SSRIs and Postpartum Haemorrhage: A Signal Evaluation

Introduction

A pharmacovigilance signal does not become a confirmed adverse drug reaction simply because several studies report statistically significant associations.

The association between selective serotonin reuptake inhibitors (SSRIs) and postpartum haemorrhage (PPH) is a useful example because it contains many of the problems that make signal evaluation difficult in real-world pharmacovigilance:

• the outcome is clinically important but relatively common;
• exposure is concentrated in a biologically unusual period;
• the proposed mechanism is plausible;
• observational studies can identify an association but cannot by themselves establish causality;
• the underlying indication for treatment may itself be associated with the outcome;
• obstetric factors strongly influence the baseline risk of haemorrhage;
• exposure timing appears to matter;
• some studies report an association while others do not;
• the relative increase in risk may appear more impressive than the corresponding absolute increase.

This makes the signal particularly useful for teaching.

The purpose of this article is not to decide whether SSRIs "cause" postpartum haemorrhage from a single study. It is to demonstrate how a pharmacovigilance professional can evaluate the signal systematically, distinguish what is established from what remains uncertain, and reach a proportionate conclusion.

The example is deliberately based on a drug class rather than a proprietary product. Individual SSRIs differ pharmacologically, clinically and in the amount of available evidence, and a class-level signal should not automatically be interpreted as proof that every individual product carries the same magnitude of risk.

What Is the Signal?

The signal evaluated here is:

Exposure to an SSRI during pregnancy, particularly close to delivery, may be associated with an increased risk of postpartum haemorrhage.

The first important question is whether the outcome itself has been defined consistently.

Definitions of PPH have changed over time and differ between studies. The American College of Obstetricians and Gynecologists has defined maternal haemorrhage as cumulative blood loss of at least 1,000 mL, or blood loss accompanied by signs or symptoms of hypovolaemia within 24 hours after birth. 0

Older epidemiological studies may therefore use lower blood-loss thresholds or administrative definitions that are not directly interchangeable with contemporary clinical definitions.

This matters.

If two studies use different outcome definitions, a difference in reported risk may reflect differences in case ascertainment rather than a difference in the underlying drug effect.

A signal evaluation should therefore begin with a precise phenotype.

Why This Is a Plausible Pharmacovigilance Signal

At first sight, the association has biological plausibility.

Platelets do not synthesise serotonin in the same way as neurons, but they take up serotonin from plasma and store it in dense granules. Serotonin released during platelet activation contributes to platelet aggregation and vascular responses involved in haemostasis.

SSRIs inhibit serotonin reuptake. Consequently, chronic SSRI exposure can reduce platelet serotonin concentrations and may impair aspects of platelet-mediated haemostasis.

This mechanism has been proposed as one explanation for the association between serotonin reuptake inhibitors and abnormal bleeding in other clinical settings. 1

There is therefore a biologically coherent pathway:

SSRI exposure → reduced platelet serotonin availability → altered platelet-mediated haemostasis → potentially greater bleeding under conditions of haemostatic stress.

However, this is where signal evaluation needs discipline.

What is established?

It is well established that platelets use serotonin in haemostatic processes and that serotonin reuptake inhibition can alter platelet serotonin biology. SSRIs have also been associated with bleeding outcomes in observational literature outside pregnancy. 2

What is plausible?

It is biologically plausible that impaired platelet function could contribute to greater blood loss during delivery.

What is not fully established?

It is not established from the mechanism alone that SSRI exposure causes clinically important PPH in pregnant women.

The obstetric haemostatic environment is much more complicated than platelet function alone. Uterine atony, placental abnormalities, retained placental tissue, operative delivery, trauma, coagulation disorders and other factors can dominate the risk of PPH.

Mechanistic plausibility therefore supports investigation of the signal.

It does not establish causality.

Signal Evaluation

The EMA's GVP framework describes signal management as a structured process involving signal detection, validation, analysis, prioritisation and assessment before recommendations for action are made. 3

The SSRI–PPH example illustrates why those stages should not be collapsed into one question.

Step 1: Is There an Association?

There is substantial observational evidence suggesting an association.

An early systematic review published in 2015 identified four studies and concluded that the evidence was inconclusive. Two studies reported increased PPH while two did not; the authors noted that, if an association existed, the absolute increase in risk was likely to be small. 4

A later systematic review and meta-analysis published in 2016 included eight studies involving more than 40,000 PPH cases. Antidepressant exposure was associated with a pooled RR of 1.32 (95% CI 1.17–1.48). For SSRIs specifically, the pooled RR was 1.20 (95% CI 1.04–1.38). Current and recent exposure were associated with increased risk, whereas past exposure was not clearly associated with PPH. 5

This is an important observation.

The signal appears stronger when exposure is temporally close to delivery.

That pattern is more informative than a simple association between "ever used an antidepressant during pregnancy" and PPH.

A Large Observational Study

One influential cohort study examined antidepressant exposure close to delivery in a large population of women in the United States.

Current serotonin reuptake inhibitor exposure was associated with an adjusted RR of 1.47 (95% CI 1.33–1.62) compared with no exposure. The investigators also performed high-dimensional propensity-score analyses, which produced a similar estimate.

Importantly, the investigators estimated an adjusted excess risk of approximately 1.26 percentage points for current serotonin reuptake inhibitor exposure, corresponding to a number needed to harm of approximately 80 under the assumption of a causal relationship. 6

That distinction between relative and absolute risk is central.

A relative risk of approximately 1.5 can initially sound large.

An absolute increase of approximately 1–2 percentage points may produce a different clinical interpretation.

Both numbers are true descriptions of the same study.

Neither should be presented without the other.

The Importance of the Comparator

The same study also produced an important methodological observation.

Increased risk was not restricted to serotonin reuptake inhibitors. Other antidepressant classes were also associated with PPH.

That finding does not necessarily invalidate the SSRI association.

But it weakens a simple interpretation that the observed association must be caused specifically by serotonin reuptake inhibition.

The study authors themselves noted the possibility of residual confounding by factors associated with depression and antidepressant use, including behavioural factors that were incompletely captured in the available data. 7

This is a classic signal-evaluation lesson:

If an apparently drug-specific association also appears with mechanistically different comparator drugs, the causal interpretation deserves closer examination.

The observation may indicate a class effect.

It may also indicate confounding.

The data alone do not automatically distinguish the two.

Timing of Exposure

Timing is one of the most informative features of this signal.

The 2016 meta-analysis found increased risk with current and recent antidepressant exposure but not clearly with past exposure. 8

The more recent systematic review of controlled observational studies, covering literature through September 2023, identified 20 studies. Most focused on SSRIs or SNRIs, and the majority of studies reporting a statistically significant association involved exposure late in pregnancy, particularly exposure within 30 days of delivery. 9

This provides some temporal coherence.

If the hypothesised mechanism involves impaired haemostasis at the time of delivery, exposure immediately before delivery is more biologically plausible than exposure many months earlier.

However, temporal coherence is not the same as causality.

Late pregnancy is also exactly the period when many obstetric characteristics change and when treatment decisions may be influenced by maternal health status.

Thus:

late exposure strengthens the signal, but it does not eliminate confounding.

A Study That Illustrates the Problem

A prospective observational analysis of women with mood disorders provides a useful counterpoint.

Among 263 women, the occurrence of estimated blood loss greater than 600 mL was 9.8% among those taking serotonin reuptake inhibitors during the third trimester and 8.5% among those not exposed. The difference was not statistically significant. 10

The study was substantially smaller than the large administrative database studies.

It therefore had less statistical power.

But it has another useful characteristic: exposure was assessed prospectively and the blood-loss outcome was determined from obstetric records by a blinded medically trained reviewer.

This illustrates an important principle in literature evaluation:

A negative study is not necessarily evidence that the signal is absent. Its value depends on study power, exposure ascertainment, outcome definition, comparator selection and residual bias.

Likewise, a large positive study is not automatically proof of causality.

Study size and study validity are different concepts.

A Hospital-Based Cohort With a Large Apparent Effect

Another hospital-based cohort reported a substantially larger association.

Among women exposed to SSRIs, the reported absolute risk of PPH was 18.0%, compared with 8.7% among non-users. In women having vaginal non-surgical deliveries, SSRI exposure was associated with an odds ratio of approximately 2.6 for PPH. 11

At first glance, this appears much stronger than the large population-based studies.

But the study also demonstrated important baseline differences between exposed and unexposed women. For example, SSRI users differed in BMI, smoking and epidural analgesia exposure. 12

This does not prove that the finding was confounded.

It demonstrates why adjustment is necessary and why apparently large crude differences should not be interpreted as drug effects without examining the underlying populations.

What Does the Totality of Evidence Show?

The literature has evolved from relatively sparse and conflicting evidence toward a more consistent observation of a small association, particularly with antidepressant exposure near delivery.

The 2024 systematic review is especially useful because it incorporates studies published after the earlier reviews.

It identified 20 controlled observational studies. Most studies focused on SSRIs or SNRIs, and most reported statistically significant associations with late-pregnancy exposure. However, the authors concluded that the extent to which the association was causal remained uncertain because of possible non-pharmacological factors, including maternal indication. They also noted that relatively few studies examined severe, well-defined PPH or dose changes. 13

The conclusion is therefore not:

"SSRIs cause postpartum haemorrhage."

A more defensible interpretation is:

Late-pregnancy antidepressant exposure, including SSRI exposure, is associated with a small increase in the risk of PPH in observational studies, but the extent to which this association represents a direct pharmacological effect remains uncertain.

That is a pharmacovigilance conclusion rather than a rhetorical one.

Confounding by Indication

One of the most important challenges is confounding by indication.

SSRIs are prescribed for depression and anxiety disorders.

Those underlying conditions may be associated with characteristics that also influence pregnancy and delivery outcomes.

For example, psychiatric illness may correlate with:

• smoking;
• substance use;
• obesity;
• socioeconomic factors;
• differences in antenatal care;
• comorbid medical conditions;
• concomitant medicines;
• obstetric risk factors;
• differences in delivery management.

Some of these factors may be measured and adjusted for.

Others may be incompletely measured.

Some may not be captured at all.

The Swedish national register-based cohort is particularly instructive because it examined both SSRI exposure and psychiatric illness without current treatment. The study found increased PPH risk among women treated with SSRIs and also among women with prior or current psychiatric illness. Adjustment for multiple confounders did not eliminate the association. 14

This finding neither proves nor disproves causality.

But it demonstrates that the underlying psychiatric condition itself may be part of the causal-inference problem.

Could the Association Be a Class Effect?

Another important question is whether the signal follows the pharmacology.

If serotonin reuptake inhibition is responsible, one might expect stronger or more consistent associations with drugs that have substantial serotonin reuptake inhibition.

Yet some studies have reported associations with non-serotonergic antidepressants as well. 15

That observation creates several possibilities:

1. serotonin reuptake inhibition contributes to the effect;
2. several antidepressant classes influence risk through different mechanisms;
3. antidepressant exposure is a marker for other maternal or clinical characteristics;
4. the outcome definition or residual confounding contributes to the apparent class-wide association.

These possibilities are not mutually exclusive.

The available literature does not justify collapsing them into a single mechanistic explanation.

Pharmacological Plausibility: What We Know and What We Do Not

A disciplined signal assessment should separate biological evidence from clinical evidence.

Established or strongly supported

SSRIs inhibit serotonin reuptake.

Platelets depend on serotonin storage and release as part of normal platelet function.

Serotonin reuptake inhibition can alter platelet serotonin biology and has been associated with abnormal bleeding in other settings. 16

Biologically plausible

Reduced platelet serotonin availability could impair haemostatic responses during tissue injury associated with delivery.

This provides a plausible pathway for increased blood loss.

Less certain

The extent to which platelet effects materially increase clinically significant PPH is not established solely by pharmacology.

PPH is multifactorial, and uterine atony, placental pathology, operative delivery, trauma and other obstetric factors may have much larger effects on risk.

Not established

The evidence does not establish that the pharmacological mechanism accounts for all, or even most, of the observed epidemiological association.

This distinction is essential.

Mechanism can strengthen causal inference.

It cannot substitute for epidemiological evidence.

Alternative Explanations

A good signal evaluation should actively try to explain the signal without assuming the drug is causal.

Possible explanations include:

Residual confounding

Important characteristics may not be completely measured or may be measured inaccurately.

Confounding by indication

The condition for which the drug is prescribed may itself be associated with the outcome.

Detection or ascertainment differences

Women taking antidepressants may have different healthcare utilisation patterns, which could influence identification or recording of PPH.

Outcome misclassification

Administrative PPH definitions may not correspond precisely to contemporary clinical definitions.

Treatment-selection bias

Women who continue treatment late in pregnancy may differ systematically from women who discontinue treatment.

Concomitant medication

Analgesics, antithrombotic drugs and other medicines may influence bleeding risk.

Obstetric confounding

Mode of delivery, parity, prior caesarean delivery, fetal size, induction, labour complications and placental factors can substantially alter PPH risk.

A signal evaluation becomes more convincing when these alternative explanations have been actively tested rather than merely mentioned.

Why Absolute Risk Matters

Relative measures are useful for detecting and characterising signals.

Clinical interpretation requires absolute risk.

Suppose the baseline probability of the outcome is approximately 5%.

A relative risk of 1.4 would correspond approximately to a risk of 7%, an absolute increase of 2 percentage points.

That is a very different clinical message from saying only:

"The drug increases the risk by 40%."

The actual baseline risk varies substantially depending on population, definition, delivery characteristics and study methodology.

For that reason, absolute risk estimates should always be taken from the individual study rather than mechanically applying a pooled relative risk to an arbitrary baseline.

In the large US cohort discussed above, the adjusted excess risk associated with current serotonin reuptake inhibitor exposure was approximately 1.26 percentage points. 17

The clinical interpretation of that increase is therefore different from what the relative risk alone might suggest.

What Would Strengthen the Causal Case?

Several types of evidence would increase confidence that the association is pharmacological.

1. Consistent exposure-response relationship

Higher exposure or stronger serotonin reuptake inhibition associated with greater risk would support causality.

2. Consistent timing relationship

A stronger association immediately before delivery, with diminishing risk after discontinuation, would fit the proposed mechanism.

3. Class coherence

Multiple SSRIs showing comparable associations would support a class effect, although it would not eliminate confounding.

4. Appropriate negative controls

An outcome that should not plausibly be affected by SSRI exposure can help detect residual confounding.

5. Active comparators

Comparing SSRI users with patients receiving treatments for similar indications can reduce confounding by indication.

6. Better measurement of psychiatric severity

This could help distinguish the effects of treatment from the effects of the underlying condition.

7. More precise outcome definitions

Studies focusing on severe PPH, transfusion, surgical intervention or clinically important blood loss may provide greater clinical relevance than broad administrative PPH definitions.

8. Mechanistic biomarkers

Evidence linking SSRI exposure to measurable platelet-function changes during pregnancy could strengthen biological coherence, although such evidence would still not establish clinical causality by itself.

What Would We Expect If the Signal Were Not Causal?

This question is equally important.

If the association were primarily due to confounding, we might expect:

• similar associations with pharmacologically unrelated antidepressants;
• stronger associations in populations with greater residual confounding;
• attenuation when active comparators are used;
• association with psychiatric illness even without antidepressant exposure;
• weak or inconsistent exposure-response relationships;
• inconsistent associations with severe PPH;
• associations with negative-control outcomes.

Several observations in the literature already point in this direction, particularly the association observed with psychiatric illness and the findings involving antidepressants outside the SSRI class. 18

These findings do not refute a pharmacological contribution.

They make a simple causal interpretation less secure.

Signal Prioritisation

Would this signal warrant pharmacovigilance attention?

Yes.

There are several reasons:

• PPH is clinically important;
• the signal is biologically plausible;
• multiple observational studies have reported an association;
• the association appears stronger with exposure close to delivery;
• there is evidence of bleeding effects with serotonin reuptake inhibition in other settings;
• the outcome occurs at a clinically critical time.

But prioritisation does not mean confirmation.

The appropriate pharmacovigilance response depends on the totality of evidence, the existing product safety information, exposure in the relevant population, the magnitude and severity of the potential risk, and whether additional regulatory or risk-minimisation action would meaningfully reduce harm.

What Would a Pharmacovigilance Team Do With the Signal?

A practical evaluation would include several parallel activities.

Individual case review

Review relevant spontaneous reports for:

• timing of exposure;
• timing of delivery;
• estimated blood loss;
• clinical definition of PPH;
• transfusion;
• uterine atony;
• placental abnormalities;
• operative delivery;
• concomitant medicines;
• prior bleeding history;
• dechallenge where meaningful;
• relevant laboratory or clinical information.

Literature review

Characterise:

• study design;
• population;
• exposure definition;
• timing of exposure;
• outcome definition;
• comparator;
• adjustment strategy;
• effect estimate;
• absolute risk;
• missing data;
• residual confounding;
• consistency across studies.

Aggregate review

Assess whether the signal is:

• increasing;
• stable;
• restricted to particular exposure windows;
• concentrated in specific populations;
• associated with severe outcomes;
• consistent across products in the class.

Product-information review

Determine whether the existing product information adequately reflects the evidence and whether any new information would materially change prescribing or monitoring decisions.

Benefit-risk assessment

Consider the potential bleeding risk alongside the consequences of untreated or inadequately treated maternal depression or anxiety.

This final point is particularly important.

A signal should not be evaluated in isolation from the benefits of treatment.

A Two-Level Conclusion

Evidence conclusion

The available literature supports an association between antidepressant exposure, particularly exposure late in pregnancy, and an increased risk of PPH. The association is generally modest in relative terms, although estimates vary considerably between studies. 19

Causal conclusion

The evidence is insufficient to conclude that the observed association is entirely or predominantly caused by the pharmacological effect of SSRIs. Residual confounding, particularly confounding related to maternal psychiatric illness and associated characteristics, remains an important alternative explanation. 20

A reasonable pharmacovigilance interpretation is therefore:

The SSRI–PPH association represents a credible safety signal with biological plausibility and substantial observational support, particularly for exposure near delivery. However, the magnitude of any direct pharmacological effect remains uncertain because of residual confounding and heterogeneity in study populations and outcome definitions.

That conclusion is stronger than saying simply "there is a signal" and more defensible than saying "SSRIs cause PPH."

What This Example Teaches About Signal Evaluation

This signal illustrates several general principles that can be applied to other drug-event pairs.

A statistically significant association is the beginning of evaluation, not the end.

A signal requires structured assessment.

Biological plausibility is supportive evidence, not proof.

A mechanism can explain an association without demonstrating that the association is causal.

Timing matters.

Exposure immediately before the event can provide stronger evidence than remote exposure, particularly when the proposed mechanism is short-lived or exposure-dependent.

Comparator selection matters.

If pharmacologically unrelated drugs show similar associations, confounding deserves increased attention.

Negative studies need to be interpreted in context.

Power, outcome definition, exposure ascertainment and residual bias all influence their evidentiary value.

Relative risk should be accompanied by absolute risk.

A modest absolute increase may have a different clinical meaning from the relative measure alone.

The underlying disease is part of the causal model.

For drugs used to treat psychiatric disorders, separating treatment effects from disease-associated factors can be particularly difficult.

Signal evaluation should end with a calibrated conclusion.

The correct conclusion may be:

• confirmed;
• strongly supported;
• plausible but uncertain;
• weakly supported;
• not supported;
• or requiring additional evidence.

There is no requirement that every signal evaluation produce a binary yes-or-no answer.

Key Takeaways

The association between SSRI exposure and postpartum haemorrhage is a useful example of a credible but causally complex pharmacovigilance signal.

The principal findings are:

1. Multiple observational studies report an association between antidepressant exposure and PPH.

2. The association appears more consistent when exposure occurs late in pregnancy, particularly close to delivery. 21

3. A biological mechanism involving platelet serotonin and haemostasis is plausible. 22

4. The evidence is not sufficient to attribute the entire association to SSRI pharmacology.

5. Confounding by maternal psychiatric illness and associated characteristics remains an important alternative explanation. 23

6. Associations reported with other antidepressant classes make a simple SSRI-specific interpretation less certain. 24

7. Absolute risk is important when translating an epidemiological signal into clinical significance.

8. The most defensible conclusion is that late-pregnancy SSRI exposure may represent a small risk factor for PPH, while the magnitude and causal contribution of the direct pharmacological effect remain uncertain.

9. Current evidence therefore illustrates why pharmacovigilance signal evaluation must integrate literature, mechanism, timing, confounding, clinical severity and absolute risk rather than relying on a single study or statistical association.

References

1. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module IX – Signal management. EMA/827661/2011. 25

2. American College of Obstetricians and Gynecologists. Practice Bulletin No. 183: Postpartum Hemorrhage. Obstet Gynecol. 2017;130:e168-e186. Reaffirmed 2024. 26

3. Jiang HY, Xu LL, Li YC, et al. Antidepressant use during pregnancy and risk of postpartum hemorrhage: a systematic review and meta-analysis. J Psychiatr Res. 2016;83:160-167. doi:10.1016/j.jpsychires.2016.09.001. 27

4. Bobo WV, Moore KM, Betcher HM, et al. The Association of Antidepressants in Late Pregnancy with Postpartum Hemorrhage: Systematic Review of Controlled Observational Studies. J Clin Psychopharmacol. 2024;34(10):428-446. doi:10.1089/cap.2024.0085. 28

5. Palmsten K, Hernández-Díaz S, Huybrechts KF, 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. 29

6. Skalkidou A, Sundström-Poromaa I, Wikman A, et al. SSRI use during pregnancy and risk for postpartum haemorrhage: a national register-based cohort study in Sweden. BJOG. 2020;127(11):1366-1373. doi:10.1111/1471-0528.16210. 30

7. Andrade C, Sharma E. Selective Serotonin Reuptake Inhibitor Use in Pregnancy and Risk of Postpartum Hemorrhage. J Clin Psychiatry. 2022;83(2):22f14455. doi:10.4088/JCP.22f14455. 31

8. Andrade C, Sandarsh S, Chethan KB, Nagesh KS. Serotonin reuptake inhibitor antidepressants and abnormal bleeding: a review for clinicians and a reconsideration of mechanisms. J Clin Psychiatry. 2010;71(12):1565-1575. doi:10.4088/JCP.09r05786blu. 32

9. Andrade C, Sharma E. Serotonin Reuptake Inhibitors and Risk of Abnormal Bleeding. Psychiatr Clin North Am. 2016. doi:10.1016/j.psc.2016.04.010. 33

10. Palmsten K, et al. Selective serotonin reuptake inhibitor use during pregnancy and postpartum hemorrhage and anemia: observational cohort evidence. 34

11. Cohen LS, et al. Is third trimester serotonin reuptake inhibitor use associated with postpartum hemorrhage? J Clin Psychiatry. 2016. 35