A Facebook friend “shared” an article from the New York Times this morning, and I decided to take the bait. After a couple hours’ work, I wrote the author and the public editor regarding the misleading headline. Here is my letter:
Dear KJ Dallantonia (and Margaret Sullivan),
Facebook is abuzz with your post. Your headline is being “shared” hundreds of times. Readers see the headline and assume it states the facts.
But your headline does not state what the original research article says. Autism has increased from 3 in 10,000 in 1980 to 1 in 88 today. That’s about 35-40 times as many cases. You correctly note that the research article is “cautiously worded.” That was a red flag to me, and I read the original article. There are a couple points of which you may not be aware:
First, in statistical studies such as these, the use of a huge database (as in 100,000 people) creates a problem: Statistical regression is such a powerful tool that it can find very tiny associations with such large datasets. In this case the correlation was so small (less than .08) that we would normally ignore it, even when the results are “statistically significant.” As a colleague of mine was fond of saying, “Just because it is significant doesn’t mean it’s meaningful.” A correlation of .20 would “explain” 4% of cases; .30, 9%; .40, 16%–and this assumes that causality has been demonstrated (that it can be shown that A causes B).
Second, the conclusions you use in your headline do not reflect the authors’ statements, even in their summary conclusion. On facebook, after a friend “shared” the news, I read your article, then scanned over the research article, and I posted the following response:
The abstract notes: “Assuming an unconfounded, causal association, antidepressant use during pregnancy explained 0.6% of the cases of autism spectrum disorder.” This means that, if SSRIs DIRECTLY cause autism (which this study cannot show–the authors are careful to note that, for example, depression itself could be causing the effect), then, out of 167 autistic children, one case could be attributed to antidepressants. This is a problem with the use of regression in very large databases–associations can be found that are statistically significant, but so small as to be meaningless. In fact, it is possible to conclude from this that the dramatic rise in autism that we have seen recently is NOT (and could not possibly be) from antidepressant use.
When I returned to the research article, I noticed that they stated essentially the same thing in their summary conclusion: “However, assuming causality, antidepressant use during pregnancy is unlikely to have contributed significantly towards the dramatic increase in observed prevalence of autism spectrum disorders as it explained less than 1% of cases.” [italics added] Even for the authors to make the conclusion “unlikely to have contributed significantly” they must first “assume causality”–which they acknowledge they cannot do!
This is because “correlation does not imply causation” and this was a regression study (correlation is another form of regression). We have no evidence here that antidepressants cause autism, and this study cannot answer that question–it can only find a statistical association, or correlation. The association that has been discovered is so weak that, in my opinion, this study demonstrates conclusively that the 4000% increase in autism cannot possibly be from antidepressants. Well, maybe 0.6% of the 4000% if and only if SSRI’s “cause” autism. That’s all this study can and does conclude.
Your take on the article (especially your headline) is alarming to those who are convinced that pharmacological treatments are the root of all evil, and they repost and share without reading further, as my friend did this morning. My take on it is that, while we may need to be cautious, this is a spurious finding resulting from an overeager application of regression to a massive database, fueled by the “publish or perish” environment of academia.
I hope you will reconsider your headline and your writeup of this research.