Research shows that lead paint is A-OK!

I am very late to this party, but I am stopping by anyway. I started this post the day I read Judith Warner's screed, but it takes me forever to write anything non-bullet-point-y. Also, becoming a doctoral candidate has had an unexpected effect on me -- I find myself thinking thoughts like, "Should I really hit publish before I understand how they calculated the area under the curve in the lead study? Do I need to be clearer that neither of the studies I'm talking about can claim to be the last word on the topic?" Which is probably a little silly, so here's the post I wrote.

Judith Warner has jumped on Hanna Rosin's bandwagon, where the tune they're playing goes something like this: "I have zero public health credentials but I'm going to broadcast my ignorance to the widest possible audience anyway." There are already plenty of bloggers reacting to Rosin's piece, so I'm going to focus this on one maddening misperception in both pieces: the idea that breastfeeding has no significant impact on cognition.

Rosin interviewed Michael Kramer, the Canadian physician who has done something really cool in Belarus: he found a way to include randomization in a study of breastfeeding. (You probably know already that this is a limitation of many studies of breastfeeding. Women who breastfeed in this country tend to be educated middle-class women, and it's hard to untangle the advantages conferred by breastfeeding from the advantages conferred by those other confounding variables.) Kramer found more than 17,000 women who were planning to breastfeed and randomized them to different healthcare providers: either hospitals and clinics where the staff had been through Baby Friendly training, or hospitals and clinics without BFHI certification. Clued-in HCPs make an enormous difference to mothers who hope to breastfeed, as confirmed by Kramer's findings on breastfeeding exclusivity and duration. He's been following the families for 7 years now and has published a number of articles about his results.

Because of that randomization, his work is untroubled by the problems that plague observational studies. At the same time, the great strength of his study is also a limitation: every baby in the study was breastfed. It is an examination of the impact of less-breastfed vs. more-breastfed. It is not designed to describe the effects of breast vs. bottle.

That makes it all the more startling, then, to read his article on cognition [.pdf link to full text], in which the more-breastfed kids outscored the less-breastfed kids: 7.5 points for verbal IQ, with smaller effects observed in other domains. It's not about their mothers' vocabulary or education -- that's where the randomization is important. I have read at least a dozen articles looking at this question, and another few dozen looking at closely related questions, and I find Kramer's results both astonishing and persuasive.

Rosin, despite interviewing Kramer at some length, dismisses his findings as "a bump." Warner, even more egregiously, describes the effect of breastfeeding on cognition as "a more or less meaningless couple of points."

Let's be very clear: half a standard deviation is not a bump. If you call that "more or less meaningless," you demonstrate incontrovertibly that you're not qualified to write about it. In the annals of environmental influences on cognition, half a standard deviation is a big deal. An example: suppose you live in an older home with some paint in poor repair. Let's say your baby's blood lead level is 3 when he's 6 months old. It spikes way up at a year, because he's been crawling around in the lead dust without your realizing it, to 21. This is a nightmare. The health department gets involved. They tell you to do $30,000 worth of lead abatement work to your home, that you can't have a child living there. You blanch. You panic. You work out a compromise involving sealing the windows shut, cleaning and washing hands like a maniac, and continuing to monitor the blood lead levels. They fall: at 18 months the level is 15, at 24 months it's 10, at 30 months it's 7, at 36 months it's down to 4. Everyone is relieved but still a bit concerned. (Please note that I have simplified, perhaps oversimplified, the calculation of a mean blood lead level.)

The reason for all this hullabaloo? 7.4 IQ points. That's right-- less than the effect observed by Michael Kramer. And hey, Rosin and Warner have told us how little that matters. Bring back the lead paint!

No, wait, don't. Because while 7 points may not matter much to an individual -- I'd never claim that I could judge reliably between a person whose IQ was 97 and a person whose IQ was 104 -- it makes a whopping difference in a population. As Philip Landrigan (of Mt. Sinai's Children's Environmental Health Center) has observed, “a 5-point drop in population mean IQ reduces by 50% the number of gifted children (IQ above 120) and increases by 50% the number with borderline IQ (IQ below 80). These are differences that have profound consequences for the intelligence and the accomplishment of a society.”

In any discussion of breastfeeding it is imperative to separate the conversation about individuals from the conversation about populations. Hanna Rosin doesn't want to breastfeed? Her baby, her breasts, her call. She gets no grief from me. Hanna Rosin wants to publish a sloppy 5000-word dismissal of the wide array of epidemiologists who have concluded that human babies do in fact do better when they are fed human milk? I call foul. "I don't want to" ≠ "so you shouldn't bother." "I am seething because my husband has no breasts" ≠ "it's overrated anyway." And "I hated pumping" categorically does not mean that no one should do it, as Warner concludes. (All you NICU babies, you can just deal with the bowel necrosis. What doesn't kill us makes us stronger, you know. We wouldn't want to offend Rosin's delicate sensibilities with a piece of equipment that reminds her of Josef Mengele, after all.)

When I had one baby I thought it was simple: everyone ought to breastfeed. What broadened my perspective was the process of lactation consultant certification. (I find this ironic, since LCs are the HCPs most likely to be labeled, sight unseen, as nazis.) There are so many reasons why women don't breastfeed. I will never assume that I know how another woman should feed her baby. My job is to be her ally, not her judge.

At the same time (here's that individuals/populations thing again), from a public health perspective I find the evidence for breastfeeding too compelling to disregard. Kramer's work says to me that women should be clamoring for breastfeeding-savvy HCPs. (Yet another irritating inaccuracy from Rosin's article-- the intervention he studied was not about "strongly encouraging women to nurse" but about providing adequate support to women who already wanted to do so.) Rosin looks at Kramer's work and says, "Meh. Doesn't matter." In doing so she has made not a case against breastfeeding but a case against her own journalistic competence.

Comments

[Lilian]

YAY! You did post it! Thank you so much for once again using all your scientific training and your best persuasive writing to debunk these toxic myths that these women in the media keep putting out there! What a level headed breath of fresh air! I'll be linking to you soon.

[Karen]

Beautifully done. thanks for the deep thinking and the clear writing. Please send to the NYT and the Atlantic Monthly. ASAP!

[rachel]

Thank you!

[JenA]

Love it, Jamie!!

[Luo Lin]

I've been waiting to read your response. Thank you!

[Karen]

Because of that randomization, his work is untroubled by the problems that plague observational studies.

Do you really think this? Wow. No, just no. It alleviated some of the confounding issue, but hardly all of them, and certainly not the ones that most people think are the driving factors.

[Jamie]

I am editing this comment a few weeks later to clarify a couple of things. I considered writing about the covariate question in this post, but I decided not to given the length of the post and its intended audience. Kramer assessed for between-groups differences in maternal and paternal education and didn't find anything significant. It is certainly possible that factors other than the intervention Kramer was studying explain the cognitive outcomes he described. It seems more plausible to me, however, that Kramer's conclusions are correct: breastfeeding plays a part in cognitive development. The remainder of this comment is my original reply to Karen, whose comment appears immediately above this one.

"Untroubled" is too sweeping, but the scale of the issue is much, much smaller in a randomized study than in an observational study. What concerns do you have about Kramer's approach to covariate control? It's not as if he ignored the issue.

I also haven't seen published responses taking Kramer to task for any failure to address important confounding variables. Der & colleagues wrote in, unsurprisingly, but I think their focus was more on the difficulties of assessing cognition in a multilingual population. I hope you'll write to the editor if you think Kramer has overlooked something important.