Dubious Science: Downplaying the Risks of Antidepressants in Pregnancy

A number of research studies have found that use of antidepressants during pregnancy increases the risk of neurodevelopmental disorders in the child. At the same time, medical websites often downplay the risks, and an examination of two studies reveals how dubious science discounts them.


Mad in America has published a resource page that provides a long list of studies that tell of how antidepressants in pregnancy may pose substantial risks to a newborn and the child’s long-term health. Given the role that serotonin plays in fetal development, it is the most common antidepressants—the SSRIs and SNRIs—that could be expected to pose the biggest risks to the fetus.

Serotonin is best known for its role as a key neurotransmitter involved in signaling between neurons in the brain’s serotonin system. Serotonin and its receptors also regulate breathing, blood flow and clotting, bladder control and bowel motility, sexual function, and more. If you get hungry—be it for food, air, sleep, sex, drugs, or a bathroom break—serotonin helps tell you so.

Serotonin also plays a key role in fetal development. It communicates between cells in embryos over the first eight weeks of rapid development, helping cells proliferate, migrate, and develop or die as needed, all over the body and brain. It provides the “cellular language” in the embryo-building blueprint.

Given serotonin’s important role in fetal development and in post-natal health, it is perhaps not surprising that medications that alter serotonin levels may pose risks to the fetus and to the child’s health. The risks include miscarriage, major birth defects including heart and face malformations,  intrauterine growth retardation, prematurity, low birth weight, small for gestational age births, newborn intensive care unit (NICU) admissions, low Apgar scores (indicators of overall newborn health), developmental delays, speech/language disorders, and neuropsychiatric disorders including autism, depression, and ADHD.

Indeed, there are direct links between known drug mechanisms, cell behavior changes, resultant embryo abnormalities, and children with increased risks of birth defects and abnormal behavior, including apparently impaired emotional processing.

Antidepressant use in pregnancy may also expose newborns to drug-withdrawal symptoms, since they, in essence, may go “cold turkey” after birth. The symptoms, which are sometimes called a neonatal abstinence syndrome, neonatal withdrawal syndrome, or poor neonatal adaptation syndrome, can include breathing problems or tremors. Researchers found such withdrawal symptoms in more than 60% of newborns whose mothers took serotonergic drugs at the time of delivery, though they typically resolve within a few days.

More seriously, the use of SSRIs late in pregnancy doubles the risk of persistent pulmonary hypertension. Although the absolute risk is low, 3 per 1000 liveborn infants versus a background incidence of 1.2 per 1000 liveborn infants, 10% to 60% of affected newborns die, and about 25% of survivors experience long-term problems including delays and disabilities.

Finally, newborns tend to be less responsive if their mothers took SSRIs during pregnancy, and that gap grows over the first month.

Downplaying the Risks

What is clear from this body of research is that the use of antidepressants during pregnancy does increase the risk that the newborn, in some way or another, will be negatively affected by that maternal drug exposure. Various studies have found that around 8% of pregnant women in the U.S. take antidepressants at some point during pregnancy.

This information should be provided to women and to the public. Yet, widely trusted sources of information downplay antidepressant risks, and instead often put the blame for such risks on “untreated depression.”

For example, the Johns Hopkins Medicine website features an “Antidepressants and Pregnancy” page that quotes Lauren Osborne, assistant director of the John Hopkins Women’s Mood Disorders Center, with this message for the public:

Women who take antidepressants, such as selective serotonin reuptake inhibitors (SSRIs), during pregnancy may worry about whether the medications can cause birth defects.

There is good news on this front. Osborne says that there is generally no need to taper off medications during pregnancy. “We can say with strong confidence that antidepressants don’t cause birth defects,” says Osborne. She adds that most studies finding a physical effect on babies from antidepressants taken during pregnancy fail to account for the effects of the mother’s psychiatric illness.

The Mayo Clinic Website, while not completely dismissing the risks, nevertheless downplays concerns about them. The website says birth defect risks associated with maternal antidepressant usage are “very low”; it describes newborn respiratory distress as a temporary withdrawal symptom, and it states that “the connection between antidepressant use during pregnancy and the risk of autism and attention-deficit/hyperactivity disorder in offspring remains unclear. But most studies have shown that the risk is very small, and some studies have shown no risk at all.”

As the Mayo Clinic Website states, some studies have concluded that there is no risk. The authors of such studies have done so even though their own “crude” data showed the risk of such abnormalities or harms were substantially higher in the children whose mothers took an SSRI or SNRI antidepressant during pregnancy. These authors make “adjustments” to the data to account for possible confounders, and such adjustments may decrease the risk to a level where it is no longer deemed a “statistically significant” difference. These adjustments may introduce more bias into estimates than they correct for due to a lack of methodological rigor in these studies. Specifically, their models may suffer from “collider bias,” because they did not first perform appropriate causal diagramming to ensure that any causal relationship between antidepressant use and neurodevelopmental risk is not distorted through controlling for factors that are influenced by antidepressant use.

Here are two studies of this type.

In a 2017 study published in JAMA, Brown et al conducted a retrospective study of the outcomes of 35,906 children born to mothers receiving public prescription drug coverage during pregnancy in Ontario from 2002 to 2010. There were 2837 children in this group whose mothers were given at least two consecutive prescriptions for an SSRI or SNRI during their pregnancy, and 33,069 had no exposure to an SSRI or SNRI during pregnancy. The researchers identified the number of children in each group diagnosed with autism spectrum disorder after the age of two.

In the SSRI exposed group, 58 of the 2837 children had been diagnosed with autism spectrum disorder after age two (2%). In the non-exposed group, 335 of the 33,069 children had been so diagnosed (1%). This showed a doubling of the “crude” risk of being so diagnosed in the antidepressant group, which was a “statistically significant” difference.

The researchers then re-calculated the risks for each group by adjusting estimates based on a matching technique using “high-dimensional propensity scores.” This is a statistical method used to create groups that appear otherwise similar on a large number of potential confounders, to estimate the effect of a treatment. Thus, this analysis is designed to adjust for patient differences between the two groups. After making this analysis, the researchers determined that the risk for a child being diagnosed with autism spectrum disorder was 1.61 times higher for those who were exposed in utero to antidepressants.

However, while the relative risk remained notably higher after this adjustment, it was barely no longer a “statistically significant difference” at a P value less than or equal to .05. And with this no “statistical significance” finding in hand, the researchers then drew this conclusion in the discussion section of their paper:

In children born to mothers receiving public drug coverage in Ontario, Canada, in utero serotonergic antidepressant exposure compared with no exposure was not associated with autism spectrum disorder in the child.

However, as a number of researchers have noted, a finding of “statistically significant” or not at a P value of .05 should not be understood as drawing a line that separates an effect into two categories, either practically significant or non-significant. Instead, a P value defines the “compatibility” between the observed evidence and what we would expect to see if we knew the whole truth, or the “confidence” that the finding falls within a range of possible outcomes.

This range is calculated as a compatibility or confidence interval. In this instance, the 95% compatibility interval ranged from .997 to 2.59, meaning that there was a 95% chance that if this study were conducted numerous times, the relative risk each time would fall within this interval. At the low end of .997, there is no difference between the two groups, while at the high end the risk is 2.59 times greater for the children whose mothers took an SSRI or SNRI during pregnancy. As one statistician noted, this compatibility interval should be seen as evidence that a harm may indeed be associated with the treatment. Here was the statistician’s explanation for a compatibility interval of this type:

If the 95% CI for an RR (relative risk) is 0.95–2.20, the traditional interpretation would have been “not significant,” but a better interpretation would be that the results are mostly compatible with an increase in risk.

So you can see in this study, once the compatibility interval is reviewed, that even after their adjustments, a relative risk of around 1.61 for the antidepressant- exposed children, rather than being evidence of no increased risk, was instead a data point “mostly compatible with an increase in risk.”

It is also notable that, in this study, the researchers conducted several subanalyses. Even after adjustment, the increased risk of autism spectrum disorder associated with SSRI or SNRI use remained statistically significant in women with “mood and anxiety disorders,” for exposure to SSRIs as a class (as opposed to SNRIs), and for “second or third trimester antidepressant exposure.”

Such findings clearly didn’t warrant a definitive conclusion that their results showed there was “no association” between in utero exposure to SSRI and SNRI antidepressants and the risk of the child being diagnosed with autism spectrum disorder.

In a 2022 study published in JAMA Internal Medicine, Suarez and colleagues analyzed outcomes in children from two databases—one of Medicaid patients and one of privately insured patients—that totaled 3 million births from 2000 to 2015. They compared outcomes in children whose mothers had taken an antidepressant in the second half of their pregnancy (145,702) to children who weren’t exposed to an antidepressant during pregnancy at all (3,032,745), and counted the number of children in each group who were diagnosed with a neurodevelopmental disorder–ADHD, autism spectrum disorder, learning disorder and 4 other disorders—by age 12.

Once again, they found that the “crude” rate of neurodevelopmental disorders in children that were exposed to antidepressants during this latter stage of pregnancy was 1.76 times higher than in the unexposed group. By age 12, 47% of exposed children in the Medicaid database, and 25% of exposed children in the private database, had been diagnosed with a neurodevelopmental disorder of some kind. In the unexposed group, the figures were 31% and 15% respectively.

These data showed a substantial possible difference in the relative risk between the two groups—“up to a doubling in risk of neurodevelopmental outcomes associated with antidepressant exposure.” The absolute risk was also so substantial that the researchers stated this study found that “Neurodevelopmental disorders were common among children of individuals with antidepressant exposure.”

However, the researchers then ran through what they described as an “extensive list of measured potential confounders,” and that led to a much lower “adjusted” risk of .97-1.18 for the exposed children. The researchers then compared outcomes of children whose mothers took antidepressants in late pregnancy with those whose mothers had taken them “in the 90 days prior to last menstrual period (LMP), but not the 30 days prior to LMP.” This brought the estimated risk down to .94-1.24. This definition of “discontinuers” would bias their findings toward zero if antidepressant exposure during the preconception period carried risks. Next, the researchers restricted their analysis to a small part of the whole sample where they could compare neurodevelopmental outcomes of different children within families where mothers had taken antidepressants during one pregnancy and not another. That brought the estimated risk down to .60-1.23. Then, the researchers concluded that “the results of this cohort study suggest that antidepressant use in pregnancy does not increase the risk of neurodevelopmental disorders in children.”

It appears that Suarez and colleagues, like Brown and colleagues, adjusted for a huge number of things that might matter (confounds), without first drawing out causal relationships to be sure that they weren’t controlling for factors that were influenced by antidepressant use (colliders). This risked introducing more bias than it corrected for through “collider bias,” distorting any causal relationship between mothers using antidepressants during pregnancy and their children suffering neurodevelopmental harm.

Yet, even after adjustments, the authors’ analyses showed that there are most likely increased neurodevelopmental disorder risks including of autism associated with mothers taking antidepressants during pregnancy, although not sharply increased risks. In an adjusted model, the 95% compatibility interval ranged from .96-1.17, meaning that there was a 95% chance that if this study were conducted numerous times, the relative risk each time would fall within this interval. At the low end of .96, there is no difference between the two groups, while at the high end the risk is 1.17 times greater for the children whose mothers took antidepressants during late pregnancy. The HDPS adjusted model produced a similar CI of .97-1.18.

Again, the main point is that the evidence presented in this study originally shows substantial possible risks. The adjustments that lessen the authors’ subsequent estimates of these risks lack methodological rigor, because the researchers skipped the essential first step of diagramming causal relationships to ensure that their attempts to correct for confounders did not distort their results by controlling for factors that were themselves influenced by antidepressant use.

There is a guild interest present in this study. Details of the adjusted models, such as the exact number of covariates in the high-density propensity score adjusted analysis, were not disclosed. The estimates resulting from these adjustments, and their misrepresentation as part of a cumulative picture from this study of evidence of no risk, serve the interests of a medical community that prescribes antidepressants to pregnant woman. Four of the authors also disclosed financial ties to pharmaceutical companies.

The Bottom Line

I am not arguing that research on this issue proves a causal link between antidepressant use in pregnancy and increased risk of harms to the fetus. But the research literature does provide reasons to worry. Such exposure appears to substantially increase risks of autism and other neurodevelopmental disorders, even though the “absolute risk” of these disorders in antidepressant-exposed children remains low (around 2-9% for autism). In addition, there are many other elevated risks related to use of SSRIs and SNRIs in pregnancy, including miscarriage, facial and heart defects, preterm birth and respiratory distress.

My point in critiquing these studies is that when popular websites, such as the Johns Hopkins Medicine and Mayo Clinic websites, downplay the possible risks of antidepressant use in pregnancy, they are ignoring the evidence and instead relying on conclusions of “no risk” that aren’t warranted. In the two studies critiqued here, the data still provide cause for concern.

Informed consent demands that women—and the public—be informed of research showing substantially elevated “crude” or unadjusted rates of risks for children exposed in utero to SSRIs and SNRIs. It also mandates they be informed that there isn’t science establishing that these drugs do not cause such risks. Patients and the public can then grapple with uncertainties present in the research literature on this issue, which is crucial to protecting children’s long-term health from the start.


Mad in America hosts blogs by a diverse group of writers. These posts are designed to serve as a public forum for a discussion—broadly speaking—of psychiatry and its treatments. The opinions expressed are the writers’ own.

Leave a Comment