The use of the gold standard in pharmacologic research, the randomized controlled trial, is restricted in pregnant women due to ethical concerns.
42 Thus, to a greater extent than in other populations, research on the effects of medications in pregnancy has relied on observational studies, which are unable to establish causation due to the limits of their design.
43,44 Historically, early observational studies of antidepressants or other psychiatric medications in pregnancy did a poor job of controlling for confounding factors, especially psychiatric illness itself, leading to erroneous associations and conclusions. For example, initial comparator groups were often healthy women versus women with mental illness taking medications,
rather than women with mental illness not taking psychiatric medications being compared to women with mental illness taking psychiatric medications. Controlling for mental illness severity also remains a challenge—women with less severe psychiatric illness tend to take less or fewer medications, especially in pregnancy; may be better equipped to engage in healthy behaviors and psychotherapy; and may be at less risk for adverse obstetrical outcomes than women with severe mental illness, whether or not the latter are taking medications. Additionally, studies finding no association between perinatal medications and negative fetal outcomes are less likely to be published due to bias in favor of positive findings.
45 Meta-analyses, which are higher powered to find associations by virtue of combining multiple prior studies, have no ability to change the design flaws already present in the studies they use and, therefore, may amplify confounding,
46 leading to false accuracy and certainty.
47 Finally, the use of pharmacy databases and filled prescriptions to denote adherence to medications is particularly problematic given high rates of women stopping psychotropic medications suddenly in pregnancy.
48 Newer statistical advances that have improved the ability to control for confounding in observational studies, such as propensity score matching, have begun to correct some of the erroneous associations found in earlier studies and should continue to advance the field.
44,47,49