Strengths and Limitations
One strength is that we quantified the impact of exposure
misclassification, applied methods to deal with time-varying exposure
and confounders, missing data, and examined ADHD risks from a diagnosis
and symptom perspective.7 We attempted to limit
confounding by indication by including only women with clinical
depression/anxiety during pregnancy, and measured their symptom severity
at two time points in pregnancy via a validated
instrument.17 We carried out several sensitivity and
sub-analyses to explore the robustness of our findings, as well as the
role of confounding by maternal psychiatric indicators prior to
gestation. In addition, we attempted to overcome the limitation of
averaged HRs by estimating period-specific hazards; however, the
built-in bias of differential selection on these results cannot be
excluded. In addition, we cannot rule out the role of residual
confounding by depression severity, genetic, environmental or familial
factors, or even chance, on our findings.
Several limitations need mentioning. Symptoms of depressive and anxiety
were not measured at baseline, but only at two time points in pregnancy.
We relied on maternal self-report of depression/anxiety during or prior
to pregnancy, which cannot replace a clinical diagnosis. Information on
dosage is not available in MoBa. The ADHD symptom measure was
mother-reported. Although the risk of outcome misclassification cannot
be ruled out, this was probably non-differential, and the
depression-distortion bias had a negligible impact on our effect
estimates. Also, the internal consistency of the CPRS-R was high. The
MoBa study has a low response rate (41%), with a possible
self-selection of the healthiest women into the
cohort.10,11 Although association measures have been
shown to be valid in MoBa in relation to immediate birth
outcomes,34 the impact of selection bias on
longer-term outcomes cannot be excluded.35 Our small
sample size precluded analyses of SSRI and SNRI as separate groups, or
for individual antidepressants, as well as sibling-design analysis. We
could, however, take into account familial risk of ADHD using parental
ADHD-medications use, as well as parental self-report ADHD symptoms in a
subsample.