[Skip to Navigation]
Comment & Response
March 2016

Sibling Comparisons and Confounding in Autism Epidemiological Studies—Reply

Author Affiliations
  • 1The Irish Centre for Fetal and Neonatal Translational Research (INFANT), Department of Obstetrics and Gynaecology, University College Cork, Cork, Ireland
  • 2Department of Epidemiology and Public Health, University College Cork, Cork, Ireland
JAMA Psychiatry. 2016;73(3):303. doi:10.1001/jamapsychiatry.2015.2882

In Reply We published a study on mode of delivery, more specifically birth by cesarean section (CS), and the development of autism spectrum disorder (ASD).1 In it, we found that CS was associated with ASD in the population. However, the association was attenuated when we used a sibling design, and we concluded that CS was likely not a cause of ASD and a more probable explanation of the association was confounding.

Add or change institution
1 Comment for this article
Comment on the exchange between Schendel and Parner and Curran et al.
Craig Newschaffer | AJ Drexel Autism Institute Drexel University
The exchange between Schendel and Parner(1) and Curran et al (2,3) on the employment of sib-pair and non-sibling designs in examining the potential association between cesarean delivery and ASD risk reflect a favorable trend toward increased focus on causal inference in investigating associations between pre- and perinatal factors and ASD risk. However, Schendel and Parner’s statement that “risk estimates from sibling comparisons will be valid only in the absence of confounding from unshared factors… [and] more bias may be created [in sib-pair versus non-sibling designs] if unshared confounding factors are more important than shared factors”1 diminishes the potential benefit of sibling designs by over-emphasizing the subtle and minimizing the obvious.

The obvious here being that both sibling and non-sibling designs can be threatened by confounders that are not shared by siblings and both designs must endeavor to measure and adjust for them. If this is properly done in a sibling study, then the additional bias that Schendel and Parner emphasize in their comment is a non-issue. Furthermore, for the sib-pair design to have the potential of introducing additional bias from uncontrolled unshared confounders, the exposure needs to be correlated in siblings, which then opens a backdoor path when conditioning on discordant pairs in the sibling design.(4) It is not obvious how familial factors would have a direct effect on mode of delivery but there could be indirect familial effects though mediating factors like fetal growth, which is associated with cesarean delivery and likely influenced by familial genetics. Adjusted for such factors in the analysis of sib-pair data (as Curran et al did with fetal growth) block the opened backdoor path and eliminate any excess bias due to other unmeasured, unshared confounders. However, it is the residual confounding due to the unmeasured, unshared factors that is likely more of a causal inference concern and would be a worry even in a non-sibling design. Moreover, if the mediating factors are associated with autism risk, then they are confounders themselves demanding appropriate statistical control regardless of design (sib-pair or non-sibling design).

One can’t argue with Schendel and Parner’s conclusion that strengths and limitations of study designs should be carefully considered and compared, but the sib-pair design does bring with it strength over non-sibling designs with respect to controlling for shared confounders and, at the same time, users of non-sibling designs should be as concerned with the proper control of unshared confounders as users of sibling designs.


(1) Schendel DE, Parner E. Sibling Comparisons and Confounding in Autism
Epidemiological Studies. JAMA Psychiatry. 2016 Mar 1;73(3):302-3.

(2) Curran EA, Dalman C, Kearney PM, Kenny LC, Cryan JF, Dinan TG, Khashan AS.
Association Between Obstetric Mode of Delivery and Autism Spectrum Disorder: A
Population-Based Sibling Design Study. JAMA Psychiatry. 2015 Sep;72(9):935-42.

(3) Curran EA, Kenny LC, Khashan AS. Sibling Comparisons and Confounding in Autism Epidemiological Studies-Reply. JAMA Psychiatry. 2016 Mar 1;73(3):303.

(4) Frisell T, Öberg S, Kuja-Halkola R, Sjölander A. Sibling comparison designs:
bias from non-shared confounders and measurement error. Epidemiology. 2012