Of Milk and Misery

Diagramming the causal logic of breastfeeding and postpartum depression

For the last few years, I’ve been thinking a lot about selection effects (e.g., health predicts health) and interpretive problems (e.g., starvation harms versus breastfeeding benefits) in infant feeding research. This post builds on that thinking and its big-picture causal diagram (DAG) from this recent post…

… Zooming in on one of many areas where the breastfeeding myth says one thing, but reality says another: postpartum depression. 

On the basis of bad science, experts bill breastfeeding as prevention and cure for many things to which it actually causally contributes, at least in the way it’s currently practiced (the exclusive breastfeeding paradigm). Then, medical researchers subject vulnerable new mothers to research that puts breastfeeding first. But far from preventing postpartum depression (and other adverse outcomes), such interventions may cause it. 

This research risks unintentionally doing harm by design by not putting causality first, and appearing to not consider biological realities. Here’s how.

Breastfeeding and Postpartum Depression DAG

X - breastfeeding choice
BF - breastfeeding
MH - maternal health 
BFI - breastfeeding insufficiencies
NS - neonatal starvation
C - excessive infant crying
P - pain
S - sleep deprivation
CH - child health
PPD - postpartum depression
LS - lactation support

Like its big-picture analogue above — like all my DAGs, possibly like all DAGs? like all research? —this DAG “is of course incomplete and possibly wrong… But an incomplete model is still an improvement over no model at all” (Deffner et al). It’s just a step forward in thinking structurally about causality.

It expresses the idea that maternal health is a confound in the relationship between breastfeeding and depression. Maternal health appears to be a fork influencing factors — known ones like inflammation, disordered glucose metabolism, and endocrine disorder (i.e., PCOS), and probably unknown ones, too — that may contribute to both breastfeeding insufficiencies and postpartum depression. This shows why frequently observed, weak inverse correlations between BF and PPD — e.g., in Ip et al's 2007 meta-analysis and Dias and Figueiredo's 2015 review, but not Chowdhury et al 2015's meta-analysis — should not be interpreted as evidence of a causal protective effect of breastfeeding against depression, as leading researchers are careful to explicitly acknowledge. First, causal diagramming should precede these analyses. And second, it and they must account for this confounding. 

In a tragic historical accident, breastfeeding the way it’s now normally promoted and practiced usually causes neonatal (newborn period) starvation in the two full days before mothers’ milk typically comes in. That starvation may risk permanent neurodevelopmental harm and death. As if evolved to sound an alarm about such risks, starving babies cry. A lot. And excessive infant crying and sleep deprivation contribute to PPD risk. (Not to mention sleep deprivation’s contribution to other, less common but severe problems like postpartum psychosis.)

In addition, common breastfeeding insufficiencies also cause neonatal starvation when milk is delayed even longer coming in, or when supply remains insufficient. In both cases (BF and BFI), neonatal starvation can cause or worsen child health problems (e.g., jaundice, underweight) that may also contribute to PPD. When your child hurts, you hurt, too.

Breastfeeding commonly causes pain (e.g., nipple pain in 50% of mothers); pain also contributes to depression risk. Add to that the well-established fact that mental health problems including depression can also contribute causally to sleep problems (more on this later).

Meanwhile, at the beginning of all this, other things that factor into breastfeeding choice (X) — like socio-economic status, social support, and stress — also affect depression risk. Mothers at lower risk for depression for these other reasons are also more likely to breastfeed, something many researchers recognize that we struggle to account for (e.g., Reading 2001, Seimyr 2004, and Borra 2015). The same factors may also influence CH and LS.

Overall, this is a grim picture for the idea that we should look to breastfeeding and breastfeeding support interventions — practices that aim to increase breastfeeding rates, not optimize individual maternal or child health — to prevent PPD. It highlights the need for structural thinking about causality to prevent bad inferences and make better science. Avoiding bad inferences is not just about applying the rules of logic. It’s ultimately about not accidentally causing harm through inappropriate healthcare interventions that apply bad inferences, in this case to particularly vulnerable patients (new moms and their infants)…

Pre-registered Cochrane review: A critical view

Bad inferences like this:

In a UK cross‐sectional self‐report survey to examine the relationship between specific reasons for stopping breastfeeding and depressive symptoms in the postpartum period, multiple reasons for stopping breastfeeding were associated with higher depression scores (Brown 2016). However, in a regression analysis, the only reason for stopping breastfeeding that predicted postpartum depression was physical difficulty and pain with breastfeeding. These results suggest issues with pain and physical breastfeeding were most indicative of depression in comparison to psychosocial reasons, highlighting the importance of breastfeeding support early after delivery.

- Lenells et al, planned Cochrane review, “Breastfeeding interventions for preventing postpartum depression.”

In other words, Lenells et al argue that healthcare professionals should push breastfeeding to prevent depression, even though they acknowledge that the pain it causes is associated with depression, and not being able to do it is, too. 

This is backwards. The sad thing is that a lot of women report experiencing intense guilt and shame about breastfeeding having not worked for them and their babies. Research like this ignores these mothers’ experiences under the auspices of supporting them. It also fails to apply the rules of logic. 

The authors don’t appear to consider that exclusive breastfeeding biologically fails for a substantial minority of women and infants, saying “Support for women who wish to breastfeed can work in different ways for different women. What is required is timely, skilled assistance to avoid or overcome breastfeeding problems that may lead to early discontinuation.” The planned outcome measures do not include anything about neonatal starvation or health, or mother’s perceptions of breastfeeding problems including delayed or insufficient milk. There is no recognition that some (common) breastfeeding problems are biologically insurmountable — or that the basic needs of new mothers (e.g., to sleep) and their infants (e.g., to eat) matter more than breastfeeding targets, and the two might conflict.

This is “If you believe it, you can achieve it” breastfeeding promotion — the usual flavor of the dominant myth.

Back in the land of logic, researchers should consider what proportion of women will likely still experience physical difficulty and pain with breastfeeding after support interventions, since breastfeeding support of this flavor seems unlikely to decrease and may actually increase these women’s postpartum depression risk. It would be even better to start with causal diagramming, and design entirely different research from there. Like what?

Alternate future research

This points to an alternate future research agenda in women’s health. One that puts listening to women and logic before ideology, acknowledges the partly biological bases of common problems like breastfeeding insufficiencies and postpartum depression, and better protects some particularly vulnerable human subjects to whom researchers owe a heightened duty of care. There’s a brave new world of promising interventions to explore when it comes, for example, to lowering inflammation, improving glucose metabolism, and normalizing endocrine function through diet and lifestyle (e.g., Metformin, the ModiMed diet, and exercise).

These sorts of interventions may benefit many women before as well as during and after pregnancy by lowering inflammation, improving glucose metabolism, and normalizing endocrine function— lowering depression risk while also improving the odds that breastfeeding will work by treating biological problems that can undermine lactation, an intensive metabolic act. This puts women’s health first as an end unto itself, breastfeeding rates be damned. Why, exactly, aren’t we already doing that?

It’s hard to imagine a study explicitly comparing feeding newborns with starving them in order to measure the relative effects of the two infant feeding practices on maternal mental health. Such a study would be wildly unethical. Yet, in the days before most mothers’ mature milk comes in, newborns go hungry under current breastfeeding guidance to withhold supplementary milk. This says something about infant feeding practices today, and about research that uncritically reviews literature on these practices.

Adding injury to injury

Researchers must critically review literature on current infant feeding practices, because they appear to do common and preventable harm to both infants and their mothers. For example, look at the part of the DAG where lactation support (LS) contributes to sleep problems (S), thus contributing to PPD. S is a collider, because it’s influenced by both LS and PPD. Thus analyses of LS effects on PPD need to account for this, and simple stratification on sleep is a poor choice because it risks introducing more bias into analysis than it reduces (collider-stratification bias).

The very simple, real-world problem here is that healthcare professionals frequently advise mothers with common breastfeeding problems (e.g., insufficient milk or mastitis) to just do it more. For example, standard advice including from field leaders in breastfeeding medicine and lactation consultancy is to nurse, bottle-feed, and pump on a continuous three-hour rotation to increase low supply. This promotes maternal sleep deprivation.

It is odd that Lenells et al omit this from their review of breastfeeding interventions for preventing PPD, although they do include a range of interventions (e.g., oxytocin, lactation cookies, fenugreek supplementation) that are supposed to increase breastmilk supply — bracketing actual evidence for these claims, which tends to be lacking —and that tend to have neither intended nor proven effects on PPD. Current medical practice poses substantial possible risks to maternal as well as infant health. Therefore, research that looks at breastfeeding interventions and PPD cannot just assume that the possible effects are all positive, without even considering the possibility of iatrogenesis (harm from breastfeeding support).

Myth-based medicine brainwashes women into starving their newborns, then pathologizes them for being distressed — and valiantly offers the expert assistance of medical professionals to save the day. Dubious science launders this insanity into something resembling respectable knowledge. But on closer examination, it is not.

In the bigger picture of women’s health, this looks like one case study among many where standard care hurts women. The same is also true of abortion care, as I’ve written previously. Standard prenatal depression screening can be similarly criticized. When we put biology on the board and triangulate qualitative with quantitative evidence using the tools of visual causal thinking, it opens up a world of better medicine and science.

So why not do that? Is this just part of that great dumpster fire of complex modern society — the collective action problem of spin science? Or is it also, in this context, because women’s bodies are still somehow too scary to look at in the light? Women’s voices too disreputable to be believed? Is this sexism, or are we all screwed?

Toxic feminism

Why choose? After all, today’s women can have it all!

While both those narratives (sexism and doom) are plausible, I think there’s also a simpler, psychosocial story here of fear-driven stasis. It creates cognitive dissonance for experts to consider that they may be harming the people they intend to help. It’s easier to keep doing what you’re doing, and go with a story that supports that. This isn’t just a problem in women’s health, but it does have unique political dimensions here. “You’ve come a long way, baby” — to be gaslit by a new specialist class of women’s health experts whose common practices actually undermine women’s well-being.

This is toxic feminism, a modern Western phenomenon where apparently feminist intentions underpin regressive interventions (separate post).

In this context, medical practitioners should be aware that the scientific evidence on maternal and infant health effects of exclusive breastfeeding is insufficient to establish causality when it comes to postpartum depression. To the contrary, apparent possible benefits may reflect downstream consequences of better maternal health and other factors causing both more successful breastfeeding and healthier mothers and offspring (the selection effects story); and this spurious correlation may mask substantial costs of breastfeeding the way it’s currently promoted and practiced, including increased PPD risk. So what?

Healthcare professionals should encourage supplementary formula-feeding (mixed feeding) from birth until breastmilk supply is established in order to prevent neonatal starvation and promote maternal health. They should advise women who wish to become pregnant on lifestyle changes that might facilitate better outcomes of many kinds, including lactation. We have the evidence base to screen, treat, and counsel pregnant women with occult or frank disease that places them at heightened risk of both breastfeeding problems and PPD — common problems like overweight/obesity. So maybe someone could work on constructing review-based guidelines for that kind of clinical intervention on a harm prevention model, instead of assessing whether the literature provides any evidence that lactation cookies prevent depression in new mothers, or do much of anything at all. (Spoiler alert: Nope.)

The beautiful mess

This is what I like about putting biology on the board in women’s health: It takes health problems out of women’s heads — where they have historically been treated as “hysteria” — and puts them firmly in physical reality, while also keeping it messy. In this case, maternal health (physical and previous mental health, socio-economic and social, historical and situational) likely influences both breastfeeding and maternal mental health outcomes. But it could still be the case that behavioral differences in breastfeeding drive biological changes causally affecting depression.

As Sapolsky notes, "our behavioral biology is usually meaningless outside the context of the social factors and environment in which it occurs." Putting biology on the board does not reduce us to Cartesian machines. It doesn’t necessarily simply things at all. It just puts the basic pieces we know might matter into place for structural thinking about what’s going on here in this beautiful mess we call life that is physical and psychological, behavioral and chemical, individual and social, all at the same time. If we’re going to try to piece together health puzzles, having all the pieces out seems like a good start.

Building on Sapolsky’s insight, I have to say that breastfeeding could very well be part of a virtual/vicious cycle dyad in which biopsychosocial feedback loops mean that it has positive health effects when it goes well, and negative ones when it goes poorly (many subtopics). The point is not that this story is implausible. It’s that the breastfeeding myth only tells the positive half, and does that without properly mooring it in fuller causal logic according to available evidence. Science and medicine can do better. Let’s.