I immediately had a bad feeling about this story re. autism diagnosis via brain scans which was reported by various media outlets including the Guardian earlier this week. Firstly, as many different, unrelated causes can lead to autism-like phenomena, a sample of 20 adults with autism is much too small. They may cover a non-representative sample of the autistic spectrum, and thus not tell us much about autism in general. And if the brain scan is so simple and easy, why haven't the researchers scanned 2000 people with and 2000 without autism?
Secondly, the statistical "success" of the study isn't all that good, as is explained here.
Oh, and lastly, looking at adults with autism and claiming the results can serve in early childhood diagnosis is just ridiculous. Look at children before they show worrying behaviour, and come back when you can predict which ones will develop autism.
Which brings me back to something that I've said before (on twitter, I think): Rather than paying Ben Goldacre to make fun of the competing media outlets that misrepresent science, the Guardian should pay him to read their own science stories before they get released. Or, maybe if they find the time and funds, they could do both. But the quality control would be a much greater public service. The people who get misinformed by bad science stories (including those appearing in the Guardian) will not normally read Goldacre's column to correct their information.
In other words, preventing the disease from spreading is much more valuable than pointing a finger at the people affected.