Abstract
The strong relationship found by Lewis et al. [1] between in utero exposure to antidepressants and neonatal growth outcomes is a cause for concern. However, incomplete statistical analysis combined with recruitment bias makes it impossible to know the extent to which the relationship is due to antidepressant exposure and to what extent it is due to social variables strongly correlated with antidepressant usage.
Table 1 reported that 11 (40.7%) of the 27 subjects in the medication group were ‘Employed full/part time’ compared to 4 (14.8%) of the 27 subjects in the control group, and that 3 (11.1%) of the subjects in the medication group were ‘On leave part or full time’ compared to 15 (55.6%) of the control group. This equates to 21.4% (3/14) of the medication subjects with a job being on leave compared to 78.9% (15/19) of the control subjects with a job being on leave. The highly significant x2 (13.034) and p (0.004) values for the ‘Employed full/part time’ variable were recorded in Table 1 but not the x2 and p values for the ‘On leave part or full time’ variable.
I am very surprised these highly statistically significant variables, strongly correlated with the outcomes and intuitively appealing as contributing in the expected way to the outcomes, are not mentioned anywhere in the study other than Table 1. I would be very interested to know the strength of the association between these variables and the outcomes and the extent they, compared to medication status, contribute to variance. I would be particularly interested in the comparison between the subjects who continue to work in pregnancy and those who choose to take leave.
While medication subjects were recruited via referrals from private psychiatrists and medical staff at the Mercy Hospital for Women and were already taking antidepressants, the control group was recruited via antenatal appointments at the same hospital but were not taking antidepressants. Depressive symptom severity as captured with the Beck Depression Inventory, second edition (BDI-II) would not be a valid measure of the true differences between the groups. One would not expect the control subjects to have the same complex psychiatric, relationship and social problems as subjects taking antidepressants and referred by psychiatrists and other doctors, irrespective of BDI-II scores.
Finally, although no significant associations between levels of depression in pregnancy or at birth and growth outcomes were found, the authors report neither presence nor absence of significant associations at one month. Was there an association with levels of depression which the authors found inconvenient with respect to the conclusions they seemed so eager to draw?