The conclusion section of that very study says "...further research is needed to confirm these associations and determine causality and mechanisms" and "A causal relationship is plausible..." [emphasis added].

In other words there is an association, but the study is not able to prove (or even suggest) causation. For example, it does not exclude the possibility that other factors that actually cause autism and Tylenol use are themselves linked. So Tylenol use could be correlated with autism but not a cause of it. In that case, pregnant women who would otherwise use Tylenol not doing so are not reducing the chance of autism. And as the study points out, failing to treat conditions that warrant Tylenol usage can also have negative pregnancy outcomes.

I'm extremely suspicious of anything that looks at 'an increase in autism rates' without considering the same factors that have gone into there being 'more' left-handed people over the past 50 years or so. There are a lot of people in the past who 'weren't left handed' because they were punished as children if they showed left-handed behavior, and that's still a whole lot less stigmatized than autism.

Has the study stratified for improved diagnosis? When I was a child in the 80's teachers weren't looking for the telltale signs, but you better believe there were autistic poeople, they just weren't diagnosed.

> If you think about it though, someone knowledgeable enough to write this paper is exactly the kind of person you'd want to serve as an expert witness.

Maybe, but the same was also true of the now disgraced Andrew Wakefield, although his conflict of interest was even greater since he didn't reveal his funding before publishing his original paper.

Admittedly I haven't read the whole study. But how do they account for the confirmation bias in their review – the fact that null results typically aren't published. Studies that look at an intervention (Tylenol) and fail to find an association are less likely to include it as a variable in their model or less likely to publish a null finding.

Reading the paper, I'm wondering why they didn't do an actual meta-analysis, to estimate average effect size or likelihood of publication bias. As it is, it's more of a systematic review.

The studies in general they include are case control and prospective cohort studies, predicting neurobehavioral outcomes from paracetamol use retrospectively or prospectively.

The most interesting ones to me are siblings control studies where they compare siblings with and without exposure or case status to control for unmeasured confounds like genetic or family environment variables.

In those studies they reviewed there is still a link but it's much weaker, mostly limited to mothers using paracetamol for a month or more, and on measures not necessarily reflecting autism per se.

That pattern to me is equally suggestive of something other than paracetamol being the causal factor. It could be reflective of a dose response relationship, but you also have to wonder about what else might be going on among women who feel compelled to take paracetamol for over a month at least during pregnancy.

Maybe a paper to call for further better research but not exactly a clear causal link.

> One objection that I've seen is that the lead author, Dr. Baccarelli, has a conflict of interest because he was an expert witness in a lawsuit about acetaminophen and neurodevelopmental disorders.

Every charlatan researcher grifting on bogus autism data really is just copying Andrew Wakefield's homework