Diagnostic substitution and autism prevalence

I don't know if it is still part of the autism debate, but I remember a while back there was a lot of interest in whether or not diagnostic substitution was a factor in the increasing prevalence of autism spectrum conditions.
The argument went something like this: 20-odd years ago, children were diagnosed with intellectual disability (ID) even though they may have presented with either autism or autistic features. In more recent times, said children who would have been diagnosed with ID would now be classified as having autism or an autism spectrum condition as a primary diagnosis. The shift is due to either a greater awareness of autism, changing diagnostic criteria for autism or clinicians getting better and more accurately diagnosing autism.
I have to say that I always had mixed feelings about this argument. Mixed feelings because this would suggest that despite autism being included in the diagnostic manuals for quite a few years, clinicians were basically, pardon my language, either crap at diagnosing it during the 80s and 90s (perhaps into the noughties) or were so swayed by the diagnostic 'fads' of the time that they did not diagnose it. I know a few clinicians and have to say that I can't buy this being such a universal phenomena. If anything else what does this tell us about the criteria for autism being used?
On the other side of argument is the various research which has more than hinted at a reduction of ID diagnoses corresponding with an increase in autism diagnoses. It is not just in the US that this trend has been noted, but also in Canada. The 'Californa data' has been pivotal in this argument / debate.
The common consensus is that diagnostic substitution has probably contributed to the increase in autism prevalence (at least in the USA and Canada); as a percentage roughly ranging from anywhere between a quarter to about a half of cases.
I say all this because an interesting paper has emerged on PLoS ONE titled: Autism and intellectual disability are differentially related to sociodemographic background at birth. The full-text of the paper is happily available here.
You are right - the paper does not, from the title, seem like it is going to deal with the issue of diagnostic substitution, but look further, particularly at Figure 1 and there are some interesting numbers to crunch.
I think it is worthwhile stating at this point that this paper looked at Australian trends in diagnosis so we cannot really say too much other than this is what happened to Australian trends. I have talked about Australian prevalence trends in a previous post.
From Figure 1: mild-moderate ID diagnostic prevalence (without autism) by birth year peaked in 1992 and by 1999 was quite a few orders lower. Severe ID (without autism) by contrast seemed to show a quite unstable pattern throughout the whole period of study (1984-1999) with lots of peaks and troughs.
Autism (with or without ID) grew in prevalence up to 1996. At that point autism and no ID seemed to drop off (I wonder if this was a change point where Asperger syndrome became more 'fashionable' a diagnostic label to use?) whilst autism with ID continued its steady climb upwards.
The conclusion: well it is a complicated picture. It does appear as though there are some 'opposite' trends in ID and autism diagnoses but the relationship is not completely straight forward.
This is of course just my own interpretation of the figures; complete with my own biases et al. Others have their own take on the data and results. I would encourage readers to take a look at the figures themselves and draw their own conclusions on this very complicated part of the autism debate.