SSRIs and Postpartum Haemorrhage: A Signal Evaluation

Purpose and Scope

Selective serotonin reuptake inhibitors (SSRIs) are widely used antidepressants. Postpartum haemorrhage (PPH) is clinically important bleeding after childbirth, but it is also a multifactorial obstetric outcome with a substantial background frequency. The proposed association between SSRI exposure near delivery and PPH is therefore an instructive pharmacovigilance problem: an observed association may be real without the medicine being the sole, or even dominant, cause of the outcome.

The regulatory question is not whether one positive observational study can prove causality. It is whether the totality of evidence supports a sufficiently credible association to justify product-information or risk-management action, and how strongly that evidence supports a causal interpretation.

In October 2020, the EMA Pharmacovigilance Risk Assessment Committee (PRAC) considered observational evidence on postpartum haemorrhage with several SSRIs, SNRIs and vortioxetine and recommended product-information updates for relevant products. This regulatory conclusion is important evidence that the issue warranted communication; it should not be confused with a claim that every episode of PPH in an exposed woman is attributable to the antidepressant.

Defining the Safety Question

A useful signal question specifies four elements: the medicine or class, the outcome phenotype, the exposure window and the population. Here the question is whether maternal exposure to an SSRI, especially close to delivery, is associated with an increased risk of PPH.

That precision matters because studies have used different definitions of PPH and different windows for antidepressant exposure. A study defining exposure as any use during pregnancy is not answering exactly the same question as one examining treatment during the final month before delivery. Likewise, administrative diagnostic codes, estimated blood loss and severe haemorrhage endpoints are related but not interchangeable phenotypes.

The first task in evaluation is therefore to determine whether apparently concordant studies are actually measuring comparable exposure-outcome pairs.

Biological Plausibility

Platelets acquire serotonin from plasma through the serotonin transporter and store it in dense granules. Serotonin released during platelet activation contributes to haemostatic responses. By inhibiting serotonin uptake, SSRIs can reduce platelet serotonin content and may impair aspects of platelet-mediated haemostasis.

This creates a coherent hypothesis: SSRI exposure close to delivery could modestly increase bleeding tendency at a time of major haemostatic challenge. The same pharmacology is consistent with bleeding associations reported in other clinical contexts.

Mechanistic plausibility strengthens a signal because it provides a pathway connecting exposure and outcome. It does not establish the magnitude of any clinical effect, and it does not remove competing obstetric explanations. PPH can result from uterine atony, placental abnormalities, retained tissue, genital-tract trauma, operative delivery and coagulation disorders. A plausible platelet mechanism must therefore be integrated with, rather than substituted for, clinical epidemiology.

Why Timing Matters

If platelet effects contribute to PPH, exposure near delivery is more relevant than remote exposure earlier in pregnancy. Observational reviews have generally found the signal to be more evident with current or recent antidepressant exposure than with past exposure.

This temporal pattern is supportive because it aligns the pharmacological hypothesis with the clinical event. However, late-pregnancy exposure also identifies women who continued therapy despite changing pregnancy circumstances. They may differ from women who stopped treatment in disease severity, comorbidity, healthcare use, concomitant medicines and obstetric characteristics. Timing therefore improves biological coherence while simultaneously creating potential treatment-selection bias.

Interpreting the Observational Evidence

The literature is dominated by observational studies because randomising antidepressant treatment around delivery for the purpose of studying haemorrhage would not be an appropriate way to answer this safety question. Observational evidence can estimate associations in large populations, but its interpretation depends on the quality of exposure ascertainment, outcome definition, comparator selection and confounder control.

Meta-analytic evidence has generally suggested a modest relative increase in PPH with antidepressant exposure, particularly recent exposure. Large database studies have also reported increased relative risks. These findings are relevant because they are reproducible across more than one data source, but statistical consistency alone is not proof of a pharmacological effect.

Relative and Absolute Risk

A central teaching point is the distinction between relative and absolute risk. A relative increase can appear substantial while corresponding to a comparatively small increase in the number of affected women because PPH already occurs in unexposed pregnancies. Conversely, even a small absolute increase can matter clinically when an exposure is common and the outcome can be serious.

A pharmacovigilance assessment should therefore present the study-specific baseline risk and absolute risk difference whenever those data are available. It should not apply a pooled relative risk mechanically to an arbitrary baseline population.

Confounding by Indication

SSRIs are prescribed principally for depression and anxiety disorders. The treated population may differ from untreated pregnant women in smoking, obesity, substance use, socioeconomic factors, comorbid disease, concomitant medicines, healthcare utilisation and obstetric characteristics. Some factors can be measured and adjusted for; others may be poorly captured in routine data.

This is confounding by indication and by factors associated with the indication. It is particularly important when increased PPH risk is also observed among women with psychiatric illness who are not currently exposed to an SSRI or among users of antidepressants with different pharmacology. Such findings do not disprove an SSRI effect, but they weaken a simple explanation in which serotonin reuptake inhibition accounts for the whole observed association.

Comparator Choice

An untreated general population comparator maximises contrast in medicine exposure but may also maximise differences in underlying illness. An active comparator can improve similarity in indication while introducing different pharmacological effects. Comparing continuers with discontinuers may reduce some indication differences but can introduce bias because continuation itself reflects clinical characteristics.

No comparator is automatically correct. The strongest assessment asks whether the association persists across reasonable comparator strategies and whether the direction and magnitude remain clinically coherent.

Integrating Mechanism and Epidemiology

The signal is stronger than a purely statistical association because there is a plausible platelet mechanism and a temporal pattern compatible with that mechanism. At the same time, PPH is sufficiently multifactorial that residual confounding remains a serious alternative explanation.

The appropriate conclusion is therefore graded rather than binary. The evidence supports an association between late-pregnancy exposure to serotonergic antidepressants and a modestly increased risk of PPH. It is reasonable for product information to communicate this risk. The evidence does not justify attributing all of the observed excess to a single pharmacological mechanism or assuming an identical effect size for every SSRI and every obstetric population.

Practical Pharmacovigilance Assessment

For an MAH assessing this signal, the literature review should be connected with the wider safety database. Relevant questions include whether serious or medically significant PPH cases contain informative timing, whether other bleeding risk factors were present, whether individual products show materially different evidence and whether the risk is already adequately described in product information.

Spontaneous reports can contribute clinically rich narratives but cannot estimate incidence because the denominator and reporting probability are unknown. Observational studies can provide comparative estimates but are vulnerable to confounding. Mechanistic evidence can support causal coherence but cannot quantify clinical risk. The value of the assessment comes from combining these evidence types according to what each can and cannot establish.

Benefit-Risk Context

A safety conclusion must also preserve the benefit side of the decision. Abrupt or clinically inappropriate discontinuation of effective antidepressant treatment during pregnancy can create risks associated with relapse or undertreated psychiatric illness. A product-information warning should therefore support informed clinical management rather than imply that exposure near delivery should automatically be stopped.

This illustrates a general PV principle: recognising a credible adverse-reaction risk does not by itself specify the correct clinical action. Risk communication should enable prescribers and patients to weigh the risk in the context of treatment need and available alternatives.

Failure Modes in Signal Evaluation

Several analytical errors can distort this signal. Treating statistical significance as causality ignores confounding. Presenting only a relative-risk estimate obscures clinical magnitude. Combining all pregnancy exposure windows can dilute or mischaracterise a time-dependent effect. Treating all antidepressants as pharmacologically interchangeable can create a false class conclusion. Conversely, dismissing the signal because PPH has many non-drug causes ignores the possibility that a medicine contributes modestly to a multifactorial outcome.

A further failure mode is regulatory over-interpretation. A PRAC recommendation to update product information means that the evidence supported regulatory communication. It does not convert an observational association into a precisely quantified causal effect applicable to every patient.

Inspection and Governance Considerations

An inspector reviewing the handling of this type of signal could reasonably ask how the organisation identified the relevant literature, defined the outcome and exposure window, assessed confounding, compared individual products and classes, considered spontaneous cases, documented the benefit-risk conclusion and translated any regulatory recommendation into product-information action.

The evidence trail should make it possible to reconstruct why particular studies were given greater or lesser weight. The QPPV's role is system oversight: ensuring that important safety information is identified, evaluated and acted upon through the pharmacovigilance system. EU rules do not create a universal requirement for the QPPV personally to author or approve every literature assessment.

Summary

The SSRI–postpartum haemorrhage issue demonstrates how pharmacovigilance moves from an observed association to a calibrated conclusion. Biological plausibility and the stronger association with exposure near delivery support the signal. Repeated observational findings add weight. Yet confounding by psychiatric illness, obstetric factors and treatment selection limits simple causal interpretation, and the clinical meaning is best understood through absolute as well as relative risk.

The resulting lesson is broader than this drug class. In common, multifactorial outcomes, signal evaluation is strongest when phenotype, timing, comparator choice, mechanism, confounding and clinical magnitude are considered together rather than allowing any single positive study to determine the conclusion.

References

1. European Medicines Agency. PRAC recommendations on signals adopted at the 28 September–1 October 2020 PRAC meeting: citalopram, desvenlafaxine, escitalopram, fluoxetine, fluvoxamine, milnacipran, paroxetine, sertraline, venlafaxine and vortioxetine — postpartum haemorrhage. EMA/PRAC/495737/2020.
2. Jiang H-Y, et al. Antidepressant use during pregnancy and risk of postpartum hemorrhage: a systematic review and meta-analysis. J Psychiatr Res. 2016;83:160–167.
3. Palmsten K, et al. Use of antidepressants near delivery and risk of postpartum hemorrhage: cohort study of low income women in the United States. BMJ. 2013;347:f4877.
4. Bruning AHL, et al. Antidepressants during pregnancy and postpartum hemorrhage: a systematic review. Eur J Obstet Gynecol Reprod Biol. 2015;189:38–47.
5. European Medicines Agency. Good pharmacovigilance practices, Module IX — Signal management.

Regulatory Note

This article is an educational reconstruction of signal evaluation, not product-specific prescribing advice. The EU regulatory position should be checked against the current authorised product information for the individual antidepressant and Member State concerned. The 2020 PRAC signal recommendation supports product-information communication of postpartum haemorrhage risk for relevant serotonergic antidepressants; it should not be interpreted as a universal numerical risk estimate for all products or pregnancies.