Researchers began with a group of 5,914 newborn babies in 1982, and were able to follow-up with 3,493 of them three decades later. That's a follow-up rate of 59 percent. (Researchers consider their follow-up rate to have been 68 percent because they count 325 subjects who are known to have died and several for whom they didn't have complete IQ or breastfeeding information.) Though one of the study's co-authors, Bernardo Lessa Hort, characterized the rate as "quite good," many scientists would say it's suboptimal.
In general, a follow-up rate has to exceed 60 percent to be considered acceptable, and reach 80 or 90 percent for it to be considered good. "A good rule of thumb," wrote the author of a 2011 study about loss to follow-up in clinical trials, "is that less than 5 percent loss leads to little bias, while more than 20 percent poses serious threats to validity."
Even with small rates of loss, significant biases are possible. And the number of people who drop out of a study is actually less important than how those departures are related to the findings. The contributing factors that make someone lost to follow-up are often the same factors that would affect the outcome of the study.
In the context of a study about breastfed babies, the obvious example is socioeconomic status. That's because researchers tracked key socioeconomic markers—income levels, duration of schooling—that could also influence whether a person is findable for a follow-up. So, in theory, the breastfed babies who wouldn't have ended up earning as much or performing as well on an IQ test could have been the ones lost to the study. Which is why these sorts of losses to follow-up are considered one of the main threats to validity in clinical trials, and they rarely occur randomly. "The only way to ensure that losses to follow-up have not biased study results," wrote the authors of a 1998 study about tracking research cohort, "is to keep all losses to an absolute minimum."
Even among the study subjects whom researchers were able to follow for the full 30 years, there are questions. Like the fact that mothers participating in the study were required to recall their breastfeeding habits years after their babies were born. Researchers first followed up with them in 1984, two years after enrollment in the study.
Relying on a research subject's memory can be a major threat to validity, and breastfeeding mothers have provided inaccurate estimates to scientists before. "When mothers are asked retrospectively in breastfeeding studies about breastfeeding duration, the reports become increasingly inaccurate with increasing time since cessation," according to a 2008 paper about research biases by Lars Fadnes. "Mothers who had breastfed their infants for longer systematically underreported their breastfeeding duration; while in contrast, those who had breastfed for a shorter time over-reported the duration."