Sibling Comparison Studies

Abstract

Unmeasured confounding is one of the main sources of bias in observational studies. A popular way to reduce confounding bias is to use sibling comparisons, which implicitly adjust for several factors in the early environment or upbringing without requiring them to be measured or known. In this article we provide a broad exposition of the statistical analysis methods for sibling comparison studies. We further discuss a number of methodological challenges that arise in sibling comparison studies. Expected final online publication date for the Annual Review of Statistics and Its Application, Volume 9 is March 2022. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.

Labor epidural analgesia and subsequent risk of offspring autism spectrum disorder and attention-deficit/hyperactivity disorder: a cross-national cohort study of 4.5 million individuals and their siblings

“…These models were adjusted for sex, birth year (each year as a separate category), and birth order (categorized into 1, 2, 3, and >4). To further account for the timestable unmeasured familial confounding shared between full siblings (ie, their shared early-life environments and an average of half of their cosegregating genes), 30 we fitted analogous stratified Cox regression models, which allowed for the baseline hazards to vary across families, thus implying that the risk comparisons were made within families and between differentially exposed full siblings. 30,44 To increase the precision of the estimates, we subsequently pooled the country-specific estimates using the inverse variance weighted fixed-effects metaanalytic model, which weighs the estimates from each country by their relative sample size.…”

“…To further account for the timestable unmeasured familial confounding shared between full siblings (ie, their shared early-life environments and an average of half of their cosegregating genes), 30 we fitted analogous stratified Cox regression models, which allowed for the baseline hazards to vary across families, thus implying that the risk comparisons were made within families and between differentially exposed full siblings. 30,44 To increase the precision of the estimates, we subsequently pooled the country-specific estimates using the inverse variance weighted fixed-effects metaanalytic model, which weighs the estimates from each country by their relative sample size. 45 In complementary sensitivity analyses, we excluded the offspring born by cesarean delivery and those born prematurely (gestational age <37 weeks) and defined individuals as having ASD or ADHD only if they had been diagnosed with each condition (or dispensed ADHD medications) at 2 separate instances.…”

“…29 This approach allowed the researchers to account for a portion of the genetic differences shared between siblings. 30 Although these studies have consistently demonstrated that populationwide associations between labor epidural analgesia and offspring ASD (odds/hazard ratio range, 1.05e1.32) were fully attenuated in the adjusted within-mother models, they have lacked sufficient statistical power to confidently exclude the possibility of a small and potentially causal association.…”

The Association Between Cognitive Ability and Body Mass Index: A Sibling-Comparison Analysis in Four Longitudinal Studies

“…Another possibility is that our within-family estimates are biased and overly conservative. Sibling designs can introduce bias through three channels: measurement error of the exposure, confounding via non-shared factors, and carryover effects [46,67,68]. Measurement error is an unlikely explanation here as the cognitive tests used have high reliability [69,70].…”

“…To our knowledge, no studies have used a sibling design to examine associations between early life cognition and BMI during adulthood. In this study, we combined sibling data from four cohort studies to examine the within-family association between adolescent cognition and adult BMI; large samples are required to achieve adequate statistical power in sibling designs [46]. Given existing evidence that associations are stronger for obesity than BMI, we used a novel statistical approach – Residualized Quantile Regression [47] – to examine (within-family) associations with cognitive ability across the BMI distribution, rather than just the (conditional) mean.…”