Sunday, October 10, 2010

Are autistic people natural born criminals?
Associations between autism and notorious violent crimes are easy to find--they seem almost automatic. Here is one example, and another, and one more. There seems to be an entire book on this theme, though I haven't read it.

In the scientific literature, you can find powerful deficit models of autism at work in predictions that autistics should disproportionately be violent and prone to criminal behavior. For decades now, examples and claims (just a few here, here, here, here) fitting this prediction have been highlighted, while the few dissenting views (e.g. here and here) have had little effect.

Then there's political expedience. Unless we receive the usual lobbied-for interventions, autistics will disproportionately display criminal behavior and wind up in jail--or so it is claimed. A few examples from the usual autism politics here, here, here.

Kathrin Hippler and her colleagues deserve high praise for noticing how well-placed they were to investigate these kinds of popular claims. Their recent paper takes advantage of their access to information about the large group of individuals--177 of them--who are former patients from Hans Asperger's clinical practice in Vienna.

Most (93%) of these former patients are male. All were assumed to score at least in the normal range of intelligence as children, but for most there are no recorded IQ scores. They were born between 1938 and 1979, and on average were diagnosed at age eight (range 3-21 years). In 2010 their average age would be about 50. It is unlikely that most of this cohort would have undergone the usual lobbied-for autism interventions as children.

Hippler et al. obtained information from the Austrian Penal Register about all criminal convictions registered, as of 2002, in this cohort. They found 33 convictions for a total of 8 individuals, resulting in 23 "custodial sentences" ranging from 2-30 months and 11 fines.

They also checked whether, compared to the general population, their cohort had a higher rate of newly-registered convictions for the years 1998-2002. Here is what they found:
the average proportion of convictions found in our sample (1.30%) is very comparable to that in the general male population (1.25%)
As to kinds of crimes:
By far the most common convictions in Asperger’s former patients were for property offences [...] Offences against life and physical integrity were rare.
And while data for the general population were limited:
qualitative assessment of offence types in Asperger’s former patients suggests that they do not differ radically from those in the general public
Because this point needs to be underlined, here is more from Hippler et al.'s discussion:
the findings from our study do not suggest an over-representation of certain offence types. In the case records spanning 22 years and 33 convictions, there were only three cases of bodily injury, one case of robbery and one case of violent and threatening behaviour.
Again to their great credit, Hippler et al. also provide data broken down according to Asperger's system of diagnosis. The 177 former patients were divided into an AP group ("autistic psychopathology," N = 73) and an AF group ("features of autistic psychopathology," N = 104).

Some guesswork is involved, but the AP group would fall under current criteria either for Asperger syndrome or the specific diagnosis of autism. Hippler et al. conservatively estimate one-third would be specific-autistic, but it's not difficult to find researchers who would guess a higher proportion (here, for example).

While a minority of the AF group might meet Asperger syndrome criteria, according to Hippler et al., others might be PDD-NOS, or in the (nonautistic) broader autistic phenotype. The AFs are described as "former patients at the less extreme end of the spectrum" and were included by Hippler et al. because:
Asperger believed that ‘autistic psychopathy’ was a heritable condition blending into "normality", which is reflected in the case descriptions of these children in the sense that the core features were the same but symptoms were less severe or could be compensated for better.
According to near-universal assumptions, it would be far better to be AF (less "severe" or "extreme") than AP. Transforming AP-types into AF-types is a major goal of the usual lobbied-for autism interventions. But Hippler et al. found that most of the registered criminal convictions in their cohort belonged not to the "more autistic" AP group but to the "less autistic or not autistic at all" AF group.

Indeed, of the 33 convictions found registered as of 2002, only three convictions of two individuals were found in the AP group. The remaining 30 belong to six individuals in the AF group, with two in this group contributing 22 convictions.

In the comparison with rates of newly-registered convictions in the general population, the AP rate was 0.6% while the AF rate was 1.7%. These figures are lower and higher, respectively, than for the general male population, while the AP rate is comparable to the general population, females included (0.7%).

Hippler et al. provide a competent overview of the relevant literature (including this recent finding), as well as a fair discussion of their study's limitations. Under the banner "Wider Implications," they write:
There is a public perception that individuals with mental health diagnoses in general, and Asperger’s syndrome in particular, present a threat to the general public. We contend that, based on the follow-up data from Asperger’s original cohort, as well as other studies, this perception is wrong.
Even wider implications include the neglected question of how being regarded as just naturally violent and dangerous to others, as natural born criminals, has affected the outcomes of autistics.


Hippler, K., Viding, E., Klicpera, C., & Happé, F. (2009). Brief Report: No Increase in Criminal Convictions in Hans Asperger’s Original Cohort Journal of Autism and Developmental Disorders, 40 (6), 774-780 DOI: 10.1007/s10803-009-0917-y

Postscript: This post has been included in the 43rd edition of Scientia Pro Publica.

Tuesday, August 31, 2010

Are you high or low functioning? Examples from autism research
If you are autistic and ever venture or are pushed into public, a near-certainty is that you will publicly be ranked and classified by total strangers.

For example, you will be assigned to the "high end" or the "low end" of the autistic spectrum, according to whether you are claimed to have a good or bad outcome (I've been claimed to have both). Non-political observers may notice how ethically and scientifically problematic this is, but there are few discussions, formal or informal, in which autistics aren't automatically assigned "high-functioning" and "low-functioning" rankings as definitive and permanent aspects of ourselves: there are LFAs and then there are HFAs.

If the whole area of autism were not so politicized, it might have long been seen as important to check the scientific basis for these ubiquitous rankings. It might even be seen as an ethical obligation. It might be noticed that there is something wrong with discarding and therefore dishonouring the contributions to research made by the countless autistic children and adults who have been recruited as participants in autism research, and without whom there would be no autism research at all.

I've never been impressed by autism politics, on whatever "side," so I'll start with some basic information about levels of functioning as reported in autism research, then provide several examples pulled from some of the numerous papers I've recently looked at.

What does level of functioning mean in autism research?

In autism research, autistics' level of functioning is most often judged according to scores on specific tests of IQ (e.g., Wechsler) or developmental level (Mullen, Bayley, sometimes the Vineland) at a specific time.

The reported threshold dividing "high" from "low" functioning ranges from 50 to 90--at least in papers I've read so far; the actual range might be even greater. Those are IQ or IQ-type scores with a mean of 100 and a standard deviation of 15. So the threshold, the line dividing "high" from "low" functioning in autism research, is almost three SDs wide. Fall into that impressive span, and you may be high or you may be low functioning, depending on who you ask.

If 90 is the threshold, then about 25% of the entire population (autistic, nonautistic, everyone) is low-functioning. If the more common threshold of 85 is chosen, then about 16% of everyone is low-functioning.

70 is the threshold often considered to be standard even if in reality, if you read the literature, the threshold varies dramatically. But there are different tests and within commonly-used tests there are different ways to set a threshold, even when the threshold is numerically set at 70.

With this in mind, here are some examples.

Example 1

In Farley et al. (2010) the threshold is indeed 70 and autistics with lower baseline (circa age 8yrs) IQs were excluded. As the authors write, "all participants had baseline IQs in the nonimpaired range."

In fact the full-scale baseline IQs of autistics deemed to meet the 70 threshold, and included as participants, range down to 36. An IQ of 36 is near the current threshold for severe intellectual disability. An IQ of 36 (if you were under age ~4yrs) would exclude you from Ivar Lovaas' famous ABA study--you are too low-functioning.

But in different circumstances, which are specified by Farley et al., a full-scale IQ of 36 can also be considered to grossly underestimate your abilities and to put you in a study of participants who would widely be ranked as high-functioning.

Example 2

Klin et al. (2007) is another study where the autistic participants (ages 7-18yrs) are all considered high-functioning and the threshold is 70. In one of two samples, verbal, performance, and full-scale IQs run from 52 to 150. Then there are Vineland scores. While Vineland is a test of adaptive behavior, in autism research Vineland scores can also be used as measures of intelligence ergo level of functioning.

Vineland composite scores in this sample range down to 25. Two of the Vineland domain scores range down to 20. In a recent population-based study (ages 9-14yrs), an assigned Vineland score of 19 was used to represent profound intellectual disability. But in Klin et al., you can have a Vineland composite score of 25, or Vineland domain scores of 20, and you are high-functioning. You are in the same category as someone with a verbal IQ of 150.

Example 3

In Akshoomoff et al. (2004), there is also a threshold of 70, but autistics falling on either side of this threshold are included then divided into lower and higher functioning groups. Performance IQ scores in the lower-functioning group range up to 128. That is not only higher than the individual scores of the entire higher-functioning group, it is higher than all but ~3% of the entire population. If a PIQ of 128 ranks you as low-functioning, as it does in Akshoomoff et al., then almost everyone is low-functioning.

Example 4

In view of the above it might seem wise to abandon the 70 threshold and try something completely different. Annaz et al. (2009) created high-functioning and low-functioning autism groups (ages 5-11yrs) by incorrectly using the CARS (Childhood Autism Rating Scale), which is supposed to be a measure of autism "severity." This study features two measures of intelligence, one verbal (British Picture Vocabulary Scale, a verbal IQ equivalent) and one non-verbal (a subtest from the British Ability Scales).

One result is that you can have a verbal IQ of 62 and be classified as high-functioning, and a verbal IQ of 111 and be classified as low-functioning.

The non-verbal measure is reported in age-equivalents only, but the two autistic groups are remarkably well-matched on age. And the mean non-verbal age equivalents are both (1) the same for the two autistic groups, and (2) very nearly the same as mean chronological age, again for the two autistic groups.

In other words, another result is that both high- and low-functioning autistics here represent groups whose mean measured non-verbal intelligence is the same as the general population mean, and of course vice-versa. Setting aside possible differences in distribution, now everyone is both high- and low-functioning.

Conditional examples 5 and 6

This is a bit of a digression, but if you adhere to the common political or ideological prejudice that "Kanner's autism" is "classical autism" is "low-functioning autism" you are then required to achieve an IQ of over 140 (from Kanner's original 11) or 150 (from Kanner's 1956 follow-up) to be ranked as high-functioning. That means you have to be better than the 99th percentile.

Now what?

The above doesn't nearly convey the arbitrariness in the existing autism literature. As I wrote, I've provided just a bunch of examples, among many others, you will find if you read a lot of papers. Also, I've left out entire major areas, like changes over time and results from deliberate comparisons between different tests of intelligence.

While this was not intentional, the examples above might be construed as exploiting the atypically high variability characterizing individual autistics and in addition characterizing autistics as a group. Then the question is whether it serves the interests of autistics, and whether it advances autism research, to diminish, misrepresent, trivialize, denigrate, obscure, or deny this characteristic variability.

And as usual, the above can be verified by reading the existing autism literature. If there are any factual errors, as is always possible, please let me know.

Futher reading

The late Ivar Lovaas expressed his views about levels of functioning in autism in one of the major ABA manuals; this is quoted here. From a different point of view, this article concisely applies ethical consideration to the issue of level of functioning, among others. In autism politics, the dimension of autism "severity" and the different dimension of "level of functioning" are often wrongly confused or conflated. You can find science-based information about autism "severity" here and here.


Akshoomoff N, Lord C, Lincoln AJ, Courchesne RY, Carper RA, Townsend J, & Courchesne E (2004). Outcome classification of preschool children with autism spectrum disorders using MRI brain measures. Journal of the American Academy of Child and Adolescent Psychiatry, 43 (3), 349-57 PMID: 15076269

Annaz D, Karmiloff-Smith A, Johnson MH, & Thomas MS (2009). A cross-syndrome study of the development of holistic face recognition in children with autism, Down syndrome, and Williams syndrome. Journal of experimental child psychology, 102 (4), 456-86 PMID: 19193384

Farley MA, McMahon WM, Fombonne E, Jenson WR, Miller J, Gardner M, Block H, Pingree CB, Ritvo ER, Ritvo RA, & Coon H (2009). Twenty-year outcome for individuals with autism and average or near-average cognitive abilities. Autism research : official journal of the International Society for Autism Research, 2 (2), 109-18 PMID: 19455645

Klin A, Saulnier CA, Sparrow SS, Cicchetti DV, Volkmar FR, & Lord C (2007). Social and communication abilities and disabilities in higher functioning individuals with autism spectrum disorders: the Vineland and the ADOS. Journal of autism and developmental disorders, 37 (4), 748-59 PMID: 17146708

Postscript: This post has been included in the 40th edition of Scientia Pro Publica.

Wednesday, August 18, 2010

Making autism research history: The Preschool Autism Communication Trial
Some months ago, Jonathan Green and his colleagues simultaneously published (in the Lancet) and presented (at IMFAR 2010) their multi-site RCT of an early autism intervention. In conducting and reporting the Preschool Autism Communication Trial, they have made autism research history. They have done so simply by applying to autistics scientific standards that are well-established in non-autism non-ABA areas. The upshot is an autism early intervention trial whose results are worth serious consideration and which is unprecedented in this respect.

Via the PACT website, you can find the Lancet paper reporting the PACT RCT, the peer-reviewed PACT intervention "manual" (which refers to another, more detailed manual), the PACT RCT protocols, references for related papers including a 2004 pilot RCT, a list of investigators and funders, and so on.

I've previously written about the PACT as an anomaly in autism intervention research. The PACT entails an early autism intervention that was not widely promoted as effective or essential before it was fairly tested. That is a first in the history of autism research. It may also be the largest RCT of any kind of autism intervention ever published (if I'm wrong, I'm sure someone will tell me).

In short the PACT RCT is an internationally-registered multi-site trial of a manualized intervention targeting the parents of preschool (ages 2-5yrs) autistic children. The PACT intervention spans one year and involves 12 twice-monthly 2hr clinic-based sessions followed by 6 monthly sessions. Parents are expected to apply what they learn, in interacting with their autistic child, for half an hour per day.

For the PACT RCT, 152 children who met criteria for the specific diagnosis of autism were randomly assigned to receive either the PACT intervention plus treatment as usual or TAU alone. PACT and TAU children were well-matched at intake across several measures, and finished well-matched in other treatments received. Parents were not as well-matched. A detailed description of the PACT intervention is here.

Dr Green and colleagues were successful both in recruiting an unprecedentedly large number of preschool autistic children for an RCT--in fact, expectations for recruitment were exceeded--and in minimizing drop-outs. In the Lancet, you will find an outstandingly clear CONSORT flowchart of participants as well as an intention to treat analysis and mixed results.

Results favouring the PACT group were mainly in the area of parent-child interaction, as assessed through videos. The PACT group also outperformed the TAU children in parent-reported vocabulary and social measures, results carefully downplayed by Green et al. due to the parents not being blinded to intervention status. Post-treatment-only measures of adaptive behaviour reported by teachers did not differ between groups.

The main measure was the ADOS used as a scale, and here too there was no result. In my view, using the ADOS as the main measure is a major error but an informative one. The ADOS, as used here (awkward tweaks and all), has thresholds only and doesn't work like a scale. And where is the evidence that lower ADOS scores in very young autistic children in themselves lead to better eventual outcomes? On the other hand, the strong design of the PACT RCT is such that it calls into question what exactly the ADOS is assessing--a result which deserves a lot of attention.

Using ADOS thresholds, 27% of the 146 children who completed their assigned treatment had migrated from the specific diagnosis of autism to the lower-threshold "ASD" category, while an additional 6% no longer met criteria for any autistic spectrum diagnosis. No group differences were found. These findings lend perspective to the widely-publicized diagnostic category results from another recent RCT involving a small sample of much younger children receiving more than two years of intervention.

As with all studies, it's possible to list a range of flaws, potential concerns, and caveats. The ADOS is not only wrongly used as a measure, it is incorrectly used to stratify the sample for randomization. The ADI-R is used in a slightly unconventional way, possibly to accommodate children who otherwise would be excluded by its limitations. There was no screening ergo no exclusion for genetic syndromes, which arguably makes the findings more difficult to interpret. There was poor agreement between raters for two of the three parent-child interaction measures, which in my view is fascinating and telling: nonautistic raters are not good at agreeing about what child initiation and shared attention consist of, when a child is autistic. If you look at the Lancet paper, depending on how your brain works, you should rapidly find at least one minor error in the data (I can't help it...).

The PACT intervention itself has questionable aspects. The fundamental idea is sound: train parents to better respond to their autistic children's communication. But the PACT is a rigidly developmental six-step approach based on how nonautistics develop. In addition, and like so many other autism interventions, the PACT imposes extreme rationing of information and materials. Everything is cleared and shut away except a few toys or items. Anything in which autistics regardless manage to show strong interest will be taken away and replaced with something less interesting. What this does to an autistic child's communication and learning is not considered. The PACT explicitly requires every effort to identify and cater to the parents' different ways of learning, but no such consideration is extended to the autistic children.

The above is a not-close-to-complete list of the PACT's strengths, many of them unique, and its weaknesses--which due to its strengths serve as important information for the future of autism intervention research. Because the trial is so strongly designed compared to the rest of the literature, it is that much more important to learn from and build on. This may not serve the interests of various lobby groups or service providers or political leaders or advocates, but it will serve the interests of autistics. Here is a suggestion from Dr Green and colleagues in the Lancet:
These findings suggest that the optimistic results from other studies should be reassessed.
Yes--good experimental design means that you risk not having your biases confirmed. But taking this risk is currently the only demonstrated way to fairly test interventions and treatments, and for this reason, the only way to conduct ethical research.

There is a Lancet editorial about the PACT RCT, written by two NIH scientists. Here is one thing they point out:
Thus, in a field in which minimum study standards have made it difficult to even look for literature to answer what works for autism, this study is an achievement.
Here is another, and they get the last word:
This study furthers the field by setting a new bar for the minimum standards of rigorous methodology needed in trials that have potentially far-reaching service and policy implications.


Green, J., Charman, T., McConachie, H., Aldred, C., Slonims, V., Howlin, P., Le Couteur, A., Leadbitter, K., Hudry, K., & Byford, S. (2010). Parent-mediated communication-focused treatment in children with autism (PACT): a randomised controlled trial The Lancet, 375 (9732), 2152-2160 DOI: 10.1016/S0140-6736(10)60587-9

Spence, S., & Thurm, A. (2010). Testing autism interventions: trials and tribulations The Lancet, 375 (9732), 2124-2125 DOI: 10.1016/S0140-6736(10)60757-X

Monday, May 17, 2010

The circadian prison

This post was chosen as an Editor's Selection for
I had no idea my sleep-wake cycle was pathological until I saw a presentation a few years ago by the geneticist Thomas Bourgeron. One of his interests is clock genes in autism. In fact only by speaking with him did it dawn on me that I lack a circadian rhythm.

As it turns out, autistics are considered to have circadian clocks "gone bad." Two reviews (Bourgeron, 2007; Glickman, 2010) cover the evidence with respect to autism and circadian rhythms, most of it in the direction of comprehensive atypicalities. Glickman (2010) summarizes the problem:
Disturbed sleep-wake patterns and abnormal hormone profiles in children with autism suggest an underlying impairment of the circadian timing system.
And autistic children's bad timing, when it comes to sleeping and waking, is considered disruptive or worse, both to themselves and their sleep-deprived families. Bourgeron (2007) raises the further concern that typical sleep-wake patterns are essential for "appropriate" development, for the functioning of memory, and so on. Apparently there is nothing good about autistics' bad circadian clocks.

Autistics whose sleep-wake cycles carry on independently from environmental and social cues are said to be "freerunning." My form of freerunning is extreme. I can sleep whenever I want--a few hours here, twelve hours there, as necessary, no matter the time of day or night. I enjoy sunlight but its absence has no effect on my mood, or on my sleep. I can work through nights, no problem, and shift to days in a blink. I've never suffered through jet lag. And so on. Until I ran into Bourgeron I had no idea this was considered pathological, at least in an autistic.

The usual response to freerunning in autism is to see this as an autism-related sleep disorder. There is very preliminary evidence that freerunning autistics can be successfully treated with melatonin. Bourgeron (2007) refers to a short case study about an autistic whose free-running was remediated by melatonin treatment.

Of course nonautistics take melatonin too. They put themselves through sleep hygiene routines and they go to sleep clinics in droves. They suffer terribly from shift work and jet lag, and inadequate efforts to eradicate this suffering are everywhere, from new drugs to new iPhone apps.

There is an enormous market for products and services that can treat individuals who stray from the proper expected sleep-wake cycle. Remarkably, the goal is never to free people from their circadian prisons but to return them there and keep them there. The schedule may be changed, with effort, but the prison remains the same.

The next time I spoke with Bourgeron, I told him there should be a sort of anti-melatonin which would allow people to freerun like me, ergo avoiding the ills of jet lag, shift work, and what-all. It seems possible no one has seriously thought of this.

I suggested the anti-melatonin idea to some nonautistic colleagues. They burst into stories about destroyed careers, disrupted lives. In constrast with autistics like me, my colleagues have perfectly "good" circadian clocks--which turn out to be incompatible with their challenging lives. They suffered, for being normally chained to their normal circadian rhythms.

If there was a pill that could give them my very dysfunctional clock so they could freerun to the extreme, they would buy it in bulk. But there isn't. Instead there is the assumption that everyone has to live in a circadian prison.

Glickman (2010) speculates that some autistics' failure to chain our sleep-wake cycles to environmental cues (you can find information about entrainment, as this process is called, here) may arise from our atypical perception. My totally wild guess might be that an extreme freerunning phenotype in autism may be contributed to in part by cognitive versatility in autism, which would result in perceived environmental cues affecting sleep-wake cycles in an optional rather than mandatory way.

The renowned Bora Zivkovic, who unlike me knows all about circadian clocks, recently wrote that--unlike me--typical people have "have complex circadian systems that are easy to get out of whack" with dire consequences:
In the state of almost permanent jet-lag that many of us live in, a lot of things go wrong. We get sleeping disorders, eating disorders, obesity, compromised immunity leading to cancer, problems with reproduction, increase in psychiatric problems, the Seasonal Affective Disorder, prevalence of stomach ulcers and breast cancer in night-shift nurses, etc.
BoraZ was writing enviously about reindeer who live with drastic changes of season yet have escaped this kind of misery, possibly by having a "low-amplitude" clock. They just stroll out of their circadian prisons, whenever it's adapative. How can poor suffering human beings do this, is the question--finally. It's a very good question.


Bourgeron, T. (2007). The Possible Interplay of Synaptic and Clock Genes in Autism Spectrum Disorders Cold Spring Harbor Symposia on Quantitative Biology, 72 (1), 645-654 DOI: 10.1101/sqb.2007.72.020

Glickman, G. (2010). Circadian rhythms and sleep in children with autism Neuroscience & Biobehavioral Reviews, 34 (5), 755-768 DOI: 10.1016/j.neubiorev.2009.11.017

Postscript: This post has been included in the 32nd edition of Scientia Pro Publica.

Wednesday, February 17, 2010

Oxytocin versus autism: A cure for altruism
The widespread message arising from Andari et al. (in press) is that the hormone oxytocin "may be a powerful weapon in fighting autism" or words to that effect.

The heart of this study is a computer game version of catch which appears to involve four human players. When a player is thrown the ball, he must then throw it to another player of his choice. Every time a player receives the ball, he receives a bit of money.

In Andari et al. (in press), small groups of autistic and nonautistic adults ("P") individually play this game with three strangers ("A" "B" "C"). Much is done to persuade the autistic and nonautistic participants that the strangers, who in fact are elaborately programmed, are actual, present, proximate (in adjacent booths), decision-making human beings.

The three strangers start by equally distributing the ball to the other players. Then the elaborate program kicks in: A and C increasingly favour one player (P and B, respectively) while shunning the others. B's behaviour does not change.

In a prototypical display of us-vs-them thinking, the nonautistic Ps responded by ganging up with A to reap the social and monetary gains of a close alliance founded on the exclusion of half the players.

The autistic Ps in contrast displayed no such selfish and discriminatory behaviour. They continued to throw the ball to the other players in equal proportion, ignoring their self-interest in favour of keeping all players equally included.

Further, subjective ratings of the other players revealed that the autistic Ps did not have the kinds of biases that are routinely called hypocrisy. They did not judge C, who shared the most with B, as worse or less trustworthy than A, who shared the most with themselves.

According to Andari et al. (in press), autistics "cannot understand or engage in social situations," as evidenced by autistics' outstandingly altruistic performance in this game. It is this profound social deficit, this altruistic autistic behaviour, that was targeted for treatment.

And indeed, the treatment was successful. Autistics randomly administered a nasal mist containing oxytocin, rather than a saline placebo, significantly improved. They became willing to work with one of the players in an effort to shun and discriminate against the other two, and thereby get more than their fair share of money and attention. They became willing to see the player who shared with them as good and trustworthy, and the player who shared with someone else as bad and untrustworthy. They learned and displayed selfishness and hypocrisy and us-vs-them thinking. Their objectivity, fairness, and altruism were--temporarily--cured.

Then this finding was replicated in a second small group of autistic adults who performed the same task but without monetary rewards.

Success! Cue the avalanche of blogging and media stories. Uta Frith says, "This could be revolutionary."

There is more to Andari et al. (in press), which also reports on two tasks involving face images, about which a few quick notes:
1. There is little ecological evidence that autistics avoid face images or do not make eye contact with face images.

2. Numerous other studies (a few examples here, here, here, here) which feature stronger designs (e.g., use of fixation crosses) have found typical duration and distribution of visual fixations to human face images, and/or typical attention to the eye area, in autistics.

3. For the nonautistic participants, the reported total visual fixation time to the face images--the crucial measure--well exceeds the total exposure time (both tasks), which has me, for one, scratching my head.

4. As yet there is no evidence that the reported oxytocin-related measured changes in visual fixation times to face images would overall be more beneficial than harmful to autistics.

5. In other studies (examples here and here) faster saccades and shorter visual fixation times, presumed by Andari et al. (in press) to be a very bad thing, have been associated with enhanced performance in autistics.
All this too is telling in its way, but it cannot equal the reported spectacular findings with respect to autism and altruism. In this respect Andari et al. (in press), even with its obvious limitations in design, is an enormously valuable paper. It has much to reveal about how autistics are regarded, and about the nature and consequences of the fight against autism.

For more about us-vs-them thinking in autistics and nonautistics, I recommend Tyler Cowen's prescient book, Create Your Own Economy, as well as this related video.

Interpretations of Andari et al. (in press) which in no way resemble the above can be found here, here, and here.


Andari, E., Duhamel, J., Zalla, T., Herbrecht, E., Leboyer, M., & Sirigu, A. (2010). Promoting social behavior with oxytocin in high-functioning autism spectrum disorders Proceedings of the National Academy of Sciences DOI: 10.1073/pnas.0910249107

Postscript: This post has been included in the 22nd edition of Scientia Pro Publica.

Friday, February 12, 2010

What is severe autism?
We have to wait, patiently, for the DSM-V people to cough up their system for ranking and classifying all autistics according autism "severity." In the meantime, some recently reported data are worth mulling over.

First, here is the most recent unofficial DSM-V autism "severity" ranking-system proposal, and here is my response, including information about instruments commonly claimed to measure autism "severity." The current official DSM-V void in this area can be located here.

An increasingly prominent measure of autism "severity" is the Social Responsiveness Scale. Developed by John Constantino, it now exists in different age-range versions. Its purpose encompasses quantifying what are presumed to be autistic traits, from none on up, across the full range of humanity.

The SRS is a 65-item questionaire most often filled out by parents or teachers, whose ratings classify children according to "severity of autistic symptomology." Higher scores, above an established threshold, indicate greater autism "severity," and are therefore considered worse.

In a recently epublished high-profile paper (Roberts et al., in press), SRS scores were reported for 25 autistic children who were recruited then divided into two groups according to performance on a popular test of language abilities.

The children were about 10 years old. Sixteen autistic children classified as non-language-impaired achieved an average language ability score of ~100, right at the mean for the administered test. In contrast, nine autistic children classified as language-impaired scored ~65, more than two standard deviations lower.

I'm going to ignore the main purpose of this interesting study, which used magnetoencephalography to compare auditory evoked responses in autistics and nonautistics. Instead I'll concentrate on how this divided sample of autistic children was characterized.

Apart from language scores, the authors reported what could be considered verbal and performance intelligence, as per indexes from the latest child version of the Wechsler scales. Here the non-language-impaired autistic children scored significantly higher, with ~20-point and ~16-point advantages, respectively, over the language-impaired autistic group.

As I wrote above, Roberts et al. (in press) also reported SRS scores, described here as "dimensional symptom severity ratings."

SRS raw scores are commonly used in research, but the SRS has also been standardized, providing T-scores which account for variables such as gender and differences between raters (e.g. parents vs teachers). T-scores are standard scores with a mean of 50 and a standard deviation of 10.

While the SRS is not yet considered a diagnostic instrument, a T-score of 60, one SD above the mean, is currently the SRS-based cut-off for an autistic spectrum diagnosis. Standard scores from 60 to 75 currently indicate SRS-based "mild to moderate" autism; 76 or higher is the SRS-based "severe" autism range.

As might be predicted, the two autistic groups in Roberts et al. (in press) differed significantly on SRS T-scores; that is, they differed on autism "severity." One group scored on average 70.89, falling into the "mild to moderate" range. The other scored 81.44, a full standard deviation higher, and crossed the threshold into "severe" autism.

As wouldn't be predicted at all, the non-language-impaired autistics, who had advantages not only in language but on measures of intelligence, were the "severe" autistics, while the language-impaired children were the "mild to moderate" ones.

My wild guess is that if an autism intervention was shown in fair tests to consistently produce a reduction in SRS scores of a full standard deviation, and what's more, transported groups--however small--of autistic children from the "severe" to the "mild to moderate" SRS range, this would be considered tremendous progress.

The sample of autistic children in Roberts et al. (in press) is of course very small, as is charateristic of the bulk of the autism literature, and much can happen within such small sample. But these data aren't inconsistent with numerous other existing reports, and deserve at least a look while we wait, patiently, for the DSM-V people to proclaim on autism "severity."

You can find an uncritical short blurb about the SRS, which used to be called the "Social Reciprocity Scale," here, and a more extensive, critical description here.

Many thanks to Jennifer Stevenson for (patiently) answering my questions about the SRS. Any factual errors are entirely mine and should you find any, you should alert me immediately.


Roberts TP, Khan SY, Rey M, Monroe JF, Cannon K, Blaskey L, Woldoff S, Qasmieh S, Gandal M, Schmidt GL, Zarnow DM, Levy SE, & Edgar JC (2010). MEG detection of delayed auditory evoked responses in autism spectrum disorders: towards an imaging biomarker for autism. Autism research : official journal of the International Society for Autism Research PMID: 20063319

Postscript: This post has been included in the 21st edition of Scientia Pro Publica.

Wednesday, February 10, 2010

Proposed new autism criteria: the DSM-V

Here they are and here are ten off-the-cuff thoughts:

1. The most sit-up-and-take-notice change is the total removal of Rett syndrome from the DSM. It is gone. The DSM-V people are saying, "genetic syndromes don't belong in our book," or words to that effect, and I agree.

2. Indeed the vast majority of named neurodevelopmental disabilities do not appear in the DSM, past, present, or future. This raises the question of why autism is there. Rett's being excluded is not going to immediately result in Rett's individuals being catastrophically deprived of recognition or assistance. Instead it may result in these individuals being regarded more accurately, to their great benefit. Removing autism from the DSM would have the same beneficial effect.

3. Another change to grapple with: CDD (Childhood Disintegrative Disorder) is now lumped in with autism, which in turn has a single vaguely phrased onset criterion. To cover CDD (Volkmar et al., 2005), that criterion will have to cover onset at age 5 (fairly common in CDD) and up to age nine (rare, but happens).

4. In addition, autism and CDD have very different cognitive profiles. This is one of many ways in which the DSM-V, even more than its predecessor, is running away from the productive and beneficial--to autistics--notion of autism as a cognitive phenotype.

5. The headline-making but most predictable--and most predictably responded to--change is the loss of Asperger's and PDD-NOS, which have both always been considered part of the autistic spectrum, as distinct-from-autism diagnoses. Whatever their shortcomings, the loss of these diagnoses is another signal that autism is, officially and more so than ever, merely a series of deficits in overt typical behaviour.

6. At the very least, the DSM-V strongly discourages any view of autism as an atypical cognitive phenotype involving relative (to nonautistics) cognitive strengths and weaknesses.

7. The changed criteria, which collapse the DSM-IV social and communication domains, overlook any role for manual and oral motor abilities in these two areas. And whose definition of the now-mandatory social reciprocity criterion will prevail? Here is John Constantino's one-way-street definition:
Reciprocal social behavior refers to the extent to which a child engages in emotionally appropriate turn-taking social interaction with others.
The closer-to-equal time, so to speak, now granted the previously-relegated RIRB (restricted interests and repetitive behaviours) domain could be seen as progress, ditto the disappearance of the "nonfunctional" assumption. But autistics will no longer have DSM-IV unusually focused and intense interests (a strength), we will instead be pathologically fixated.

8. The vagueness and subjectivity of the criteria should later be elaborated on with explanatory text which may add or subtract clarity. And the highly-publicized DSM-V effort to rank and classify all autistics according to a rigid hierarchy of "severity" is as yet missing. How the DSM-V will handle aspects of the by-far most important distinction within the autistic spectrum--the idiopathic vs etiological autism distinction (and see genetic syndromes, above)--remains to be seen.

9. While the DSM-V has enormous political clout, what might change in actual diagnostic practices is unclear. Clinicians and entities currently employing anything-goes or free-for-all and/or expedient-type standards are unlikely to change in this respect. And in many ways DSM-V autism is autism altered to conform to the current "gold-standard" autism diagnostic instruments (see the role of Catherine Lord in both), whose predominance, weaknesses and limitations have come to determine what autism is and isn't.

10. Organizations which (a) have little use for basic autism research (the kind I'm involved in) or actually oppose it; and (b) promote political views of autism, including that more services are always better--will likely be happy with the DSM-V changes. You can see ASAN fulfill this prediction here.


Volkmar, F.R, Koenig, K., & State, M. (2005). Childhood Disintegrative Disorder. In: Volkmar, F.R., Paul, R., Klin, A.,Cohen, D. (Eds.), Handbook of Autism and Pervasive Developmental Disorders . Hoboken, NJ: Wiley.

Postscript: Another view of the new autism criteria is here. An overview of the proposed DSM-V in many areas is here.

Sunday, February 07, 2010

Very early autism intervention: the Early Start Denver Model
Early autism intervention research carries consequences for all autistics, and for a lot of nonautistics as well. DawsonG et al. (2010) is a newly-published randomized controlled trial of a newly-developed manualized very early autism intervention and as such deserves a close look.

Developed by Sally Rogers and Geraldine Dawson, the Early Start Denver Model has been the subject of a handful of previous papers, none involving anything approaching a major trial. ESDM is reported to combine the Denver Model with Pivotal Response Training, two interventions developed in the 1980s which individually lack good quality evidence for their benefits or harms. If you are interested in ESDM and its possible fall-out, I suggest reading the existing DM, PRT, and ESDM literatures.

A book chapter, possibly written circa 2006, describes the ESDM RCT, apparently before all the results were in (SmithM et al., 2008), and is essential reading, as is the riveting recently-published ESDM manual (Rogers & DawsonG, 2010).

My first exposure to the foundations of ESDM was a 2003 presentation by Sally Rogers. She showed a video of an adult picking up two blocks and banging them together and putting them down. In response a very young nonautistic child automatically picked up the blocks, banged them together, and put them down. A very young autistic child immediately noticed that the blocks were very interesting-shaped (and really they were--these were incredibly interesting blocks), picked them up, and inspected them closely seeing as they were so darn interesting.

You can guess which child's response was deemed wonderful and praiseworthy and "right," and which child's response was deemed defective and unfortunate and totally wrong. With this in mind, here are some comments:

About the ESDM

1. According to current definitions, ESDM is an "eclectic" autism intervention, combining ABA-based with non-ABA-based approaches. The manual specifies that ESDM "most closely resembles" RDI, DIR/Floortime, Hanen, responsiveness training, and SCERTS. ESDM also "has clear ties to" PRT, incidental teaching, and milieu teaching. And ESDM "has in common" some elements of ABA-based interventions as per Lovaas. Speech and occupational therapy are also thrown in. PROMPT too. The wide range of intervention approaches constantly available to children receiving ESDM is emphasized in all its available descriptions.

2. On principle and on the basis of small poor-quality trials, major behaviour analysts have claimed that "eclectic" autism interventions are ineffective; two recent examples here and here. The Association of Professional Behavior Analysts has thoroughly condemned "eclectic" approaches to autism, to uncritical applause by the Association for Behavior Analysis International's autism special interest group.

3. The reported overarching theoretical basis for ESDM is a 1985 book by the well-known psychoanalyst, Daniel Stern. Other psychoanalytic influences have been reported for the DM component of ESDM.

4. ESDM is further premised on autism being (from the manual) "at its heart a social disorder." Autistics' profound (entirely one-sided) failures in displaying proper interpersonal relationships and proper social motivation produce a cascade of purely negative effects disrupting all areas of proper development and learning.

5. The ESDM goal: autistics must become as "normal" (from the manual) as possible. Every waking hour, autistics must work towards the one ESDM-recognized proper way to relate, learn, play, develop, etc.

6. Autistics' tendency to learn in atypical ways is treated as though disordered and harmful--as evidence for our primary social deficits, which in turn deny us all the proper, typical "crucial learning experiences" (e.g., "birthday party games!"--from the ESDM manual).

7. The only proper ESDM way to learn is through typical interpersonal interaction involving proper, typical use of developmentally-proper highly-rationed highly-controlled materials which must only be regarded and manipulated by the child in the one proper socially-related ESDM manner.

8. Keeping information away from autistic children who are persistently at risk of learning from it in non-ESDM-approved ways is a theme throughout the ESDM manual. While it is claimed that ESDM does not require a therapy room, in fact there are many instructions as to how such a room should be set up. This is representative: "Ideally the room should be able to be arranged with nothing in it except a table and chairs and a closed or covered cabinet or shelves." This is called "the natural environment."

9. While a great advantage of ESDM is reported to be that it is "fun to do!" in fact the manual sternly declares, when contemplating autistics who may enjoy learning in the wrong non-ESDM way: "Our job is not to keep children happy; it is to teach them their objectives." And that very limited hierarchical range of objectives must be rigidly achieved in the one rigid proper manner and the one rigid proper order.

About the design of the ESDM RCT (DawsonG et al., 2010)

1. The trial spanned 2003-2008 and was registered more than a year after it started. Many of the reported planned measures (see SmithM et al., 2008), particularly those related to brain activity, go unmentioned in DawsonG et al. (2010).

2. Basic description: 48 autistic children within the age range of 18 to 30 months (actual range is not reported) were randomized to two groups, to recieve either a minimum of 2yrs of ESDM or of services available in the community. Children were evaluated with four main measures (Mullen, Vineland, ADOS, a scale of repetitive behaviours) and according to their diagnostic status, after about one then two years. At the two-year evaluation, children had received an average of 29 months of intervention. Children had to be at least 48 months at the 2-year evaluation; therefore children younger at intake received more intervention than children older at intake.

3. Numerous and strict, but not atypical, exclusion criteria were applied. For example, children considered to fall into the "severe" and "profound" ranges of intellectual disability were excluded, as has been done in many popularly-cited ABA controlled trials.

4. In this trial at least, the ESDM is not an intensive intervention. The authors aimed for 20hrs/wk, but achieved only 15.2 (standard deviation 1.4), which falls into the range of low intensity as currently defined.

5. Parents were expected to apply ESDM principles for an additional 5hrs/wk, but in fact reported (how accurately is unclear) more than triple this amount, indicating great enthusiasm and high expectations. ESDM parents also reported using non-ESDM services for about 5hrs/wk.

6. By the rock-bottom standards of the ambient autism literature, the ESDM trial design has many notable strengths. The intervention is manualized and attempts were made to establish treatment fidelity for the ESDM group. Currently-recognized gold-standard diagnostic instruments were used. By non-autism non-ABA standards, this is a very small trial, but it improves on an abysmal literature rife with even tinier trials. Most important, against the strong trend denying autistics even the possibility of good experimental design, this is an inexcusably rare RCT.

7. However, by very well-established standards in non-autism non-ABA areas, the ESDM RCT is in many respects poorly reported, which makes it difficult to properly asssess its design.

8. The control condition lacks virtually any useful description much less a rationale. Unspecified individuals provided the control group parents with unspecified information (what "resource manuals"? what "reading materials"?) including about unspecified services. The authors briefly trot out a few generic types of available services without indicating whether they were recommended or chosen. Children in this group received on average ~9hrs/wk of unspecified individual therapy plus the same of unspecified group-level therapy. Only mean intensities are reported--no SDs much less ranges. We don't know whether some children received huge amounts of intervention while others received none.

9. In my view, the authors' documented astonishing lack of interest in their own control group speaks to the very human expectations and biases applied to this group, which would be unlikely not to affect their outcomes.

10. Crucial information about who or what generated the randomization sequence, and whether or how allocation was concealed, is missing.

11. Remarkably, 51 autistic children were originally randomized to the two groups. Three disappear off to the side somehow, no explanation given, after random assigment to groups but without showing up in either group. Two assigned to the ESDM group rejected this intervention due to its requirements. The third, of unreported group assignment, was found to be not autistic, but Rett's. But none of these randomly allocated children shows up in their assigned group.

12. Even with the 48 children deemed to be assigned to a group, there is no intention to treat analysis. Data were ultimately lost from 3 children in the control group, no reason given.

13. The pre-post use of the ADOS as a scale of "severity" raises numerous questions. The first preliminary attempt at creating such a scale was not accepted for publication until the end of 2008 (and not epublished until 2009). This scale is further based on the revised ADOS algorithm, which is not cited by DawsonG et al. (2010) and was not accepted for publication until mid-2006.

14. Blinding is notoriously difficult in these kinds of trials. Parents cannot be blinded to intervention group, but some measures rely on parent report. While evaluators are described as "naive to invervention status," they are not described as blind to the nature and purpose of the study or of ESDM. Those involved in providing ESDM would in addition be aware of the study's purpose, importance, and outcome measures.

About the reported results of the ESDM RCT (DawsonG et al., 2010)

1. The ESDM manual provides results claimed to be from DawsonG et al. (2010), which is referred to both as "in press" and as a published 2010 Pediatrics paper. Inexplicably, information and data in the manual reported as being from DawsonG et al. (2010) differs somewhat from what is reported in the paper. There are also contradictions about the RCT within the ESDM manual regarding, for example, number of children in the ESDM group and intensity of the intervention they received.

2. There is an obvious error in the data in DawsonG et al. (2010). I spotted it immediately on first reading, as should anyone--you don't need to be able to do math in your head--who actually looks at the data. For two of the Vineland subscales (daily living and motor skills), some data are apparently reversed from Table 1 to Table 2, such that data in one table does not correspond to data in the other. The persistence of such an obvious error through copy-editing and so on raises questions about whether there are less obvious errors.

3. The main result is in the Mullen composite. On this score, the ESDM group finished at ~79, representing an ~18 point change from baseline but almost no change from the first year score of ~76. There is a just-significant 11-point change score advantage for the ESDM group over their controls at the two-year evaluation, an advantage which diminished from first to second year evaluations. The Mullen ESDM advantage is concentrated in the expressive and receptive language subscales.

4. Most striking re the Mullen composite is the huge increase in the scatter of ESDM scores. First and second year SDs are ~23 and ~24, much higher than baseline (SD ~9) and much higher than in the control group. Regardless, ranges of scores are not provided. But clearly one effect of ESDM is to dramatically increase variability of outcomes at a very young age. The ~11-point year two ESDM group advantage is drowned in their ~24-point SD.

5. While the ESDM group outperformed their controls on the Vineland composite, this results from the control condition reducing Vineland scores. These scores depend on unblinded parent report, which would plausibly reflect the disinterest and low expectations, obvious in how this study is reported, conveyed and applied to the control group. The ESDM group, regardless of their important advantages in this respect, did not experience any improvement in their Vineland scores over two years.

6. In addition, the ESDM group's Vineland scores show a huge increase in scatter, not found in the controls, by year two. The SD balloons from ~6 to ~16 (and from ~8 to ~22 in the communication sub-scale). Again, the ESDM's effect is to greatly increase variability in outcomes.

7. The authors dedicate a lot of space to claims of group differences in subjectively-determined diagnostic status. Those changes were limited to moves between autism and PDD-NOS--none of the children were judged nonautistic. However, on the ADOS, the only reported quantitative measure which offers the possibility of differentiating PDD-NOS from autism, the groups did not differ in outcome. Also, diagnosis was remarkably stable in this study, compared to previous findings regarding children in the 30-month-or-younger age range.

8. Apart from not differing on ultimate ADOS "severity" scores, the groups did not differ on a repetitive behaviours scale. The ESDM did not make autistic children measurably "less severely autistic" by the usual standards.

9. The premise of ESDM is that only through displaying proper typical social interaction and behaviour can autistics begin to learn anything worth learning. In the two reported measures of social abilities (ADOS, Vineland socialization), the ESDM children did not do any better than their controls at either evaluation. They did not improve (by the usual standards) pre-post, with losses in Vineland socialization.

10. The ubiquitous claim that "earlier is better" accompanying the publicity for this study is unjustified. The authors made no attempt to correlate outcomes with their 12-month possible age-range at intake. To the contrary, the authors enacted the view that younger children at intake must receive more intervention in order to be judged alongside children who were older at intake--the contrary position to "earlier is better."

11. The ESDM can only be used up to 60 months, with objectives targeted at children up to 48 months. At the end of this RCT, the ESDM children had a possible age range (actual range not reported) from 48 to 54 months. All were still autistic or PDD-NOS. Given the enormous increase in scatter of Mullen and Vineland scores, some had done extremely poorly by the usual standards. Against the unfounded claims of "earlier is better," questions arise as to how being so judged at such a young age within a highly publicized intervention will affect these children.

12. In reporting the ESDM RCT, the ESDM manual states that "ESDM needs to be independently replicated before it can be considered an empirically supported treatment." But a recent paper whose senior author is Sally Rogers claims that ESDM is "evidence-based" and that its components are "empirically supported."

A few last comments

Eric Fombonne has published a Pediatrics eLetter about the ESDM RCT. He concurs with some points I make, not with others, and he failed to spot the error in the data.

If you rummage around in the site you will find an in-progress multi-site RCT of the ESDM where intake age range has been dropped to 12 to 24 months. You will also find a multi-site RCT, targeting 5-12 month old children considered "at risk" for autism, which will likely use ESDM-type intervention in an effort to prevent autism.

As a large body of work, the ESDM manual, RCT, and other related materials, including the formal and informal work of Geraldine Dawson and Sally Rogers, read like a total condemnation of what I've done in the above. I'm not supposed to notice details and errors, contradictions and anomalies--or if I inappropriately do, I should forget about them and concentrate on emitting socially-appropriate responses (pick up the blocks, bang them together, put them down--don't look at them). I'm not supposed to notice other discrepancies, for instance between recognized scientific and ethical standards and those imposed on autistics. I'm supposed to direct all my effort towards the developmentally-appropriate equivalent of "birthday party games!"

The above should be verifiable, through reading the supplied sources. As usual, if I've made any factual errors, always a possibility, please let me know.


Dawson, G., Rogers, S., Munson, J., Smith, M., Winter, J., Greenson, J., Donaldson, A., & Varley, J. (2009). Randomized, Controlled Trial of an Intervention for Toddlers With Autism: The Early Start Denver Model PEDIATRICS, 125 (1) DOI: 10.1542/peds.2009-0958

Rogers, S.J., & Dawson, G. (2010). Early Start Denver Model for Young Children with Autism. New York, NY: Guilford Press.

Smith, M., Rogers, S., & Dawson, G. (2008). The Early Start Denver Model: a comprehensive early intervention approach for toddlers with autism. In: Handleman J.S., Harris S.L., eds., Preschool Education Programs for Children With Autism, 3rd ed. Austin, TX: Pro-Ed, pp. 65–101.