occurred at what level of accuracy?
00:33:05
Speaker
Let's talk about the kinds of biosensing that's available now. And you've done a considerable amount of this in the past, too. This is not your first experiment using biosensing. That's right. It's easy to get a Fitbit that you can measure heart rate variability. I think people are more familiar with this because you can do a lot with just a Fitbit. So what kinds of sensing is useful to tell you something about a flight or flight response? Yeah.
00:33:32
Speaker
I'll tell you what we've used and I'll tell you why. There may be other additional sources of information in the future that may enrich the signals that we're getting now, but we are measuring peripheral autonomic nervous system signals. So this is heart rate and heart rate variability. So the speed and the timing of the heart beating.
00:34:02
Speaker
electrodermal activity or what some- Okay, so heart rate, heart rate variability, when you just pure speed of a heart rate might tell you something about being amped up, what does heart rate variability tell you? So let me go back for a sec. So right now, one of the challenges has been how do you get a sensory sensitive, socially hypervigilant child
00:34:31
Speaker
How do you record physiology from them? You know traditionally what that would mean is a 12 lead ECG tethered to a machine sitting chair Don't move at all. You're like just act like nothing's happening and we'll get your data that I spent a lot more uninterested in that. Yeah. Yes
00:34:46
Speaker
So then we had halter monitors. So now you could start to wear ambulatory devices. We got, that got a little bit better, but many children with autism still find that to be too foreign and wouldn't want to wear it. And then, so when I was in the media lab was working with Rosalind Picard and the affective computing group, we built a early.
00:35:07
Speaker
a device, we called it iCom, and we published this in IEEE Transactions. That was one of the first devices that I'm aware of that was built and then had some scientific validity of some of the measures I'm talking with you about today that could be packaged in a wrist-worn, almost looks like a watch. And this would, I mean, we're very fortunate, we didn't know this was going to happen,
00:35:34
Speaker
was just before Fitbit, Fuel, Up, Nike, Garmin. There was polar monitors before that. And then we had this punctuated equilibrium in Moore's law. And all of a sudden, we had consumer wearables everywhere. So has that made it a lot easier to be doing this stuff because the technology has been constantly doing it? Exactly right. So what we didn't have 20 years ago that we have now are commercially available consumer
00:36:01
Speaker
biosensors. We have mobile phones with way more CPU than computers had, you know, desktops had 20 years ago. We have the cloud and we have communication protocols that let us now collect and stream data device to cloud, incredible advancements in machine learning and predictive modeling.
00:36:28
Speaker
and then real-time communication abilities via telephones and text, that ecosystem did not exist 20 years ago. Well, it exists now. And really, this paper is kind of a demonstration now of if we put all of that into practice, trying to see if we can predict an aggression before it occurs, is it possible? So if I come back now to the sensing part,
00:36:57
Speaker
the autonomic nervous system is peripheral to the central nervous system, the brain, right? So when we engage in fight or flight responses, we have two branches of the autonomic nervous system. We have the sympathetic, which is kind of the fight or flight, and we have the parasympathetic, which is the rest and digest. Think maybe about for ease, sympathetic is the accelerator, parasympathetic is the brake.
00:37:23
Speaker
Our vagus nerve, tenth cranial nerve goes down our spine and we have projections of sympathetic and parasympathetic to the eyes, the mouth, the lungs, the heart, our vascular system, our stomach and our reproductive organs.
00:37:39
Speaker
And what the biosensor is letting us do is look at how fast is the heart beating? What's the regular timing interval? So heart rate variability, to answer your earlier question, as we have more irregular distancing between beats, that is a measure typically of health, elasticity in the system, adaptability quicker to be able to respond than slower heart rate variability or fixed heart rate variability.
00:38:10
Speaker
And so what the device is measuring is blood volume pulse. It's using photoplasmography, PPG, it's shining optics, red and green lights at the underside of your wrist. And because oxygen is in our blood and has a different density, as it's traveling slower or faster through those light arrays and the refraction rate, we can estimate from blood volume pulse, interbeat interval,
00:38:40
Speaker
with which you can calculate heart rate and heart rate variability. Then there are two other electrodes that are recording skin potential of sweating, electrodermal activity or galvanic skin response. So what exactly is this telling you? I think of it, I mean, this is something, you know, lie detectors, right? Use something like that just to tell how nervous you are from little micro-sweats that you have on the surface of your skin.
00:39:06
Speaker
Similar set of measures. So you're getting, so same that were used in polygraphs. I think it's informative to know that you cannot use a polygraph in a court of law and there is not scientific evidence that you can't game those systems. So I do not believe that polygraphs are necessarily correct, but they are same premise that, you know, if it's more effortful and the stakes are higher to tell a lie than to tell the truth,
00:39:35
Speaker
that you should see more sympathetic nervous system arousal when you are having to think harder or keep a poker face or tell something untruthful. The principles are the same.
00:39:51
Speaker
But there have been cases of people, you know, put a tack in your shoe and every time they ask you a true question, step on the tack. That way when you tell the lie and not step on the tack, you can't differentiate which is truth and which is lie.
00:40:09
Speaker
There's a whole thing about the polygraph that is, I don't want to say it's sham. The application of it, I think, is pseudoscientific. The measures itself... It's indicating some reaction, but it's not indicating... Not that level of specificity. You need much more control over the environment and context in order to say that.
00:40:28
Speaker
But similar measures, it also has something called thermopile. So that is giving you skin surface temperature. And then it's got a three-axis accelerometer. And in some now, we have gyros. That's telling you about motion as a function of gravity in space and time. And so collectively, the biosensor off the shelf that we use
00:40:53
Speaker
is giving you heart rate, heart rate variability, electrodermal activity, motor activity, and skin surface temperature. A priori, what would you have predicted the time course would be like for say an episode where it could be predicted? Good question. It's going to be very different for different people in different settings and different background factors. I was going to jump for joy if we could see it a minute in advance.
00:41:21
Speaker
Yeah, that would seem like a minute would seem like a long time to predict something like this in advance. Our first published attempt at this, we got a minute. We needed three minutes from the past. We could make a prediction one minute into the future at about 74% average accuracy for everybody.
00:41:43
Speaker
We in the more recent study have seven, sorry. And that was, how long ago was that? I think that was about three or four years ago. That was one clinical inpatient site with 20 individuals with autism and only looking at aggression to other people.
00:42:01
Speaker
The recent study is 70 individuals, none of whom were included in the prior study, at four different clinical inpatient sites. And we were looking at aggression to other people plus looking at self-injury and also looking at meltdowns or tantrums. And the results this time around suggest we can make a prediction three minutes in advance
00:42:29
Speaker
with 80% average accuracy for all participants and all behaviors. And what gets very exciting to me about this is three minutes is enough time to do something, preemptive. That's a long time. I mean, it really is surprising that three minutes is something's rebuilding up for that long. Yeah, the impact of that is amazing. I mean, I was noting in the paper
00:42:57
Speaker
that in 497 hours of observation, you saw 6,665 aggressive behaviors. That's more than 10 an hour. That's correct. That's a lot. A lot. And it's partly this impatient setting, right? That's why they're being sent there. I knew that. I knew we would see high frequency so you can collect, you can get a lot of instances in a shorter amount of time. I mean, that's still 500 hours of data collection. This is an enormous task.
00:43:23
Speaker
of a very large team of people gathering that data set.
Future of Biosensing Technology
00:43:27
Speaker
But we get a lot of instances in a short period of time. The next, I mean, there are two major next, three probably major next steps of this work going forward. But the next is, most immediate, is reproducing these same experiments in an outpatient setting.
00:43:45
Speaker
Whereas something might be a lot less frequent, you might be more worried about false alarms, that you might be setting it off too often. Exactly right. Exactly right. And, you know, the really, I'd say dangerous aggressive behaviors are the ones that incur that occur infrequently, but, but with very high intensity, the infrequency gets people off their guard. And but the consequences are very high because it's a very extreme or dramatic
00:44:14
Speaker
outbursts, typically, that nobody's prepared for. And that's when stuff really gets challenging. So the outpatient is going to do a few nice things for us. This is now a collaboration with the Marcus Autism Center at Emory University. It's a NIHRO-1, so it'll be four years. We're in year two now.
00:44:35
Speaker
Atlanta has a much larger population and a much more diverse population than the clinical sites that we've been gathering data in before. So I think we'll have greater demography, geography, ethnicity represented.
00:44:52
Speaker
It's also, our data is being collected, the biosensor data is being collected in the context of functional analysis of behavior. So these are trained behavior analysts who are doing several, I think they'll do up to like 15 different repeated sessions where they are intentionally manipulating the environment to see if they can elicit the behaviors so that they can determine whether they are socially mediated.
00:45:22
Speaker
or automatically maintained, which has a different function of the behavior, which also has different intervention approaches that have differential effectiveness depending on
00:45:36
Speaker
What's the generating function of the behavior? And so one of the things I'd like to do is see if we incorporate that clinical knowledge about automatically maintaining to socially mediated or sensory mediated or object preferentially mediated or escape or avoid a demand mediated.
00:45:57
Speaker
If we can incorporate those as features in our machine learning models, can we do a better job of making predictions for different types of individuals with autism who engage in aggression than treating them as a single class one size fits all. It also means, and it's the other thread of the work that is not, it's alluded to in the paper, but it's not described in the paper is collecting this data, running machine learning on it in real time.
00:46:25
Speaker
and pushing just-in-time adaptive intervention alerts in the moment pre-escalation. And so we've built a software infrastructure to do that. And if our models get better in the outpatient setting and increased
00:46:47
Speaker
accuracy of the predictions are made because we can incorporate clinical decisions about the function of the behavior. And we can send real-time alerts prior to onset of the behavior. I want to send those system home
00:47:02
Speaker
with the families where the clinician has provided a behavior management plan based on their functional analysis and trained the parent how to implement that on their own and see what is the benefit of alert notification reminding the families
00:47:20
Speaker
before they have to deploy the behavioral intervention to get ready, will delivering the notification versus suppressing it have a measurable impact on the success the parent has at avoiding or reducing the length and harm of the aggressive behavior?
00:47:38
Speaker
Well, I mean, intuitively, you would think that the earlier the earlier you intervene, the more effective it's going to be. Right. Yes. Yeah. Like you kind of you break the chain, right? You reduce the behavioral momentum. Right. Exactly. Before it gets out of control and sort of. That's exactly right. Yeah. In terms of that, I mean, let's say you had three minutes of a warning that something was about to happen, an aggressive behavior. What kinds of interventions
00:48:06
Speaker
are thought to be effective or hypothesized to be effective. Yeah, so I won't differentiate them, but I'll allude to some of both classes. And some of this is blue sky. We'll see. We'll have to test and we'll really know.
00:48:23
Speaker
Remember, too, that we live in a world with a lot of smart devices. By smart, I mean they're connected to the internet. They're on local networks. They know that there are other technologies that are a part of the same household. There are if, then, that protocols that you can have technologies respond contingently to other technologies. I just want to give a little context because the answer may sound different than the one that you're anticipating.
00:48:51
Speaker
say you've got the sensor, say you've got a phone or a computer that's receiving the data, it's connected to the internet, it's pushing the data and any behavioral observations that a family has made to the cloud, the classifiers are running, then they push in response an alert that there's three minutes. Okay, so the doors of the house could automatically lock so a child can't run into the street.
00:49:19
Speaker
lights could get brighter or dimmer to signal to somebody in another room come here or to try to reduce the visual load of the person who's in the room.
00:49:30
Speaker
Music could come on your sonos and play sounds that you've determined before or soothing or maybe a favorite song that the kid likes to break the chain from whatever they were focused on to now a more desirable reinforcing stimulus. Lexa could come on and start recording the environment or giving some verbal instructions.
00:49:58
Speaker
a communication to the parents' phone who may be in a different room. You should probably start heading down to the playroom where your child is at because they're starting to escalate. So they're going to stop whatever they're doing, go be physically present with their child, giving cues to take a deep breath and relax, doing tasks that are mastered, that are easy. You're trying to break the attention
00:50:27
Speaker
and the momentum away, you're trying to de-escalate or to redirect. We have evidence that
00:50:37
Speaker
de-escalation, reinforcing incompatible behavior, redirection, all of those reduce the impact or giving someone who is nonverbal speech communication opportunities where you can help them point and articulate in a picture that maybe they need to take a break or my tummy hurts or you have to remember they can't tell you what their life experience is like on their own.
00:51:04
Speaker
So these are all things that I would seek to bundle the sensing and the alerting with. Ultimately, it'll be the clinician and the parents who tell us what configuration of
00:51:18
Speaker
how to give that alert, what can change in the environment automatically and what we can do with our child given that period of time will give us a sense of, you know, even if you don't resolve it in three minutes, I would like to think that maybe you can get three more minutes. You know, maybe you could keep delaying the, that final outburst.
00:51:41
Speaker
Well, in a sense, it seems like some of this is quite similar to biofeedback, right? It'd be like other enabled biofeedback. I mean, there's plenty of things we could do. You know, when I have this emotion, or I think this thought, all of a sudden the doors lock, or Alexa comes on, or all of this stuff, right? It's all a consequence of that internal state, not necessarily something we've done, but something that they've, you know, they just have this internal state. Now, this gets hard, right?
00:52:08
Speaker
Because everything I've said, most people I talk to say, I totally see, like, I'm excited that we're going to likely see some benefit come out of this. But the peripheral autonomic nervous system is part of a generalized response that our bodies have to a myriad of things that may not have to deal with aggression, right? We regulate our
00:52:34
Speaker
Blood pressure, we regulate our temperature. We experience pain. There's other noxious stimuli. We get in bad moods. We eat things that don't make us feel good. So there's an ongoing question of sensitivity and specificity of these measures and those behaviors. And then we have to think about time. The farther you get away,
00:52:59
Speaker
from the onset of the behavior, the more intervening factors in life can happen where your changes in physiology may not be related to that ultimate behavior. So while I gave you a little bit of a picture where everything is automated, I feel very strongly this data and the actions people take have to involve a human in the loop.
Potential of Emotion Recognition and Biofeedback
00:53:28
Speaker
Do you think there might be some room for an automated response in higher functioning individuals who still may have behavioral issues that can integrate this information and sort of understand what Q-IT provides?
00:53:45
Speaker
Yeah, very good question. I've done focus groups with more verbally able individuals with autism about this work. And they say, I would love it if this was like my personal assistant on my phone of saying, you're starting to trend. Your body, my body is starting to trend in a direction that last time when I didn't deal with it, it went really bad for me.
00:54:13
Speaker
And so this is a cue to raise my consciousness, take a break from what I'm doing, communicate to other people, give me a little space while I kind of gather myself. It could be self-monitoring and self-management, 100%. The individuals with profound autism that are in this inpatient setting and really more in the outpatient setting, many of them do not have the verbal and the cognitive
00:54:42
Speaker
abilities that make it clear to me that they could do that on their own. It would be something I think we should always try. That's why I focused a little bit more on the caregivers and the support staff for that segment of the population.
00:54:57
Speaker
But yeah, know thyself, quantify thyself, and be proactive. But self-regulation can be harder for lower functioning individuals, of course. But I know that you have done in prior research you've worked with, and we can talk, I'm sure you have tons to talk about other research too,
00:55:17
Speaker
emotion recognition for higher functioning, autistic individuals? That was also a really nice kind of early start enabling technology in the Media Lab. So half of us were working on the biosensing. The other half were working on facial expression detection and kind of computer vision and affect.
00:55:43
Speaker
And if you remember back in your early psychology classes and studying emotions, one of the leading models was the circumplex model of affect two dimensions. We have arousal, high to low, and we have valence positive to negative. So if you're looking at a quadrant and now you could kind of plot each discrete emotion is kind of an elliptical.
00:56:10
Speaker
somewhere around the center of that quadrant going out. So positively valence, high arousal, negatively valence, high arousal, low arousal, positive valence, high arousal, low valence. And you could start to sort out what is anxious from relaxed, what is stressed from excited. So in the absence of someone telling you about their affective state, their emotional state,
00:56:39
Speaker
One of the better objective ways is to look at facial expressions and look for valence and look at peripheral physiology for arousal, for physiological arousal. So we were looking at like, could we combine those two together or single modality? How well could you help a more cognitively able individual with autism
00:57:07
Speaker
learn how to recognize facial expressions in other people where a small tablet with an outward facing camera is finding faces in the natural field, plotting action units on their face, running a facial expression detection algorithm. We would then send kids out on emotional treasure hunts. We'd say, go see if you can find instances of other people looking
00:57:36
Speaker
happy, confused, afraid, disgusted. And then we would turn this into social stories where we'd say, now go see if you can elicit those facial expressions in other people. And we were trying to play with the affinity for computing and watching videos in that segment of the population, where they're learning something about perception, production,
00:58:01
Speaker
language to communicate emotional things with the idea of like, let's gamify what is otherwise a really boring way to teach people social emotional understanding. Now, have you kept up on any of that stuff as neural networks have been improving and facial recognition has been getting much better? That's all moving forward. I have also had better mentorship, more time to read and think.
00:58:29
Speaker
and really impressed by some other scientists' research showing that emotions are much more complex than we have traditionally come to think they are. And just a computer detecting them by the face is only gonna get you so far.
00:58:47
Speaker
We construct emotions. We perceive our bodies. We selectively attend to things in the environment. We have histories. We have variation in experience.
00:59:00
Speaker
emotional language, emotions are very context dependent, being able to just have a computer tell you what the feeling state of a person is based on their face. I now am more conservative about the accuracy and benefit of that than I was quite early on.
00:59:20
Speaker
I don't mean to say that it doesn't have a role to play, but I'm seeing less applications of that technology in what I would call quality health science. I'm seeing it more in marketing and advertisement and maybe like driver safety.
00:59:38
Speaker
Well, this research is fascinating and really important work as well. I think the application of this is tremendously, the potential for this is incredible. And there's such a need out there as well. I guess as a last question, Matt, what are you excited about in terms of next steps? What's the next big thing that you're really excited about? Yeah, thank you for that question. There's sort of two things, I think.
01:00:09
Speaker
So this was published on December 21st. That's only a couple of weeks ago. It's already been viewed and downloaded over 4,000 times. That tells me that there are people out in the world who are interested in this. We also published a supplement that details every step we took to produce our machine learning classifier.
01:00:34
Speaker
And the Simons Foundation has agreed to host our de-identified data set, and I am going to give the data that we used to run all the experiments in the paper freely available to other scientists to see if they can outperform us.
01:00:51
Speaker
crowdsource this data, the data will be de-identified. But if those investigators are willing to share back their successes and failures like I will continue to do, I'm excited that we might continue to improvement that gets to families faster. On that second side, I am increasingly interested in developing software that a caregiver and a family member can use
01:01:19
Speaker
that provides a direct benefit to them day to day, but also, so it makes it sticky. They're going to keep using it because it's, it's providing a clinical benefit and a mass increasingly larger, more diverse data set, and then bring real world data back into academia and have my students doing theses and dissertations on human lives day to day innovating in,
01:01:48
Speaker
new machine learning or new sources of data collection that are going to kind of power what would then be even better and more broadly available products and services that help parents better manage this class of behavior so that these individuals are
01:02:12
Speaker
not reliant on too few, too costly professional services. It's really like, can we empower the families and communities to have less reliance on mainframe, expensive hospital? Okay, last question. Interested families who might hear this and wonder how they can keep up with any new developments in the field, anywhere you can point them.
01:02:38
Speaker
Oh, boy, I need to get better organized because I want to I hear from families a lot and I love it. I don't this is not a product and a service right now. I can't provide it to somebody. This is research. This is being maintained by faculty and graduate students who will get your Kickstarter up and going. So that's a good that's so that's where this needs to go. If there are people out there who have a child that
01:03:08
Speaker
meets a description, thinks that this would be a potentially good solution. I wanna know who you are and I wanna ask more questions to figure out what the common needs are. And I wanna see if we can get a critical number of people to help find investments so that we can really focus on delivering this as a service. I'm being very careful.
01:03:36
Speaker
and very slow to make sure scientifically that whatever claims we're gonna make about these technologies will be accurate. But eventually we're not gonna know until we really put it into practice with a lot of people. And so that'll be the next hump is getting this out of academia only translating this into community science. And then ultimately we're gonna live with
01:04:01
Speaker
live with these technologies and really see their
Future of Autism Research and Community Engagement
01:04:03
Speaker
benefit. And once we see that benefit, I'm gonna do whatever I can in regulatory, in evidence-based to get FDA to approve this, to see if we can get insurance companies to pay for this. The people who really need it can't afford it. So this is, when that will happen, how that will happen, I don't know yet, but that's gonna be the next frontier. How do we deliver this?
01:04:31
Speaker
Well, Matt Goodwin, this has been absolutely great having you on. We really appreciate you spending some time with us. We've been wanting to have you on the show for a while, so thanks a lot for hanging out. Cool. I really appreciate it. I enjoyed talking to you guys, and I appreciate you spreading the good word. Thanks, Matt.