Reclassifying Psychopathology

Dimensions instead of diagnoses  • Day-to-day variance • Potential and promise 

The Diagnostic and Statistical Manual of Mental Disorders (DSM) has faced many criticisms over the years. Mental illnesses don’t neatly fit into its categories, boundaries for diagnoses can seem arbitrary, and reliability from clinician to clinician, or even for one patient over time, is low.  

“These diagnoses don’t do a good job at capturing people’s symptoms,” said Miri Forbes, an associate professor at Macquarie University, in an interview with the Observer. “They don’t map onto the way people experience symptoms in a clean way.” 

The current system is both too specific and too general, said Aidan Wright, a professor at the University of Michigan. On the one hand, certain diagnoses, like depression and anxiety, co-occur more often than would be expected based on chance, suggesting that the categories are not as distinct as the DSM assumes. On the other hand, because of the DSM’s checklist nature (people must meet some but not necessarily all symptoms to receive a diagnosis), two people with different symptoms might end up with the same diagnosis. Their underlying issues and thus appropriate treatment plan might be different.  

“One diagnosis doesn’t always capture the full range of psychopathology,” said Wright. “Maybe we should stop defining it in these neat categories that don’t seem to hold up.”  

Both the variation between individuals within a diagnostic category and the co-occurrence of multiple diagnoses make identifying treatments difficult. “Those diagnoses haven’t cracked open the answer to effective treatments,” said Forbes.  

Studies are often focused on trying to get a clean look at one particular issue, such as anxiety. To do so, people with more than one issue—anxiety and depression, for example—are excluded from the study. Any resulting treatment won’t work in the clinic because the included participants “are unicorns,” as APS Fellow Koraly Pérez-Edgar, a professor at The Pennsylvania State University, said in an interview with the Observer. “The people you’re dealing with in the clinic aren’t just pure anxiety and nothing else.”  

But getting treatments covered by insurance, or even recruiting participants to study, requires labels and categories. “At some point, someone’s going to ask, ‘Do you meet criteria?’” said Pérez-Edgar. “The only way to create criteria is to have a formula of sorts. The question is, what does that formula look like?” Recently, more researchers have asked that question and are trying to devise ways to better classify mental illnesses in the hopes of fixing some of these issues.  

“There’s just been this groundswell from different corners of the field saying, ‘Hey, we need to be doing things differently. The old way isn’t moving us forward,’” said Wright.  

Dimensions instead of diagnoses 

To address some of these issues, in 2017 a group including Wright and Forbes created the Hierarchical Taxonomy of Psychopathology (HiTOP). The idea was to replace traditional diagnoses with dimensions and to view mental illnesses on a spectrum rather than as clear-cut categories. Broader dimensions (such as internalizing) are at the base of the hierarchy, related issues (such as major depressive disorder and generalized anxiety disorder) are clumped together in the middle, and specific symptoms make up the tip.  

But HiTOP was built using the DSM constructs. “The more time that I spent in that area, the more that I thought, ‘Well, it’s pretty ironic that we’re building this framework on the back of the DSM diagnoses with the rationale that we should stop using DSM diagnoses,’” said Forbes.  

In order to truly get away from the DSM constructs and start from scratch, Forbes used individual symptoms as the building blocks. Her team created a spreadsheet with each diagnostic criterion from every disorder as a separate item, removed any duplicates, and ended up with a list of over 600 unique symptoms. For every symptom, they created a self-report measure. The measures were randomized and scrambled into surveys of various lengths, with almost 15,000 participants and around 7,500 responses to every item. The items were then reorganized according to how they covaried.  

The resulting model, published in Clinical Psychological Science, is strikingly similar to HiTOP (Forbes, Baillie, et al., 2024). “When I break it down into individual symptoms and build it back up, all those same domains are still there,” said Forbes. “I think that it speaks to the robustness of those dimensions as organizing constructs to help us understand people’s symptom presentations.” 

But Forbes thinks more large-scale studies need to be done and new perspectives that weren’t available when HiTOP was created need to be added. “The way that I think about it is not that we need to move away from HiTOP, but that we need to revise it,” she said. 

For example, the model needs to be validated across ages, cultures, and identities. In another 2024 Clinical Psychological Science study, Forbes examined the structure of psychopathology in youth (Forbes, Watts, et al., 2024). By combining data from six samples, her team created a model based on over 18,000 11- to 17-year-olds. Most of the dimensions were similar to HiTOP, but Forbes also found a dimension covering uncontrollable worry, obsessions and compulsions, and generalized anxiety that is not its own domain in the adult version. Additionally, there was nothing specific to psychosis (such as a thought-disorder category). These findings may be a result of symptoms presenting differently in children and teenagers, or they may be due to the types of questions asked of young participants; regardless, more work needs to be done.  

Forbes wants to be sure researchers are thinking about people through many different lenses and across the age span, as well as considering people with different lived experiences. Next, she is hoping to use her adult sample that was organized according to symptoms rather than diagnoses to compare people across cultural and linguistic backgrounds as well as across gender identities and sexual orientations.  

“That will actually strengthen the model because we can understand what’s stable, what’s robust across all these different ways of looking at people’s experiences,” she said. “And what’s different? Where do we need to have some flexibility or some different ways of thinking about the dimensions that are most useful for different groups?” 

Day-to-day variance 

Another problem with the existing models is they assume traits are stable. A person labeled as anxious, for example, is assumed to be anxious all day, every day, in every context. But in reality, these traits fluctuate from day to day and in different situations. Studying these fluctuations can provide clues to identify needs and find treatments, said Pérez-Edgar. But by only asking questions retrospectively, providing surveys every few months, and averaging the resulting data, “those clues have been overlooked.” 

In an upcoming article in Current Directions in Psychological Science, Pérez-Edgar outlined how changing the time scale examined—making it shorter and taking more frequent measurements—can reveal how even traits assumed to be stable, like temperament, change. For example, delta–beta coupling is the correlation between the delta and beta oscillations of an electroencephalogram (EEG) signal. On average, positive coupling is correlated with behavioral inhibition, a temperament characterized by a fear of novelty and a strong predictor of anxiety. However, when you look moment by moment within individuals, individuals occasionally show negative coupling and the relationship to behavioral inhibition is not as strong.  

“It’s the same measure and it’s the same biological relationship, but depending on how you manipulate the data, you’re going to get different answers,” said Pérez-Edgar. “And they give you different predictions.” 

Individual variance is also lost by averaging across people. Wright is trying to create a more personalized approach to psychopathology, where a model can be created for an individual that takes into account these fluctuations in symptoms and day-to-day variance. It’s not only about a patient’s specific symptoms, but under what conditions they manifest, he said. 

Trying to parse out those conditions requires intensive data collection from each individual. A lot of the longitudinal work with such collection methods has been diagnosis-specific, comparing individuals with depression to controls, for example. “While this is valuable and it gives us good information on some level, on another level, it goes back to this fundamental limitation of selecting people on these diagnoses that have sort of limited validity,” said Wright. 

In a 2025 article in Clinical Psychological Science, Wright and colleagues attempted to bring together both the research on the basic dimensions of psychopathology and ways to sample people intensively and repeatedly over time (Wright et al., 2025). The resulting inventory is suitable for studying differences both between individuals and within individuals, over multiple HiTOP domains, and intensively in real time.  

“A criticism of HiTOP is that it’s often focused on between-person differences at one time point or collapsing over time points,” said Wright. “This is an effort to really respect that kind of research but also take it into daily life and look at it as a dynamic process that unfolds over time.” 

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

Asking these questions now is possible because of both the ubiquity of the smartphone and advances in computational approaches. Traditionally, clinicians ask for retrospective analysis: What led to a behavior or feeling? What happened before and after? But memory is foggy and flawed and many things may be missed because they didn’t seem important at the time.  

“Something is changing on a daily basis in between [sessions],” said Pérez-Edgar. Having a smartphone available at all times to record in-the-moment data helps capture these missing pieces. “The technology has helped. It’s given us the data,” she said. 

But all those data require processing. For even a small number of participants, creating individual models and then looking for similarities among them could take weeks, months, or even years, said Wright.  

“Computational approaches have emerged that have allowed us to be able to look for those types of commonalities,” he said. “A lot of this stuff is becoming much easier to deal with than it was in the old days.”  

Potential and promise 

There are still limitations to this new strategy. The constantly pinging smartphone can be burdensome; some participants drop out to avoid answering so many questions. Truly moving away from the DSM system would come with a huge cost and is unlikely unless these new models are justifiably better.  

“It’s a really exciting idea with a lot of potential and promise to help researchers and clinicians and clients,” said Forbes. But it’s not yet ready for prime time. “We just have so much still to learn.” There needs to be large-scale testing in the clinic asking if and how it helps with treatment. Do patients think it does a better job capturing their experiences? Do clinicians feel it gives them a better understanding? 

Ultimately this new approach holds a lot of promise. “It stops us from having diagnostic blinders on where you’re trying to fit symptoms into a box. Instead, it encourages people to think very broadly, to think across all these different dimensions and areas of symptoms and functioning,” said Forbes. It also helps reduce stigma. Putting people in boxes makes them feel fundamentally different from people not in that box. But here, there are no boxes. “Everyone sits somewhere on all of those dimensions,” she said. 

It can also help people who are subclinically anxious or depressed and enable intervention with young children before diagnoses are given. “These approaches, I think, have an opportunity to pull in more people, to pull in more ways, to apply our knowledge prior to someone coming in with real impairment,” said Pérez-Edgar. “A lot of times by the time we get them into the clinic they’ve been suffering for a long time.” 

“My overriding sentiment is of optimism,” Forbes said. She added that she thinks this approach will help researchers find treatments that work for specific symptoms in specific people and contexts. “I’m really excited about the potential and promise of this direction. And impatient for us to get there.” 

Back to top

Feedback on this article? Email apsobserver@psychologicalscience.org or login to comment.