Simply put, I love this article. It addresses a significant, but difficult to answer, question, with superb study design, analysis, and reporting. What a pleasure to read!
Source: Brown HK, Ray JG, Wilton AS, et al. Association between serotonergic antidepressant use during pregnancy and autism spectrum disorder in children. JAMA. 2017;317(15):1544-1552; doi:10.1001/jama.2017.3415. See AAP Grand Rounds commentary by Dr. Dan Doherty (subscription required).
Investigators in Canada used an administrative database to analyze over 35,000 births, including 2837 pregnancies where mothers were exposed to antidepressant medication and the 1.1% of those 35,000+ births where a diagnosis of autism spectrum disorder (ASD) was made. After extensive analysis with corrections for confounding variables, the authors found no association between in utero serotonergic antidepressant exposure and ASD in their offspring (adjusted hazard ratio* 1.60 with 95% confidence interval 0.69 - 3.74).
I want to highlight just a few items with this article. First of all, know that a retrospective cohort analysis cannot prove that there is no causal role for antidepressants in childhood autism, it can only show whether or not there is an association. It's impossible to "prove" a negative in this type of study.
The authors included 2 very important features in their study:
1. They used high-dimensional propensity analysis (HDPA) to attempt to correct for about 500 covariates that might confound results. Regular readers will recall we discussed propensity analysis recently. This is a great help in retrospective studies, though of course not a perfect answer for confounding variables.
2. They also used sibling controls for their children with ASD. With any evaluation of causes of neurodevelopmental outcomes, it's always difficult to isolate the possible causative agent (maternal antidepressant use, in this example) from all the other variables that can affect neurodevelopment. One method to help with this "nature and nurture" conundrum is to look at sibling controls, where the other sibling is discordant for the agent of interest (e.g. did not have in utero exposure to antidepressants).
Use of these 2 features lends greater strength to the authors' conclusions.
In addition to the above, I also liked the researchers' discussion of study limitations. These included
1. The population under study consisted of women in a publicly funded drug plan; this population may have important differences from the general population in terms of other risk factors for their infants' neurodevelopmental outcomes.
2. The study could only collect information on whether prescriptions for antidepressants were provided, not whether the women took the medication as prescribed.
3. The diagnosis of ASD was determined by data related to pediatrician or psychiatrist diagnoses, so would have missed ASD diagnosed by family practitioners or psychologists.
4. They had no information about the infants' fathers or their postnatal environment, both of which could impact ASD diagnosis.
5. The diagnostic coding system itself might be biased in how covariates are listed; thus, someone using a different coding system might come up with different variables and different conclusions.
I could go on and on about this study. Certainly, I will keep it on hand to use in my future EBM teaching sessions. For now, I think we can offer mothers a fair amount of reassurance that use of serotonergic antidepressant medication during pregnancy is unlikely to be an independent risk factor for ASD in their children.
*The link for this explanation is a site that is new to me, called "Students 4 Best Evidence." It looks interesting, I plan to check it out a few more times. I'm always happy when I see students using EBM!