Increasing breastfeeding rates to reduce infant illness at the community level

Objective. Although breastfeeding is associated with lower rates of a variety of infant illnesses, skeptics have suggested that much of the association is attributable to confounding, even after appropriate statistical adjustment. This article utilizes a novel design to investigate changes in infant illness at the community level after a successful breastfeeding promotion program. Methods. In this population-based cohort study, the medical records of all infants born in one Navajo community the year before a breastfeeding promotion program (n = 977) and the year during the intervention (n = 858) were reviewed. Outcomes assessed include changes after the intervention in: proportion breastfeeding and/or breastfeeding exclusively; incidence of common infant illnesses in the first year of life; and feeding-group specific incidence of illness. Results. The proportion of women breastfeeding exclusively for any period of time increased from 16.4% to 54.6% after the intervention. The percent of children having pneumonia and gastroenteritis declined 32.2% and 14.6%, respectively, after the intervention. Feeding-group specific rates of these illnesses were unchanged, indicating that the decline observed was attributable to the increased proportion of infants breastfeeding. In contrast, rates of croup and bronchiolitis increased after the intervention among those fed formula from birth, suggesting a viral epidemic which was limited to those never exclusively breastfed. Finally, sepsis declined in both formula-fed and breastfed infants after the intervention, suggesting that other factors affected this illness outcome after the intervention. Conclusions. Increasing the proportion of exclusively breastfed infants seems to be an effective means of reducing infant illness at the community level. The experimental design suggests that the increased incidence of illness among minimally breastfed infants is causally related to lack of breast milk, rather than being attributable to confounding.