von Ehrenstein et. al Respond to "Are Sibling Comparison Designs Worth the Effort?".
von Ehrenstein et. al Respond to "Are Sibling Comparison Designs Worth the Effort?".
复制标题
冯·埃伦斯坦等。
DOI：
10.1093/aje/kwaa184
复制
发表时间：
2021
影响因子：
5
通讯作者：
Ritz,Beate
中科院分区：
文献类型：
--
作者：
vonEhrenstein,OndineS;Cui,Xin;Yan,Qi;Aralis,Hilary;Ritz,Beate
We thank Frisell for his insightful commentary that highlights the strengths and many challenges of the siblingcomparison design in general and, in particular, when assessing a habitual or addictive behavior during pregnancy, smoking, and a rare neurodevelopmental outcome, autism in the offspring (1, 2). We agree that the siblingcomparison design needs to be critically evaluated because of the particular biases pointed out by Frisell and that it nevertheless can contribute information to triangulation when assessing causality in observational studies (1, 3). The sibling-comparison design is indeed intuitively appealing as it provides a natural “matching on family” that in theory removes the influence of hard or impossible to measure and unknown shared family-level confounders. We agree that causal inference is hampered by the potential biases that affect a within-pair–based estimate, and these might explain why a sibling-pair–derived estimate could differ from the full cohort (source population), specifically including, as pointed out by Frisell (1, 3), amplification of nonshared confounders (eg, potentially stress in pregnancy), amplified attenuation from random measurement error (eg, smoking recording), and biases from cross-sibling interactions (eg, changes in smoking behavior due to an earlier born sibling with autism). Additionally, a sibling-comparison design needs sufficient statistical power, which in the end we did not have, even though we started with over 2,000,000 births and thousands of ASD children with siblings—but smoking prevalence in pregnancy was rare (2.5%)(2). A recent Danish study, comparable in size and spanning an over 20-year period of births, had more power, likely because of its almost an order of magnitude higher prenatal smoking prevalence (21.5%)(4). Thus, we appreciate the encouragement Frisell expressed for reporting results from a sibling-comparison design, if not for triangulation purposes then to avoid publication bias of null results (1). Our study also clearly underscores the limitations of the efforts to apply this design to a rare disease, such as autism, when the exposure prevalence is low or the exposure of interest is unlikely to change between 2 pregnancies. Frisell appears to answer his question “Sibling comparison designs, are they worth the effort?” with a restrained “yes” and encourages publishing study findings like ours, given that there are good reasons for epidemiologists to employ unusual designs if data permit and are ready for use in such a design. It allowed gaining a more in-depth understanding of the usefulness of registry data for triangulation by design, including identifying limitations and opportunities for using statewide registry data with innovative approaches, such as sibling-comparison analysis, when exposures are more frequent and/or likely to change between pregnancies, which might guide and inspire future studies. We provided not only a useful hands-on example for the design’s statistical power needs but hope to inspire future development of user-friendly power-estimating methods for it. We believe publishing and discussing our study is therefore informative in a broader sense, and we appreciate this forum in the Journal. Finally, we also agree that from a public health perspective, the lack of consensus on maternal smoking and ASD in the end might not be a real problem because there is strong evidence for other adverse health outcomes (such as asthma, growth restriction, and child cancer) and maternal smoking. Even in the absence of conclusive evidence on the risks of ASD, encouraging women not to smoke in pregnancy is a valid and highly recommended public health strategy. In …
