Tuesday, 13 August 2013

The "link" Between Autism and Induced Labour Could Be Just Noise (and in any case is being reported poorly)

Again with the autism links. This time, a study out of Duke University (which is unfortunately paywalled) showed a statistical link between autism and induced or augmented childbirth.

There are, of course, a few qualifying comments. The study authors note that their result doesn't imply that induced or augmented childbirth causes autism, even some of the time. There could be other factors at work. As a hypothetical example, if there was an unknown foetal developmental issue that both caused an increased risk of autism and interfered with the signalling that starts labour (which is still not well understood), such an issue would explain the study results--especially since the authors did not separately consider mothers who had labour induced because they went past term and mothers who had labour induced at term or before for other reasons. Most importantly, the authors stress, in such hypothetical case it would be useless from the point of view of autism prevention and downright harmful from the point of view of general maternal health for parents to refuse a medically recommended induction on the basis of this study.

There's another issue that the authors don't address, and it's worth mentioning: the study could just be picking up noise. The study reports a number of different results for various models, but they all are just barely significant. For example, the odds of an induced only (as opposed to induced or augmented) baby developing autism are 10% larger than that of an un-induced childbirth--with a confidence interval of 9%. 10+/-9 isn't very precise; the confidence interval is almost as large as the effect. Further, the confidence intervals reported are 95% confidence intervals, meaning there is a 1 in 20 chance of seeing an effect that large by chance.

A 1 in 20 chance is fine for most studies (though, if you think about the hundreds of studies published each year, it means an appalling number are wrong through sheer randomness). It's dangerous, though, in look-back studies of this type. I've written before about the problems here, but it's worth summarizing again. The key point is that if researchers are working on a problem with many possible connections, 1 in 20 suddenly seems quite poor. The question that needs to be asked is, how many other connections with autism were, at the start of the study, as likely as that between autism and induced labour?

Or to phrase it another way, was there some reason to think a priori that induced labour was more likely to be linked to autism than, say, breastfeeding vs formula, or the mother's food choice, or early life air quality, or lack of any number of vitamins, or abundance of any number of vitamins, or certain types of stimulation, etc? If there's not, than 1 in 20 isn't very good. This would be obvious if you actually tested all the possibilities: if you did a study that looked at 40 different possible causes for autism and came back with a positive result for one of them, with a 1 in 20 chance that the result was just due to random fluctuations, that obviously wouldn't mean anything. Not checking the other 39 possibilities doesn't make your statistics better.

Of course, we know that look-back studies like this are the most popular type reported in the media, and we've seen that even the study authors were either unaware or ignored the statistics of such studies. So it's not surprising, though it's still disappointing, to see all of the all of the nuance and caveats go out the window in the way this study has been reported.

“Pregnant women who have procedures to induce or encourage labor might have an increased risk of bearing a child with autism, according to a new study,” reads the opening line of the Wall Street Journal's article. Readers who finish the article will learn that at best this a gross simplification, but assuming that readers who are both busy and lacking a science background will realize that the opening sentence here is at odds with the quote is either a naive estimation of the audience or a deliberate confusion for the sake of an attention-grabbing hook.

The write-up in the NY Daily Mail has this helpful caption below their related picture: "Among autistic boys in a new study, one third of mothers had labor induced or hastened, compared to 29% of boys without autism." This is head-scratching on multiple levels: most obviously, why compare a percent with a fraction? People hate fractions. Second, they're reporting the numbers in the form of a probability the mother had induced labour given that child has autism. But this is entirely backwards, since what we actually want is the probability a child gets autism given that mother had induced labour, relative to the probability a child gets autism under any circumstance. The former is a needlessly confusing way of reporting the numbers, given that the latter form is available in the study, and is what's actually discussed in the study's conclusions.

(As an aside, the NY Daily Mail also wins the look-back study reporting award, with the following "Related" links scattered throughout their article: "STUDY LINKS IVF TO SMALL RISK OF MENTAL DISABILITY"; "ANOREXIC GIRLS ALSO HAVE AUTISTIC TRAITS: STUDY"; "HIGH LEVELS OF AIR POLLUTION LINKED TO AUTISM RISK".)

We also have the BBC News report. I should mention here that the BBC News story was the most responsible of the ones I looked at in stressing the limitations of the study and the importance of consulting with your doctor about delivery decisions. In it, though, we learn that "Children whose mothers needed drugs to start give birth are slightly more likely to have autism, US researchers say." Beyond the caveat noted above, that this actually could be just noise, there's an additional, subtle problem. Suppose the statement itself is true. Then is the statement, "If you have an induced childbirth, your child is slightly more likely to have autism," true? The answer is no.

Another example is helpful here. Suppose you and your spouse both have dark hair. Suppose also that your parents both had dark hair, and their parents, all dark hair stretching back to the dawn of recorded history. Suppose this is true of your spouse as well. Now let's imagine that the two of you move to Sweden together for work, and while you're there you have your first child. As you're being taken to the delivery room, the doctor says to you, "Did you know that 85% of children born in Sweden are blond? Since your child is a child born in Sweden, there's an 85% chance he or she will be blond."

Obviously this is ridiculous; moving to another country doesn't change your DNA, and if neither you nor your spouse is carrying blond genes, your baby isn't going to get them, no matter where you live. So even though it's true that 85% of children born in Sweden are blond1, and it's true that your child will be born in Sweden, it's not true that your child has an 85% chance of being born blond.

The exact same structure carries over: Even if it's true that children whose mothers had induced childbirth are 10% more likely to develop autism, and even if it's true that you had an induced childbirth, it is not necessarily true that your child is 10% more likely to develop autism.

This last point needs to be stressed, because it would be tragic if any mothers or infants were harmed out of a belief that inducing labour increased the chance of the child developing autism.  I will most likely come back to it in another blog post, because the conflation of general odds with individual odds is at the heart of many of the most controversial misunderstandings we face.

In the mean time, while this blog doesn't exactly have the readership of the NY Daily Mail, I'll do what I can to spread what should be the take-home message here: at the moment there is no reason to believe that YOUR child is at a greater risk of autism if YOU have an induced labour. It's a message that would be clearer if news outlets and study authors were more focussed on getting the message right than on getting it out.

1Full disclosure: I have no idea what the actual rate of blond babies is in Sweden. It's just an example.

No comments:

Post a Comment