Signal Evaluation: SSRIs and Persistent Pulmonary Hypertension of the Newborn

Introduction

The historical association between selective serotonin reuptake inhibitors (SSRIs) and persistent pulmonary hypertension of the newborn (PPHN) is a particularly useful example of how a pharmacovigilance signal can change substantially as evidence accumulates.

The signal began with an epidemiological observation that appeared large enough to warrant immediate attention.

A 2006 case-control study reported an approximately six-fold increase in PPHN following SSRI exposure after the twentieth week of pregnancy. The study included 377 infants with PPHN and 836 matched controls; 14 cases had late SSRI exposure compared with 6 controls, producing an adjusted odds ratio of 6.1 (95% CI 2.2–16.8). 1

That finding was clinically important because PPHN is a serious neonatal condition associated with substantial morbidity and mortality.

It was also biologically interesting because serotonin has recognised effects on pulmonary vascular biology.

But the signal immediately raised difficult questions.

Was the observed association causal?

Was the risk specific to SSRIs?

Was it specific to exposure late in pregnancy?

Could maternal depression or other characteristics associated with antidepressant treatment explain part of the association?

And, most importantly for clinical decision-making, how large was the absolute increase in risk?

Subsequent studies produced conflicting findings. A 2014 systematic review and meta-analysis found an increased association with late-pregnancy SSRI exposure, but estimated the absolute increase to remain small. 2

A much larger 2015 US cohort study then examined 3,789,330 pregnancies and found that the apparent association was substantially attenuated after restricting to women with depression and adjusting for measured and proxy confounders. For SSRI exposure, the adjusted odds ratio was 1.10 (95% CI 0.94–1.29). 3

The history of this signal therefore provides an unusually good demonstration of a core pharmacovigilance principle:

The strength of an initial signal is not necessarily the strength of the final causal conclusion.

A signal should evolve as better evidence becomes available.

1. Define the Signal

The signal can be expressed as:

Maternal SSRI exposure during pregnancy → persistent pulmonary hypertension of the newborn.

The important refinement is timing.

The strongest early concern related to:

exposure during late pregnancy.

This distinction matters because fetal development and pulmonary vascular adaptation change throughout gestation.

The historical evidence did not establish that all prenatal SSRI exposure carried the same risk.

Indeed, the original 2006 study found an association with exposure after completion of the twentieth week but not with exposure before that point. 4

Therefore, the initial signal was more appropriately framed as:

Late-pregnancy SSRI exposure may be associated with an increased risk of PPHN.

That is a more precise signal hypothesis than simply stating:

SSRIs cause PPHN.

2. What Is Persistent Pulmonary Hypertension of the Newborn?

PPHN occurs when the normal fall in pulmonary vascular resistance after birth does not occur adequately.

The newborn therefore retains abnormally high pulmonary vascular resistance, resulting in impaired pulmonary blood flow and right-to-left shunting through fetal circulatory pathways.

The clinical consequence can be profound hypoxaemia and respiratory failure.

PPHN is therefore a serious outcome even though it is uncommon.

This combination is important in pharmacovigilance:

low frequency + high clinical severity

can justify investigation even when the number of cases is small.

3. Why Timing of Exposure Matters

The original signal was not a general association between:

any SSRI exposure

and:

PPHN.

The reported association was concentrated in late pregnancy.

That temporal specificity is important because it provides a potentially testable biological hypothesis.

If a drug effect is causally related to a developmental or physiological transition occurring late in gestation, the risk might reasonably be expected to vary according to exposure timing.

Conversely, if the association appears equally strong regardless of timing, that would weaken a narrowly timing-dependent biological explanation.

Timing therefore serves two roles:

1. it helps define the clinical signal;
2. it provides evidence relevant to biological plausibility.

4. The Initial Signal: 2006

The key historical study was published by Chambers and colleagues in the New England Journal of Medicine in 2006.

The investigators conducted a case-control study involving:

• 377 women whose infants had PPHN;
• 836 matched control women.

Maternal medication exposure was assessed through blinded interviews.

Fourteen infants with PPHN had been exposed to an SSRI after completion of the twentieth week of gestation compared with six control infants.

The adjusted odds ratio was:

6.1 (95% CI 2.2–16.8).

In contrast, neither SSRI exposure before the twentieth week nor non-SSRI antidepressant exposure was associated with increased risk in that study. 5

The result was striking.

It was also based on a relatively small number of exposed cases.

That distinction matters.

5. Why the Initial Result Was Important

A six-fold association with a serious neonatal condition is sufficient to generate regulatory concern.

The result had several features that made the hypothesis worthy of further evaluation:

• the outcome was serious;
• the exposure was biologically active;
• the association was temporally specific;
• the result was statistically significant;
• and a plausible biological mechanism could be proposed.

However, the study could not answer every important question.

It could not establish:

• whether the association was reproducible;
• whether it applied to all SSRIs;
• the precise absolute risk;
• the extent of confounding by maternal disease;
• or whether the magnitude of association was substantially inflated by study design or residual bias.

Signal generation and signal confirmation are different stages.

6. Regulatory Attention

The initial evidence prompted regulatory concern.

The US Food and Drug Administration issued a public health advisory in 2006 concerning the potential association between SSRI exposure during pregnancy and PPHN.

The regulatory position subsequently changed as additional evidence accumulated.

By 2011, the FDA had reviewed additional studies with conflicting findings and concluded that it was premature to reach a conclusion about a possible link between SSRI use during pregnancy and PPHN. 6

This is an important historical pharmacovigilance lesson.

A regulatory warning or advisory does not necessarily represent the final scientific conclusion.

Regulatory communication may change as the evidence changes.

7. Conflicting Epidemiological Evidence

After the original 2006 publication, subsequent studies did not produce a consistent result.

Some studies reported an association.

Others did not.

Several factors made the evidence difficult to interpret.

These included:

• small numbers of PPHN cases;
• differences in PPHN definitions;
• different exposure windows;
• differences in control groups;
• differences in adjustment for confounding;
• differences in ascertainment of SSRI exposure;
• and differences in control for neonatal or maternal risk factors.

A signal that produces different estimates under different study designs requires careful epidemiological assessment rather than simple vote-counting.

8. Why Confounding Matters

Pregnant women who receive antidepressants are not necessarily comparable to pregnant women who do not.

They may differ in:

• psychiatric diagnoses;
• severity of depression;
• smoking;
• obesity;
• diabetes;
• substance use;
• healthcare utilisation;
• obstetric history;
• socioeconomic characteristics;
• and use of other medicines.

Some of these factors may themselves influence neonatal outcomes.

This creates the possibility of confounding by indication or by the underlying illness.

The critical question therefore becomes:

Is the observed association caused by SSRI exposure, the maternal condition associated with SSRI use, other characteristics associated with treatment, or some combination?

9. Depression as a Confounder

Depression is particularly important in this signal.

If:

depression → increased probability of SSRI treatment

and

depression or associated factors → increased probability of PPHN

then an apparent:

SSRI → PPHN

association may be partly confounded.

This does not prove that depression explains the association.

It means the analysis must attempt to separate:

treatment effect

from:

underlying disease and associated characteristics.

That distinction became increasingly important as larger epidemiological datasets became available.

10. The 2014 Meta-analysis

A 2014 systematic review and meta-analysis examined seven eligible studies.

The analysis found no significant association between early-pregnancy SSRI exposure and PPHN:

OR 1.23 (95% CI 0.58–2.60).

For late-pregnancy exposure, however, the pooled estimate was:

OR 2.50 (95% CI 1.32–4.73).

The authors estimated an absolute risk increase of approximately:

2.9–3.5 additional cases per 1,000 births

for late-pregnancy exposure, corresponding to an estimated 286–351 women treated in late pregnancy for one additional associated case under their assumptions. 7

This represented a substantial reduction from the six-fold association reported in the original study.

But the association remained statistically significant.

The signal had therefore not disappeared.

11. Relative Risk Versus Absolute Risk

This case is an excellent demonstration of why pharmacovigilance communication should not stop at relative measures.

Suppose the baseline risk is approximately:

2 cases per 1,000 births.

An approximate doubling of risk would produce:

4 cases per 1,000 births.

The relative increase sounds substantial.

The absolute increase is approximately:

2 additional cases per 1,000 births.

Both statements can be correct.

But they communicate very different clinical impressions.

A responsible benefit-risk assessment should provide both where possible.

12. Why the Meta-analysis Did Not Settle the Question

Meta-analysis increases statistical power.

It does not eliminate weaknesses in the underlying studies.

The 2014 investigators explicitly identified important limitations, including:

• residual confounding;
• inconsistent adjustment for risk factors;
• uncertainty concerning exposure ascertainment;
• heterogeneity in outcome definitions;
• inability to assess some potential modifiers;
• and the fact that association does not establish causation. 8

The analysis therefore strengthened the evidence for an association but did not convert an observational association into proof of causality.

13. The Large 2015 Cohort Study

The next major development was the large study by Huybrechts and colleagues, published in JAMA in 2015.

The study examined:

3,789,330 pregnancies

from the US Medicaid population.

Of these:

• 102,179 women used an SSRI late in pregnancy;
• 26,771 used a non-SSRI antidepressant.

The investigators deliberately examined the effect of progressively stronger confounding control.

This was particularly important because it allowed readers to see how much of the apparent association was explained as the analysis became more rigorous. 9

14. The Unadjusted Result

Initially, the SSRI association appeared positive.

The unadjusted odds ratio was:

1.51 (95% CI 1.35–1.69).

This is important because it demonstrates that even a very large dataset can initially reproduce an association.

The key question is what happens after accounting for differences between treated and untreated populations.

15. Restricting to Women With Depression

The investigators first restricted the cohort to women with a diagnosis of depression.

This is conceptually important.

Instead of comparing:

women receiving SSRIs

with:

all women not receiving SSRIs,

the analysis compared women with a relevant underlying condition.

The association became smaller.

This supports the hypothesis that some of the original association was related to differences between women treated and untreated for depression.

16. Propensity Score Adjustment

The investigators then used propensity-score methods to address differences between treatment groups.

The adjusted SSRI odds ratio became approximately:

1.12 (95% CI 0.95–1.31).

Using high-dimensional propensity-score adjustment produced:

OR 1.10 (95% CI 0.94–1.29). 10

The confidence interval included 1.

This does not prove that there is no causal effect.

It means that after the available confounding adjustment, the study did not demonstrate a statistically significant overall association of the magnitude suggested by the original study.

17. Why the 2015 Study Was Important

The study did something particularly valuable for signal evaluation.

It demonstrated how much the estimate changed when the investigators attempted to address confounding.

The sequence was approximately:

unadjusted → depression-restricted → propensity-adjusted → high-dimensional propensity-adjusted

and the estimated association progressively weakened.

This is a powerful teaching point.

When evaluating an observational safety signal, ask:

How stable is the association when plausible confounding is progressively addressed?

A finding that disappears after reasonable adjustment deserves a different interpretation from one that remains robust.

18. The Non-SSRI Comparator

An additional useful feature of the 2015 study was the analysis of non-SSRI antidepressants.

The unadjusted association for non-SSRI antidepressants was also positive.

After adjustment, however, the association was approximately null:

OR 1.02 (95% CI 0.77–1.35). 11

This is informative.

If an association appears with multiple antidepressant classes before adjustment but becomes substantially attenuated after adjustment, that pattern raises the possibility that characteristics of the treated population contribute to the observed association.

It does not completely exclude a drug-specific effect.

But it changes the evidentiary balance.

19. Primary PPHN

The 2015 investigators also examined a more narrowly defined outcome.

For primary PPHN, the adjusted SSRI estimate was:

OR 1.28 (95% CI 1.01–1.64). 12

This estimate is important because it prevents an overly simple conclusion that the study demonstrated "no risk."

The broader analysis was compatible with no statistically significant association.

The narrower primary-PPHN analysis retained a small association.

Therefore the scientifically honest conclusion is more nuanced:

The large study substantially reduced the estimated magnitude of risk, but it did not completely eliminate uncertainty about a small residual association.

That is a much better signal-evaluation conclusion than either:

SSRIs cause PPHN.

or:

SSRIs do not cause PPHN.

20. Biological Plausibility

A biological mechanism was proposed involving serotonin.

Serotonin has recognised effects on pulmonary vascular tone and vascular smooth-muscle biology.

The hypothesis was that increased fetal serotonergic exposure might contribute to pulmonary vasoconstriction or pulmonary vascular remodelling.

This provides biological plausibility for the association.

However:

biological plausibility is not proof of clinical causality.

The existence of a plausible mechanism tells us that the hypothesis is scientifically credible.

It does not tell us how large the effect is.

21. What Is Established?

Several points can be separated from the more speculative mechanistic discussion.

Established

• SSRIs cross the placenta.
• PPHN is a biologically plausible outcome for investigation.
• An association between late-pregnancy SSRI exposure and PPHN was reported.
• Subsequent observational evidence has been inconsistent in magnitude.
• Large studies suggest that any absolute increase in risk is small.
• Confounding materially affects the estimated association.

Plausible but not fully established

• fetal serotonergic effects may contribute to pulmonary vascular changes;
• susceptibility may vary according to gestational timing;
• particular clinical or biological subgroups may carry different risks.

Not established

The evidence does not establish that:

• every SSRI carries the same risk;
• every late-pregnancy exposure causes clinically significant pulmonary hypertension;
• the original six-fold estimate represents the true causal effect;
• or a causal effect, if present, can be separated completely from all maternal and obstetric factors.

This distinction is important in authoritative pharmacovigilance writing.

22. Other Causes of PPHN

PPHN is not a single-cause disorder.

Potential contributors include:

• congenital heart disease;
• pulmonary hypoplasia;
• prematurity-related pulmonary pathology;
• meconium aspiration;
• infection;
• structural lung abnormalities;
• and pulmonary vascular abnormalities.

Maternal and obstetric factors may also influence risk.

Therefore, a study that compares SSRI exposure without adequately accounting for these variables can produce a misleading association.

The 2014 meta-analysis highlighted this problem, noting that several recognised PPHN risk factors were inconsistently controlled across the underlying studies. 13

23. The Problem of Caesarean Delivery

Caesarean delivery is a particularly interesting example of why causal diagrams matter.

It may be:

• associated with antidepressant use;
• associated with PPHN;
• a consequence of another factor related to maternal illness;
• or, depending on the causal structure, potentially lie on an intermediate pathway.

Simply "adjusting for everything" is not necessarily correct.

The investigator needs to determine whether a variable is:

confounder

versus:

mediator

versus:

collider.

The interpretation of an adjusted estimate depends on that causal structure.

This is one reason why pharmacovigilance signal assessment benefits from epidemiological expertise rather than purely statistical processing.

24. Maternal Depression and Treatment

There is another important clinical issue.

The decision is not:

SSRI versus no risk.

The real decision is often:

SSRI exposure versus untreated or undertreated maternal psychiatric illness.

Maternal depression can itself have consequences for:

• maternal functioning;
• prenatal care;
• substance use;
• nutrition;
• sleep;
• adherence;
• obstetric outcomes;
• and postpartum health.

Therefore, even a confirmed small neonatal risk would need to be interpreted within the overall maternal and fetal benefit-risk balance.

This is why regulatory communication should avoid implying that stopping an effective antidepressant is automatically the safest option.

25. A Hypothetical Case

Consider a pregnant woman receiving an SSRI for major depression.

She continues treatment into late pregnancy.

Her infant develops PPHN shortly after birth.

How should the case be assessed?

The evaluator should document:

• exact medicine;
• dose;
• duration;
• gestational timing;
• adherence;
• maternal indication;
• severity of depression;
• other medicines;
• smoking;
• diabetes;
• obesity;
• obstetric complications;
• gestational age;
• mode of delivery;
• meconium aspiration;
• congenital abnormalities;
• neonatal infection;
• diagnostic confirmation of PPHN;
• treatment;
• and outcome.

The fact that:

SSRI exposure preceded PPHN

establishes temporal compatibility.

It does not establish causality.

26. A More Informative Case

Now consider a case with:

• continuous SSRI exposure through late pregnancy;
• term delivery;
• no congenital heart disease;
• no meconium aspiration;
• no neonatal infection;
• confirmed PPHN;
• and no major alternative explanation.

This case provides stronger support for the signal hypothesis.

But even here, the case does not quantify risk.

A case report can strengthen:

case-level evidence.

It cannot independently establish:

population-level incidence or relative risk.

27. A Confounded Case

Now consider a different patient.

She receives an SSRI during pregnancy and has an infant with PPHN.

The pregnancy is complicated by:

• severe obesity;
• diabetes;
• pre-eclampsia;
• caesarean delivery;
• and neonatal respiratory complications.

The case still deserves evaluation.

But causal attribution to the SSRI becomes more difficult.

The correct conclusion may be:

The event occurred during SSRI exposure, but multiple established maternal and neonatal risk factors provide alternative explanations and substantially limit the strength of the drug-specific causal inference.

This is more useful than assigning an arbitrary causality category without explaining why.

28. What Would Strengthen the Signal?

Evidence that would strengthen the causal hypothesis would include:

• consistent associations across well-designed studies;
• a reproducible relationship with late exposure;
• evidence of a dose-response relationship;
• biologically coherent timing;
• persistence of the association after rigorous confounding adjustment;
• consistent findings across different populations;
• and evidence that competing explanations do not account for the association.

A particularly important finding would be a consistent association that survives comparison with an appropriate active comparator.

29. What Would Weaken the Signal?

Evidence weakening the hypothesis would include:

• disappearance of the association after controlling for depression;
• similar associations with unrelated antidepressant classes;
• strong relationships with maternal or obstetric factors;
• lack of temporal specificity;
• inconsistent results across large well-designed studies;
• and substantial attenuation after adjustment for confounding.

The 2015 study provided several of these weakening features.

30. Why the Six-Fold Estimate Should Not Be Used Uncritically

The original:

OR 6.1

is historically important.

But it should not be presented today as though it were the established magnitude of risk.

Later evidence produced considerably smaller estimates.

The 2014 meta-analysis estimated approximately:

OR 2.5

for late-pregnancy exposure.

The 2015 large cohort study produced an adjusted estimate around:

OR 1.10

for overall PPHN after extensive confounding adjustment, with a somewhat higher estimate for primary PPHN. 14

This progression is exactly why historical signal evaluation matters.

The number attached to a signal can change dramatically as the evidence base matures.

31. Signal Evaluation Should Track the Evidence Over Time

A useful historical timeline is:

This is how a signal should be understood.

Not as a single number.

As an evolving body of evidence.

32. The Importance of Absolute Risk

The 2014 meta-analysis estimated an absolute risk increase of approximately 2.9–3.5 per 1,000 births for late SSRI exposure under its assumptions. 15

The 2015 study found PPHN rates of approximately:

• 20.8 per 10,000 among unexposed pregnancies overall;
• 31.5 per 10,000 among SSRI-exposed pregnancies.

After restricting to women with depression, the corresponding rates were approximately:

• 24.9 per 10,000 unexposed;
• 33.8 per 10,000 SSRI-exposed. 16

These numbers illustrate how much the apparent excess changes when the comparison population becomes more clinically appropriate.

This is why absolute risk should be presented alongside adjusted relative measures.

33. What Happened to the Regulatory Interpretation?

The historical regulatory story is important because it demonstrates uncertainty being communicated rather than hidden.

The FDA initially communicated a potential increased risk following the 2006 study.

After reviewing conflicting evidence, the FDA revised its position in 2011 and stated that it was premature to conclude that SSRI exposure during pregnancy caused PPHN. 17

This is not regulatory inconsistency in the pejorative sense.

It is an example of evidence-responsive pharmacovigilance.

A regulatory conclusion should be capable of changing when the evidence changes.

34. Reading Product Information Historically

When conducting a historical signal evaluation, product information should be treated as an evidence trail.

A useful exercise is to compare:

older SmPC wording

with:

later SmPC wording.

The evaluator should ask:

• When was PPHN first mentioned?
• Was it described as a possible risk or established adverse reaction?
• Was the timing of exposure specified?
• Was absolute risk communicated?
• Did the wording change after major epidemiological studies?
• Did regulators distinguish PPHN from other neonatal adaptation effects?
• Was the warning maintained despite uncertainty?

The exact wording should always be checked against the relevant historical product-information version rather than reconstructed from memory.

35. Why SmPC Evolution Is Valuable

The evolution of product information can reveal how regulators interpreted the totality of evidence at different points in time.

For a historical signal series, this gives us three separate evidence streams:

1. scientific literature;
2. regulatory assessment;
3. product-information evolution.

When these are compared chronologically, the reader can see how evidence became regulatory knowledge.

That is considerably more informative than simply stating:

PPHN is a known risk of SSRIs.

36. Signal Evaluation Versus Signal Confirmation

This case also demonstrates why the word "signal" should be used carefully.

The 2006 publication provided evidence that justified concern.

It did not prove causality.

The subsequent evidence changed the estimated magnitude.

A mature conclusion therefore distinguishes:

signal detection

from:

signal validation

from:

signal assessment

from:

regulatory action.

EMA's GVP Module IX explicitly describes signal management as a structured process involving detection, validation, analysis and prioritisation, assessment, and recommendation for action. 18

37. How a QPPV Should Review a Similar Signal

A QPPV reviewing an SSRI-PPHN signal should ask:

Case evidence

• How many cases?
• How many are medically confirmed?
• What was the gestational timing?
• What alternative causes were present?

Epidemiology

• What is the background incidence?
• Which comparator was used?
• Was the analysis restricted to women with depression?
• Were important confounders controlled?
• Does the estimate survive sensitivity analyses?

Mechanism

• Is there biological plausibility?
• Is the proposed mechanism established or hypothetical?
• Does the timing of exposure fit the mechanism?

Clinical relevance

• What is the absolute risk?
• How severe are the cases?
• Are there identifiable high-risk groups?

Regulatory status

• What did regulators conclude?
• When did the conclusion change?
• What risk-minimisation measures exist?
• Has product information evolved?

This framework is applicable far beyond antidepressants.

38. The Most Important Analytical Lesson

The most important lesson from this signal is not that:

SSRIs cause PPHN

or that:

SSRIs do not cause PPHN.

It is that the estimated association changed substantially when the evidence became larger and confounding was addressed more rigorously.

That tells us something about the nature of observational pharmacovigilance evidence.

An initial association can be:

• real;
• exaggerated;
• partly confounded;
• or a combination of all three.

The task of signal evaluation is to determine which explanation best fits the totality of evidence.

39. Overall Signal Assessment

The historical evidence supports a cautious interpretation.

An association between late-pregnancy SSRI exposure and PPHN was initially reported with a large effect estimate.

Subsequent studies produced conflicting results.

A 2014 meta-analysis continued to support an association with late exposure, estimating a pooled odds ratio of approximately 2.5, while emphasising that the absolute risk remained low and that residual confounding was an important limitation. 19

A much larger 2015 cohort study found that the association was substantially attenuated after accounting for depression and other confounding factors. The adjusted estimate for overall PPHN was close to the null, although the analysis of primary PPHN remained compatible with a small increase. 20

Therefore, the evidence does not support treating the original six-fold estimate as the established magnitude of risk.

Nor does it justify claiming that a small causal effect has been definitively excluded.

The most defensible interpretation is:

Late-pregnancy SSRI exposure has been associated with PPHN in observational studies, but the magnitude of any causal increase appears substantially smaller than suggested by the original report and is difficult to separate completely from maternal, obstetric and other confounding factors. If a causal increase exists, the absolute excess risk is likely to be small.

That conclusion is appropriately cautious without dismissing the signal.

40. Final Perspective

This signal illustrates why pharmacovigilance is fundamentally an evidence-integration discipline.

The initial study provided a reason to investigate.

The subsequent studies challenged the magnitude of the initial association.

The meta-analysis increased precision but could not remove limitations in the underlying observational evidence.

The large 2015 cohort demonstrated the importance of controlling for the underlying indication and other confounding factors.

The biological mechanism provided plausibility but did not establish causality.

And the regulatory position evolved as the evidence evolved.

For QPPV practice, the lesson is straightforward:

A signal should not be evaluated by asking only:

"Is there an association?"

The more useful questions are:

1. How strong is the association?
2. How consistent is it?
3. Does it survive appropriate confounding control?
4. Does the timing make biological sense?
5. What alternative explanations exist?
6. What is the absolute clinical risk?
7. How serious are the outcomes?
8. Which patients may be particularly susceptible?
9. What did regulators conclude at each stage?
10. Has subsequent evidence changed the original interpretation?

That is the difference between identifying a signal and actually evaluating one.

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. N Engl J Med. 2006;354:579–587. doi:10.1056/NEJMoa052744.
2. 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. doi:10.1136/bmj.f6932.
3. 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. doi:10.1001/jama.2015.5605.
4. U.S. Food and Drug Administration. Public health advisory and subsequent communication concerning antidepressant use during pregnancy and persistent pulmonary hypertension of the newborn, including the 2011 reassessment.
5. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module IX — Signal Management.
6. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module VI — Collection, management and submission of reports of suspected adverse reactions to medicinal products.
7. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module V — Risk management systems.
8. Huybrechts KF, Bateman BT, Hernandez-Diaz S. Maternal antidepressant use and persistent pulmonary hypertension of the newborn — correspondence and methodological discussion. JAMA.
9. Kieler H, Artama M, Engeland A, et al. Selective serotonin reuptake inhibitors during pregnancy and risk of persistent pulmonary hypertension in the newborn: population-based studies from the Nordic countries.
10. Källén B, Olausson PO. Maternal use of selective serotonin re-uptake inhibitors and risk of persistent pulmonary hypertension of the newborn.