For the video version of this post, click here.
Theres a reason why so few medications are recommended during pregnancy, and for the most part its not because we have evidence that they cause fetal harm. No, the reason so few drugs are recommended in pregnancy is because theyve never been tested in pregnant women. With the exception of a few medications explicitly designed for pregnancy, pregnant women are almost universally excluded from clinical trials.
So to determine if a drug may be harmful in pregnancy, we have to use observational data, and that means adjustment. This week, we can use a study appearing in the Journal of the American Medical Association as a perfect example.
Do anti-depressants, when used in pregnancy, increase the risk of persistent pulmonary hypertension of the newborn? PPHN is a very rare, but highly morbid condition whereby the fetal circulation persists after birth, leading to hypoxia, respiratory failure, and in about 10% of cases, death.
To determine if anti-depressants increase the risk of PPHN, a group of Harvard researchers used data from the Medicaid eXtract database, comprising almost 4 million pregnant women. Of those 4 million, about 130,000 had filled a prescription for an anti-depressant towards the end of pregnancy, mostly SSRIs. The rate of PPHN was 20 out of 10,000 births among those not taking anti-depressants, and 30 out of 10,000 births among those who did. Case closed?
Well, no. These women werent randomized to take anti-depressants, they took them for a reason, and lots of factors play into this: depression, obviously, but also personal beliefs, access to care, and any number of other confounders.
Enter adjustment. The authors worked very hard to identify a slew of confounders, and after accounting for them, the relationship between anti-depressants and PPHN went away. So it would seem that women who are more prone to take anti-depressants are more prone to have a child with PPHN, but the medications are not causal.
Unless...there was over-adjustment. Over-adjustment occurs when you statistically account for something that lies on the causal pathway between the exposure and outcome of interest. As an example, what if anti-depressants increase the risk of premature birth, which increases the risk of PPHN. Adjusting for prematurity would make it look like the medications were safe, when in fact they werent.
It can be hard to tease out. But theres a really important fact that we shouldnt forget here. All medications have risks, and all medications have benefits. Depression in a mother is a much, much greater risk to a newborn than PPHN.
Personally, I think the signal of harm is small enough to essentially be insignificant. For women with depression, these medications may, in fact, dramatically benefit both mother and baby.