Clever Analysis of Nonheritable Risk Factors: Maternal Use of Antidepressants during Pregnancy and Risk of ASD

Background: Few nonheritable risk factors for ASD have been concretely established, in part due to methodological challenges of conducting observational epidemiology studies. Observational studies are subject to multiple biases and limitations that hinder causal inference. The alternative of randomized controlled trials is often not feasible for various reasons, such as ethical problems in studying randomized exposures in pregnant women. Thus, scientists attempting to identify nonheritable risk factors of ASD must extract as much insight as possible from observational studies.

Objectives: The objective of this presentation is to discuss and critique three recent epidemiology studies in Denmark that performed clever analytic techniques to examine whether maternal use of antidepressants during pregnancy increases risk of child ASD, or if the increased risk is actually due to confounding by indication.

Methods:  We examine three published studies by Sorensen et al. (2013), Hviid et al. (2014), and Gidaya et al. (2014) drawn from the same Danish register data. The different strategies used by the studies included: stratification on the indication; sibling analyses; negative controls; and simulation approaches for measurement error correction.

Results: Using traditional analytic approaches, all three studies found that maternal use of antidepressants during pregnancy increased risk of child ASD. However, after application of clever analytic approaches, only one study suggested an increased risk of ASD. All analytic approaches had their own difficulties, and such difficulties likely influenced to varying degrees the utility of these approaches. We offer insight as to what further studies on this topic can do to improve the state of science. 

Conclusions: While clever analyses are often good ideas in theory, they are challenging to carry out in practice, and may yield flawed conclusions.