There's little evidence that antidepressants are useful in reducing repetitive behaviors in autism - but there is evidence of bias in the published literature. That's according to Carrasco, Volkmar and Bloch in an important report just out in Pediatrics: Pharmacologic Treatment of Repetitive Behaviors in Autism Spectrum Disorders: Evidence of Publication Bias
They looked at all of the published trials examining whether antidepressant drugs (mostly SSRIs, like Prozac) were better than placebo in reducing repetetive behaviours in children or adults with an autism spectrum disorder (ASD).
A meta-analysis showed that there was a statistically significant benefit of the drugs overall, but it was marginal, with a small effect size d=0.22, and it was driven mainly by two old, very small studies that found big benefits. One of them only had 12 subjects. By far the largest study, King et al (2009) with 149 people, showed zero effect.
This plot shows all of the studies, with the red line being no benefit of drug vs placebo. The further to the right of the line, the bigger the benefit, but the grey horizontal lines show the uncertainty. As you can see, two small, messy studies found big effects, the others didn't.
Worse yet, although there were 5 published studies, the authors also found that there had been 5 studies that had been completed, but never published. Carrasco, Volkmar and Bloch wrote to the people in charge of those studies and asked for data; only one out of 5 replied. The data showed no benefit.
We don't know what the other 4 unpublished studies found, but the way science works means they probably came out negative. If we assume that they did, then even the small benefit seen in the published studies disappears.
This paper, incidentally, is great example of why trial registration is a great thing. Without mandatory pre-registration on clinicaltrials.gov, no-one would know about the 6 unpublished trials at all. It would have been even better if the researchers had been forced to make public the results, as well as the existence, of the unpublished trials; but it's a lot better than nothing.
Finally, the authors of this paper stress that this doesn't mean antidepressants don't help at all in autism - just that they probably don't help with repetitive behaviors.
Carrasco, M., Volkmar, F., and Bloch, M. (2012). Pharmacologic Treatment of Repetitive Behaviors in Autism Spectrum Disorders: Evidence of Publication Bias Pediatrics, 129 (5) DOI: 10.1542/peds.2011-3285