This whole study has a bit of a hokey feel to it and I wouldn't put much credence in it.
First of all, it involved only a total of 110 kids and they knew, going in, which would be more likely to develop autism.
Then, after they were all diagnosed with autism, the researchers went back and looked at their data and then identified the pattern "predicting" autism retroactively.
I'm not saying it's completely full of crap, but I'm reserving a good deal of skeptism. Let's see some success actually predicting autism in children, rather than going back and finding the indicators after the diagnosis.
It’s always good to be skeptical before results are checked and replicated plenty of times. I don’t think the methodology seems too bad though. They collected data, waited to find out who the affected subjects would be, and then combed back through the data to find a common characteristic that may be a predictor. Seems solid, in broad strokes at least, to me.
As you say, though, the real test will be to see if they really can predict it now.