Rabbit, Duck, or Moonwalking Bear?
A causal diagram (DAG) for breastfeeding, and reflection on enduring mysteries
Last week, I revisited my first breastfeeding article’s core methodological criticisms of most infant feeding research. It’s uncertain of what most breastfeeding studies measure an effect. One major problem is that they tend to vary more than one thing at a time. Another is that they tend to not measure important variables. Then researchers draw unwarranted conclusions from correlations, when we don’t know what the correlations mean because of these problems.
This is also true of PROBIT, an oft-cited 1990s Belarussian cluster randomized trial of exclusive breastfeeding (EBF) promotion protocols that’s often misunderstood as a randomized trial of breastfeeding. This confusion leads to strange situations, like infant feeding reform activists whose voices I cherish sometimes, accidentally, repeating breastfeeding myths (e.g., “PROBIT showed breastfeeding boosted child IQ”) as if they were proven facts. But when we drill down into what PROBIT actually did, and what most breastfeeding studies do, we see that the findings can be interpreted as representing two very different pictures…
One way to view PROBIT’s findings is as showing the “rabbit” of newborn starvation periods, a big picture in which the EBF paradigm probably decreased average starvation period (time after birth until first feeding) as compared to the prior era’s factory-feeding hospital nursery model — but simultaneously worsened three other forms of newborn starvation: (a) starvation in the two full days before most mothers’ mature milk comes in, (b) continuing starvation when that milk supply is still not enough, and (c) starvation when that supply cuts out later or becomes insufficient. Alternately, another way to interpret PROBIT’s findings is as showing the “duck” of the EBF paradigm itself, a big picture in which breastfeeding proponents often claim benefits from a bunch of associated new norms including denying infants supplementary milk or water, keeping mothers and infants together instead of separating them after birth, and encouraging immediate skin-to-skin contact and breastfeeding.
So does the evidence from oft-cited breastfeeding studies including PROBIT show the rabbit or the duck? Neither. We have to interpret it. Or — and this is better scientific practice — diagram out the causality in play here, and say what we don’t know from available evidence given the relevant causal logic.
The unbearable lightness of being (a regular dope who happens to be doing science)
Uncertainty is a relatively unpopular conclusion.
It denies the public the security blanket of experts proclaiming certainty to give them control over their health (and their children’s) if they just engage in the right rituals. It re-codes science from secular religion to progressive, fallible, logical, chaotic, man-made practice. Culturally, this is not the way we usually think of science. Instead, we tend to talk as if “science says” what’s best. It does not and can not.
Sometimes, though, scientists pretend that it does. They are “the man behind the curtain,” making certainties of uncertain evidence — some of them putting on more of a show about it being Science talking, than others. This is a huge problem in medical journals, where statistical analyses are “frequently misleading or even wrong,” and reform leaders recently declared "It is time for the research community to... actively fight for accurate plain-language reporting of results by challenging referees and editors who continue to force destructive conventions on authors" (Greenland, Mansournia, Joffe, Preventive Medicine 2022).
A breastfeeding DAG
In my last post, we left off holding onto the cliff-face of reality, wondering where the next hold might be to climb higher seeking more truth (or at least less mouth-gravel). When I feel this way, my new impulse is to draw a DAG (causal diagram), because that’s a hold. By thinking structurally about causality, the hope is that we can get farther in interpreting the evidence in line with logic instead of pre-existing belief and other common cognitive biases. Here’s a breastfeeding DAG — again (as in my recent abortion DAG) not something I claim is The Right Structure, but my best effort to balance the heuristic and the realistic to take a step forward along this path.
X - Exclusive Breastfeeding Choice
NS - Neonatal Starvation
EBF - Exclusive Breastfeeding
MH - Maternal Health
CH - Child Health
This DAG expresses the idea that EBF interacts with maternal health (MH) in powerful ways, and then indirect effects of neonatal starvation can be activated by early withholding of formula and/or difficult EBF. Effusive thanks cannot repay Richard McElreath for his kindness and generosity in helping me redraw the DAG, restate the argument, and think more about all this. And for pointing out that it’s a problem that DAGs don’t show moderating effects, but powerful moderating effects are part of the argument.
The DAG shows complex relationships between most of these variables. EBF can affect neonatal starvation (NS), as when colostrum is insufficient and newborns starve. NS can affect EBF, as when newborns are too weakened by days of starvation to successfully breastfeed when mothers’ mature milk comes in. MH can affect EBF, as when exhaustion, disordered glucose metabolism, and/or endocrine disorder contribute to breastfeeding problems. EBF can affect MH, as when it contributes to inflammation and exhaustion directly from the metabolically intensive biological labor of lactation, and/or indirectly from sleep deprivation. NS can affect child health (CH), as when it contributes to jaundice and its progression with possible substantial downstream consequences including permanent neurodevelopmental disorder risk. CH can affect NS, as when poor newborn health impairs feeding. MH can affect can affect CH, as when healthier women produce healthier children. CH can also affect MH, as having a sick child may impact a mother’s physical and mental health. The question in most breastfeeding literature is what the arrow means between EBF and CH. (It just happens to be widely misrepresented as an answer, when we don’t really know.) And it’s well-recognized that CH can also impact EBF, as when newborns are too sick for traditional breastfeeding, and sometimes cup-feeding is considered an alternative (see, e.g., Lang et al 1994, McKinney et al 2016).
This is my favorite part, because I got it wrong at first: The DAG also shows X, exclusive breastfeeding choice. In the PROBIT dataverse, this would include randomized treatment assignment to a hospital promoting EBF. In another dataverse, it would still include whether the hospital/healthcare practitioners promote EBF. It may be influenced by whether a woman’s mother or other female kin or intimates breastfed, or what social networks told her about it otherwise. It may be influenced by other unmeasured aspects of the healthcare setting (usually a hospital) or the woman. For example, multiple researchers and clinicians have noted that it tends to be first children of upper/middle class parents who really want to do things right, whose “exclusively breastfed” newborns wind up rehospitalized with horrific complications of starvation. It’s also widely noted that African-American women tend to formula-feed instead at much higher rates than other racial groups. So there’s a broad range of factors, many unmeasured, that may influence whether a mother chooses to try EBF in the first place.
What does this mean?
When we see these sorts of complex causal relationships, we should think twice before declaring that we know what’s “best.” Even in terms of interpreting correlations that come out of experimental data like PROBIT’s. The DAG includes things we can’t know, because they aren’t generally measured in breastfeeding studies: factors influencing mothers’ exclusive breastfeeding choice, neonatal starvation, and maternal health. So when we see associations, we can’t decide what they mean.
The hat trick I’m hoping to pull off now is convincing people that this is great news. Uncertainty — the existential fact that the mystery is unfolding — is part life, and part of science. It need not be a disappointment. It is not an imperfection. Not a distraction. The mystery is unfolding — whoa!
There’s a collection of famous selective attention experiments and videos showing that, when we look too hard for one thing — like basketball passes, or a rabbit, or a duck — we can miss other happenings (spoiler alert)... Like a moonwalking bear. One of the cognitive-emotional challenges of doing marathon rabbit-holing like science is to keep pulling back to listen for what you don’t know what to listen for, see what you don’t know to look for, relax and let something new come — when you want to go forward, forward, forward (especially me). But there is no forcing flow, our higher intelligence at work in spite of ourselves; and you want flow in awareness to see the big picture. Otherwise, you don’t know which way is up, so you don’t know which way is forward. There is no brute-forcing any of the best things in life: art, love, insight.
So part of the job of science reform is to make wondering, wonderful again. I wonder what causal effect breastfeeding has on child health. I don’t know. You don’t, either, and that’s not a criticism, per se. Everyone (mostly) wants children to be healthy, and everyone wants to find the truth. But these things are not ours to control with ritual or performative certainty. The mystery is unfolding.
In the meantime, not starving babies probably remains a Very Good Idea (TM).