Neonatal Risks of Maternal Treatment With Mood Stabilizers—Reply

In reply

We note with interest comments by Einarson et al and Olfson and Marcus. Einarson et al raised questions about the Abstract. We believe that the Abstract is consistent with data presented in the article, and specific details were omitted because of space limitations. In response to the questions raised, we offer the following comments:

First, the rates of small for gestational age (SGA) in our sample may have been lower than what would have been expected for a population; this, however, should not invalidate or detract from our findings that the proportion of propensity score–matched infants of depressed mothers treated with SSRIs was significantly greater than for infants of mothers in the depressed or control groups. We used an American standard reference for SGA¹ and our sample from British Columbia did not have the same incidence of SGA. Moreover, we used the same standard for the treatment and comparison groups and the use of an American standard should not bias the estimate of the difference. Why our infants during the study period would be heavier at birth remains to be determined.

Second, propensity score matching was used to account for the impact of maternal illness severity. It is a valid and widely^{2 - 4} used technique to account for nonrandomized confounding variables. Our study was directed at addressing an increasing awareness that there is a need to distinguish the effects of SSRI exposure from the effects of the disease it treats,⁵ which is itself a well-recognized determinant of neonatal outcomes in this setting. Further, because double-blind randomized controlled studies of the impact of either antidepressant medication alone and maternal depression alone are not ethically or medically feasible, we believe our approach was an appropriate use of population-based data that, while not optimal, does takes us one step closer to distinguishing the impact of SSRI exposure from exposure to maternal depression.

Third, we agree that we were not able to independently verify that this cohort was indeed depressed; however, using British Columbia Medical Services Plan records and their International Classification of Diseases, Ninth Revision codes, we were able to identify women who had been given a diagnosis of depression and did not receive a prescription for an antidepressant during their pregnancy.

Fourth, we agree that adjustment for multiple testing is a key part of hypothesis testing; however, we intentionally did not make this adjustment because our focus was on testing preestablished hypotheses that had been previously observed in multiple separate studies of the impact of prenatal exposure to SSRIs and maternal depression,⁶ and thus, adjustment for multiple tests may, in this setting, be unnecessary. Furthermore, by reducing the chance of a type I error (incorrectly finding a difference or effect due to chance alone), we may increase the risk of a type II error (ie, that noneffect was found when indeed there was an effect).

We did acknowledge that differences in some neonatal outcomes were small and that clinical correlations are needed; however, our findings are consistent with numerous cohort studies⁵ showing reduced birth weight and poor neonatal adaptation following prenatal SSRI exposure. The test of relevance here should not be “clinical significance” but, rather, “population health significance.” That is, what will be the burden of morbidity in the population given the level of use of SSRIs? In this case, a relatively small effect size applied to a potentially large patient base can, indeed, have a considerable effect on population health.

In their letter, Olfson and Marcus point out that pregnant mothers treated with SSRIs may have been more likely to receive mood stabilizers (anticonvulsants, lithium) than those who were not treated with antidepressants and that it is possible that in utero exposure to mood stabilizers could partially explain the observed association between maternal SSRI treatment and the adverse neonatal outcomes. Our cohort of depressed mothers treated with SSRIs was selected from women who did not receive a prescription for benzodiazepines or antipsychotic medication during their pregnancy. While we are not able to control for the use of lithium, we were able to look at Miscellaneous Anticonvulsants (American Hospital Formulary Service code 28:12:92), which includes valproate sodium and carbamazepine. The number of mothers receiving these drugs was small and excluding them did not affect the results.

The treatment of depression during pregnancy is an urgent health concern for mothers and their infants, and our findings in no way suggest that nontreatment is an option. Quite the contrary, our findings illustrate how both depression and SSRI medications influence neonatal health; and thus, the real question remains one of what role do SSRI medications have in a treatment approach that thoughtfully balances risks to both the mother and infant.

Correspondence: Dr Oberlander, Early Human Experience Unit, Centre for Community Child Health Research, Room L408, 4480 Oak St, Vancouver, BC V6 3V4, Canada (toberlander@cw.bc.ca).

Financial Disclosure: None reported.