Infant feeding and growth: putting the horse before the cart

Previous observational studies have consistently shown slower weight and length gains in infants with prolonged breastfeeding than in those who were formula-fed from birth or breastfed for a shorter duration. These studies inferred that prolonged breastfeeding causes slower growth in infancy. We compared infant growth associated with a parts per thousand 12 mo of breastfeeding with a shorter duration of breastfeeding on the basis of 3 different analytic approaches to the same data from a randomized trial: intention-to-treat (ITT; "as randomized"), observational ("as fed"), and instrumental variable (IV; by using randomization as an "instrument" to achieve a parts per thousand 12 mo of breastfeeding). This was a cluster-randomized trial of a breastfeeding-promotion intervention. Anthropometric measurements were obtained at birth and at 1, 2, 3, 6, 9, and 12 mo. The 3 analytic approaches yielded different results. The ITT approach showed more rapid growth in the first 2 mo among infants randomly assigned to the breastfeeding-promotion intervention than among control infants, with a decreasing difference over the ensuing months and nearly identical weight, length, and body mass index by 12 mo. The observational analysis showed a different trend: higher weight and length in infants who were breastfed a parts per thousand 12 mo than in those who were breastfed < 12 mo during the first 3 mo and no difference by 6 mo, while infants who were breastfed < 12 mo showed increasingly higher weight and length from 6 to 12 mo. The IV analysis showed a temporal pattern that was similar to that seen in the ITT analysis, but with larger (and less precise) differences between infants breastfed for a parts per thousand 12 compared with < 12 mo. We observed major differences in experimental (ITT and IV) compared with observational approaches to analyzing data obtained from the same children. These approaches lead to opposite causal inferences about the relation between infant feeding and growth and underline the importance of ensuring that the postulated cause (feeding) temporally precedes its hypothesized effect (growth). This trial is registered at ext-link-type="uri" xlink:href="http://www.isrctn.org/">http://www.isrctn.org/ as ISRCTN37687716.