In this Papers Podcast, Dr. Tom Frazier discusses his JCPP Advances paper ‘Categorical versus dimensional structure of autism spectrum disorder: A multi-method investigation’ (https://doi.org/10.1002/jcv2.12142). Tom is the first author of the paper.
There is an overview of the paper, methodology, key findings, and implications for practice.
Discussion points include:
- What is meant by ‘categorical’ and ‘dimensional’ in the context of neuropsychiatric diagnoses?
- The evidence that suggests that autism is ‘categorical’ instead of ‘dimensional’.
- The implications of this study for researchers and those searching for major drivers of the autism spectrum disorder (ASD) phenotype.
- The implication of findings for clinicians and other child and adolescent professionals.
- Whether these findings could change the way children and young people are assessed for autism.
- The implications of the results on the way autism is classified.
In this series, we speak to authors of papers published in one of ACAMH’s three journals. These are The Journal of Child Psychology and Psychiatry (JCPP); The Child and Adolescent Mental Health (CAMH) journal; and JCPP Advances.
Subscribe to ACAMH mental health podcasts on your preferred streaming platform. Just search for ACAMH on; SoundCloud, Spotify, CastBox, Deezer, Google Podcasts, Podcastaddict, JioSaavn, Listen notes, Radio Public, and Radio.com (not available in the EU). Plus we are on Apple Podcasts visit the link or click on the icon, or scan the QR code.
Dr. Frazier is a licensed clinical psychologist who received his Ph.D. from Case Western Reserve University in 2004. He joined Cleveland Clinic in 2006 and from 2013-2017 was the director of the Cleveland Clinic Center for Autism and the Lerner School providing behavioral intervention to children and young adults. In 2017, he was hired as the Chief Science Officer at Autism Speaks and is currently a member of the Autism Speaks national board and chair of the medical and science advisory committee. He is also a Professor of Psychology at John Carroll University and a Research Professor in Pediatrics and Psychiatry at SUNY-Upstate.
Over the last decade, Dr. Frazier has maintained active clinical research programs focused on the evaluation and treatment of autism and other related neurodevelopment conditions. He has published more than 150 scientific papers and his recent work has focused on measuring key neurobehavioral processes relevant to early identification and monitoring response to intervention.
[00:00:01.709] Jo Carlowe: Hello, welcome to the Papers Podcast series for the Association for Child and Adolescent Mental Health, or ACAMH for short. I’m Jo Carlowe, a Freelance Journalist with a specialism in psychology. In this series, we speak to authors of papers published in one of ACAMH’s three journals. These are the Journal of Child Psychology and Psychiatry, commonly known as JCPP, the Child and Adolescent Mental Health, known as CAMH, and JCPP Advances.
Today, I’m interviewing Professor Tom Frazier of the Department of Psychology, John Carroll University in Ohio and Research Professor in the Departments of Pediatrics and Psychiatry at SUNY-Upstate Medical University in the USA. Tom is the first author of the paper “Categorial Versus Dimensional Structure of Autism Spectrum Disorder: A Multi-method Investigation,” recently published in JCCP Advances. This paper will be the focus of today’s podcast.
If you’re a fan of our Papers Podcast series, please subscribe on your preferred streaming platform, let us know how we did, with a rating or review, and do share with friends and colleagues.
Tom, thank you for joining me, welcome. Can you start with an introduction about who you are and what you do?
[00:01:16.780] Dr. Tom Frazier: Sure. Well, you did a great job introducing me, but, yeah, I’m a Clinical Psychologist. I’ve been interested in autism, really, for almost 20 years now and I have been, sort of, in the academic world, but also in the foundation world, I was the Chief Science Officer for Autism Speaks for a while and I continue to have a relationship with Autism Speaks on their National Board. So I’ve predominantly been in academic research, but also, sort of, trying to understand how do we build up the community and services. And my wife actually is a board-certified Behaviour Analyst and she runs a child mental health company that focuses on autism as well. So it’s, kind of, all in the family, but, you know – and I have a 19-year-old son with autism, so it’s been a – an adult lifelong passion for me and for my wife as well. But yeah, no, thank you for having me.
[00:02:03.649] Jo Carlowe: Yeah, you’re most welcome. So, today, we’re going to look at your paper, “Categorial Versus Dimensional Structure of Autism Spectrum Disorder: A Multi-method Investigation,” which was recently published in JCPP Advances. Tom, before we go into the detail, can you explain in lay terms what is meant by “categorical” and “dimensional” in the context of neuropsychiatric diagnoses?
[00:02:27.600] Dr. Tom Frazier: Sure. First, I get that this sounds very esoteric to a lot of people, but it does actually have practical implications and that’s why I did the study. As a Researcher, I’m not really interested mostly in theoretical stuff, I’m more interested in practical applications. But when you think about categorical versus dimensional, what we’re trying to figure out here is, is autism or autistic traits or autism symptoms, however you want to describe that, is that, really, truly a distinct pattern from neurotypical behaviour or is it contiguous? Is it really just one large dimension?
And so what I use to try to describe that to people is, you know, in medicine or just in life, we have some examples of what we would call true categories. So, pregnancy, you’re either pregnant or you’re not pregnant, right? Haemorrhagic stroke, you know, you either had a haemorrhagic stroke or you didn’t have a haemorrhagic stroke, right? For a mental health world, it would be, like, hallucinations, so you’re either having active hallucinations or you’re not having active hallucinations.
Dimensions, I think, make more sense to people, just intuitively, but, like, I’ll give some examples. So, lung capacity, we all have variations in lung capacity, right, and it’s, kind of, very dimensional in nature. Probably depends on where you live and how high above sea level you live and things like that, right? And then another example, from the psychiatry world, would be, like, anxiety. So, most forms of anxiety, not all forms, but most forms of anxiety, especially, like, worry or social anxiety are going to be more dimensional in nature, right? You can have, like, a middle amal and just, kind of, be along with that, if that makes sense?
[00:04:07.180] Jo Carlowe: Hmmm, absolutely, and very, very helpful, thank you. Can you give us an overview of the paper, what did you look at and why?
[00:04:13.170] Dr. Tom Frazier: Yeah, so this – it’s a fairly simple question here, we’re just trying to figure out is autism or autism symptoms or autism traits, is that more dimensional or categorical? And, you know, like we just described, we’re just trying to put it into these buckets of, like, is it more like pregnancy and haemorrhagic stroke or is it more like lung capacity? And the reason why this is valuable is that ultimately, this has implications for how we do research and it also has implications for how we identify autism. And so that’s, kind of, the practical reason for doing it. As I said, it’s – I’m not terribly theoretically interested in that stuff, I just wanted to know practically, how do we proceed here?
[00:04:57.500] Jo Carlowe: And what was the methodology you used for the study, if you can tell us a little about that?
[00:05:02.199] Dr. Tom Frazier: We put together some very large datasets, not all of them were large, but many of them were large. So, we actually had a total of six datasets. Some of these are, like, really big datasets. So, like, NDAR, the National Database of Autism Research, or they call it NDA now, I think, and that’s a very large dataset collected by the National Institute of Health in the US. And anybody who’s funded through NIMH, National Institute of Mental Health, and really funded for any autism research is going to put their data into NDAR, so it’s a very large dataset.
We also had the Healthy Brain Network dataset. We had the Simons Simplex Collection dataset. We also had Genetic Resource Exchange dataset and then my own datasets that include eye-tracking measures. So, all of the other datasets, other than my own, were really focussed on autism symptom measures. So measures like the Autism Diagnostic Observation Schedule, the Social Responsiveness Scale, the Social Communication Questionnaire. And then my datasets were focused on using measures of gaze to social stim.
And so, digging into the methodology a little bit here, we wanted to put together these large datasets because to answer – to ask and answer this question about dimensional versus categorical, you really have to have big datasets and to get the, kind of, power that you need to look at these kinds of things. So, in general, they recommend 300 – sample sizes of 300 or larger, and in some of our datasets, you know, we were able to put together combined datasets that were quite large.
We had some datasets that were over 300, but just – you know, not too much over 300. And then we had some datasets that were literally in the range of 13,000 published. We had one dataset in the range in 16,000. So, we had some very large datasets and all we did was, for each dataset, we identified a set of autism symptom measures or a set of gaze measures. And the idea here is that, especially for the autism symptom measures, you can use a set of empirical methods called taxometric procedures. And what taxometric procedures are, they’re a set of statistical methods that are blind to the autism diagnosis. So you don’t put autism diagnosis into the statistical procedure, you just put the symptom measurements in. And this stati – these taxometric procedures will essentially – and there’s several of them within that category, will essentially take different ways of looking at whether or not the data itself represent a latent category or a latent dimension.
And I’m not going to get into too much of the details there because it actually is quite math heavy, but suffice to say that these procedures have been shown through simulation studies, very large simulation studies, to be incredibly accurate at identifying whether or not the data are telling you that autism symptoms are a category or a dimension. For example, in these simulation studies, they are more than 99% accurate in identifying whether or not data are categorical or dimensional. So, these procedures are very accurate and they’re very good at doing this. Their whole job is really to figure out is this data measuring a category or a dimension? And so we applied those procedures to every single indicator set that we had, right? So we ended up having ten different indicator sets. Four of them were based upon gaze measures and six of them were based upon autism symptom.
So, I’ll give you some examples of that. So, for example, for our autism symptom measures, we had one set that was based upon the Autism Diagnostic Observation Schedule, ADOS, and there eight total items that we included in that analysis. Another one had the Social Responsiveness Scale, Original Five Subscale, so there were five indicators in that analysis. Another one of our indicator sets had seven different gaze measurements to these social stimuli. And so for all ten of these indicator sets, we applied the same statistical procedures. And we’re looking for a convergence here, we’re looking for internal replication, we want to see that every indicator set is going to give us the same answer and whatever direction that is.
We don’t care if it’s telling us it’s dimensional or categorical, we just want to know what is the answer and is it showing the same answer across the different datasets? Does that, sort of, give you a flavour?
[00:09:46.180] Jo Carlowe: Yeah, absolutely, it was really clear, thank you. I want to come to the – some of your findings, but before then, I notice in your paper you note that most psychiatric disorders and constructs are best represented as dimensional, but that autism is a possible exception. So, I’m wondering what has been the evidence for this?
[00:10:06.330] Dr. Tom Frazier: Yeah, that’s a great question. So, first of all, there was a group independent of us, years ago, that looked at a small sample, small dataset, but used taxometric procedures and they also found a categorical distinction. And so that was, like, one of our first hints that may be autism isn’t dimensional.
And then in 2010, I published a paper where we had slightly larger datasets, but still not even close to the order that we’re talking about today in this paper, right? So, this is more than ten years ago, we didn’t have as much data back then, and we applied taxometric procedures and we also saw categorical structure. And then I did another paper when DSM-5 came out, we were really interested in, sort of, does DSM-5 represent autism very well, right? So, I did another analysis, with a slightly different statistical method called factor mixture modelling, which looks at the same, kind of thing, is there a category or dimension here? And we also saw a categorical distinction in that data. So, we were building up some evidence here along the way, but, again, not these, kind of, large, multiple indicator datasets that we have today, right?
And then there was also evidence in the other direction. To be honest about it, like, we had evidence in both directions and that’s really why we wanted to do this much larger, more detailed study is because, you know, there are people, for example, on the behavioural genetic side, including a lot of folks in the UK, really great, strong work that was looking at behavioural genetics of autism and finding that maybe there wasn’t a categorical distinction, at least when you divide autism up into these three different aspects: social interaction, communication or restricted repetitive behaviour.
Now, that’s tricky in behavioural genetics. It’s not as straightforward as, kind of, just looking at the symptoms and seeing if there’s a categorical or dimensional distinction, but there was some evidence there.
I should back up and say there was other evidence too of categorical distinction too. So, let me give you just, like, a really plain and easy-to-understand example. Cathy Lord had studied autism, really, since the 80s and she had these longitudinal datasets, and she was finding that autism, when you diagnosed it, you know, the distinctions we originally made, like autistic disorder, PDD-NOS, Asperger’s, those didn’t pan out. You know, it was really autism spectrum disorder that panned out. But when you diagnosed autism spectrum disorder, it was really stable over time, right? So that’s pure – that’s consistent with categorical evidence because if autism was a dimension, you would expect people to, sort of, move along that dimension and jump – and so they would be jumping in and out of the diagnosis potentially or at least jumping out or potentiality jumping in. So, that also is, sort of, some evidence of categorical.
And then the other thing is, like, when they started looking at the DSM-5 diagnostic reliability, they were finding that autism was one of the more reliably diagnosed of the psychiatric conditions. So, again, that suggests that maybe Clinicians are, sort of, identifying a prototype or a qualitative distinction in behaviour that’s making that diagnosis a bit more reliable than some of the other [inaudible – 13:22]. There was evidence in both directions, though, and I just want to be clear about that, that it – that there – this is a controversial area and that was, kind of, the whole reason for us doing that.
[00:13:31.570] Jo Carlowe: Right, so let’s bring it back to current time with your larger, more detailed study. Can you highlight some of the key findings from your paper?
[00:13:39.980] Dr. Tom Frazier: And our findings are pretty straightforward. I mean, we found across all ten indicator sets that taxometric procedures were telling us that these data are category – are representing a latent autism category and not a dimension. There’s a dimension within the category, obviously, like, some people have more significant or less significant or higher or lower levels of symptoms, but the category itself appears to be distinguishable from neurotypical patterns of autism symptoms.
You know, we – that was – I think what’s really cool about this paper, and I know I’m the highest – and I’m an author on it, but, like, you know, we saw really strong evidence across every single indicator set that we created. And the base rate – so when you do these analyses, if you find a category, so if the data represented a category rather than a dimension, you can estimate the base rate of that category. So, like, what is the level of that category, how many people are falling in that category, right? And so that when we estimated the base rate, it was fairly consistent with the base rates of the diagnosis in these datasets. So remember, the procedures are blind to the diagnosis and so when they’re estimating the base rate, if it gets fairly close to the diagnostic base rate, then you could be pretty certain that the category that you’re identifying is pretty close to autism. Does that make sense?
[00:15:15.150] Jo Carlowe: Yeah, totally. So, your findings provide support for categorical structure of ASD, but what are the implications of this for Researchers and particularly those searching for major drivers of the ASD phenotype?
[00:15:27.980] Dr. Tom Frazier: You know, first of all, when you find a latent category, it does suggest – now, I’m not saying that this – and this is not a genetic study, so I just want to be very clear, I’ve worked with a lot of great Geneticists over the years and I don’t want to pretend, like, I’m one of them.
[00:15:43.620] Jo Carlowe: Okay.
[00:15:44.620] Dr. Tom Frazier: But it does, sort of, suggest that there’s some, kind of, underlying ideological factor that is driving this categorical distinction. Now, that doesn’t have to be genetics. I mean, we have good evidence of genetic changes that drive autism, in some cases, right? Like, we know of genetic syndromes that – where autism is much more likely to occur, right, like Phelan-McDermid syndrome and Malan syndrome and PTEN hamartoma tumour syndrome and, like, there’s a bunch of these genetic syndromes where autism is much, much, much more common then, in a population. So, we know that there are genetic drivers, but this work suggests that there are in fact strong ideologic factors that are probably driving this categorical distinction, and we should continue the process of looking for what these ideologic factors are so that we can understand those genetic effects or those other biological effects or even gene-environment effects.
And once you understand them, then you can really dive in and start to think, you know, ‘cause many people with autism have co-occurring conditions that are quite difficult for them, whether it be mood, anxiety, ADHD, you know, sleep difficulties, GI problems, etc. You can really dive into the biology then and try to figure out if there’s a way to help autistic people with those particular co-occurring conditions.
It really also has this other fairly basic idea, which is it’s okay to do group-based research where you identify an autism group and a neurotypical group. That doesn’t mean you wouldn’t also want to do quantitative trait studies with autism, it just means that it’s perfectly fine to do that because there is such a thing as an autism category, so identifying an autism group makes sense.
Now, the question becomes, and this is something we couldn’t answer in our study, is what do we do about broad autism phenotype? So, we’ve known for a while that there are first-degree relatives of people diagnosed with autism that have autistic traits that we would call broad-autism phenotype. And this – our study does not really tell us much about that, it suggests that broad autism phenotype probably is a part of the autism category, but there’s definitely a lot more research to try to understand what’s going on there. Is this just individuals that are showing some traits, but they’re sub-threshold for autism spectrum disorder or, you know, what actually is happening? So, stay tuned, you know, there’s obviously more research on that.
But, in general, for research, our work suggests that, yes, you can do group-based studies. When you do those group-based studies, though, you should be very careful to try to identify the full breadth of the autism category. And this is true, of course, of all research, you don’t want to selectively sample just some parts of it.
[00:18:29.250] Jo Carlowe: I’m also wondering what the implications are of your findings for Clinicians and other CAMH professionals. Tom, do you envisage your findings changing the way children and young people are assessed for ASD?
[00:18:41.960] Dr. Tom Frazier: Yes and no. So, I mean, in general, I think we’ve started to get a pretty good handle on what the assessment process needs to look like. We know we need observational measures. We know we need to, you know, do some interviewing of the parents. We know that we need to collect some questionnaire data. Nothing I’ve done would change that necessarily, but what our work suggests is that if you have a real autism category, if that’s a real thing, then we should be focusing on and using evidence-based medicine processes to identify that category.
In other words, instead of just using clinical judgement or just looking at scores on particular measures, we should be taking those scores and identifying how they change the probability that this person has – is autistic or has autism spectrum disorder. And so there’s a whole area of evidence-based medicine called evidence-based assessment where, for every measure you use, you can identify the likelihood that a particular score would change the probability of an autism diagnosis.
It will, sort of, pile those up or add those together and then get to a final probability. And so, evidence-based assessment is an area that I would love to see Clinicians adopt more and more, assuming that our results are accurate and that there truly is an autism categorical diagnosis.
[00:20:03.490] Jo Carlowe: And what are the implications for the way autism is classified? Do you have specific recommendations that you would like to see, some particular changes?
[00:20:12.250] Dr. Tom Frazier: Actually, you know, surprisingly, and this wasn’t necessarily something we were looking at, but our results really do support the DSM and the ICD-10, sort of, approach, right? And so our results actually suggest that these categorical diagnostic systems for autism spectrum disorder are probably doing a pretty good job. We have other work looking at the, sort of, symptom patterns that we see that would suggest that there probably are some tweaks to DSM-5 that we should make, particularly around not lumping together sensory interests with sensory sensitivities, like, those seem to be two different things.
And then also in the DSM-5, they, kind of, lump social motivation or affiliation, kinds of, social differences with some of the other social stuff like social communication or with reciprocity or prospective [inaudible – 21:01] kinds of things. So there’s some changes I think we should make there. In general, our work – this paper, really does support a categorical diagnostic approach.
[00:21:17.980] Jo Carlowe: Were you surprised by that?
[00:21:19.870] Dr. Tom Frazier: Yes and no. So, in 2010, when I did the first study, before I even ran the analyses, I pulled up a Word document and I named the document “Autism Continuum” because my hypothesis was that this would be a dimension. And then I get the analyses and I had to rename the paper and I had to change the title, right, because we found categorical distinction. But after that point is when I started to say to myself, “You know, the data are getting stronger here suggesting that this – that autism really is a categorical distinction.” So, I wasn’t terribly surprised by that at this point, but I was surprised at how strong and consistent the findings were.
And the other point I really should mention here, which is the reason why we put the gaze indicators into the analysis is because those are objective measures. The symptom measures are all filled out by a Clinician or by a parent, so those are – by nature, those are subjective measures and they focus entirely on our conceptualisation of what autism is.
The gaze measures don’t do that, they actually just look at how are you gazing at social stimuli, what are you looking at? Are you looking at the more socially and emotionally relevant aspects or are you checking out more technical or non-social aspects of the stimuli? And we used a whole bunch of different stimuli too, and because gaze measures are objective measures, it was very important to us to do the study with objective measures as our primary outcome because, you know, it’s possible that you can gain the system a little bit with subjective measures. Like, you might call for a category over a dimension in some cases or even in some cases you might call for a dimension over a category.
Those subjective measures are very, very useful. They aren’t necessarily going to seal up the whole replication that you need to do. These objective measures were incredibly important for us to put in here. And the fact that the objective measures are showing us the exact same pattern as the subjective measures, really confirms for us that we didn’t have some kind of methodological problem.
The last thing I want to say about it is, one of the things I am most proud of in this paper is that, I think I’ve published more than 150 papers now, I mean, this is the paper that I would point to that has the absolute strongest internal replication of any paper I’ve ever published. I feel the most confident that we did our best, at least, as Researchers to try to replicate the results across indicator sets, across subjective and objective measures, across different types of subjective and gaze measures, and that we really did our best to make sure we weren’t fooling ourselves in some.
[00:23:58.990] Jo Carlowe: It sounds a really robust piece of research. So, Tom, before we finish, I just want to find out, sort of, what’s next really, are you planning any follow-up research or is there anything else in the pipeline?
[00:24:09.669] Dr. Tom Frazier: We are collecting, right now, a much larger dataset of gaze measures. And so I want – the part that I – that we couldn’t replicate in this is our gaze samples were small. And so now we’re collecting a much, much larger dataset of gaze measures. And we’re doing it through the webcam actually, so these measures were collected using an actual eye tracker. But the technology has advanced so much from when we collected our original data that now we can use a webcam and we can actually train parents how to collect the data at home, so we can get really large samples. So, we’re collecting some very cool, very large samples of this data. And once these samples get over 1,000 individuals then I’m going to dive back in and basically rerun the same kinds of analyses, but on a much larger data sample and make sure that we see the same thing again.
[00:25:01.140] Jo Carlowe: Hmmm hmm, excellent, and finally, Tom, what is your take home message for our listeners?
[00:25:07.169] Dr. Tom Frazier: That autism is really a distinct qualitatively different patten of behaviour. That it is actually consistent with the zeitgeist of many autistic people in the neurodiversity movement where they view themselves as being distinct individuals with particular strengths and that within that group, there is a wide spectrum of individuals that have varying levels of strengths and challenges. But that group really does seem to be qualitatively distinct from neurotypical behaviour and that we should continue to study that group in its totality, so that we can truly understand how best to support and help those individuals to maximise their lives.
[00:25:52.669] Jo Carlowe: Tom, thank you ever so much. For more details on Professor Tom Frazier, please visit the ACAMH website, www.acamh.org, and Twitter @ACAMH. ACAMH is spelt A-C-A-M-H, and don’t forget to follow us on your preferred streaming platform, let us know if you enjoyed the podcast, with a rating or review, and do share with friends and colleagues.