Re: Parental depression, maternal antidepressant use during pregnancy, and risk of autism spectrum disorders: population based case-control study
We thank Glukhov (1), Morgan (2), Grant (3), and Bratt (4) for their comments on our paper (5).
There are unique challenges in human research related to long term offspring outcomes of in-utero exposures, particularly when exposures and outcomes are relatively rare and measured several years apart. As discussed in the paper, robust and adequately powered experimental data in humans are unlikely to be ever available on this topic and the only alternative is to use observational data.
Our study was nested within a longitudinal design in a population based sample, covering all known pathways of autism diagnosis and care in Stockholm County, where free and universal access to care is offered. This reduces the possibility of selection bias. Since the data were recorded prospectively, before the birth of the child, the possibility of recall and attribution biases is also eliminated. The design suggested by Bratt (5) may be prone to such biases.
It is also important to mention that a low statistical power in our study was not because of the lack of ‘affected individuals’, it was because the exposure was very rare despite relatively large numbers (we had 4429 individuals had an autism spectrum disorder in our main sample, of which 1679 had data on maternal medication use during pregnancy but antidepressant use during pregnancy). Therefore the database with 700 families that Bratt suggests (4) is unlikely to be able to address this limitation since it is less than half the size of our subsample with medication data. Glukhov also wondered if one of the p-values in Table 1 was incorrect (1)- we can confirm it is accurate, and as noted in the table footnote, was derived after conditioning on age and sex to account for the matched design.
Confounding is a major problem in observational studies. We considered a number of potential confounders and adjusted where possible. Morgan correctly mentions that this adjustment may not be very effective given the low power (2). Of more concern is the possibility of residual confounding by factors that cannot be measured or adjusted for and the possibility of confounding by indication. Glukhov suggests the possibility of ascertainment bias concerning health service use (1) and we note other potential limitations in our discussion (5).
As a result we do not claim any causal associations. On the contrary, we expressly warn against jumping to such conclusions, particularly when making clinical decisions (5).
One area of harmonious consensus between all correspondents (1,2,3,4) and authors is that there is much more work required to understand these associations. We sincerely hope our paper will prompt others to attempt to replicate, refute or improve understanding of these findings. The last word has certainly not been said, and Croen et al’s (6) and our study should be viewed as just the beginning in the understanding of this topic in human populations.
(1) Glukhov V. http://www.bmj.com/content/346/bmj.f2059/rr/642022
(2) Morgan T. http://www.bmj.com/content/346/bmj.f2059/rr/642417
(3) Grant ECG. http://www.bmj.com/content/346/bmj.f2059/rr/643182
(4) Bratt AM. http://www.bmj.com/content/346/bmj.f2059/rr/643508
(5) Rai D, Lee BK, Dalman C, Golding J, Lewis G, Magnusson C. Parental depression, maternal antidepressant use during pregnancy, and risk of autism spectrum disorders: population based case-control study. BMJ 2013;346:f2059
(6) Croen LA, Grether JK, Yoshida CK, Odouli R, Hendrick V. Antidepressant use during pregnancy and childhood autism spectrum disorders. Arch Gen Psychiatry 2011;68:1104-12
Competing interests: No competing interests