We aimed to explore why, in population studies, the positive association between normal-range birth weight and intelligence becomes negative at the highest birth weights. The study population comprised 217,746 Norwegian male singletons born at term between 1967 and 1976. All had data on birth weight and intelligence quotient (IQ) score at the time of military conscription; 137,574 had data on sibling birth weights; and 62,906 had data on male sibling birth weights. We estimated associations between birth weight and IQ score by ordinary least squares regression for the total study population and by fixed-effects regression for comparisons of brothers. The crude mean IQ score was 1.2 points (95% confidence interval (CI): 0.3, 2.2) lower for those with birth weights of 5,000 g or more compared with the reference group (with birth weights of 4,000-4,499 g). This difference leveled off to 0.0 (95% CI: -0.8, 0.9) in multivariable ordinary least squares regression and reversed to 2.2 points (95% CI: 0.3, 4.2) higher in fixed-effects regression. Results differed mainly because, at a given birth weight, participants who had a sibling with macrosomia had a lower mean IQ score. Nevertheless, within families with 1 or more macrosomic siblings, as in other families, men with higher birth weights tended to have higher IQ scores. Thus, a family-level confounder introduces a cross-level bias that cannot be detected in individual-level studies. We suggest ways in which future studies might elucidate the nature of this confounder.
Keywords: Norway; birth weight; cross-level bias; fetal macrosomia; intelligence; multilevel analysis; siblings.
© The Author 2014. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.