The main problem I see with the study design is that they used data from existing patients rather than recruiting fresh new patients. That suggests to me that they had already seen a few anecdotal examples of Chiari malformations in their existing group of known patients, and that's why they decided to quantify it and publish it. Why else would you choose to investigate association with Chiari malformation of all things? They never explain that in their paper. I don't think it was malicious - I can imagine noticing that you have a handful of patients with Chiari malformations and when you investigate all potential exposures, you find one that just happens to be present in the group. I'll bet if you did a study where you just checked random patients who you've never met before, you wouldn't find any association.
Another factor that leads me to suspect the same thing is that they found an 18% prevalence of Chiari malformations in their SSRI group and 2% in the other group. Those are both way higher than the general population prevalence of 0.1%. That suggests that even their "normal" patients were selected from an abnormal group. It also doesn't fit with our clinical experience... nobody has ever noticed an association here, which would be quite strange if Chiari malformations were anywhere near as common as 18%. I think that this is just an issue with confirmation bias... if you have hundreds of bits of information about 100 patients with developmental disabilities (which is how they found the data in this study), you'll be able to find a correlation between one of those developmental problems and some sort of prenatal exposure.
They even admitted in their intro section that they noticed a handful of patients with Chiari malformations in another study, and chose to include those data points in this particular data set even though it obviously biases the data. If you took 1000 random patients who you've never seen before, I'd be willing to bet good money that you won't find any increase in Chiari malformations over baseline.
