A New Approach to Understanding Psychopathology: Insights from the HiTOP Model

Are traditional mental health diagnoses missing the bigger picture? How can a new model help us better understand and treat mental disorders? What does recent research say about how we categorize psychopathology in youth? 

In this episode, APS’s Özge Gürcanlı Fischer Baum interviews Miri Forbes of Macquarie University. Together, they address how traditional models like the Diagnostic and Statistical Manual of Mental Disorders (DSM) categorize mental health disorders and explore Forbes’ recent study published in Clinical Psychological Science highlighting the more nuanced and dimensional approach that the emerging Hierarchical Taxonomy of Psychopathology (HiTOP) offers. The strengths and shortcomings of these systems, challenges of diagnosis, and evolving landscape of mental health research are discussed. 

Send us your thoughts and questions at  [email protected] 

Unedited Transcript

[00:00:00.270] – APS’s Özge Gürcanlı Fischer Baum

How do mental health professionals identify psychopathological categories? What are the diagnostic criteria they rely on. Are there alternative approaches to consider? I am Özge Gürcanlı Fischer Baum with the Association for Psychological Science. In this episode of Under the Cortex, APS welcomes Miri Forbes from Macquarie University, who recently published an article in Clinical Psychological Science, Exploring the Symptom-Level Structure of Psychopathology in Youth. Together, we discuss the strengths and limitations of the traditional models and compare them with the hierarchical taxonomy of Psychopathology Model, which possibly offers a more dimensional and integrative approach to understanding mental health. Miri, welcome to Under the Cortex.

[00:00:57.170] – Miri Forbes

Thank you so much. Thanks for having me.

[00:01:00.130] – APS’s Özge Gürcanlı Fischer Baum

Miri, let’s start with our first question, like we always do. What type of psychologist are you?

[00:01:05.870] – Miri Forbes

I would say I’m a quantitative psychologist, so I really rely on using statistics to find patterns in data. It’s the bulk of the work that I do.

[00:01:16.110] – APS’s Özge Gürcanlı Fischer Baum

Let’s start with the basics. The DSM has had some criticisms over the years. Why don’t people like how the DSM works?

[00:01:26.130] – Miri Forbes

I think there’s probably a lot of reasons for different people, depending on why they use the DSM, but I can definitely speak to a couple that come up a lot in the academic literature. The kinds of things that have been spoken about for quite a while now are things like comorbidity, right? That the disorders tend to co-occur at rates much higher than chance, which suggests they’re not really distinct categories that exist in nature, like truly organize people into separate groups. There’s things like symptom overlap. So we tend to see the same symptoms coming up again and again in different disorders. They remixing the same information into lots of different categories. An example would be insomnia is a symptom of 22 diagnosis, difficulty concentrating of 17, depressed mood of 15, that idea that we see these same symptoms coming up. And that really blows the boundaries in a different way inflating that surface, a similarity of different disorders, not knowing where different symptoms belong, so to speak, in the classification system. The third one might be heterogeneity in the diagnosis. So you might have two people with the same diagnostic label that have even no symptoms in common, even though they have this same diagnosis.

[00:02:39.040] – Miri Forbes

And that can happen because of the checklist approach in the DSM. So major depressive disorder, for example, has nine diagnostic criteria. You have to have five to meet full diagnostic criteria. But because some of them are things like insomnia or hypersomnia, people can meet five. Another person can meet another five, and they can have no symptoms in common in their profile. And then another key one is the low reliability of the diagnosis. So for example, different mental health professionals tend to assign different diagnosis to the same person or to the same case. And because of the categorical threshold that once you reach a certain number of symptoms, you have a diagnosis. And if you drop below that threshold, you don’t have a diagnosis, you can also see unreliability in those diagnosis over time. Those are just a sampling of some of the things that often come up in literature.

[00:03:31.180] – APS’s Özge Gürcanlı Fischer Baum

Yeah, thank you. There is also this thing that we call the HiTOP framework. Could you please explain what that is before I move on to my questions about that?

[00:03:44.930] – Miri Forbes

Yeah, sure. Hitop is short for the Hierarchical Taxonomy of Psychopathology. And basically, instead of taking a committee consensus-based approach to deriving and organizing diagnosis, it’s based on following the patterns in data. So it’s an empirically-based classification system. And so the data supports the idea that most domains of psychopathology are dimensional, not categorical. And so it has these statistically-derived dimensions that organize people’s experiences of psychopathology, their emotions, thoughts, feelings, behaviors, physical symptoms, and it organizes them into a hierarchy. So there’s maybe, I would say, six core spectra in the middle that are a lot like the big five of personality, where Everyone sits somewhere on every dimension. And then you can go higher in the hierarchy to organize psychopathology into broader constructs, and then going narrower as we go down the hierarchy, we can drill all the way down to individual signs and symptoms. Basically, it organizes ways of explaining how people experience mental illness.

[00:04:53.740] – APS’s Özge Gürcanlı Fischer Baum

Miri, can we say that this approach is an improvement from the current usage of the DSM?

[00:05:02.660] – Miri Forbes

I think that it does overcome a lot of those limitations we talked about. For example, the early versions of the dimensions were based on modeling patterns of comorbidity, so using it as a signal rather than a hassle that gets in the way of the way that the classification system works. Things like the symptom overlap, you would only have symptoms repeating in the structure where that’s statistically indicated, where we see that in the way that people experience symptoms rather than as a result of insomnia being a common symptom when people are experiencing psychopathology. So it’s popping up in a lot of different diagnosis in DSM. And then in terms of heterogeneity, like we were just talking about, that’s covered because everyone is somewhere on every dimension. So we’re explaining people using the same set of constructs. And then reliability, that’s just an inherent advantage that dimensions have over categories, that it’s a more reliable way of measuring things, particularly over time.

[00:05:59.740] – APS’s Özge Gürcanlı Fischer Baum

Have you You’ve already talked about the limitations of the DSM. Are there any limitations for the high TAP framework?

[00:06:06.760] – Miri Forbes

Yeah, definitely. I could speak for a long time about the limitations, but the two that I think are most relevant to the study that we’re talking about today are that it was largely built on DSM diagnosis in the first instance. So really the foundational or a lot of the foundational studies were based on modeling those patterns of comorbidity between DSM disorders. And that’s ironic because the whole premise of high top is that we need to move beyond those constructs as our units of analysis or as the constructs that frame our research. And the other is that the lower levels of the framework, the detailed levels, are really underdeveloped. So there are detailed constructs in the official HiTOP model, the current model that we’re working with. But all of them are subscales from existing self-report measures. So they’re not really from systematic work on the structure of psychopathology with that aim in mind. Those were two key limitations that I wanted to work to address in this study.

[00:07:06.210] – APS’s Özge Gürcanlı Fischer Baum

What were you hoping to learn from your design of your study?

[00:07:11.240] – Miri Forbes

I guess I wanted to know if we got rid of the diagnosis, if we just broke all of that down and boiled it down to the symptoms that are in the DSM 5, which are based on definitely decades, arguably centuries of clinical observation, we’ve got this really rich description of the ways that people experience psychopathology. And so what I wanted to do is to say, Well, instead of sitting down and rationally deriving how those symptoms go together, if we follow the patterns in the data in how people report experiences processing those symptoms, what would a model like Hitop look like built from those individual building blocks? That was really the key thing that I was excited to answer.

[00:07:54.830] – APS’s Özge Gürcanlı Fischer Baum

Let’s talk about your surveys a little bit. I saw that You used four types of surveys, and it is not common to offer different length of surveys to your participants. It was a really interesting design. Can you talk a little bit about that?

[00:08:13.970] – Miri Forbes

Yeah, great question. I was really pleased with how the design of the study worked out. Because we started with so many items, I think we had 640 items going into the primary data collection. We knew that not everyone would to or be able to complete every item in that survey battery. And so we designed, like you’ve mentioned, different length surveys. So what we did was we randomized those items, split them into 12 blocks, and then we let people choose. Do you want to answer one block, three blocks, six blocks, or all 12 blocks at the start of the survey they could opt in? And then what we did was we randomized the order of the blocks so people would get blocks selected at random, and the order that the blocks were presented in was selected at random. And the items in each block were re-randomized for each participant. So like as randomized as you can get, which allowed us to build this massively missing at random design so that if people didn’t complete all of the items, some of them because they selected a shorter survey length from the outset and some of them because they gave up because there were too many questions, then we can still use their data because the questions that they answered are not related to which survey they selected or whether they dropped out.

[00:09:29.780] – Miri Forbes

So the missing data is missing at random. And in the end, we ended up having, I think it was 45 % of people selected to complete the full length survey, and a big chunk of those people completed every single item that’s in there. So In total, we had about 15,000 people in the analytic sample. And I think that we had about seven and a half thousand complete responses for every item that was in the survey. So it worked pretty well for what we needed in our design.

[00:10:00.270] – APS’s Özge Gürcanlı Fischer Baum

I just want to highlight that this is what I love about your study. Your sample size is fantastic, and I think we can learn a lot from your data. Let’s go back to your surveys a little. How did you pick what items to include in each survey?

[00:10:18.570] – Miri Forbes

I’ve got a photo of my older daughter when we started this project. She was about six months old, and she is about to turn seven. This has been a very long-running project that we’ve been working on. And basically we broke down the DSM into its individual symptoms. So first we literally turned it into a spreadsheet of pulling out from each diagnostic criterion the individual items that are in those criteria. And then we boiled down that list of all of the symptoms into what are the unique items in that list, which is much shorter because of that repetition between diagnosis we talked about. And then we turned every possible symptom that we could into a first person retrospective report, an item on that. And that involved working with a lot of experts in psychiatrics and in the content domains that we were working in. But essentially the criteria for ruling items out. So the criteria for ruling items in is it’s in the DSM, right? So we tried to measure everything that is described in the diagnostic criterion, but we didn’t measure items that were only relevant for children. So there was a is often truant from school is an example of an item that we would have excluded.

[00:11:34.100] – Miri Forbes

Things that required standardized testing, like IQ testing or polysynography, that those kinds of needing to be able to report tests that people might not have had. And also things that could only be observed by others. An example is cyanosis during sleep, like getting a bit of a blue color to your skin. But we wanted to include as much as possible. We really tried to have it as a high bar for not measuring an item. And Essentially, it’s the whole DSM, give or take.

[00:12:04.200] – APS’s Özge Gürcanlı Fischer Baum

Yeah, that’s right. Let’s talk about the key constructs that came out of this big sample size. What were they?

[00:12:13.520] – Miri Forbes

I was really struck by how similar the higher levels of the model looked to what is already in high top. That wasn’t necessarily a given because we started with totally different building blocks from what high top is derived from at the moment. So we saw familiar constructs like internalizing, which is a propensity for negative affect or distress, externalizing, which is more like behaviorally disinhibited things and antagonism, thought disorder, which is schizophrenia spectrum psychosis kinds of symptoms, somataform, which is those physical symptoms that people experience. They were really key familiar domains from high top. We also saw some things that were different. So we saw mania and detachment went together in mania and low detachment, excuse me, went together into something that looked a bit like extraversion in the literature. Harmful substance use is part of externalizing in high top, but was separate here. And similarly, eating pathology is part of internalizing in high top, but was separate here. And then a really key difference at this broad level of the organizing constructs was that there was a whole new branch of neurodevelopmental and cognitive difficulties, which isn’t something that’s represented in high top at the moment.

[00:13:28.990] – Miri Forbes

And then So I guess if you think about what are the key constructs that you found, we really wanted to flesh out the full structure from individual symptoms all the way up to if there was a single overarching factor idea. And so there’s a lot more detail in the model that we have a lot more subfactors, a lot of clusters, like highly, highly correlated syndromes thing, how symptoms hang together. And so there’s a lot of differences in there. But I guess that that that similarity really struck me as noteworthy in these results compared to high top.

[00:14:02.980] – APS’s Özge Gürcanlı Fischer Baum

In your paper, you note that some of the greatest differences between your analysis and the DSM-defined cost-reacts were in the diagnosis with the deepest historical roots, like schizophrenia. Is this a sign that DSM is just widely out of date?

[00:14:25.410] – Miri Forbes

Part of me wants to agree with that, but I think maybe not necessarily out of date, so much as this would suggest it’s not strongly empirically supported, the way that the symptoms are organized in the DSM isn’t the way that we see the symptoms are covarying in how people report experiencing them, I guess, to be strictly technical and fair in terms of what these results say. But I think that what we’re really seeing is the consequence of the fact that there’s no standardization in how DSM disorders have been derived. So So historically, there’s disorders that are based on a single, like an individual patient or a case study of patients or drawing on other fields of research, like education, research and psychology that was looking at kids struggling in class who don’t have cognitive impairment. That’s where ADHD came from. We’ve got the roles of psychodynamic theory and lobbyists, right? Like lobbying groups. And I guess that there’s a lot of different paths for things to get into the DSM, and it’s not standard standardized how they’re defined or what criteria they have to meet to be included historically. Now there’s a really clear system on how changes are made from where we’re at, but everything is inherited from the past versions of the DSM.

[00:15:46.160] – Miri Forbes

And so I think that what we’re seeing is all of those different approaches haven’t converged on a reliable set of constructs that really hang together in the data that we’re observing. So I guess I would also highlight One of the key things that really struck me is that there was not a single DSM defined disorder that hung together as its own syndrome in the results. There were a couple of examples of really tightly bound disorders, like insomnia disorder, where all of the symptoms were in a tightly bound syndrome, but they were always mixed with symptoms of other disorders in those couple of examples. And then when we did see the most homogeneous, the most empirically coherent coherent versions of DSM disorders we tended to see sat under one of those broad spectra we were talking about before. So for example, things like intermittent explosive disorder, conduct disorder, those kinds of behavioral dysregulation disorders, All of those symptoms sat under externalizing, but they didn’t bind together into one of those really tightly bound syndromes. Similarly, we saw symptoms of sexual dysfunctions, some of the eating disorders, some of the anxiety disorders. We tended to see that they would hang together under a single spectrum.

[00:17:04.170] – Miri Forbes

So they were pretty empirically coherent. And then, like you said, there were examples of disorders that were essentially the opposite, that the symptoms did not bind together based on those patterns of co-variation in how people reported experiencing the symptoms in our sample. And so the symptoms split apart. They would be under five different spectra. And so that’s suggesting that there’s a lot of heterogeneity in the symptom sets being used for those diagnosis. The ways that people experience those symptoms just varies a lot. That’s what we’re seeing in the data.

[00:17:37.850] – APS’s Özge Gürcanlı Fischer Baum

Yeah. Miri, let me ask you a short, yet loaded question. How do we move forward from here as a field?

[00:17:48.400] – Miri Forbes

Yeah, that is a loaded one. There isn’t one answer. There’s so many different directions we can take this work, and so many questions that we still have to answer. So it’s something that I’ve been thinking a lot about recently is the cost of making a large scale change to classification would be enormous in terms of the infrastructure costs, the retraining costs to say, All right, everyone, the DSM is done. We’re going to use this new framework. Everyone has to learn it. All the systems have to be like that idea of making that change at this stage is a bit silly to suggest that we’re going to tuck the DSM out and move on. And so I think that if we want to really capitalize on these advantages that we can get in this system, we need to show not just that this system does as well at the outcomes that we care about, but that it does better, that it does better enough to justify making those changes. And so that’s an area that I’m really excited about working in personally. I think that moving towards testing these constructs, these empirically derived dimensions and syndromes, for example, and comparing them head to head with DSM diagnosis, saying which one of these is doing a better job at capturing, let’s say, signals in neuroscience?

[00:19:08.970] – Miri Forbes

Which ones of these are aligning more reliably with specific neural substrates? So I And I think that’s the direction that I really want to head in, is starting to test out which of these constructs is doing better and in what situation. But I think that more broadly than that, it’s just about we’re all on the same team, us, everyone everyone working on classification, everyone working on reconceptualizing psychopathology. There’s a lot of different approaches. And I think it’s about speaking to each other, having communication about what we’re finding, what’s working, what’s not working, thinking about how they fit together and not losing sight of the bridges between us. We’re at the stage now where we know that we need a change, but we need to be in it together. That’s my very optimistic take, is that It’s about team science. It’s about failing early and communicating that to our colleagues. It’s about shared resources, open science and data, those sorts of things, I think, will help us move in this direction of tackling what is a really big problem for the field.

[00:20:18.410] – APS’s Özge Gürcanlı Fischer Baum

Yeah, I agree. It is definitely about shared resources and team science and having a numeric approach to all these diagnosis. Miri, this was It’s my pleasure. I personally learned so much, and we are hoping to keep in touch with you and with your research.

[00:20:38.370] – Miri Forbes

Yeah, thank you so much. Thanks again for having me.

[00:20:43.430] – APS’s Özge Gürcanlı Fischer Baum

This is Özge Gürcanlı Fischer Baum with APS, and I have been speaking to Miri Forbes from  Macquarie University.