Invited Commentary: The Use of Sibship Studies to Detect Familial Confounding
American Journal of Epidemiology
The authors discuss how the sibship design can be used to detect and control for familial confounding. Familylevel confounding is especially problematic when estimating modest individual-level effects in the presence of familial confounders with large effects. This circumstance arises frequently in studies which relate indicators of fetal growth, such as birth weight, to outcomes that are strongly associated with parental socioeconomic status and genes. The study by Eriksen et al. in this issue
... of the Journal (Am J Epidemiol. 2010;172(5):530-536) uses the sibship design to capture the relation between birth weight, gestational age, and intelligence score among Norwegian males born as singletons at 37-41 completed weeks' gestation during 1967-1984. Their study illustrates how valuable the design can be in this kind of scenario. It also illustrates the potential complexity of sibship studies and the challenges they present for appropriate interpretation. birth weight; cohort studies; fetal development; gestational age; intelligence; Norway; siblings Abbreviation: IQ, intelligence quotient. For well more than half a century, epidemiologists have sought to elucidate the relation between birth weight and general intelligence or intelligence quotient (IQ). The question is important, because we need to understand the relation of fetal growth to subsequent experience over the life course. Since birth weight and IQ have both been recorded for numerous large populations, one might wonder why the question has not yet been resolved. It has proven extremely difficult, however, to disentangle a modest effect of birth weight from the larger effects of confounders such as parental socioeconomic status and genes. In addition, many investigators have focused only on the relation of low birth weight to IQ, which is also important but does not address the broader question of the relation of birth weight to IQ in the majority of the population. The design adopted by Eriksen et al. (1) exemplifies how this question can be resolved. In order to minimize confounding, they studied male siblings, for all of whom birth weight and gestation were recorded in the Medical Birth Register of Norway and a proxy for IQ was recorded by the National Conscript Service. They focused sharply on the broader question, by limiting their study to gestational ages of 37-41 weeks and to birth weight z scores (z scores for birth weight standardized to gestational age) between the 10th and 90th percentiles. They analyzed the relation of birth weight z score to IQ within as well as between families. The extremely large sample size made it possible to achieve precision for modest effects.