Sunday, March 2, 2014

Breastfeeding: the Sibling (study)

The study is here: 

Background:  I'm a Bayesian -- I feel strongly that an analysis of the foundation of evidence on which a study rests and what the prior probability of a claim is fundamental to evaluating the claim.
Therefore, here's a brief analysis of the prior studies:
It is well accepted that breastfeeding is important in developing countries and premature infants because of the decreased risk of necrotizing enterocolitis and diarrheal illnesses [citation left as an exercise to the reader.]  

However, although breastfeeding is extremely culturally emphasized with the "breast is best" mantra and the "babyfriendly" hospital initiative restricting access to formula, the prior literature showed little support for a clinically significant impact of breastfeeding on term infants in the developed world.  In addition, it was strongly suspected that the majority of effect in published studies was related to the association bias.  These facts are absolutely necessary to know in determining how likely you are to find the Colen, et al. study to be valid.  Therefore, I include here a limited literature review to emphasize the point.  (It is also important to note that benefits of breastfeeding have also been exaggerated by the publication bias: i.e. negative studies are unlikely to be published, especially given the current breastfeeding politics.)

  • Intelligence: Studies consistently support an effect of breastfeeding on intelligence.  This effect is consistently so small as to be likely considered trivial to the average person.  For instance, Erikson, et al. found that breastfeeding has the same effect on intelligence as the sex difference, less effect than irregular breakfast and would not have had a significant p-value had the authors used a Bonferroni correction. (!po=46.4286)  The extremely hyped study last summer in JAMA pediatrics ( found the effect size to be 0.35 verbal IQ and 0.29 nonverbal IQ points per month breastfed. This is considered to be the premiere study on intelligence and breastfeeding.
  • Asthma: The evidence is very limited.  The currently most cited study is, which finds that infants who are exclusively breastfed for 2 months or more are less likely to wheeze by preschool age, but equally likely to go on to develop asthma and infants who are exclusively breastfed and then supplemented are more likely to wheeze than those who transition to formula and the effect size and the p-values both are not very significant.  Nonetheless, there are several, similar quality studies making the same claim.  Plausible to have a very small effect.
  • Obesity:  The "it study" is this Japanese study, which showed that among 7-8 year olds, only 7-8% of those exclusively breastfed for up to 6 months were overweight (with 7% at 6-7 months, and 8% for 2 months), compared to 9% of formula fed children and 1.5-2.9% obese vs. 3.5-3.9% obese.  However, a randomized control trial of pro-breastfeeding interventions that succeeded in substantially increasing breastfeeding in the intervention group did not find any difference in weight
So, in summary, in starting the study there is some evidence of a very small positive effect of breastfeeding in reducing asthma and obesity and increasing intelligence.  This effect is statistically small and clinically small and may be completely accounted for by sociodemographic differences.

Introduction:  What does this study add?  The goal of this study is to use paired sibling controls to eliminate sociodemographic differences from consideration as a potential confounding factor.  

  • The study chose to look at 11 outcomes: BMI, obesity, asthma, hyperactivity, attachment, compliance, math ability, memory ability, vocab, reading comprehension and "scholastic competence"
  • Subjects were recruited between ages 4-14 in 13 waves over a 24-year period 
  • Results such as obesity and asthma were coded at the age that they developed 
  • Information was gathered through maternal and child interviews
  • There were 4,071 families (8,237 children) recruited
    • 3,153 families in the sibling cohort (7,319 children)
    • 665 families in the discordant sibling cohort (1773 children)
  • sibling was defined as sharing the same mother
  • Breastfeeding was binary: any breastfeeding v. no breastfeeding
  • The following confounding variables were measured: survey year, maternal age at time of birth, patient age, ethnicity, marital status, region of the country, maternal education at time of interview and birth, income at time of birth and interview, maternal employment at time of birth and interview and insurance (private, public v. none), birth order, preterm status, smoking during pregnancy, alcohol during pregnancy, and prenatal care in the first trimester
It's worth noting in their demographic data that this is old and therefore, only 10-11% of women worked full time in the year after their child was born and a whopping 30-31% smoked during pregnancy -- so probably demographically very different than the modern woman.

Beyond that side note, it is notable that in the discordant sibling sample, there is slightly higher percentage of woman of color, slightly lower educational attainment (12.5 mo v. 13) and slightly lower income (62k v. 68k at time of birth)

 When they compare outcomes between breast and bottlefed infants, they find large benefits in breastfeeding in their intelligence, behavior and obesity measures, which is almost completely nullified, and in some cases even reversed by evaluating paired siblings.  After this step, no statistically significant differences are seen

What this study adds most importantly is it validates the strongly suspected hypothesis that differences between breast and bottlefed infants were largely accounted for by confounding variables.  

The major limitations are worth noting; however,
Exclusivity: The very loose criteria for being "breastfed" was simply having received any breastmilk.  It could be argued that there is a subpopulation of breastfed children who breastfed for longer/higher percentage of nutrition who performed better and are buried in the sample.
Rarity of discordance: I would argue that discordant breastfeeding among siblings is rare and a woman who formula feeds one or more of her children is likely not representative of the average breastfeeding mother.  We can speculate that they are more likely to supplement, and the formula fed child is more likely to be higher in the birth order, which gives them an intelligence and behavior advantage, thus confounding the results.  This can also be seen in that the discordant group was relatively disadvantaged to begin with, so even less likely to have been exclusively formula fed and possibly subject to more effect from other environmental factors.