What are the real benefits of breastfeeding? Statisticians weigh in.

A few days ago I pondered the ethical dimensions of breastfeeding given a recent article trumpeting its astounding benefits for infants and mothers. Those ethical considerations took as given that the claims trumpeting in the article were more or less true.

Today, I want to point you to an examination of those very claims by Rebecca Goldin (Director of Research, Statistical Assessment Service, Assistant Professor, Mathematical Sciences at George Mason University), Emer Smyth (Assistant Professor of Pharmacology at Univ. of Pennsylvania), and Andrea Foulkes (Assistant Professor of Biostatistics at Univ. of Massachusetts, Amherst). Will it surprise you that the data don't seem to support the conclusion that breastmilk has miraculous powers?

Here's just a taste of what Goldin, Smyth, and Foulkes found:

We start with the scariest question first: Will babies die if they are not nursed? According to the American Academy of Pediatrics (AAP), breastfeeding leads to a 21 percent decrease in the death rate of babies in an age range over one month and under one-year old.

But turn to the AAP's source. The scientific study used to support this claim found that babies who are nursed are less likely to die... of injuries!

Is this because breasts are softer than formula cans ... or is this one of those "don't mistake correlation for causation moments?

Many of the papers referenced were done in the 1980s, when medical care, daycare, and social contexts were significantly different from those today. Perhaps more importantly, studies done in the 1980s did not control for all the factors (such as whether the parents smoke) that we now know have an important impact on infant health. Other studies - especially more recent papers - simply didn't find what AAP claimed they did.

Let's take the article " Differences in morbidity between breast-fed and formula-fed infants," published in 1995 by the Journal of Pediatrics. This article tried to assess the health impact, if any, for affluent families in the United States. These were highly-educated families with access to medical care. The authors compared those babies who were breast fed almost exclusively (no more than 120 ml of formula per day until 12-months old) with those who used formula almost exclusively (defined as having never been nursed, or stopped nursing before three-months old).

In this article, the researchers found that "There were no significant differences in rates of respiratory illness," which contradicts the AAP's claim that there were decreased upper and lower-respiratory illnesses for nursed babies. The article adds that "Morbidity rates did not differ significantly between groups in the second year of life." So whatever protective effects nursing has don't seem to prevent illness in Year Two.

Hey, isn't science supposed to keep integrating the results of new (especially more carefully controlled) research into its body of knowledge? 1995 is more recent than the 1980s, right?

The Times takes the concept that an indictment is as good as a conviction to new heights. It makes such claims as breast-fed babies "appear to be at a lower risk for autoimmune diseases like... juvenile diabetes." And yet it turns out that this question is only now being asked in a large multinational study. So, in fact, no benefit has hitherto been shown.

Of the studies cited by the AAP as indicating a benefit in this area, one was based on babies in Chile, another on Indians in Peru, and a third only found results for children exposed to food. Infant formula wasn't even considered!

Appearances will do until the data is in? How can this be presented as scientific knowledge if the relevant studies haven't actually been done?

There are, of course, loads of potentially relevant variables that contribute to the health of an infant. Controlling all of them to clearly establish the benefits of breastmilk over formula is a pretty formidable task (and, as Goldin, Smyth, and Foulkes point out "it's virtually impossible to carry out the gold standard of research on this issue - a case-controlled study in which mothers are randomly assigned whether to nurse or not").

In the absence of definitive results about the precise benefits of breastmilk, it seems like people raising infants might want to play it safe -- to opt for the feeding option the research so far seems to favor. But, given that breastfeeding is not without costs (financial, physical, emotional, and social ones), and given that healthy babies have been raised on formula, trying to push women towards exclusive use of breastmilk on the basis of "scientific knowledge" that is rather less black-and-white than the reporting would suggest seems intellectually dishonest and unfair.

More like this

It is another classic case of taking from the left something that can be used as a weapon against the left. This has mommy wars written all over it. The problem is that the subtext of these breastfeeding discussions approaches feminism as something that occurs purely within the current system and not something that is trying to change the system.

At the same time, when we had to choose between having our children fed by bottom-line watching, anti-regulation, corporate boobs or the other kind attached to mommy, we were fortunate to be in a position to be able to choose the ones that didn't put earnings reports before our kids' welfare. But then our socio-economic situation, my wife's health and diet, and any number of other factors made it an easier choice for us than it is for many others.

wow, an examination of the literature shows that breastmilk isn't necessarily better? I'm shocked, I tell you, shocked.

By Frumious B (not verified) on 20 Jun 2006 #permalink

OK, I've read the article by Goldin et. al., and I have a few comments. First of all, I am not a statistician, and I habitually get in the back seat when a statistician speaks about research. However, I am not happy with the depth of their analysis.

First, in order to make a pronouncement about the entirety of the literature on a subject, one would have to gather all the relevant articles and read them. I do not see any evidence that they have done so. One would have to rank the studies in terms of quality. The date of the study is one factor, but it is not really a measure of quality. Maybe they did all that, and did not tell us, but I don't think we can be confident in their conclusions without knowing.

Second, I have to give them credit for finding evidence that some of the people involved in this might be biased. That is important to know. But it is circumstantial, and does not really help us get to the truth of the matter. It only cautions us to look more closely at the research methods.

Furthermore, their own article is not free of evidence of bias. For example, they write rather dismissively about the seriousness of the illnesses: "if the only adverse consequence of not nursing is that babies get a few more colds..." They do not mention necrotizing enterocolitis, (Lucas A, Cole TJ: Breast milk and neonatal necrotising enterocolitis. Lancet 1990 Dec 22-29; 336(8730): 1519-23) which has a significant mortality rate.

Finally, they conclude with this:

Our lives are filled with risks, small and large. Not nursing is a small risk, the real question is what it costs (or benefits) you.

That is true. However, when you consider that every single person on the planet has either been breastfed, or not, whatever risk you find, no matter how small, has to be multiplied by six billion. Take a small number and multiply by six billion, and the product just might turn out to be significant.

The authors point out, correctly, that a full analysis of the topic would be highly complex. They are critical of the AAP for being too casual in their analysis. Then they toss another casual analysis on the heap.

Perhaps they are correct in saying that an important public health announcement should have a stronger scientific basis. But it would be more helpful if they would go on to say how the science could be strengthened, e.g. by a Cochrane review of the existing data...

...and, as it turns out, there is a Cochrane review that is pertinent, albeit not comprehensive; it only examines a narrow piece of the overall question. Among their conclusions:

Infants who are exclusively breastfed for six months experience less morbidity from gastrointestinal infection than those who are mixed breastfed as of three or four months, and no deficits have been demonstrated in growth among infants from either developing or developed countries who are exclusively breastfed for six months or longer.

If a careful review of 2,668 unique citations leads to a conclusion that exclusively-breastfed infants do better, then I think we need to pay attention. Granted, it is not a comprehensive analysis, but it also is not "just a throw back public health campaign based on voodoo science."

I'll agree with them that some of the studies cited in the New York Times are insufficiently rigorous. I'll also agree with them that a comparison to smoking is unwarranted. But if what we are trying to do is get a fair and balanced look, then let's take a look at the studies that are rigorous.

As a professional statistician, the Goldin et al. article is embarrassing. Excuse me whilst I rant...

1. They look at this paper: Chen & Rogan (2004) Pediatrics 113: e435-e439.

Goldin et al. make a big thing of breast feeding apparently reducing death due to injuries. They then go on to down-play the effect on SIDS:

But the association is not statistically significant (the 95 percent confidence interval is .67-1.05).

i.e. we can ignore it. But looking at the same table (Table 3), we see the CI for death by injury: 0.43-1.05. So, that is also not statistically significant, but does not get down-played.

Shit! (sorry) It's even worse. The estimates of the effect size for death due to SIDS are statistically significant. The original paper calculates two effect sizes. One is from the original data, the other is SUDAAN-adjusted. The difference is which counter-factual you use. The esimate from the original data gives a comparison between two individuals, otherwise identical, one of whom is breast-fed, the other is not. The SUDAAN-adjusted values, if I've understood this correctly, use as a counterfactual two random individuals drawn from the US population. The estimate from the original data is the genuine estimate of the effect size: the counterfactual is correct. The SUDAAN-adjusted values are better for estimating the national effect of the intervention: i.e. how many children's lives could be saved (IOW they're more for prediction than inference). I think it's dodgy ethically to only present one figure without explaining the full story (but I'll bow to expert opinion on this :-)): I think their statements are misleading.

2. Goldin et al. comment that many studies in the 1980s weren't controlled for other factors. OK, but whre's the evidence that this would have an effect? It might make the relationship stronger! Oh, and Chen & Rogan (2004) do correct for the other factors.

3. Goldin et al. then mention a paper from 1995. They don't give a reference (naughty!), but a quick search brings up this:

Dewey, K.G, Heinig, M.J. & Nommsen-Rivers, L.A. (1995). Differences in morbidity between breastnext term-fed and formula-fed infants. The Journal of Pediatrics. 126: 696-702.

The paper show few significant effects. But we'd expect that: the sample size is less than 100 (46 breast fed, 41 formula fed). This is as opposed to Chen & Rogan (2004) who have a sample size of 8944, and do get statistically significant results (but still with wide confidence intervals). With such a small sample size, it's not surprising that there are a lot of non-significant results. This is basic statistics: eve biologists have been known to understand it.

So, they rubbish a large study (actually, they could have just cited the conclusions, which admit that they can't totally separate out breast-feeding from other factors), and instead cite a study that has poor power but which supports their view (I'm not criticising Dewey et al., BTW. Good on them for publishing negative results). It makes me wonder if they work for this company.

Bob

Thanks, Bob, for your expert input on the Goldin et al. It's good to have this kind of methodological critique from someone who does this kind of thing for a living. I take it this adds more weight to Joseph's critique, which I note here.

I'm struck at a bit of irony here: The Goldin et al. paper was posted on the Statistical Assessment Service (STATS) at George Mason University website, whose mission is described thusly:

Our goals are to correct scientific misinformation in the media resulting from bad science, politics, or a simple lack of information or knowledge; and to act as a resource for journalists and policy makers on major scientific issues and controversies.

My impression from the feedback here is maybe they have some work to do still to meet that goal!

I fully support the idea that the scientific method helps us understand the world we live in. It can also help us understand the risks and benefits of different life choices we make. All good stuff.

I'm also pretty sure that biological systems are quite complex and that we just don't fully understand how and why they work - we're still nibbling at the edges of a very big cookie.

I'd go so far as to say that evolution trumps human knowledge - give live a few billion years to work out the details and systems exist that work quite nicely.

So...human milk for human babies MUST be better than cow milk, which is made for cow babies. I'm not even sure if soy babies suckle, but if they do, soy milk is for them.

How much better? Who knows? It's obvious we can raise human babies on formula.

We're also seem to be seeing more kids with allergies (where the heck did this nut allergy thing come from???) and asthma. Are they related? Not known.

Given the option of formula vs breast, doesn't it make sense to opt (whenever possible) for the breast?

Doug

I found the STATS.org article to be such an appalling load of spin I've spent the past several days writing a two-part post on all the problems I found with it. Part 1 is at and Part 2 at , if anyone's interested.