El Marroun and colleagues’ findings

El Marroun et al suggest an association between prenatal SSRI exposure and autistic traits in children, as well as an association between prenatal depressive symptoms without SSRI exposure.^{Reference El Marroun, White, van der Knaap, Homberg, Fernández and Schoemaker1} However, the paper concludes that long-term drug safety trials are needed before evidence-based recommendations are possible. There is no doubt that a well-designed randomised controlled trial would be the best tool to answer the question about SSRI exposure in pregnancy and autism. Such a trial would account for both measured and unmeasured confounding and if we detect an effect, we can feel fairly certain that it is a causal effect. One question is whether it is feasible. First, if there already is some evidence that suggests harm to the unborn child, would it be ethical to conduct such a trial? Second, a trial of this kind requires a very large sample size (as autism is a relative rare outcome) and may struggle with recruitment, adherence and experience some difficulties in following children up several years after birth. In that respect, the numbers from the study by El Marroun et al speak for themselves. Out of nearly 9000 eligible pregnant women, 69 received an SSRI at some point in pregnancy and these numbers rapidly dwindled when it came to the measurement of the outcome. Going through the thought process of designing a safety trial on antidepressants in pregnancy and autism it soon becomes apparent that it would be very difficult if not impossible. Nevertheless, it offers us the opportunity to explore the value of observational studies and highlight the areas where these are likely to be subject to biases.

El Marroun et al compare the children of a small group of women who were reported to receive SSRIs treatment in pregnancy against children of women with no exposure to SSRIs and a low score of maternal depressive symptoms. They also compare the latter group against women who were depressed, but not on antidepressants. The question is whether these are meaningful and unbiased comparisons.^{Reference Petersen, Gilbert, Evans, Man and Nazareth7} Table 1 in El Marroun et al reveals that not only do these groups differ in terms of exposure to antidepressants, but also on a range of measured covariates. For example, a quarter of the women on SSRIs were also prescribed benzodiazepines and over 50% were smoking in pregnancy. In contrast, less than 1% of the women in the non-depressed comparison group were prescribed benzodiazepines and less than 25% were smoking. A multivariable regression analysis may appear to account for some of these differences, but only for those variables that were measured and included in the analyses. We cannot ignore the fact that those who continue antidepressants in pregnancy may also be women with a history of more severe depression.

Although the paper goes to some lengths in evaluating parental reports on the outcomes there remains the possibility that the outcomes in this study are subject to reporting bias. Moreover, the absence of figures of absolute risks of autism makes it difficult for the reader to assess the numbers of children who actually experienced pervasive developmental problems. Further, if these numbers are small, the results of multivariable regression analysis can be very sensitive to misclassifications of the outcome. El Marroun et al suggest that their results ‘demonstrated an effect of SSRIs on autistic symptoms in young children’.^{Reference El Marroun, White, van der Knaap, Homberg, Fernández and Schoemaker1} They support this statement with evidence from animal models. However, there is an alternative interpretation of the results of this study.

A different interpretation of the results

El Marroun et al showed that compared with children of women without depressive symptoms, in children of women with elevated depressive symptoms the odds ratio of pervasive developmental problems was 2.02 (95% CI 1.53-2.66) and the odds ratio for those prescribed SSRIs in pregnancy was 2.58 (95% CI 1.46-4.54). Indirectly, this suggests that children of women with elevated depressive symptoms and women treated with SSRIs in pregnancy were equally likely to have pervasive developmental problems (Model I, Table 2). A head-to-head comparison of the effect estimates for autistic traits of the SSRI cohort against the cohort of women with depression suggested that children of the SSRI cohort rated slightly higher on all three domains. This, however, does not prove a causal relationship between SSRI exposure in pregnancy and autistic symptoms. Rather the results may suggest that there is an association between maternal depression and childhood autistic symptoms. Hence, some animal studies have established an association between maternal stress and autism.^{Reference Howerton, Morgan, Fischer and Bale10} Only a small proportion of women continue antidepressants in pregnancy. Thus, prescription of SSRIs in pregnancy may be a mere marker of the severity and other characteristics of the underlying depressive illnesses. Indeed, this study confirmed that the vast majority of women with depressive symptoms are left untreated in pregnancy. This may potentially pose a far greater risk to the welfare of the mother and development of the child. We fear that the proposed link between SSRIs and childhood autistic symptoms is yet another red herring with potentially detrimental consequences. It may fuel further anxiety and guilt among women, who are faced with depression in pregnancy and possibly leave some women without treatment.