Signal Evaluation: SSRIs and Persistent Pulmonary Hypertension of the Newborn

The association between selective serotonin reuptake inhibitors (SSRIs) in pregnancy and persistent pulmonary hypertension of the newborn (PPHN) illustrates how an apparently strong early signal can become smaller and more uncertain as better-controlled evidence accumulates. The initial 2006 case-control study reported an approximately six-fold association with SSRI exposure after the twentieth week of pregnancy. That result was clinically important and appropriately prompted regulatory attention.

Later studies did not reproduce a six-fold effect consistently. Meta-analyses generally supported a possible association with later-pregnancy exposure, while a very large US cohort showed substantial attenuation when analyses were restricted to women with depression and adjusted more rigorously for measured and proxy confounders. The remaining possible increase was much smaller in relative terms and small in absolute terms.

This history teaches three linked pharmacovigilance concepts: exposure timing can define the signal, confounding by the underlying maternal condition can materially change an estimate, and relative risk is not sufficient for communicating a rare neonatal outcome.

Clinical and Developmental Context

PPHN occurs when pulmonary vascular resistance does not fall adequately after birth. Blood continues to shunt away from the pulmonary circulation through fetal channels, producing hypoxaemia and potentially severe respiratory failure. It is uncommon but clinically serious.

The signal was not simply “any SSRI exposure at any point in pregnancy.” The strongest early concern related to exposure later in pregnancy. That temporal specificity matters because pregnancy is not a single biological exposure period. Drug effects around organogenesis, fetal growth and late-gestational physiological adaptation can have different mechanisms and outcomes.

A well-defined signal hypothesis is therefore:

maternal SSRI exposure in later pregnancy → possible increased risk of PPHN in the newborn.

This is more scientifically useful than the unqualified statement “SSRIs cause PPHN.” It specifies both the exposure window and the outcome that subsequent studies should test.

The 2006 Signal

Chambers and colleagues published the pivotal early case-control study in 2006. The study included 377 infants with PPHN and 836 matched controls. Fourteen case mothers and six control mothers had used an SSRI after completion of the twentieth week of gestation. The adjusted odds ratio was 6.1 (95% CI 2.2–16.8).

Several features made the finding difficult to ignore:

• PPHN is serious;
• the association was large;
• the exposure window was specific;
• a biologically plausible serotonergic hypothesis could be proposed;
• the confidence interval excluded no association.

At the same time, only a small number of exposed cases generated the estimate. A large odds ratio from 14 exposed cases can be important and still be imprecise. The wide confidence interval already signalled uncertainty about the true magnitude.

The US FDA issued a public health advisory in 2006 on the potential risk based on the emerging evidence. That was a signal-management response, not a declaration that every element of causality and magnitude had been established.

Why Confounding Was Central From the Beginning

Pregnant women who receive antidepressants differ from pregnant women who do not in ways that can affect neonatal outcomes. Depression severity, anxiety, smoking, obesity, diabetes, substance use, obstetric history, other medicines, healthcare utilisation and socioeconomic factors may all differ between groups.

This creates a classic confounding structure:

maternal psychiatric illness and associated characteristics → probability of SSRI treatment

and

maternal psychiatric illness and associated characteristics → probability of adverse neonatal outcome.

If these pathways are incompletely measured, part of the observed association can be attributed incorrectly to the medicine. The fact that an exposure precedes an outcome does not solve this problem.

Confounding by indication is especially difficult in pregnancy because withholding treatment is itself not biologically neutral. Untreated or undertreated maternal psychiatric illness can affect maternal health, nutrition, substance use, antenatal care and other determinants of pregnancy outcome. Signal evaluation therefore cannot compare medication exposure with an imaginary risk-free state.

Regulatory Uncertainty Evolved With the Evidence

The FDA revisited the concern after additional studies produced conflicting results. In its 2011 Drug Safety Communication, FDA stated that the available evidence was conflicting and that it was premature to reach a conclusion about a possible link between SSRI use during pregnancy and PPHN.

This regulatory evolution is instructive. The later statement did not imply that the 2006 advisory had been unreasonable. The original association was sufficiently serious and large to warrant attention. The change reflected a more mature evidence base in which replication was inconsistent.

Regulatory communication can therefore be provisional in the scientific sense while still being actionable. A pharmacovigilance system should be designed to update its conclusion rather than defend the wording of its first signal assessment indefinitely.

What the Meta-Analyses Added

A 2014 systematic review and meta-analysis separated early- from late-pregnancy exposure. Early exposure was not associated with a statistically significant increase in PPHN (OR 1.23, 95% CI 0.58–2.60). Later exposure was associated with an increased pooled estimate (OR 2.50, 95% CI 1.32–4.73).

This result did two things at once. It supported the importance of exposure timing, but it substantially reduced the apparent effect relative to the original six-fold estimate. The authors also translated the association into absolute terms, estimating approximately 2.9–3.5 additional PPHN cases per 1,000 births under the assumptions used in their analysis.

That translation is essential. A relative increase can sound large when the underlying event is rare. Pharmacovigilance communication should therefore ask two separate questions:

1. How much higher is the relative probability?
2. How many additional events might occur in absolute terms?

Both are relevant to benefit-risk assessment. Neither should be used alone to create reassurance or alarm.

The 2015 Medicaid Study: Watching Confounding Change the Estimate

The large cohort analysis by Huybrechts and colleagues is especially instructive because it showed how the estimated association changed as confounding control became more stringent. The study included 3,789,330 pregnancies in the US Medicaid population; 102,179 women used an SSRI late in pregnancy.

The crude occurrence of PPHN was higher among SSRI-exposed pregnancies than among unexposed pregnancies. The unadjusted odds ratio was approximately 1.51. However, comparing all exposed women with all unexposed women leaves major differences in the underlying populations.

The investigators therefore restricted analyses to women with depression and then applied propensity-score approaches, including high-dimensional adjustment intended to capture additional proxies for disease and healthcare characteristics. With stronger adjustment, the overall SSRI estimate attenuated to approximately 1.10 (95% CI 0.94–1.29). The confidence interval included no association.

A narrower analysis of primary PPHN retained a modest association, with an adjusted estimate around 1.28 (95% CI 1.01–1.64). The authors therefore did not conclude that risk was impossible. They concluded that a possible increase appeared substantially smaller than earlier estimates and that the absolute risk was low.

Why this matters for signal evaluation

The most valuable feature of the study is not simply its size. It is the stability test it provides. The estimate moved toward the null as the comparison became more clinically comparable and adjustment became stronger.

That pattern raises the probability that confounding explained part of the initial association. It does not prove that all residual association is confounding. Unmeasured factors can remain, and statistical adjustment can also introduce its own assumptions. The scientifically appropriate conclusion is therefore bounded: the early six-fold estimate was not robust, while a small late-pregnancy association could not be excluded.

Later Evidence: A Small Association Remains Plausible

Subsequent syntheses continued to support a possible association but at a much smaller magnitude than the original study. A 2019 updated meta-analysis of nine cohort and case-control studies covering more than 7.5 million subjects reported a pooled odds ratio of 1.516 (95% CI 1.035–1.997). The estimated absolute risk difference was small.

The direction of this later evidence is important. It does not justify returning to the original six-fold estimate, but neither does it support describing the signal as conclusively disproven. The evidence is compatible with a modest association concentrated around later pregnancy, with residual uncertainty about the degree to which maternal illness and correlated factors contribute.

This is a good example of signal shrinkage. Initial studies of rare outcomes can produce large estimates because exposed event counts are small, study populations are selective or confounding is incompletely controlled. As larger and methodologically stronger evidence accumulates, the estimated effect may move toward a smaller and more stable value.

Signal shrinkage should not be interpreted as failure of pharmacovigilance. Detecting a serious possible association early and then refining its magnitude is exactly what a functioning system should do.

Biological Plausibility: Supportive, Not Quantitative

Serotonin has recognised effects on pulmonary vascular tone and smooth-muscle biology, and SSRIs cross the placenta. These facts make a serotonergic contribution to fetal or neonatal pulmonary vascular adaptation biologically plausible.

Mechanistic plausibility cannot tell us whether the true clinical odds ratio is 1.1, 1.5 or 6.1. It also cannot remove confounding by maternal illness. Mechanistic evidence therefore belongs in an integrated assessment after the clinical phenotype and epidemiology have been defined.

Timing can make the mechanism more informative. If the hypothesised pathway concerns late fetal pulmonary vascular adaptation, an association concentrated in later pregnancy is more coherent than one of equal magnitude across all trimesters. That coherence supports the hypothesis but remains only one element of causal inference.

Benefit-Risk Reasoning in Pregnancy

Pregnancy safety questions require particular care because the comparator is not “medicine versus no risk.” Maternal depression can itself be serious, and abrupt discontinuation or undertreatment may create maternal and fetal consequences. A pharmacovigilance article should therefore not convert a population-level association into generic advice to stop an antidepressant.

The appropriate regulatory and clinical response is individualised assessment using current product information and clinical guidance. For pharmacovigilance, the task is to characterise the evidence accurately: exposure window, outcome severity, relative and absolute risk, confounding, uncertainty and any product-specific differences.

This separation protects the article from making two opposite errors: minimising a neonatal safety concern because maternal treatment is beneficial, or presenting the safety signal as though treatment itself has no benefit or alternatives carry no risk.

Practical Evidence Matrix

Failure Modes and Governance Lessons

The first failure mode is timing dilution. Combining first-trimester and late-pregnancy exposure into a single measure can obscure a signal that is biologically and epidemiologically time-dependent.

The second is crude comparator bias. Comparing women receiving SSRIs with all unexposed pregnancies can make treatment groups appear more different than they would if the underlying psychiatric condition and associated factors were considered. The Huybrechts study demonstrates how strongly an estimate can move when the counterfactual is improved.

The third is relative-risk amplification. Quoting a six-fold or two-fold association without the underlying rarity of PPHN can create a misleading clinical impression. Conversely, quoting only the small absolute increase can understate the severity of the neonatal outcome. Both dimensions belong in a balanced assessment.

The fourth is regulatory time collapse. FDA's 2006 concern and its 2011 statement that the evidence was conflicting belong to different evidence stages. Presenting one while omitting the other distorts the regulatory history.

Finally, the signal should not be converted into generic treatment advice. Population-level pharmacovigilance evidence informs benefit-risk assessment; individual treatment in pregnancy requires current product information, clinical guidance and patient-specific evaluation.

Illustrative inspection questions

An inspector assessing governance of a pregnancy safety signal might ask:

• Is gestational timing captured accurately enough to test the defined signal hypothesis?
• Are maternal disease, smoking, obesity, diabetes, concomitant medicines and other important confounders considered when data permit?
• Are PPHN case definitions and source-data limitations documented?
• Does the aggregate assessment show both relative and absolute risk where estimable?
• Are regulatory communications presented chronologically so that changes in evidence are visible?
• Can the organisation explain why a historical signal conclusion was strengthened, weakened or retained as new studies appeared?

These are illustrative questions rather than examples of actual inspection findings or new mandated data elements.

Key Takeaways

The SSRI–PPHN history is a model of how pharmacovigilance should behave when an early strong signal becomes more modest with better evidence.

The original 2006 study reported a large association with exposure after 20 weeks and appropriately triggered regulatory attention. Later studies were inconsistent. Meta-analysis supported a possible late-pregnancy association but with a smaller effect. A very large confounding-focused cohort showed substantial attenuation, while later pooled evidence remained compatible with a modest increase.

The durable lessons are:

• define pregnancy safety signals by biologically relevant exposure windows;
• treat large estimates from small exposed-event counts as important but imprecise;
• examine how estimates change when the underlying maternal condition and other confounders are addressed;
• communicate absolute risk alongside relative measures;
• allow regulatory conclusions to evolve as the evidence base changes;
• distinguish population-level safety evidence from individual treatment decisions.

The final signal conclusion is therefore not that the original concern was either wholly correct or wholly mistaken. It is that the evidence evolved from a large initial association toward a smaller possible late-pregnancy effect whose absolute risk is low and whose interpretation remains sensitive to confounding. That is a more useful pharmacovigilance conclusion because it preserves both what is known and what remains uncertain.

References

1. Chambers CD, Hernandez-Diaz S, Van Marter LJ, et al. Selective serotonin-reuptake inhibitors and risk of persistent pulmonary hypertension of the newborn. New England Journal of Medicine. 2006;354(6):579–587. PMID: 16467545.
2. US Food and Drug Administration. Public Health Advisory: treatment challenges of depression in pregnancy and the possibility of persistent pulmonary hypertension in newborns. 2006.
3. US Food and Drug Administration. FDA Drug Safety Communication: Selective serotonin reuptake inhibitor (SSRI) antidepressant use during pregnancy and reports of a rare heart and lung condition in newborn babies. 14 December 2011.
4. Grigoriadis S, VonderPorten EH, Mamisashvili L, et al. Prenatal exposure to antidepressants and persistent pulmonary hypertension of the newborn: systematic review and meta-analysis. BMJ. 2014;348:f6932. PMID: 24429387.
5. Huybrechts KF, Bateman BT, Palmsten K, et al. Antidepressant use late in pregnancy and risk of persistent pulmonary hypertension of the newborn. JAMA. 2015;313(21):2142–2151. PMID: 26034955.
6. Ng QX, Venkatanarayanan N, Ho CYX, Sim WS, Yeo WS. Selective serotonin reuptake inhibitors and persistent pulmonary hypertension of the newborn: an update meta-analysis. Journal of Women's Health. 2019;28(3):331–338. PMID: 30407100.

Regulatory Note

The FDA communications cited here describe the evolution of a US regulatory safety concern and are not EU legal requirements. The epidemiological studies and meta-analyses are scientific evidence, not regulatory mandates.

Pregnancy treatment decisions should be based on the current authorised product information and applicable clinical guidance for the relevant medicine and jurisdiction, with individual assessment of maternal and fetal benefit-risk. The analytical framework and illustrative inspection questions in this article are explanatory recommended practice.