It is not every day that a study based on a 30 year prospective cohort is published, and when it is led by Cesar Victora, in Lancet Global Health, we had better pay attention.
The study was trying to become the first to directly explore the association between duration of breast-feeding and earnings. There had been much suggestive work showing indirect links between earnings and breastfeeding (via things like child IQ and school enrolment), but nothing directly linking breastfeeding to adult earnings.
The study tracked 5914 newborns from 1982, born in 5 maternity hospitals in Pelotas, Brazil. In 2012-13 the authors were able to do follow up analyses on 3493 members (some could not be traced, some had died, some had incomplete information on IQ and breastfeeding).
The study measured the duration of breastfeeding (21% of sample less than 1 month; up to 17% of sample greater than or equal to 12 months) and the duration of predominant breastfeeding (26% of sample less than 1 month; up to 12% of the sample greater than or equal to 4 months; defined as "the age when foods other than breast milk, teas or water were introduced". The authors say that exclusive breastfeeding was "seldom practised at the time", hence they did not analyse it).
The study controlled for potential confounders such as monthly family income, household assets, maternal education, maternal age, pre-pregnancy body mass index of mother, type of birth delivery, gestational age, birthweight and, interestingly, genomic ancestry (based on DNA samples).
The study found that adult IQ was significantly increased by duration of breastfeeding and by duration of predominant breastfeeding, as were educational attainment and adult income.
For duration of breastfeeding and IQ the key results are in the figure below. Note the association between family income at birth and subsequent IQ (this shows the importance of controlling for confounders such as income at birth). Even controlling for family income at birth (and the other confounders), you can see how IQ increases with duration of breastfeeding, with a bigger effect for the higher income group (the red triangles)
The results are even more striking, I think, for predominant breastfeeding. Predominant breastfeeding covers a shorter period and is associated with a monthly increase in income that is over 80% of the income increase associated with longer durations of (not necessarily predominant) breastfeeding. It is too bad the exclusive breast-feeding data were not suitable for analysis.
The increases in IQ and educational attainment are associated with increases in monthly income (IQ is responsible for 72% of the "effect" of breastfeeding on income).
How big are the income changes? For the difference between lowest and highest duration of breastfeeding the effects are big: nearly one year of additional education, 4 points in IQ and about an additional one third of the average income. How many education and income generation programmes would like to show that size of effect? A lot.
Is the methodology good enough? The authors are careful to emphasise this is an associational study--for causation you would have to randomly assign babies to breastfeeding and non breast-feeding groups and of course that is not ethical.
The authors check on attrition (are the people they found in 2012-13 sufficiently representative of the original set of newborns?) Yes.
They control for confounding variables using regression analysis (although unlike economic journals, medical journals never seem to give much detail on the regression methods used--hey at least they still publish these kinds of associational studies, but that is another point).
There is some recall data on duration of breastfeeding for 4% of the sample, and this is found to be of reasonable accuracy. Seasonality of monthly income is not a major issue in this urban population. IQ tests were not done for parents, but educational attainment was included as a proxy.
The study also notes that the breast-feeding mechanism (is it the content of the breast milk or is it the bonding) cannot be determined, and admits the possibility of influence of unmeasured confounders such as intellectual stimulation or simply the woman's motivation to be a good mother.
In all cases the authors do more than enough to assure me (if I had been a reviewer) that they are cognizant of the study limitations, have done all they can to address them, and that they do not undermine the fundamental finding of the paper.
This is a carefully done study that finds large and significant associations between breastfeeding, IQ, education and income. Can we be definitive that the findings are causal? No. Can we be confident that they are? Yes.
Economic policymakers, I hope you are listening.