Why the "fractionable autism triad hypothesis" is worthless nonsense

The following is based on a review I wrote of a paper that was proposed for publication but which I argued was not worth publishing (and thus wasn't).

Review by Robin P Clarke for [journal name] of: 
“Partially distinct genetic and environmental influences….”

 If this paper should be published it should be for reasons other than those the authors had in mind.  The methodology and execution of the study appear to be competent.  But I find the theoretical / conceptual basis unpersuasive.  It might be considered predictable that I would take an unfavourable view given that I originated a significantly different theoretical position nearly 30 years ago, which was eventually published in 1993 (Clarke 1993), and that these authors have failed to mention that conflicting account.  But any dismissal of my comments should surely still have to be in terms of reasoning and evidence rather than just a notion that I’m biased against the authors’ position.

 The study takes its inspiration from the fractionable autism triad hypothesis, which burst upon the world in the cited Happé et al 2006 with later reinforcements in Happé & Ronald 2008 and Robinson et al 2011.  I’ll here abbreviate it as the FATH.  Like so many concepts in psychiatry it doesn’t allow easy summary in one sentence.  When handling such relatively complex concepts, careful presentation (communication) becomes important if readers are to be clear what writers are trying to say.  Sure, writers aren’t always fully sure themselves, as I have found myself.  But I am disappointed that nowhere in the literature of FATH have I found any clear statement of the various conceptual alternatives to be tested between (and there’s quite a potential muddle here).  So I’ll put here my own reckoning as follows. 

 A. The extreme descriptive FATH:  that there is no real autistic syndrome, but only several (three?) syndromes which have got incorrectly muddled together by Kanner and successors and preserved thereafter by the self-confirming definitional criteria for diagnosis. (This conflicts obviously with Clarke 1993.)

 B. The extreme causal FATH:  that there is no aspect of causality that is shared between all autism.  (This conflicts obviously with Clarke 1993.)

 C. The moderate descriptive FATH:  that while there is an autistic syndrome, it is only weak, and the sub-syndromes of communication, social and restricted/repetitive are more useful entities to focus on.

 D. The moderate causal FATH:  that while there is a causality shared between all autism, it is relatively minor and causalities separately acting on the sub-syndromes are the more useful to focus on.

 E. The Null Hypothesis counter to descriptive FATH:  that the autistic syndrome is a totally unified amorphous clustering with no sub-clusters that could be empirically partialled out.

F: The Null Hypothesis counter to causal FATH:  that all the causality of autism (and of autistic-like functioning) can be fully accounted for in terms of one unifying aspect to the exclusion of any causality not involving that aspect. 

 Like trying to saw through a piece of jelly, it is difficult to respond to the FATH literature because of its wobbly concepts, its rarely making clear which of these alternatives they have in mind at any particular point, and conspicuously wobbling between them at times.  I will therefore instead make an examination of the six positions I have myself defined above. 

 Firstly (F) the causal null hypothesis that all the autistic and autistic-like behaviour can be entirely accounted for in terms of one channel of causality (such as antiinnatia in my own theory).  We can dismiss this possibility quite easily for the following reason.  There will obviously be various genes such as those required for (e.g.) theory of mind, such that just one mutation or a critical combination of mutations of such “theory-of-mind genes” cause lack of theory of mind (or ditto in respect of language).  That outcome would obviously manifest as an autistic-like trait, and even be usefully diagnosed and managed as autism, even  though it would obviously not be the result of a unified causality of autism.  And I don’t think anyone has ever seriously advocated that null hypothesis – I certainly haven’t. 

 Secondly (E) the descriptive null hypothesis that there are no sub-clusters that can be separated out in the autistic syndrome, or in other words all the correlations between the various characteristics are all the same.  Well, put that way it already sounds a bit improbable.  I apologise for my short memory but I vaguely recall that back in the days of musty volumes on shelves, I read a paper by someone (Lorna Wing?) who did some such correlational study in the 1970s or 80s and identified two or three clusters.  Again, I don’t think anyone takes such a null hypothesis seriously anyway. 

 Thirdly (A) the extreme descriptive FATH, that there is no real autistic syndrome, but only three sub-syndromes accidentally muddled together.  The point about circularity of evidence resulting from prior definitions is well made.  But where did the unified notion come from in the first place?  The autistic syndrome has been—correctly or otherwise—discovered not once but three times independently:  by Kanner, by Asperger, and also about 60 years earlier by Dr Down (who gave separate descriptions of early onset and regressive forms but did not apply the autism label to them).  

 Sure, it remains appropriate that this impressionally-presumed unity be subjected to systematic query.  Unpicking of unstated axioms is an important factor in scientific progress.  But here it is tricky precisely because it is easy for even an incompetent to produce the low correlations that would hopefully constitute the evidence.  So the investigators, for their results to be worthy of credibility, must take extreme care to show that the low correlations are genuinely significantly low, and not in any way just the result of any defect of measures or method.  This is a very difficult hurdle to convincingly jump, and I’m not sure it is convincing done here.  It is just too easy to get the required low correlations by using unsound inputs.  The authors identify three “fractions” of autism, of the notorious triad of autism.  We here have to get mired into a whole further conceptual minefield here (gasp!), of what diagnosis is in psychiatry.   

 People assume that diagnoses are scientific concepts.  But they are not, rather they are adminstrative concepts.  For instance a scientific study might compare people living in India with people living in Spain.  These are administrative concepts and just likewise are those with or without autism diagnoses.  The human race can be divided into those who have diagnoses and those who do not.  But it does not divide into those who “have” autism and those who do not “have” it. 

 Back in 1976, Lorna Wing published a table of about 40 characteristics associated with the autistic syndrome (= Clarke 1993 table 2). These included some most peculiar ones such as toe-walking, flapping of hands, spinning without dizziness, pronoun reversal, echolalia of whole phrases, and peculiarly symmetrical, intelligent-looking, attractiveness of appearace.  When proud scientists can’t explain some facts, they find it embarrassing and end up pretending that they were probably not really true anyway, so all those wierd facts about autism have been conveniently swept under the carpet except by the author of Clarke 1993 who found them all too harmonious with the explanation in terms of antiinnatia.  Meanwhile due to the notoriously non-existent increase of autism, more and more thousands of health professionals and carers needed a simple explanation of what autism is and how to “diagnose” it.  So the same Lorna Wing (I think it was) created an autism-for-dummies formula which is the now-famous triad of impairments on which the FATH literature has based itself. 

 That diagnostic device is rightly used by researchers for the administrative function of assigning subjects to autistic v control groups equivalent to studies comparing people in India with those in Spain.  But I fail to see why it is rightly used by researchers as a basis for their theoretical concepts.

 The widespread use of that triad as a tool does not meant there is valid real clustering within its components.  Take for instance pronoun reversal.  Many would assume that to belong in the communication cluster.  And yet I see it as more likely to be a manifestation of lack of innate concept of self (as stated in my 1993 paper), hence nothing whatsoever to do with communication per se. 

At http://sfari.org/news-and-opinion/news/2011/twin-study-suggests-autism-traits-inherited-independently you can see the comments by Jon Brock which indicate the lack of real substance behind the inputs to that particular study.  I fail to see any clear improvement in respect of the present manuscript. 

 Furthermore there is the problem of reckoning to infer the nature of autism from facts about the non-autistic population.  (Btw, the title and abstract of this paper fail to make clear that this is not a study of autistic twins.)    The present authors make a worthy attempt to forestall this objection at page 15:

 “Autistic-like traits assessed dimensionally in the general population are thought to be relevant to understanding diagnosed ASD because most autistic trait measures show good validity (individuals with ASD score at the high extreme) and relatives of individuals with ASD show elevated scores on these measures. “

 But those points do not rule out a possibility that what happens in the general population is different from that in the autistic.  In fact, in terms of antiinnatia theory, we would *expect* that antiinnatia factors would be the most prominent causal factors within autism, but be far less prominent factors in non-autism simply because there the level of antiinnatia would by definition be low and hence non-antiinnatia causations more prominent. 

 In science we are almost always looking for significantly high correlations.  These FATH studies are necessarily looking for significantly insignificant, low correlations instead.  To make a persuasive case to the wider audience, the statistical means of establishing that needs to be made clear to the reader.  

 I have some other gripes about the background literature from which this study comes.  The original Happé et al 2006 began with the sentence “We argue that there will be no single (genetic or cognitive) cause for the diverse symptoms defining autism.”  But I had already shown in Clarke 1993 that there was good reason to dismiss any single-cause model, even to the extent of showing  bimodal distributions of social class which strongly challenge any single-cause model. 

 And: “At the cognitive level, too, attempts at a single explanation have failed.”  But in Clarke 1993 I had already explained how autism is a whole-body condition (something that Martha Herbert currently thinks she has just discovered), and that there would be numerous independent primary brain abnormalities rather than just one or two underlying all the behavioural features. 

 I’m finding it hard to find grounds to recommend the publication of this study, especially in a context of so many things to read already in the autism field.  I think the authors would do better to redirect their skills and talents to a modified study that abandons its focus on the artificial triad model and instead seeks out more-empirically-based clustering within the broad syndrome of autistic conditions. And rather than trying to show how low they can get at non-correlations rather they should try to test the limits of the already-existing unifying concept that is antiinnatia.  I should point out that there are subcausalities even within that antiinnatia, in that various factors (both env and gen) add to the antiinnatia, and then in turn that antiinnatia somewhat randomly may knock out function A in child B but function C in child D.  Please note that this random causal quality shatters the concept of “shared environment” because a shared antiinnatia factor may actually create divergent outcomes in the subjects.

 There is now a decisive wealth of evidence that mercury (specifically vapor from non‑gamma-2 dental amalgams introduced from 1976) is involved in a high proportion of autism nowadays, and has been the cause of (1) a tenfold increase of autism, (2) a tenfold change of ratio of early/late onset, (3) a change from lifelong incurable to sometimes very recoverable, and (4) a change from being mostly genetic to mostly environmental (for reasons explained in the 1993 paper).  I explain in a not-yet published update why mercury functions as an antiinnatia factor.  The mercury evidence is quite well reviewed by Geier Kern Geier 2010 Acta Neurobiol Exp once you partial out their wishful thinking about vaccines. It speaks volumes that the three posh studies supposedly acquitting mercury have all turned out to be whoppingly defective. 

 I think that autism research should now focus its fractionating on fractionating out the mercury cases from the non/low-mercury cases.  A lot of what is called autism nowadays appears to me to be more just mercury poisoning ancillary to the autism. 

I urge the authors to concentrate on the themes suggested above.  [.....]