Bringing Clarity to Psychiatric Chaos

If a study aspires to explain everything about, like, schizophrenia or PTSD with three variables or a treatment that targets only two of them, says that it can have very large effects. We kind of know on its face, if we understand the world, that it's. How could that be?

Hello, I'm Dr. Farah White.

And I'm Dr. Grant Brenner.

We're psychiatrists and therapists in private practice in New York. We started this podcast in 2019 to draw attention to a phenomenon called the Doorknob comment. Doorknob comments are important things we all say from time to time, just as we're leaving the office, sometimes literally hand on the doorknob.

Doorknob comments happen not only during therapy, but also in everyday life. The point is that sometimes we aren't sure how to express the deeply meaningful things we're feeling, thinking and experiencing. Maybe we're afraid to bring certain things out into the open or are on the fence about wanting to discuss them. Sometimes we know we've got something we're unsure about sharing and are keeping it to ourselves. And sometimes we surprise ourselves by what comes out. Welcome to the Doorknob Comments. I'm Grant Brenner. I'm here with my co host, Farah White, and our guest today is Glenn Sachs. Dr. Sachs is a renowned psychiatrist, researcher, and leader in the field of traumatic stress and child psychiatry. He's best known for his work on the developmental impact of trauma and the treatment of children, adolescents, and families affected by traumatic events. Dr. Sachs is the founding director of the center for Child Trauma and Resilience at NYU Langone Health and previously served as chair of the Department of Child and Adolescent Psychiatry at NYU School of Medicine. He's known for trauma systems therapy and his work in computational psychiatry. Dr. Sachs, you have so many accomplishments. Maybe you could fill us in on some more of the highlights by way of introduction and welcome.

Well, thank you. It's great to be here. Thanks for inviting me. I think you did it pretty well. I guess the only thing I'd say is I'm very proud to develop trauma systems therapy, which is a treatment used in a lot of different places for kids and families under a lot of adversity. I'm also really happy to focus on computational psychiatry. I think it's really the future, especially an area of it called causal data science. And recently with my team, we were funded by NICHD to establish a center on causal data science for kids and families with maltreatment and called the CHAMP Center.

What does CHAMP stand for?

It's the center for Causal Data Science on CHAMP is child and adolescent maltreatment Prevention.

Thank you.

So what was, you know, the first moment that. Because I am really curious about how the trauma systems therapy came to be. And could you talk to us a little bit about what you saw in the field before you developed it and how you sort of see it being used?

Yeah. So, you know, I approach just about everything as a clinician. I'm, you know, I had good clinical training. I see the world and its complexity and what families are facing. And now this is Trauma systems therapy began to be developed 1999, 2000. I was chief of psychiatry at Boston Medical center, which is like Bellevue Hospital in Boston. And it was clear that kids and families were, it wasn't post traumatic, it was peritraumatic there. You know, all kinds of adversities. You'd be doing your best with evidence based treatment in an office and they go home and come back in worth shape. And so with my team, we wanted to figure out how to do it differently, how to address the social environment in a rigorous way. And that was the, the core of this. And then we've been on a ride since that time. Our first book was published in 05 and you know, and it's being used a lot and I'm really happy that people have picked up some of its core ideas.

What are some of the core ideas?

Well, the core ideas is everything's related to what we call a trauma system, hence trauma systems therapy, which is about two things. One, a traumatized child who can't regulate their emotional states. Two, they live within a social environment that isn't sufficiently able to help them with that regulation. And it gets way more complex. But it's built on top of those two things. In essence, what trauma sets up is the situation where kids get triggered from all kinds of events in their social environment. They get horribly dysregulated. People don't know why they're getting dysregulated, but there's patterns of it. And so you, you work to help understand and manage what stimulates the dysregulation. And then you help build the child's internal capacity to regulate when they're stimulated like that. That's in essence, it gets way more complex. You know, part of it is you have to be set up to understand that every kid and family is different. What disregulates them are very different. And so you have to be ready to assess and understand.

Terrific. I mean it sounds like this should be and could be used almost Anywhere, at any developmental stage. Is there something specific that you had in mind when you were creating it or.

Well, I created it for an inner city hospital in Boston and I thought it would be a mental health model, that sort of thing. But what surprised me actually is people began picking it up. People started in the child welfare system, thought this could be useful, the residential system, a system related to substance abuse refugees. So it kept being picked up. I did one thing that people told me I was crazy to do when we developed this, but it was the best thing I ever did with this. I didn't claim intellectual property at all. And what that meant is there were core ideas that people could innovate on top of, based on their knowledge of the world. I was very influenced by book by Eric Von Hippel called Democratizing Innovation. He and I met with him as we were designing this. He ran innovation at MIT and his idea was the open source movement. You have platforms of innovation, but then you give lead users the ability to make it work for them. And so TST became a platform of innovation. I mean, I wasn't working in residential care. I didn't, you know, but people adapted it. The idea then is that in a community people freely reveal their innovations that then get picked up by people who have similar problems. You know, it was picked up by a program in rural Kansas. What do I know about this health system in rural Kansas? All of that, but, but part of it is, you know, and if you read our second book, it's radically different than our first book. Why? Because so many innovations got subsumed into tst. And you know, I think there's a real problem with like intellectual property about these sort of things. What ends up happening is you get a treatment, then stuck in time whenever you developed it, with no, no ability to change, improve, adapt to the world based on users needs.

It's like an open source model. And I can imagine, you know, I've been, I've been playing with some AI platforms where you can, you can tell it what you want. Like you want a role playing bot or you want an app, and then you can give it parameters that it guides the underlying AI engine. You could select from a number of different AI engines and you can load it up with specific data. So you can load a textbook or a bunch of textbooks or whatnot. And then, you know, you can operationalize it. You don't need to know how to code because it does the coding for you. And so that is kind of the ultimate flexible model. Like we could upload trauma systems Therapy and, you know, then people can access it. But that's innovative because, you know, a lot of times in academia and in commercial interests, people try to maintain intellectual property ownership and it, it prevents sort of the most number of people from being helped. So.

Yeah, I mean, isn't there, Isn't emergence beautiful? What.

Yes.

What emerges out of the complexity is something you would never have predicted. Right.

Like things like locking down the ownership of the intellectual property prevents complex systems from evolving naturally. Right. Or emerging.

That was our, our experience for sure. And, you know, we've. It's been quite a ride. But. But we have people all over the country and different countries who are using it, making it work for them. We have a monthly innovation community meeting where different people share their ideas and all of that. It's become quite a platform. And part of it is recognizing the limitations of expertise. I know some things, but there's a hell of a lot I don't know. And why would I expect to know everything for what people need in all kinds of systems in all kinds of places? I mean, you know, there's an arrogance to that and, you know, opening things up, allowing people to make it work for them, creates unexpected great things.

I think that's a wonderful point. And sometimes when I'm going in to do research or look at something or I'd like to learn a little bit more about one particular modality. I don't really do sort of manualized treatments in my practice, but I have noted that some, some of these theories, and I guess I don't want to specify, but there are waiting lists for therapists to be trained in, in certain types of therapy. And I do think about how much need there is out there. And I have thought to myself, wow, I wonder, you know, how ethical it is. I don't know. I'm not a researcher. I'm not developing anything. I'm just implementing things as much as I can, you know, to help as many people as I can, which is kind of a drop in the bucket compared to, you know, these big centers. But that is something that I've come up against and I. I've thought a lot about it, but I haven't read the book that was so influential to you, so I'll check it out for sure. Yeah.

I mean, part of it is in the field, if we understand human beings are really complex and then the environments humans inhabit are awfully complex, we have to make sure that, you know, whatever we do is true to that complexity or we miss enormous things. So if a study, aspires to explain everything about, like schizophrenia or PTSD with three variables or a treatment that targets only two of them, says that it can have very large effects. We kind of know on its face, if we understand the world that it's, how could that be?

Well, I think complexity is tricky for people. Alberti, who's a Nobel laureate, he wrote a kind of a manifesto that was published in 1970 called the Crazy Ape. And he was really talking about anti nuclear war and politics. I know our, we were introduced by a colleague, Fred Stoddard, who is very involved in anti nukes. And Svengyorgi talks about how we face this modern world with what he calls a caveman's brain, which is problematic. And so, you know, a lot of your work has been to develop specific ways of capturing complexity and also sifting through the noise to find, like, what is the signal, like, what really works. And that process you described of kind of let's test two medications and see what happens, or let's test one factor that gets really tricky. And I know you've done a lot of work in causal modeling, but before we get into that, Farah and I were talking earlier, and she had talked about asking you about resilience and, and in light of your work with trauma systems therapy. And I'm, I'm thinking about intergenerational transmission of trauma, how that plays out over the generations, which, you know, in some ways is more complex, but in some ways for me is simplifying. When you see a pattern that just clearly marches from generation to generation, it kind of comes into focus much more clearly sometimes then looking at, you know, one generation. So actually, what I wanted to ask you about was kind of the flip side of that, and I don't think this is a real concept. It's me making something up. Intergenerational transmission of wisdom or healing. And I'm wondering how you think about that, what that would mean to you.

Now, that's a great question. I think, you know, I, I, I thought of that in a completely different context than psychotherapy all of, But I think that's, that's, you know, the, the idea of wisdom and what gets transmitted over the ages and the trash heap of history is filled with wisdom that fell away, but then there's wisdom that stayed over hundreds and thousands of years. And I think we ignore that at our own peril doesn't mean that just because something came to us over a long period of time, it's necessarily good. But we have to take that information that it's Lasted as something that's deeply meaningful. Yeah.

And I think that's a good point. And I do wonder if things have changed so quickly in today's world. I'm just. I don't know why it's coming to mind. But this idea of bullying. Right. And I don't know if people have sort of been able to trend it. I know it's a major risk factor for a lot of, you know, psychiatric disorders and depression, anxiety. And I wonder if with sort of the explosion of the Internet and social media, if we're finding that we almost need to translate the wisdom of previous generations into something new. And I wonder if people have taken trauma systems therapy and sort of applied it to some of those principles. I don't know.

Yeah. I mean, there's a few things I'm thinking about. Obviously, there's risks to bullying and we want to help people with that. But then there's also the expectation that childhood may go on with little stress and we should protect kids from stress. And if a kid is feeling intimidated by another kid, we have to protect them from that. There's something about helping kids learn how to manage problems, how to manage discomfort, all of that. And that was known throughout the ages. You know, the wisdom we may be talking about is wisdom of managing the crap of life, and which we can't eliminate that no matter how hard we try. But when we eliminate, kids need to deal with it and figure it out. We rob kids of certain enormously important resiliency capacities.

There's a concept of post traumatic growth which tends to occur with moderate levels of post traumatic stress. And bullying by definition, crosses some line. It's kind of a systematic set of attacks, which might be contrasted with kind of the. The way kids are. Not all kids are nice. And if you're overprotective, right, like the helicopter parent problem or the age of anxiety, then kids, you know, don't develop those muscles. It's like if you sit home all day and stare at a computer screen and you never get outside and run around. You won't. You won't develop physical fitness. So it's. It's a tricky question. However, I would say in terms of the wisdom of the ages, I don't know that any prior age has done a better job dealing with kind of people's inhumanity toward one another. And I think in a very significant way, and this is getting a little bit far afield, but baby speaks to current events. The ethical dictum might makes right still is one of the leading things that guides the world. If you're better or stronger in some area, then you're likely to prevail.

Yeah. Again, just because something may come to us from the ages doesn't necessarily mean it's good. But, but there's again. And so I, I'm not supporting that. Obviously there's been a lot of harm caused by that. It's just, you know, we also have a culture that tends to be ahistorical, that doesn't, that feels like everything true, everything I need to know is about now in my very, very, very, very, very small circle compared to the circle of all people over all time, you know, and what might be learned. And that's what I worry about.

Yeah. Things are kind of on the surface, but there's a greater breadth of information which I think is part of the problem. Like there's a way where people nowadays know a lot more about a little.

Yeah.

And then they, they think they really understand. But you don't know the history, like you said, you don't know the story, you don't know how you arrived at that point. You don't know the depth behind it. And I think the meaning gets thinned out as well.

Yeah, I mean, wisdom from culture, from religion, all of that about how people work together and, and all of that. Again, not everything's good, that's obvious. But we, we, we, we kind of flippantly dismiss old fashioned ideas at our own peril.

So how does causal data science rescue us from this problem, these problems? And, and maybe we can start with, with the problem that you've talked about of the limitations of the current psychiatric diagnostic system, which is, you know, hotly debated on social media. You've got pros, cons and everyone in the middle fighting it out.

Yeah. So I mean, the first thing is what is a cause? A cause is something that if I intervene on, it will change something else. If I want to intervene to change something else and it's not a cause of something else, I could do it for a hundred years and it won't change something else.

Okay, that's a little, I just want to point out that that's a little bit different maybe than the commonplace definition of a cause, which is more like an antecedent, something that happened in the past that like set the situation up, but it's not the thing that is keeping it going now. Whereas if you intervene, it will change it.

Well, there's many, many things antecedent to other things that are statistically associated with those, but they're not causal of it. And If I want to change any of those things, they're not going to change what I want to see change.

Yep. There's a joke in physics like, you know, what is the source of human happiness or something. And it's the laws of physics and the conditions of the universe at the Big bang is always the answer. But you're saying, like, you have to look at what's happening, something like in the here and now.

Well, not necessarily. I mean, you have to intervene on factors in the here and now. But things before can be causal of things now. Things before can be causal of the things that are causal, that proximal and distal causation, all of that. But the bottom line is, is the cause or not? And if it's a cause, how big an effect is it going to be? Because there may be causes that if you change, will only change things a little bit. But let's stick with cause or no cause. So again, and you're talking about diagnoses in psychiatry, it's either a cause or it's not. If I intervene on a non cause, it's not going to change anything. So then let's bring it into the clinic. Okay, how do I know about how I might want to intervene with the patient? It's because I use their clinical data to categorize them into a category of risk and probable response to an intervention if I believe they're at risk. So the only way I could know if they would be probably responsive to an intervention if I knew that intervention targeted the cause. Okay, so classification has to be about risk and about probable response. Intervention. That's how it works in medicine.

That's how can you, can you give an example of a cause and effect, like something which is a cause, and then if you target it, it gets better.

Just for listeners, you know, something like, say, a heart attack, you know, myocardial muscle damage, which is related to, you know, the coronary artery narrowing. That sort of thing leads to all kinds of morbidity and mortality problems, congestive heart failure, rhythmias, death, all of that. So causal, all of it. So can I, you know, reduce. Increase coronary blood flow? Can I do surgery? Could I? All of those things. That's simply true. Diagnosis in medicine is looking at causes you can intervene on.

How about in, like psychiatry? I know you've done some work on ptsd, or you could talk about depression. Just. Is there anything that you can identify there that would help illustrate the principle?

Well, but, but part of it is this. Again, if, if I have a diet, let's go with dsm. So if I have a diagnostic system based on symptoms and I don't know about causes, then I'm using that to classify people into categories of risk and probable response to intervention. But if I do an intervention based on that category, by definition, especially if we believe the disorder is multi causal, they're going to be mixed. You know, I don't know how many of them that cause is relevant. So I intervene with whatever I intervene on and it's only going to change anything if it happens to change a cause amongst a set of people in that big category of people. You know, if we believe most believe that any major mental disorder has many, many causes. So think of what that entails. It entails, let's say major depression. We categorize many, many, many people in one category. We've just admitted that category is vastly multi causal. So what can we expect?

Do you think that's part of why depression treatment isn't as robust?

That's exactly why it's not robust.

Right. Even with therapy has a pretty strong effect, but it's, it's kind of trial and error. You try different things or you look at newer treatments like tms, which stimulates the frontal cortex of the brain and it doesn't have 100% response rate, but you see patients having 60 to 80% remission. And it's not using data science I think the way you do, but it suggests that we may have identified a cause. When you see a consistent positive response to the same intervention and in that case cause would be like low activity in some brain network.

So let's. So right, that's exactly the question that you want to know. I mean your main question is how can I identify the people who would be responsive to that intervention? So then use the best information you possibly can. If there's a pattern of brain imaging findings that completely distinguish that group from everyone else, that should be your diagnostic criteria. That's what the diagnosis is for.

That would be like a future state. With tms, they're doing that. With ptsd, there are different patterns. With depression, it's not as clear, but that gets into the complexity that you're talking about. Farah, how do you think about psychiatric diagnosis? I just want to ask as you know a clinician sort of on the ground, like how you think about it and because I think part of what Glenn is saying though, you know, you're, you're being measured, you know that you, you really have some critiques of, of the way we think about it now because of the amount of sort of harm it does leaving people without treatments that could be better. Right.

Well, if I think about, think of any medical field that has had a ton of progress. Think of breast cancer, think of cardiac disease, all of that. Think of the diagnostic criteria for those disorders and say, I don't know, over the top of my head, 1980, okay, now think of the diagnostic criteria today. It's radically different. Why? Because the fields have put their most promising causal findings in their diagnostic criteria. They kept updating and updating and updating it because the only way an intervention can change an outcome if it targets a cause. Okay, now look at major depression, look at schizophrenia, 1980 with DSM 3. And today, you know, maybe you could find a little difference. Look at our, look at the tens of thousands of articles published, okay, on the ideology, risk facts, protective associates, all of that. Where are they? You know, are have. If we're touthing their treatment potential, if we're touthing their promise for patient care, why have we not decided they're so promising that we use them to classify patients? Isn't, you know, I think that's mind bog.

I agree. I don't know necessarily what we, what we can do to change it. I think that's why, you know, we're so interested in talking to you and hearing about what you're doing. I do feel really fortunate that I'm not beholden to the DSM or that it doesn't have to define how I work. But you're right, it would be lovely to have and I think Grant, you do a lot more of this with scales and in, in your practice. But I see diagnosis, causality and those types of things as really dynamic.

Yeah, a good clinician sees them as very dynamic. You want to, one, you first, you want to personalize your intervention to the, those particular causal factors that would be most relevant for a patient. So that's one thing. And then you want to. It's what clinical skill is all about. It's about like in the moment, looking at what's causing someone to do certain things. A patient missed your last appointment, now they're back. So you're worried that they're. Maybe they've been upset with you, all of that. So you're thinking, what might I say? Will that make them be more likely to engage, Less likely to engage? Well, you're predicting under intervention conditions, because whatever you might say is an intervention, and that is the definition of causality. Predicting under intervention conditions and engagement might.

Be a causal factor in therapy. Like the Therapeutic alliance is one of the few things that's been identified as being very important for outcomes. And whether or not you can engage the person meaningfully is likely to be causal.

Absolutely, absolutely. So a good clinician. Look, this is the nature of expertise. We, we have models in our head. I know you interviewed Carl Friston not long ago, and, and it's how the human beings work. They're in any human being. But an expert therapist has models in their head that guide their actions. An interpretation is an action. Anything you say could be an action. Your body, movement, language, all of that, other actions that have consequences. And you are, moment by moment, appraising unconsciously, often, sometimes consciously, the effects of your action. You're making predictions, all of that. So see that? That's good clinical care, good clinical skill. The main question is, how could science help us?

So let's look under the hood a little bit. I've read some of your papers, and I've read a lot of Carl's work, and we talked to Mike Levin recently, the biologist at Tufts, the developmental and synthetic biologist at Tufts, and it's very hard to kind of translate it. You know, I have a little bit of math and physics background, not, not anything like folks like you and Carl and Mike, but you know, enough to kind of follow the reasoning. But the kind of techniques that, that you're using seem to me to be very powerful mathematical and statistical tools for actually laying bare what is causal. And so I'm wondering if we could try to talk about that a little bit without being too mystifying.

Okay, so let's just acknowledge a certain reality. The way we were trained to infer causes is for all practical purposes, impossible. So we, we can't really do ideology, ideological experiments. Okay. We don't have our, you know, if we could, our scientific literature would be full of these sorts of experiments on certain fact. We can't do it. So we may then say, well, since we can't do it, we can't know causes. We then shouldn't have a diagnostic classification system based on causes, because that's kind of impossible.

Right, but, but it's, let's say it's also harder in psychiatry because if, to take your example of like a heart attack, you could study a heart attack, you know, in a mouse. Right. And it's going to translate to people in a lot of ways, but it's much harder to figure out human experience. Right. You can't run those experiments directly.

No, exactly. So, you know, it's like 100% what you just Said so, you know, we have less access not only to human etiological experiments, but the animal models of human psychopathology are way less, you know, approximate than in, in many medical fields. So the field has a problem. We have observational data. Okay, that's pick up any journal about a brain region, about a gene, about a molecule, any. It's observational.

It's kind of like astronomy.

Yeah, astronomy, yeah. And so far. Right. So part of it then is if this is the data that we have, what are we going to do? Because we, if we don't know causes, we're kind of at sea about how to intervene. So the first thing to think about is why our literature that's mainly observational is untrustworthy. Here's why. Because of what's called confounding, which is a common cause. Let's say I study a brain region, a gene, a molecule with a strong statistical association to any mental health outcome. Okay. And I feel, oh, this is great, all of that. Well, if there happened to be a common cause, some other variable that was causal of both, you would get, you would observe a statistical association which is meaningless for intervention, that it's confounded. That's what confounding is like.

They say correlation is not causation and.

It'S because of this. So let's say the field believes that mental disorders are multi causal. Well, what the field believes with that is that any of those causes that weren't a factor in the study may be confounding the factor in the study. Okay. So if you, if you have a brain region that you're studying with an outcome and another brain region causes its activity in the outcome, or a gene causes its activity in the outcome, or a social factor causes its activity in the outcome, you have confounding to safeguard against confounding, you have to. With hypothesis driven statistical methods we use, you're going to have to make a hypothesis about which one of those causes can plausibly affect the factor. I'm hypothesizing in the study and the outcome so I could control for that, but that's really impossible. And so we need to look at other methods. And that's how I began my journey with causal data science methods.

Farah. So we're talking about like using very advanced methodologies that Glenn is going to tell us a little bit about, but we're not really going to understand. I see pictures in my head of Markov chains and Bayes theorem. It's, you know, listeners are going to have to do a little bit of work if they want to understand it more deeply. But that fundamentally, by looking at complex data sets and using some of these techniques that Glenn is using in his lab, you can identify what are actually the causal factors and screen out the noise of the confounding factors. Right. And not all of those things are going to be interventions. Right. One might be, like a genetic factor that unless you use CRISPR or something, you probably can't change it. But another might be an environmental factor or a biological factor you can modify. I kind of want to ask you, Farah, like, because you're very good at sleuthing out causality, you know, you're highly intuitive, like, you do it quite well without, I think, using any algorithms that we've written down. So I'm just curious where you are with this now.

My patients call me witchy. I think we're, you know, onto something there.

Yeah.

And mostly I think what I'm trying to do, even though I'm not technically an analyst, Right. But I've studied a lot of analytic theory. I wouldn't call myself a psychopharmacologist, but I've studied a lot of psychopharm and I, I use it every day. But I'm just trying to understand as much of their experience as I possibly can. And I. What I love about this field is that even my own personal experience as a child, as a mother, whatever else I feel like, informs a lot of that and things that I had really never thought about. So kind of looking at how to treat patients through a certain lens has. Has illuminated a lot of parts of my personal life, too. Sorry, go ahead.

Yeah. And, you know, as a good clinician, one, you have a theory in your mind. Could be, you know, from your background growing up, integrated with psychoanalytic theory or whatever models you're using, but you're testing your theory all the time. And the evidence of whether you're onto something or not is accurate predictions under intervention conditions. So if, if you're going to be in the position of learning, then, you know, when something happens that you don't expect, you're going to wonder why and think about what you did. And so then do it a little differently. And so a good clinician is continually upd their model of the world based on their experience with, with, with their patients. And, and, and that's how it works. What? What? Without one. I think that's like the foundation of everything is good clinical care, and our science serves good clinical care. So what would add value to you from the Science. When was the last time you read a scientific article that changed what you did in practice? There are thousands and thousands of them. That is the bar about the value of a science.

Yeah, I'm more worried about the sort of the opposite is these little articles come out about like, this medication was used in 100 people. And then someone tries prescribing that medication, you know, maybe it will help, but it's not really clear whether it translates at all. So. Okay, can you dip down just for a couple of minutes into the world of math and talk about how you actually study systems? I, I was very impressed by the study of, of police trainees who gets PTSD as an example. But.

So I, yeah, and I want to, I, I actually, there's a huge, you know, it's very technical when you really get into causal data science, but there's some basic principles that help just about anyone gathered and I'll try and explain that to you as best I can. So first is what makes us so confident about randomized experiments? What makes us so confident? You know, it's that confounds are equally distributed between the groups, so we don't have to worry about them. They're randomly distributed. Okay, so then if we manipulate a factor we know the dependency of the outcome on that factor can be inferred without worrying about confounding, which would confound bias or estimates of effect.

That, that's technically cancels out the noise kind of.

Yeah. So if we are able to estimate the effect without worrying about confounding, it gives us similar knowledge about experiments. And that's what these methods do through what's called conditional independency. We usually think of conditional dependency. What conditions makes one variable depend on another variable? Conditional independency is identify a set of factors given an outcome that would render everything else independent of that outcome. Okay. That's what we do when we statistically control variables.

It's kind of like dissecting out what you want, right?

Well, that it's exactly what it is, but that, that's what, that's what causal inferences. Okay, that's what causal inference is. We come up with methods like experiments that allow us to exclude everything else so we can really estimate the effect of one thing. So there are strong methods to infer statistical conditional independency. Very, very strong methods. And basically you would mention the Markov boundary. In essence, Markov boundary is the direct causes, the direct effects, and the direct causes of the direct effects of any outcome of interest. Think of any outcome you want direct causes, direct effect. It's mathematically proven that if you know the Markov boundary, you don't. Everything else is independent of the outcome. Okay. So without getting into a lot of detail about how these methods do it and the assumptions, there are certain assumptions are testable, all of that if you, these algorithms allow you to know the Markov boundary and what you could do once you know that it includes the direct causes, so you could work your way up. So what are the causes of a variable in the markup boundary that's like directly causal? Then you could get it indirect causes, working your way up so you could build, you could understand the causal structure. And then if you've ruled out confounding and there's methods where you could detect signatures of latent confounding from unmeasured variables. So if you've ruled out confounding from anything you've measured and anything you haven't, then your estimate of effect is unbiased. And you know, I'm sure that maybe you know that that's the essence of what it is. Now again, if people are skeptical of this, everyone, we should be skeptical. We have to compare it to what we're using now. What we're using now are these studies all over our journals which are observational, which are hypothesis driven, conventional statistical methods. So and there we have not ruled out confounding. Far, far, far from it.

Yeah. For people who are interested, we'll put some links in the show notes that talk about the Markov boundary. It's really, really interesting to get into the details of it, the way these causal chains are set up and how you can look at all the different, kind of looks like a network with lots of different nodes and you identify which of those nodes are the key players in the causal sequences. And it relates to a lot of cool stuff about like agent based modeling. And there's AIs based on this stuff as well.

And then here's the missing link too. It's testing this in reality. So this gives you intervention targets you could be way more confident about. So now you can conduct clinical trials on those. I mean that's in fact what we're doing with my Champ center grant. We're building causal models, making decision support tools on different models which tell you how to intervene given clinical data and then doing clinical trials on that. So that's how you test it in the world to see. And maybe things don't work out the way your model told you. Well, you can go back and try and figure out what was wrong. Maybe measurement was wrong, maybe your data set was wrong, all of that. But now you're in the world of science, actually, and. And, you know, that's how it should be. So we if in our literature now, what factors from the study for. From our correlations are we most confident about to take into the world and test with intervention?

You know, Can I ask you about the police trainee study?

Yeah.

Quickly. So, okay, correct me if I'm. If I'm mangling this really badly, but you, you know, looked at a lot of different factors and used data science, computational modeling to look at what causes PTSD among police trainees. And there were two genetic risk factors related to stress genes. There was the intensity of their startle response. I believe there was the presence of peritraumatic dissociation, which is also one of the risk factors for the development of PTSD in observational studies. And there was one related to, I think, training parameters. Was that the fifth factor?

Yeah, that's part of it. Yeah.

Okay, so what I wanted to ask you about, I've been dying to ask you about this is the trauma wasn't one of the causal factors.

Yep. So here's a way of thinking of it. First of all, the model we published, you know, that rules out all confounding. Okay. So you might have said, well, the trauma. But you may have also said there are many other variables that we're not related to. Many of them have been, you know, published their literature, but again, they were confounded by other variables. The other thing is in causal chains that are more proximal or distal, we only looked. I think it was two steps up. So you have ptsd, you have the direct causes. That's the unmediated causes, and then you have the causes of those causes. Right. So we only looked two levels up. So there were many variables also on more distal causal chains, which have very. Have small effects. I mean, basically through Markovian processes, if you know, the more proximal one, then it renders everything else conditionally independent of the outcome, that sort of thing. So it's part of that. So, you know, trauma may have been more distal, but its effect is rendered. Its effect is related to what it sets up more proximally.

That makes sense. The other things depend on it. It makes sense clinically because, you know, a lot of people experience trauma, but a very small percentage of people actually get ptsd.

Yeah. And so if you. Let's say you knew trauma affected certain patterns of emotional dysregulation. So if you intervene on the emotional dysregulation, given the trauma, you know, that's what's needed. So that's a direct cause, all of that. So part of it is this allows you to take a enormously complex world and begin to make sense of. Really, it's about understanding the process that generated the data that that's what this is. It's about understanding whether your model corresponds to reality because reality is causal. It's how one thing affects another. So we're trying to approximate reality with this.

Right. And the DSM is not a great model. Good. So we're coming to the end. Farah, any thoughts?

It's not even a poor model.

It's a non model.

Well, it's what they call a polythetic model, which is a model composed of lots of overlapping classifications. So it's extremely fuzzy and idiosyncratic.

I think this has been terrific. Thank you so much for being here today. And how can people learn more about you and what you're doing?

You know, I'm. I'm at NYU Langone Medical Center. You know, people could look for the CHAMP center or the Trauma Systems Therapy Training Center. Perhaps we could put a link in. In this for that. But I. I love talking about this, perhaps, as you can could tell, and I love to discuss it with people who might reach out.

Yeah, I didn't get to ask you about slime molds, but maybe another time.

Okay. That's for sure.

Thanks for joining us.

Okay. Thank you for inviting me. Bye.

Bye. Remember the doorknob comments. Podcast is not medical advice. If you may be in need of professional assistance, please seek consultation without delay.