Cohort analyses suggesting that breast-feeding protects against being overweight have been criticized for inadequately controlling for confounding associated with the self-selection of feeding practices. Methods: Using nationally representative U.S. data from the National Longitudinal Study of Adolescent Health (1994 -1996), we performed traditional cohort analyses (n ϭ 11,998) using logistic regression to estimate the relation between breast-feeding and adolescent overweight (body mass index

more »

... 5 percentile, based on year 2000 CDC growth charts), controlling for known potential confounders. Breast-feeding also was assessed in a subsample of 850 sibling pairs to account for unmeasured genetic and environmental factors. Results: Among girls in the full cohort, the odds of being overweight declined among those who had been breast-fed at least 9 months; odds ratios ranged from 0.90 (95% confidence interval ϭ 0.74 -1.09) for Ͻ3 months of breast-feeding to 0.78 (0.64 -0.96) for Ն9 months. A similar effect was seen in boys, although these trends were less consistent. In contrast, an analysis of sibling pairs provided no evidence of breast-feeding effects on weight within discordant trends. Conclusion: Cohort data indicate that odds of being overweight decrease as breast-feeding duration increases, at least among girls. However, sibling analyses suggest that this relationship may not be causal but rather attributable to unmeasured confounding related to mothers' choice to breast-feed, or to other childhood risk factors for overweight. Our results illustrate the utility of sibling analyses in understanding the true effect of early life exposures (such as breastfeeding) on health outcomes over time, independent of confounding factors that may not be satisfactorily controlled using traditional prospective cohort methods.