Bleeding Risk: GI, Perioperative & Anticoagulant Co-Prescription
Fourteen antidepressant labels carry the same bleeding warning and not one of them prints a number. The only label in the class that prints bleeding percentages belongs to a drug with no bleeding warning at all.
- The warning is class boilerplate. Fourteen modern labels carry near-identical wording naming aspirin, NSAIDs, antiplatelets and warfarin, and none of them prints an incidence figure [21]. The one label with numbers is clomipramine: purpura 3% against 0%, epistaxis 2% against 0%, on 322 against 319 — and clomipramine's label carries no bleeding warning [21]. Mirtazapine's label contains no bleeding language of any kind [21].
- Relative risk is modest and consistent. Upper GI bleeding odds ratio 1.66 (1.44–1.92) across 15 case-control studies and 393,268 participants, and 1.68 (1.13–2.50) across 4 cohort studies [7]; a separate review of 22 studies and more than 1,073,000 individuals gives 1.55 (1.35–1.78) [8].
- The interaction is the actionable part. SSRI with an NSAID gives 4.25 (2.82–6.42) [7]. In the study that stratified by acid suppression, the combination ran at 9.1 (4.8–17.3) without a proton pump inhibitor and 1.1 (0.3–3.4) with one [9].
- Intracranial haemorrhage is rarer and worse: adjusted relative risk 1.51 (1.26–1.81) across 16 studies and 506,411 subjects, with no increase in subarachnoid haemorrhage at 0.62 (0.38–1.01), and an absolute excess of roughly one extra intracerebral bleed per 10,000 people treated for a year [10][13].
- On an anticoagulant the observational signal is 1.33 (1.24–1.42) and front-loaded: 1.74 (1.37–2.22) in the first 30 days, falling after six months [14]. The one randomised-trial population studied showed no significant increase at 1.16 (0.95–1.43) [15].
- No study of any design has compared stopping an antidepressant before surgery against continuing it, with a bleeding outcome. The recommendation to stop rests on exposure studies [20].
- In the largest perioperative cohort identified for this chapter, 530,416 patients across 375 hospitals, the mortality signal was 1.20 (1.07–1.36) overall and 0.86 (0.65–1.13) among patients with a documented depression diagnosis. Bleeding requiring transfusion was 1.09 (1.04–1.15) and lost significance on propensity matching at 1.07 (0.99–1.14) [20].
- Mirtazapine and bupropion are not demonstrated safer. Mirtazapine carried greater gastrointestinal bleeding risk than no antidepressant at 1.17 (1.01–1.38), and neither differed from an SSRI [35].
- No validated bleeding risk score includes antidepressant use — not HAS-BLED, ORBIT, ATRIA, HEMORR2HAGES, Glasgow-Blatchford or Rockall [39][40].
A 71-year-old man with atrial fibrillation on apixaban has been on sertraline 100 mg for four years for recurrent depression. He is booked for an elective total knee replacement in three weeks. The pre-anaesthetic clinic has written to you asking you to stop the sertraline ten days before surgery.
He has no history of gastrointestinal bleeding. He takes ibuprofen most days for the knee. His last two depressive episodes both followed an antidepressant interruption.
🏷️ One warning, no numbers
Twenty-three antidepressant labels were read for this chapter. Fourteen carry a bleeding warning. One prints a bleeding number. They are not the same drug.
The modern warning is boilerplate. Paroxetine section 5.5 reads “Drugs that interfere with serotonin reuptake inhibition, including PAXIL, increase the risk of bleeding events.” Sertraline section 5.3 is the same sentence with the brand swapped. Duloxetine section 5.5 adds one word, “may increase”. Every one of them then names the same four co-medications: aspirin, non-steroidal anti-inflammatory drugs, warfarin and other anticoagulants [21]. The section number moves between 5.3 and 5.9 depending on how many other warnings come first. Nothing else moves.
| Group | Bleeding warning | Incidence figure | Names NSAIDs, aspirin, warfarin |
|---|---|---|---|
| SSRIs: fluoxetine, paroxetine, sertraline, citalopram, escitalopram, fluvoxamine | Yes — class boilerplate in all six. Fluvoxamine's sits in an older narrative PRECAUTIONS format rather than a numbered section | None | Yes, all six |
| SNRIs and other serotonergic agents: duloxetine, venlafaxine XR, desvenlafaxine, levomilnacipran, milnacipran, trazodone, vortioxetine, vilazodone | Yes — same boilerplate | None | Yes, all eight |
| Bupropion | No warning section. Postmarketing reports list ecchymosis, gastrointestinal haemorrhage and gum haemorrhage, and altered prothrombin time or INR with warfarin | None | Warfarin only. No NSAID or aspirin language anywhere in the label |
| Mirtazapine | None. No bleeding, haemorrhage, ecchymosis, epistaxis, purpura or bruising language anywhere in the label | None | Warfarin appears for an INR interaction, not framed as bleeding |
| Tricyclics: amitriptyline, nortriptyline, imipramine, doxepin | None. Imipramine and doxepin list purpura narratively among haematologic reactions, with no bleeding-risk framing | None | No. None of them mentions NSAIDs, aspirin, warfarin or anticoagulants at all |
| Clomipramine | None | Purpura 3% against 0%; epistaxis 2% against 0%; 322 against 319, pooled placebo-controlled trials. The paediatric arm reverses: epistaxis 0% against 2% on 46 against 44 | No |
| MAOIs: phenelzine, tranylcypromine | None | None | No |
Clomipramine is the tricyclic with the highest serotonin transporter affinity in the classification quoted in section 7 [48], and the only agent among the 23 labels read whose label quantifies a bleeding reaction. It carries no bleeding warning, because its label architecture predates the warning entirely [21]. Meanwhile fourteen labels warn without a denominator anywhere.
This is the same structural problem the previous chapter found in urinary reactions, and it has the same cause: warnings were added to modern labels by class template, while numbers only ever appear in whichever trial programme happened to tabulate the term. A prescriber reading the warnings learns which drugs the regulator considers serotonergic. A prescriber reading the tables learns almost nothing at all.
The absence of numbers is not evidence of absence of effect. It is evidence that bleeding was never a tabulated endpoint in registration trials, which were powered for depression response over six to eight weeks in populations selected to exclude the comorbidity that makes bleeding likely. The epidemiology had to be done afterwards, in databases, which is why every figure in the rest of this chapter carries a design label.
📊 Three scales, three answers
The same effect looks alarming, negligible or decisive depending on which scale you read it on. All three are correct.
Left: relative estimates against no antidepressant, log scale; the subarachnoid row is the one that does not rise. Centre: the same effects expressed as extra events. Right: the antidepressant-plus-NSAID combination with and without acid suppression, from a case-control study that stratified on it [7][8][10][14][13][16][45][9].
| Scale | What it says | Source and design |
|---|---|---|
| Relative | Upper GI bleeding, SSRI against no SSRI: odds ratio 1.66 (1.44–1.92) from 15 case-control studies and 393,268 participants, and 1.68 (1.13–2.50) from 4 cohort studies [7]. A separate review of 22 studies and more than 1,073,000 individuals gives 1.55 (1.35–1.78), rising to 1.95 (1.44–2.63) when restricted to studies of SSRI exposure alone, with heterogeneity of 88.9% and 89.8% respectively [8] | Two systematic reviews and meta-analyses. Both abstract-level for the figures quoted |
| Absolute | Number needed to harm 3,177 in a low-risk population and 881 in a high-risk one, both derived from the pooled odds ratio of 1.66 [7]. See the caveat below before using either figure. On warfarin, the ATRIA cohort gives rates of 2.32 against 1.35 major haemorrhages per 100 person-years — an excess of roughly one per 100 person-years [16] | Derived figures from a meta-analysis; cohort study of 9,186 warfarin-treated patients with atrial fibrillation |
| Interaction | SSRI with an NSAID: odds ratio 4.25 (2.82–6.42) [7]. Stratified by acid suppression, the combination ran at 9.1 (4.8–17.3) in patients not taking a proton pump inhibitor or H2 antagonist, and 1.1 (0.3–3.4) in those who were. The same study reports a second interaction the chapter should not conflate with the first: a serotonergic antidepressant with an antiplatelet agent ran at 4.7 (2.6–8.3) without acid suppression and 0.8 (0.3–2.5) with it [9] | Nested case-control study, 1,321 cases of upper gastrointestinal bleeding against 10,000 matched controls in a primary-care database. In the same analysis a serotonergic antidepressant alone gave an adjusted 1.8 (1.4–2.3) — 1.6 (1.2–2.1) for the SSRIs and 2.9 (1.5–5.6) for the SNRIs — an NSAID alone 2.8 (2.3–3.5), and the two together 4.8 (2.8–8.3) before stratification |
3,177 and 881 are the two figures every summary of this literature quotes. They reproduce identically across the source abstract, an independent structured abstract, and several secondary reviews. What none of those sources publishes is the baseline event rate each was calculated from, what defines the low-risk and high-risk populations, or the time horizon — per year, per course of treatment, or lifetime. That derivation appears only in the full text, and the paper has no open-access copy anywhere [7].
An 881 read as per-year and an 881 read as per-treatment-course are different clinical objects. The figures are printed here because they are the best available and because omitting them would leave a reader with only relative measures. They are printed with this paragraph attached because a number whose denominator and horizon are unknown is not yet a point-of-care number.
Lower gastrointestinal bleeding was an open gap in the first pass and is now partly filled. A nationwide Taiwanese cohort found SSRIs increased both upper bleeding at a hazard ratio of 1.97 and lower bleeding at 2.96, while finding no significant effect for SNRIs [41]. That last part does not agree with the pooled per-agent analysis in section 7, which found its highest per-agent estimate for an SNRI. Both are printed; neither is reconciled here.
The interaction row is the one that changes prescribing. An odds ratio of 1.55 to 1.66 on a background event rate in the region of one per thousand patient-years is not a reason to withhold an antidepressant from anyone. An odds ratio of 9.1 in a patient taking daily ibuprofen without acid suppression is a reason to do something — and the something with the best supporting number is not stopping the antidepressant.
🧬 The mechanism and its cracks
Platelets do not make serotonin. They acquire it through the serotonin transporter, so blocking that transporter empties them. That much is solid. Everything downstream of it is less so.
| Step | What is demonstrated | Evidence and limits |
|---|---|---|
| Platelet serotonin falls | Clinical doses reduce platelet serotonin content by 70–90%, similarly across SSRIs and SNRIs, concordant with brain transporter occupancy imaging [1]. A cross-sectional study of 1,433 patients found a bimodal distribution with the low mode corresponding to exposure | Systematic review of platelet serotonin content studies. This is the best-established step |
| Platelet function changes | In 14 chronically treated patients against 14 matched controls, platelet serotonin fell 66% and agonist-induced aggregation fell 10–52% with ADP, collagen and epinephrine — but not with arachidonic acid [2] | Small case-control study with ex vivo aggregometry. Fourteen patients |
| …or does not | A 2025 review concludes the certainty is very low: aggregometry, flow cytometry, granule secretion and phosphatidylserine exposure give conflicting results even with similar methods, attributed to design, dose, comorbidity and the confound of depression itself [3] | Narrative review. Directly contests the row above |
| The transporter may not be the whole story | Citalopram inhibits platelet function independently of transporter-mediated serotonin uptake [4] | Preclinical study in human platelets in vitro. Complicates the depletion account |
| Non-platelet routes | Increased gastric acid secretion has been proposed to explain why the signal is gastrointestinal rather than general, alongside effects on endothelial reactivity [5] | Narrative review, proposed rather than demonstrated. No quantified human data on fibrinolysis or vascular tone with a bleeding endpoint |
| Recovery after stopping | The seven-to-ten-day figure taught everywhere is platelet lifespan arithmetic. A randomised substudy in ischaemic stroke measured aggregation on drug and after washout, but the recovery interval itself was not retrievable [6] | No measured bleeding-recovery curve was found. The interval is an inference |
Two consequences follow for the rest of the chapter. First, because platelet depletion saturates at 70–90% across agents, a clean dose-response for bleeding is not expected on mechanistic grounds — and section 7 finds that the affinity gradient is indeed weak and contested. Second, if the effect were purely serotonergic, mirtazapine should be clean. Section 8 shows it is not.
🧠 Intracranial — rarer, worse
The relative increase here is smaller than for gastrointestinal bleeding. The consequence is not.
| Outcome | Estimate | Scope and source |
|---|---|---|
| Intracranial haemorrhage | Adjusted relative risk 1.51 (1.26–1.81) | 16 studies, 506,411 subjects, comparator non-use [10] |
| Intracerebral haemorrhage | Adjusted 1.42 (1.23–1.65); unadjusted 1.68 (1.46–1.91) with heterogeneity 71% | Same study set. Both models are printed because the adjustment moves the estimate [10] |
| Subarachnoid haemorrhage | Adjusted 0.62 (0.38–1.01) — no increase | Same study set [10]. The effect is not general to intracranial bleeding |
| Absolute excess | Approximately one additional intracerebral bleed per 10,000 people treated for one year | Characterised in words rather than published as an interval [13] |
| A second meta-analysis | Adjusted odds ratio 1.30 (1.02–1.67), again with no subarachnoid increase | Scope not retrievable [13] |
The subarachnoid row is worth pausing on. If antidepressants caused bleeding by a general effect on haemostasis, subarachnoid haemorrhage should rise with everything else. It does not. Whatever is happening is not indiscriminate.
This is the figure to carry into a conversation with a patient. A relative risk of 1.5 on an event with an annual incidence measured in tens per 100,000 produces an absolute excess that most patients, told plainly, will accept. The same relative risk applied to a patient already anticoagulated, already on an antiplatelet, and already hypertensive produces a different answer — not because the relative risk changed, but because the baseline did.
There is one body of randomised evidence here, and it is easy to miss because it was collected for another purpose. Fluoxetine was given deliberately to patients with fresh brain injury in three large stroke-recovery trials, pooled in a Cochrane review [29]. One of them recorded gastrointestinal bleeding as similar between fluoxetine and placebo [30]. Randomising serotonergic antidepressants into a population with fresh intracranial injury is an experiment nobody would design for this question, and it did not produce the bleeding excess the observational literature predicts.
💊 On an anticoagulant
This is where the interaction stops being theoretical, and also where the observational and randomised literatures disagree with each other.
| Study | Design and scope | Result |
|---|---|---|
| SSRI with an oral anticoagulant | Nested case-control in a UK primary-care database, 1998–2021. 331,305 anticoagulant initiators; 42,190 major bleeding cases against 1,156,641 controls; mean age 74.2, mean follow-up 4.6 years, background rate 27.9 major bleeds per 1,000 person-years [14] | Incidence rate ratio 1.33 (1.24–1.42) against anticoagulant alone. With a direct oral anticoagulant 1.25 (1.12–1.40); with a vitamin K antagonist 1.36 (1.25–1.47). By site: gastrointestinal 1.38 (1.24–1.53), intracranial 1.56 (1.32–1.85) |
| The same study, by duration | Same cohort | Risk is front-loaded: 1.74 (1.37–2.22) in the first 30 days of combined use, elevated for up to six months, substantially lower thereafter [14] |
| Within a randomised trial population | Post hoc propensity-matched analysis inside a rivaroxaban-versus-warfarin trial. 737 SSRI users matched to 737 non-users, mean follow-up 1.6 years [15] | Major plus clinically relevant non-major bleeding 1.16 (0.95–1.43) — not significant. Rivaroxaban arm 1.11 (0.82–1.51); warfarin arm 1.21 (0.91–1.60). Major bleeding alone, warfarin arm 1.58 (0.96–2.60) |
| Warfarin specifically | Cohort of 9,186 warfarin-treated patients with atrial fibrillation; 32,888 person-years of SSRI-exposed follow-up, 45 major haemorrhages [16] | Adjusted relative risk 1.41 (1.04–1.92). Rates 2.32 against 1.35 per 100 person-years |
| By agent, on warfarin | Nested case-control in claims data; 430,455 warfarin users, 13,026 gastrointestinal bleeding cases against 653,209 controls [17] | Citalopram 1.73 (1.25–2.38); paroxetine 1.64 (1.27–2.12); fluoxetine 1.63 (1.11–2.38); amitriptyline 1.47 (1.02–2.11); mirtazapine 1.75 (1.30–2.35) |
| On a direct oral anticoagulant, against other interacting drugs | Nationwide propensity-matched cohort, 193,072 anticoagulated patients with atrial fibrillation [19] | SSRI or SNRI stratum, n=19,319: major or clinically relevant non-major bleeding 1.26 (1.17–1.35); intracranial haemorrhage 1.50 (1.25–1.81). For comparison in the same cohort: P2Y12 inhibitors 1.62, corticosteroids 1.53, low-dose aspirin 1.14, NSAIDs 1.10 |
| Clopidogrel and CYP2C19 | Propensity-matched cohort plus meta-analysis; fluoxetine and fluvoxamine are the inhibitors at issue [18] | Pooled ischaemic events 1.11 (1.01–1.22); pooled bleeding 0.80 (0.55–1.18). The clopidogrel interaction reduces antiplatelet effect; it does not increase bleeding |
| Source | Result | Design |
|---|---|---|
| Pooled across studies | SSRI with an oral anticoagulant against anticoagulant alone: hazard ratio 1.35 (1.14–1.58); direct oral anticoagulant subgroup 1.47 (1.03–2.10) [42] | Systematic review and meta-analysis |
| Spontaneous reporting | Increased reporting of serious bleeding when a serotonin reuptake inhibitor was added: reporting odds ratio 1.49 with direct oral anticoagulants, 1.37 with vitamin K antagonists, 1.38 with platelet aggregation inhibitors, 2.04 with heparins — and no signal with non-serotonergic antidepressants [43] | Global pharmacovigilance database. Reporting odds ratios are not risk estimates |
| Pharmacokinetic, not platelet | Fluoxetine and fluvoxamine carry the highest interaction risk with warfarin. All direct oral anticoagulants except dabigatran are CYP3A4 substrates, so the same two agents warrant caution there [44] | Review of drug metabolism. This is a separate mechanism from platelet serotonin and it picks out specific agents |
The pharmacokinetic route deserves one exact figure. Starting an SSRI raised the risk of a supratherapeutic international normalised ratio of 5 or more by 2.41-fold (2.01–2.89) within two months, rising to 3.14-fold (1.33–7.43) for the CYP2C9-inhibiting agents fluoxetine and fluvoxamine — although in that same study the major-bleeding signal itself tracked the platelet effect more than the interaction [57]. For a patient on warfarin, sertraline, escitalopram and venlafaxine are the agents without a statistically significant interaction [34][44]. Citalopram belongs on the other side of that line: it carries minimal cytochrome inhibition but sits at 1.73 in the by-agent table above, the highest of the SSRIs that study reported [17]. Low pharmacokinetic interaction is not the same as low bleeding risk, and on this drug the two diverge.
The pharmacovigilance row is the one that speaks to mechanism: the signal appeared with serotonergic antidepressants and not with non-serotonergic ones [43]. That cuts the opposite way from the mirtazapine findings in section 8, and the disagreement is left standing.
Inside a trial, with protocol-driven monitoring and a population screened for comorbidity, adding an SSRI to an anticoagulant produced no significant excess bleeding: 1.16 (0.95–1.43) [15]. In a primary-care database of 331,305 anticoagulant users with a mean age of 74, it produced 1.33 (1.24–1.42) [14]. Both are real answers to slightly different questions, and the difference between them is the difference between the patients who enter trials and the patients in front of you.
No dedicated antidepressant analysis exists for the apixaban, dabigatran or edoxaban trials. The rivaroxaban analysis is the only one, and it was post hoc.
The calibration in the direct-oral-anticoagulant cohort is the most useful single line here. An SSRI sits at 1.26, above low-dose aspirin at 1.14 and NSAIDs at 1.10, below corticosteroids at 1.53 and a P2Y12 inhibitor at 1.62 [19]. Clinicians co-prescribe all of those without ceremony. The antidepressant is not in a different category from the rest of the list; it is in the middle of it.
🔪 Perioperative — the advice with no study
Stop the SSRI two weeks before surgery is among the most confidently repeated instructions in perioperative medicine. No study has tested it.
A systematic search for any study — randomised or observational, any design, any surgery — comparing stopping an antidepressant before surgery against continuing it, with a bleeding outcome, returns nothing. Every study in this literature compares exposure against non-exposure: patients on an antidepressant against patients not on one. That is a different question, and it cannot answer this one, because the patients not on an antidepressant differ in the thing the antidepressant treats.
The orthopaedic meta-analysis below states this in its own text: no discontinuation-comparison study was identified, and whether stopping is advisable remains uncertain [24].
The same 530,416-patient surgical cohort under three analyses. Restricting to patients with a documented depression diagnosis removes the mortality signal and leaves the smaller bleeding signal standing [20].
The largest cohort identified for this chapter is a retrospective study of 530,416 adults across 375 US hospitals, covering spine, cardiac, vascular, gastrointestinal, gynaecological and orthopaedic procedures [20]. Its headline figures are the ones usually quoted.
| Analysis | Death in hospital | Bleeding requiring transfusion | Readmission at 30 days |
|---|---|---|---|
| Full cohort, adjusted | 1.20 (1.07–1.36) | 1.09 (1.04–1.15) | 1.22 (1.18–1.26) |
| Propensity-matched | 1.19 (1.03–1.37) | 1.07 (0.99–1.14) — no longer significant | 1.18 (1.13–1.23) |
| Restricted to patients with a documented depression diagnosis | 0.86 (0.65–1.13) — the signal is gone | 1.10 (1.03–1.18) — the signal persists | Not reported for this subgroup |
That third row is the most important line in this chapter. Among patients who took an SSRI, 41.0% had a coded depression diagnosis; among those who did not, 6.2% did [20]. When the comparison is made within the depressed, the excess mortality disappears. The bleeding excess does not. Confounding by indication plausibly accounts for the mortality finding — the frightening one — and does not account for the bleeding finding, which is small.
The same paper contains a secondary analysis that looks like the missing comparison and is not. Patients were grouped by when the drug was actually administered: given the day before or day of surgery and afterwards, against given only from the day after surgery onwards. Against a no-SSRI reference, the continued group's bleeding estimate was 1.07 (1.02–1.13) and the apparently-held group's was 1.29 (1.12–1.48), with readmission 1.18 (1.14–1.23) against 1.66 (1.49–1.84) [20].
The held group did worse, not better. But the authors caution that they cannot tell how long the drug was held, and a patient receiving no oral medication until the day after surgery is more likely to be nil by mouth, ileus-bound or sicker than to have had a considered preoperative discontinuation. No direct statistical test of held against continued was performed. This is a hypothesis-generating association, and the honest reading is that it fails to show continuing is dangerous rather than that it shows holding is.
| Setting | Evidence | Reading |
|---|---|---|
| Cardiac surgery, pooled | Meta-analysis of 7 observational studies: red cell transfusion 1.15 (1.06–1.26); reoperation for bleeding 1.07 (0.66–1.74); platelet transfusion 0.93 (0.79–1.09); fresh frozen plasma 0.96 (0.74–1.24); 30-day mortality 1.03 (0.90–1.17) [22] | Transfusion rises. Reoperation for bleeding — the hard endpoint — does not |
| Coronary artery bypass grafting | Cohort of 132,686 patients: major bleeding 0.98 (0.90–1.07); in-hospital mortality 0.93 (0.80–1.07); at least one unit of red cells 1.14 (1.10–1.18) [23] | Same pattern. Transfusion only |
| Hip and knee arthroplasty | Meta-analysis of 10 comparative cohort studies, 2,098,833 patients of whom 418,527 exposed: transfusion 1.78 (1.11–2.83), heterogeneity 97% [24] | The largest effect in the perioperative literature, on a soft endpoint, with extreme heterogeneity |
| Study | Result | Why it matters here |
|---|---|---|
| Elective colectomy, propensity-matched, 2026 | Bleeding-specific 30-day relative risk 1.29, number needed to harm 104, persisting in new initiators. The authors conclude the data support perioperative bleeding awareness rather than routine discontinuation [45] | The only number needed to harm in this chapter that arrives with its population, its endpoint and its horizon attached. Compare it with the 881 and 3,177 in section 2, which arrive with none of those |
| Meta-analysis across surgical specialties, 2015 | Increased transfusion requirement overall, but no increased requirement in coronary artery bypass grafting, with the signal concentrated in orthopaedic and other procedures [46] | Argues directly against a uniform perioperative rule. The procedure matters more than the drug |
Across every surgical setting the pattern repeats: transfusion goes up, and the endpoints that matter more — reoperation for bleeding, mortality — do not. Transfusion is a decision as much as an event, and a surgeon who knows the patient is on an SSRI may transfuse sooner.
| Body | Position |
|---|---|
| Society for Perioperative Assessment and Quality Improvement, 2022 | SSRIs, SNRIs, tricyclics and atypical antidepressants should all be taken preoperatively, including on the day of surgery [25] |
| American Society of Plastic Surgeons, 2015 | Routine discontinuation of antidepressants before surgery in the absence of a careful evaluation should be avoided, on the grounds that the risks of stopping in psychologically vulnerable patients likely outweigh any increase in complications [26] |
| European Society of Anaesthesiology and Intensive Care, 2024 | Silent. The preoperative assessment guideline was read in full and contains no recommendation on psychiatric medication [27] |
| American Society of Anesthesiologists, American Society of Regional Anesthesia, NICE, ERAS Society | No antidepressant-specific perioperative statement identified. Recorded as silence on a search basis rather than a full-text read of every document |
Every body that has taken a position says continue. The instruction to stop does not come from a guideline, and it does not come from a trial.
Across 79 studies and 21,002 participants, discontinuation symptoms occurred in 31% (27–35%) after stopping an antidepressant against 17% (14–21%) after stopping placebo — an attributable incidence of roughly 15%, or about one in seven. Severe symptoms occurred in 2.8% (1.4–5.7%) against 0.6% (0.2–1.3%) [28]. Symptoms typically begin within three days, sooner for short half-life agents.
Relapse risk specifically over a perioperative window of days to weeks has not been studied, and no study connects perioperative antidepressant cessation to postoperative delirium, pain scores or opioid requirement. Both are genuine absences rather than negative findings.
Because psychiatrists manage this procedure themselves rather than being consulted about it, it is worth stating plainly: there is no bleeding-based reason to stop an antidepressant before electroconvulsive therapy. Concomitant SSRIs modestly increase seizure duration without clinically significant harm and are associated with better postictal suppression, and a review of psychotropic co-administration during the procedure reported no serious adverse events attributable to them [56]. The logic of this section — continue rather than stop — applies here too.
⚖️ Does the agent matter?
The affinity gradient is the most repeated claim in this area and the least demonstrated. Three sources have tested it and they do not agree.
| Source | What it found | Scope |
|---|---|---|
| Nested case-control study of intracranial haemorrhage | Strong transporter inhibitors against weaker ones: 1.25 (1.01–1.54). A gradient, barely significant [11] | 1,363,990 incident antidepressant users; 3,036 cases matched to 89,702 controls; comparator throughout is tricyclics, chosen as an active comparator to blunt confounding by indication |
| Systematic review of the same question | The effect is moderate at best, with no clear dose-gradient by strength of serotonin inhibition [12] | 10 studies spanning cohorts of 136,293 to 1,363,990; 12,797 intracranial haemorrhage events |
| Anticoagulant co-prescription study | Risk did not vary by SSRI potency. Reported qualitatively; the per-stratum figures sit in a forest plot and were not retrievable [14] | 42,190 major bleeding cases |
The strongest test is per-agent rather than per-tier, and it does not follow the tiers. A 2026 systematic review and meta-analysis of individual agents found every studied SSRI and SNRI raised gastrointestinal bleeding risk, with venlafaxine highest at 1.50 (1.32–1.70), then citalopram at 1.38 (1.17–1.62) and fluoxetine at 1.38 (1.26–1.51) — and paroxetine lowest of those reported, at 1.31 (1.07–1.62) [47].
Read that against the affinity classification in circulation, which places paroxetine and fluoxetine in the high-affinity tier and venlafaxine in the intermediate one [48]. Among the agents that review reported, the one with the highest measured risk sits in the middle tier and the one with the lowest sits at the top. A large primary-care cohort adds trazodone, a low-affinity agent, with an elevated upper gastrointestinal signal [49]. A pharmacovigilance analysis puts sertraline at the head of the gastrointestinal signal, concentrated in women and in patients aged 60 and over [50].
A fifth test splits the question rather than answering it. In a cohort of 156,307 high-affinity against 102,631 low-affinity users aged 15 to 24, abnormal uterine bleeding did not differ by affinity at a hazard ratio of 1.01 (0.93–1.09) — but anaemia did, at 1.29 (1.04–1.61) [54]. The tier predicted the downstream consequence and not the bleeding itself.
The same study that found the affinity gradient also reports two strata that undercut the alarm: strong inhibitors in the first 30 days at 1.68 (0.90–3.12), and concomitant oral anticoagulant use at 1.73 (0.89–3.39). Both intervals cross one [11]. The two figures a clinician would most want to act on are the two that did not reach significance.
Note what the comparator is. That study compared SSRIs against tricyclics, not against no antidepressant. A relative risk of 1.17 for SSRIs against tricyclics does not mean tricyclics are safe; it means the two classes are close, which is consistent with tricyclics being potent transporter inhibitors themselves — a fact their labels, written before any of this was known, do not mention [21].
🚫 The drugs everyone switches to
Mirtazapine and bupropion are recommended for patients at bleeding risk on the grounds that they lack serotonin reuptake inhibition. That reasoning has been tested directly, and it did not survive.
| Comparison | Result | Source |
|---|---|---|
| Mirtazapine against no antidepressant | Gastrointestinal bleeding odds ratio 1.17 (1.01–1.38) — greater risk, not lesser | Systematic review and meta-analysis, five separate meta-analyses from studies searched to May 2017 [35] |
| Mirtazapine against an SSRI | No difference | Same review [35] |
| Bupropion against an SSRI | No difference | Same review [35] |
| Mirtazapine on warfarin | Gastrointestinal bleeding odds ratio 1.75 (1.30–2.35) — the highest figure among the agents that study reported, above citalopram at 1.73, paroxetine at 1.64 and fluoxetine at 1.63 | Nested case-control, 430,455 warfarin users [17] |
Trazodone belongs in this paragraph too. It sits in the low-affinity tier and carried an elevated upper gastrointestinal bleeding signal in a primary-care cohort of the kind used to build the tiers in the first place [49].
The authors of the meta-analysis state their conclusion plainly: it is premature to recommend mirtazapine and bupropion for patients who have a bleeding risk [35]. The authors of the warfarin study go further, arguing that mirtazapine's elevated risk — in a drug with no meaningful transporter affinity and no established warfarin interaction — is itself evidence against platelet serotonin depletion as the sole mechanism [17].
Mirtazapine's label contains no bleeding language whatsoever [21]. The switch to mirtazapine is the standard manoeuvre for a patient at bleeding risk. And the only direct test of that manoeuvre found mirtazapine carried more gastrointestinal bleeding than no antidepressant and no less than an SSRI [35].
A clean label is not a safety finding. It is a record of what was asked, when, and by which regulatory template.
🔍 Who is actually at risk
The risk factors are the ordinary ones, and the striking fact is that no instrument used to quantify bleeding risk includes the drug at all.
| Score | Variables | Antidepressant included? |
|---|---|---|
| HAS-BLED | Hypertension, abnormal renal or liver function, stroke, bleeding history, labile INR, age over 65, drugs or alcohol | No. The drugs component is anchored to antiplatelet agents and NSAIDs in the validation paper [39] |
| ORBIT | Older age, reduced haemoglobin, bleeding history, renal function, antiplatelet treatment | No |
| ATRIA | Anaemia, severe renal disease, age 75 or over, prior haemorrhage, hypertension | No |
| HEMORR2HAGES | Hepatic or renal disease, alcohol, malignancy, older age, reduced platelet count or function, rebleeding risk, hypertension, anaemia, genetic factors, fall risk, stroke | No |
| Glasgow-Blatchford, Rockall, AIMS-65 and six other upper gastrointestinal scores | Urea, haemoglobin, blood pressure, heart rate, melaena, syncope, hepatic disease, cardiac failure, age, endoscopic findings | No. A comparative review of nine such scores contains no mention of antidepressants at all [40] |
Every one of these instruments was developed or last revised after the pharmacoepidemiology in this chapter was published. None includes antidepressant exposure, and none has an antiplatelet catch-all broad enough to capture it by implication. A patient's HAS-BLED score is identical whether or not they take an SSRI.
| Factor | What is established |
|---|---|
| Concurrent NSAID | The largest modifiable multiplier: 4.25 (2.82–6.42) combined [7], and 9.1 (4.8–17.3) without acid suppression [9] |
| Concurrent anticoagulant | 1.33 (1.24–1.42), front-loaded to 1.74 in the first 30 days [14] |
| First weeks of combined exposure | Risk peaks early and falls after six months [14]. Newly combined is riskier than long combined |
| Age | Cohorts are elderly by construction — mean age 74.2 in the anticoagulant study [14]. Effect modification by age was reported as absent in that cohort |
| Concurrent corticosteroid, antiplatelet | Independent contributors in the same anticoagulated cohort: corticosteroids 1.53, P2Y12 inhibitors 1.62 [19] |
| Prior gastrointestinal bleed, Helicobacter pylori, cirrhosis, thrombocytopenia, alcohol | Standard bleeding risk factors. Effect sizes specifically in combination with an antidepressant were not identified for this chapter — an open gap, not a documented absence |
🩸 Menstrual and uterine bleeding
A reproductive-age woman on an antidepressant is the patient a psychiatrist is most likely to counsel about bleeding personally, and the literature here behaves differently from everything above.
| Finding | Result | Source and scope |
|---|---|---|
| Does transporter affinity predict abnormal uterine bleeding? | No. Hazard ratio 1.01 (0.93–1.09) for high-affinity against low-affinity agents, unmodified by duration, age or comorbidity | Primary-care database cohort, 156,307 high-affinity against 102,631 low-affinity users aged 15 to 24 [54] |
| Does it predict anything? | Anaemia, at 1.29 (1.04–1.61). The bleeding endpoint was null and the consequence endpoint was not | Same cohort [54]. This is the fifth test of the affinity gradient in this chapter and the only one that splits by endpoint |
| Heavy menstrual bleeding by agent | Wide variation between agents, with venlafaxine and duloxetine at the top of the range and bupropion at the bottom | Cohort of 1,949 women aged 18 to 35 with bipolar disorder or major depression [55]. Single-centre and smaller than the cohort above |
| The combination that matters | Adding valproate raised heavy menstrual bleeding odds across nearly every antidepressant, reaching an odds ratio of 8.48 for venlafaxine with valproate in bipolar disorder | Same cohort [55] |
Two things follow. Monitor for anaemia rather than assuming the affinity tier predicts menstrual bleeding, because it does not and the anaemia signal is the one that reached significance. And treat antidepressant plus valproate in a reproductive-age woman the way this chapter treats antidepressant plus NSAID everywhere else: each component is modest on its own and the combination is where the risk concentrates.
Postpartum haemorrhage is a real and well-studied association with serotonergic antidepressants near delivery, and it is deliberately not covered here. Perinatal safety has its own series on this platform, and splitting that material across two courses would leave a prescriber reading half of it. The short version for anyone who needs it today: the guidance from obstetric bodies is to continue treatment and prepare for the delivery rather than to taper, and the mitigation is obstetric rather than psychiatric.
🛡️ Management
One intervention has a number attached to it. The rest is reasoning — though as of 2026 the reasoning has a published framework behind it for the first time.
An expert consensus from the Association of Medicine and Psychiatry, published in 2026, is the first dedicated clinical decision algorithm for whether to initiate, discontinue or dose-adjust an antidepressant in a patient at bleeding risk [53]. It is a consensus document rather than a trial, and it preserves clinical judgement rather than issuing a rule, but it is the current reference point for the decisions in this section and it did not exist when most of the teaching in section 12 was formed.
| Option | What exists | In this condition, or borrowed? |
|---|---|---|
| Add acid suppression | The only option with a figure. Stratified by acid-suppressant use, SSRI with an NSAID ran at 9.1 (4.8–17.3) without and 1.1 (0.3–3.4) with [9]. There has never been a trial of adding a proton pump inhibitor to an antidepressant user; one dedicated review states plainly that no single study strictly examines this [33] | This condition, observational stratification only. Not a trial |
| Stop the NSAID rather than the antidepressant | No comparative study of the two strategies exists. But the NSAID is the larger multiplier, the antidepressant is treating an illness with its own mortality, and stopping the antidepressant has a measured 15% attributable discontinuation-symptom rate [28] against no measured benefit | Extrapolated, but the arithmetic favours it |
| Switch to mirtazapine or bupropion | Directly tested and not supported. Mirtazapine 1.17 (1.01–1.38) against no antidepressant; no difference against SSRI for either agent; the authors call the recommendation premature [35] | Tested and negative. This is not a neutral option |
| Dose reduction | No evidence located. Two dedicated reviews of this literature do not address it [33][34] | Absent |
| Stop the antidepressant before surgery | No study of any design. Every society that has issued a position says continue [25][26] | Absent, and advised against by the bodies that have spoken |
| Platelet transfusion or desmopressin at an active bleed | No antidepressant-specific evidence. Both dedicated reviews state it is not addressed in this literature [33][34] | Absent |
Gastroenterology is silent. The American College of Gastroenterology's upper gastrointestinal bleeding guideline, its guideline on anticoagulants and antiplatelets during acute bleeding, and its NSAID ulcer-prevention guideline were each searched in full and contain no mention of SSRIs, antidepressants or serotonin [36]. The same is true of the British Society of Gastroenterology care bundle and the European Society of Gastrointestinal Endoscopy guideline [38][37].
Geriatrics is the exception. STOPP/START version 3 carries three criteria naming this directly: SSRIs with current or recent significant bleeding, because of their antiplatelet effect; SSRIs combined with a vitamin K antagonist, direct thrombin inhibitor or factor Xa inhibitor in a patient with previous major haemorrhage; and SSRIs with significant hyponatraemia [32]. The 2023 Beers Criteria connect SSRIs to bleeding only through a warfarin drug-interaction row advising avoidance where possible and close INR monitoring, with SSRIs otherwise listed for hyponatraemia rather than bleeding [31].
Among depression guidelines, one addresses the NSAID combination; its exact wording could not be retrieved. Others were silent in the sections searched, and one could not be reached at all.
🔄 After a bleed
This is the decision clinicians ask about most. Several questions hide inside it; most now have data, one of them conflicting, and only the last is still empty.
| The question | What exists | Source and design |
|---|---|---|
| Should the drug be stopped when a patient is admitted with a gastrointestinal bleed? | No. Rebleeding after initial endoscopy was unaffected by SSRI use, and the excess mortality disappeared after adjustment for age, aspirin and NSAID use. The accompanying editorial concludes it is unnecessary to stop an SSRI on admission [51] | Cohort study, reported through an editorial in a gastroenterology journal. The editorial also notes there is no evidence that a prophylactic proton pump inhibitor reduces bleeding in SSRI users specifically, as distinct from the NSAID-interaction data in section 2 |
| Did the antidepressant cause an intracerebral bleed? | Prior SSRI use was not associated with increased risk at an odds ratio of 0.82 (0.63–1.07), and did not potentiate the risk carried by anticoagulants or antiplatelets [52] | Post hoc analysis of a case-control study, 2,287 intracerebral haemorrhage cases against 2,895 controls |
| Should an antidepressant be started fresh after an intracerebral bleed? | The one cautionary signal in this section. New initiation after the event was associated with an unfavourable 3-month functional outcome at 1.67 (1.16–2.41), attenuating to non-significance on propensity analysis [52] | Same analysis. Note what it is and is not: this is a new start after a brain bleed, not the continuation of an established antidepressant |
| Does using an antidepressant after an intracerebral bleed raise the chance of another one? | Directly studied, and the two studies disagree. A single-centre cohort of 1,279 survivors of primary intracerebral haemorrhage, followed for a median of 53 months, found a subhazard ratio of 1.31 (1.08–1.59) for recurrence with SSRI exposure — and 1.79 (1.22–2.64) in a composite higher-risk group defined by lobar haemorrhage, prior haemorrhage, APOE ε2 or ε4 carriage, or Black or Hispanic ethnicity, against 1.20 (1.01–1.42) in the rest [58]. The same exposure was associated with remission of post-haemorrhage depression at 1.53 (1.12–2.09). A Danish nationwide cohort of 3,998 post-haemorrhage SSRI users found no association with recurrence at 0.95 (0.82–1.10) [59] | Observational cohort against a nationwide administrative cohort. The conflict is left standing. Note that the higher-risk figure is a single composite group, not four separate subgroup analyses, and that a reported dose-gradient could not be verified and is therefore not quoted |
| Should an established antidepressant be restarted after discharge, and when? | Nothing. No study addresses restart timing after either kind of bleed, and none quantifies the relapse cost of not restarting over that window | Two contemporary reviews written to help clinicians with this decision both state the evidence is absent [33][34] |
The two settings diverge and the divergence is the usable finding. After a gastrointestinal bleed the evidence is reassuring and the drug should stay. After an intracerebral bleed prior use did not cause the event, a de novo start afterwards carried a worse functional outcome in the unadjusted model, and the recurrence question is genuinely unsettled between two good studies pointing opposite ways. That is a reason to be deliberate about a new start in that specific situation, and about lobar bleeds and prior bleeds in particular. It is not a reason to stop a drug the patient was already taking.
This section previously said there was no evidence in either direction after a bleed. That was too strong. After a gastrointestinal bleed the evidence is reassuring and the drug should stay [51]. After an intracerebral bleed the picture is contested rather than empty: prior use did not cause the event [52], a de novo start afterwards carried a worse functional outcome in the unadjusted model [52], and post-haemorrhage use has been linked to recurrence in one cohort and to none in another [58][59].
On guidelines, the position is narrower than it might appear. The American Heart Association and American Stroke Association's 2022 intracerebral haemorrhage guideline addresses fluoxetine, and what it says is about recovery rather than bleeding: multiple randomised trials did not confirm an earlier suggestion that fluoxetine might improve functional recovery after intracerebral haemorrhage, and fluoxetine reduced depression in those trials while increasing the incidence of fractures [60]. A search of that guideline found no discussion of haemorrhage recurrence in connection with those trials, and no recommendation reserving antidepressants for severe depression could be confirmed in its text. The restart decision after a bleed still has no dedicated guideline recommendation; what it now has is data.
⚠️ Where teaching outruns the evidence
| What is said | What the evidence shows |
|---|---|
| “Stop the SSRI two weeks before surgery” | No study of any design has compared stopping with continuing. Every society that has issued a position recommends continuing, including on the day of surgery [25][26]. Stopping has a measured 15% attributable discontinuation-symptom rate [28] |
| “Mirtazapine and bupropion are safe because they don't block the transporter” | Tested directly. Mirtazapine carried more gastrointestinal bleeding than no antidepressant at 1.17 (1.01–1.38), and neither agent differed from an SSRI. The authors call the recommendation premature [35]. On warfarin, mirtazapine was the highest of the agents reported at 1.75 (1.30–2.35) [17] |
| “The bleeding risk is clinically trivial” | Trivial alone; not trivial in combination. The interaction with an NSAID without acid suppression reaches 9.1 (4.8–17.3) [9], and intracranial haemorrhage on an anticoagulant reaches 1.56 (1.32–1.85) [14] |
| “SSRIs are contraindicated with anticoagulants” | They are not. The increase is 1.33 (1.24–1.42) observationally [14] and 1.16 (0.95–1.43), non-significant, inside a randomised trial population [15]. In a head-to-head cohort an SSRI sat between aspirin and corticosteroids for bleeding risk [19]. STOPP/START flags the combination in patients with previous major haemorrhage, which is narrower than a contraindication [32] |
| “Add a proton pump inhibitor and the problem is solved” | The stratified figures are striking — 9.1 down to 1.1 — but they come from one case-control study and no trial has ever tested adding acid suppression in this population [9][33]. It is the best-supported option and it is still observational |
| “The risk is only gastrointestinal” | Intracranial haemorrhage rises at 1.51 (1.26–1.81) [10]. Subarachnoid haemorrhage does not, at 0.62 (0.38–1.01) [10] — so the effect is neither purely gastrointestinal nor indiscriminate |
| “Tricyclics don't cause bleeding” | Their labels predate the warning entirely and mention no anticoagulant interaction [21]. Clomipramine is nonetheless the only antidepressant whose label prints bleeding percentages, and amitriptyline on warfarin gave 1.47 (1.02–2.11) [17] |
| “There's no evidence either way after a bleed” | Outdated, and this chapter said it in an earlier draft. Rebleeding after a gastrointestinal bleed is not increased and stopping on admission is not supported [51]. Prior use did not raise intracerebral haemorrhage risk at 0.82 (0.63–1.07) [52], a de novo start after one carried a worse functional outcome [52], and recurrence data conflict outright — 1.31 (1.08–1.59) in one cohort against 0.95 (0.82–1.10) in another [58][59]. What remains unstudied is restart timing after discharge |
| “Check the bleeding risk score” | No validated score includes antidepressant exposure — not HAS-BLED, ORBIT, ATRIA, HEMORR2HAGES, or any of nine upper gastrointestinal scores in a comparative review [39][40] |
🪜 Action ladder
The sertraline stays. There is no study supporting the pre-anaesthetic clinic's request, the societies that have taken a position recommend continuing through the day of surgery, and this man has relapsed twice after interruptions.
The ibuprofen is the problem. He is on an anticoagulant, an SSRI and a daily NSAID, which is the combination every figure in this chapter converges on. The knee is being replaced in three weeks, so the analgesic requirement is temporary and substitutable. Stop the ibuprofen, offer paracetamol and topical therapy, and if an NSAID proves unavoidable, add a proton pump inhibitor.
Write back to the clinic stating the sertraline is continuing, that the NSAID has been stopped instead, and why. Note the first month after any new anticoagulant or antidepressant change as the higher-risk window.
- Fourteen labels warn and none quantifies. The only label with bleeding percentages is clomipramine's, and it carries no warning [21].
- The perioperative mortality signal is 1.20 in everyone and 0.86 among patients who actually have depression. Confounding by indication is visible inside the paper's own data [20].
- Transfusion rises across every surgical setting; reoperation for bleeding does not [22][23]. Transfusion is partly a decision.
- Subarachnoid haemorrhage does not rise. Whatever the mechanism is, it is not a general haemostatic defect [10].
- Risk on an anticoagulant is front-loaded: 1.74 in the first 30 days, falling after six months [14].
- An SSRI sits between aspirin and corticosteroids for bleeding risk in an anticoagulated cohort [19]. Clinicians co-prescribe both without ceremony.
- Mirtazapine's clean label is a record of labelling practice, not a safety finding [21][35].
- No bleeding risk score includes the drug. A patient's HAS-BLED is the same with or without it [39][40].
- An antidepressant, an NSAID and no acid suppression. This is the 9.1 combination and it is the one worth acting on today [9].
- First 30 days of a new antidepressant-plus-anticoagulant combination, particularly in a patient over 75 [14].
- Any new headache, focal deficit or altered consciousness in an anticoagulated patient on an antidepressant — intracranial haemorrhage on the combination runs at 1.56 (1.32–1.85) [14].
- A pre-anaesthetic instruction to stop an antidepressant in a patient with a relapsing course. The instruction has no evidence behind it and the relapse risk is real [25][26][28].
- Melaena, haematemesis or unexplained anaemia in a patient on an antidepressant plus any antithrombotic. Investigate as a bleed, not as a drug side effect.
- A switch to mirtazapine made specifically for bleeding safety — the manoeuvre has been tested and did not work [35].
“These tablets make platelets slightly less sticky. On their own that's a small effect — roughly one extra serious bleed for every several hundred to a few thousand people treated, depending on who you are to start with.”
“What matters much more is what else you take. The anti-inflammatory is doing more here than the antidepressant is, and it's the one we can most easily change.”
“The surgery team asked about stopping it. I've looked at this carefully: there's no study showing that stopping before an operation reduces bleeding, and the groups that have issued guidance say to keep taking it. Stopping carries its own risk for you specifically, given what happened the last two times.”
“Tell me straight away about black or tarry stools, vomiting anything that looks like coffee grounds, a headache that is not like your usual ones, or new weakness or confusion. Those are the ones that need same-day attention rather than a routine appointment.”
A rapid-decision reference series for prescribers, covering the adverse effects that change antidepressant treatment decisions at the point of care.
- Anderson GM, Bruno-Pacella I. Systematic review of studies using platelet serotonin content to assess bioeffect of serotonin reuptake inhibitors at the serotonin transporter. Psychopharmacology. 2023.
- Bismuth-Evenzal Y, Gonopolsky Y, Gurwitz D, et al. Decreased serotonin content and reduced agonist-induced aggregation in platelets of patients chronically medicated with SSRI drugs. J Affect Disord. 2012.
- Mokhtarian A, Siguret V, Jourdi G. Effects of selective serotonin reuptake inhibitors on platelet functions: a literature review. Curr Opin Hematol. 2025. (Narrative review.)
- Roweth HG, Yan R, Bedwani NH, et al. Citalopram inhibits platelet function independently of SERT-mediated 5-HT transport. Sci Rep. 2018. (Preclinical study in human platelets in vitro.)
- 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. (Narrative review.)
- Kraglund KL, Mortensen JK, Johnsen SP, Andersen G, Grove EL. Antiplatelet effects of citalopram in patients with ischaemic stroke: a randomized, placebo-controlled, double-blind study. Sci Rep. 2019.
- Anglin R, Yuan Y, Moayyedi P, et al. Risk of upper gastrointestinal bleeding with selective serotonin reuptake inhibitors with or without concurrent nonsteroidal anti-inflammatory use: a systematic review and meta-analysis. Am J Gastroenterol. 2014;109(6):811–819.
- Jiang HY, Chen HZ, Hu XJ, et al. Use of selective serotonin reuptake inhibitors and risk of upper gastrointestinal bleeding: a systematic review and meta-analysis. Clin Gastroenterol Hepatol. 2015;13(1):42–50.e3.
- de Abajo FJ, García-Rodríguez LA. Risk of upper gastrointestinal tract bleeding associated with selective serotonin reuptake inhibitors and venlafaxine therapy: interaction with nonsteroidal anti-inflammatory drugs and effect of acid-suppressing agents. Arch Gen Psychiatry. 2008;65(7):795–803. (Case-control study; venlafaxine is grouped with the SSRIs.)
- Hackam DG, Mrkobrada M. Selective serotonin reuptake inhibitors and brain hemorrhage: a meta-analysis. Neurology. 2012.
- Renoux C, Vahey S, Dell'Aniello S, Boivin JF. Association of selective serotonin reuptake inhibitors with the risk for spontaneous intracranial hemorrhage. JAMA Neurol. 2017. (Nested case-control; comparator is tricyclics.)
- Douros A, Ades M, Renoux C. Risk of intracranial hemorrhage associated with the use of antidepressants inhibiting serotonin reuptake: a systematic review. CNS Drugs. 2018.
- Hemphill JC. Selective serotonin reuptake inhibitors and intracranial hemorrhage: deriving clinical relevance from an epidemiologic association. JAMA Neurol. 2017. (Editorial.)
- Rahman AA, Platt RW, Beradid S, Boivin JF, Rej S, Renoux C. Concomitant use of selective serotonin reuptake inhibitors with oral anticoagulants and risk of major bleeding. JAMA Netw Open. 2024;7(3):e243208.
- Quinn GR, Hellkamp AS, Hankey GJ, et al. Selective serotonin reuptake inhibitors and bleeding risk in anticoagulated patients with atrial fibrillation: an analysis from the ROCKET AF trial. J Am Heart Assoc. 2018. (Post hoc propensity-matched analysis.)
- Quinn GR, Singer DE, Chang Y, et al. Effect of selective serotonin reuptake inhibitors on bleeding risk in patients with atrial fibrillation taking warfarin. Am J Cardiol. 2014;114(4):583–586.
- Schelleman H, Brensinger CM, Bilker WB, Hennessy S. Antidepressant-warfarin interaction and associated gastrointestinal bleeding risk in a case-control study. PLoS One. 2011;6(6):e21447.
- Bykov K, Schneeweiss S, Glynn RJ, et al. Updating the evidence of the interaction between clopidogrel and CYP2C19-inhibiting selective serotonin reuptake inhibitors: a cohort study and meta-analysis. Drug Saf. 2017.
- Grymonprez M, Capiau A, Steurbaut S, et al. Pharmacodynamic drug-drug interactions and bleeding outcomes in patients with atrial fibrillation using non-vitamin K antagonist oral anticoagulants: a nationwide cohort study. Cardiovasc Drugs Ther. 2025;39(1):133–143.
- Auerbach AD, Vittinghoff E, Maselli J, Pekow PS, Bindman AB, Lindenauer PK. Perioperative use of selective serotonin reuptake inhibitors and risks for adverse outcomes of surgery. JAMA Intern Med. 2013;173(12):1075–1081.
- US Food and Drug Administration. Prescribing information for fluoxetine, paroxetine, sertraline, citalopram, escitalopram, fluvoxamine, duloxetine, venlafaxine extended-release, desvenlafaxine, levomilnacipran, milnacipran, bupropion, mirtazapine, trazodone, vortioxetine, vilazodone, amitriptyline, nortriptyline, imipramine, clomipramine, doxepin, phenelzine and tranylcypromine. Retrieved from DailyMed.
- Eckersley MJ, Sepehripour AH, Casula R, Punjabi P, Athanasiou T. Do selective serotonin reuptake inhibitors increase the risk of bleeding or mortality following coronary artery bypass graft surgery? A meta-analysis of observational studies. Perfusion. 2018.
- Xiong GL, Ghanem KG, Roman C, et al. Antidepressant use and risk of bleeding and mortality after coronary artery bypass graft surgery. Drug Saf. 2015.
- Serotonin reuptake inhibitors and transfusion requirement in total hip and knee arthroplasty: a systematic review and meta-analysis of comparative cohort studies. J Orthop. 2025.
- Society for Perioperative Assessment and Quality Improvement. Preoperative management of medications for psychiatric diseases: SPAQI consensus statement. Mayo Clin Proc. 2022.
- American Society of Plastic Surgeons. Statement on perioperative antidepressant use, accompanying a review published in Plast Reconstr Surg. 2015.
- European Society of Anaesthesiology and Intensive Care. Preoperative assessment of adults undergoing elective noncardiac surgery: updated guideline. 2024.
- Henssler J, Schmidt Y, Schmidt U, Schwarzer G, Bschor T, Baethge C. Incidence of antidepressant discontinuation symptoms: a systematic review and meta-analysis. Lancet Psychiatry. 2024.
- Legg LA, Rudberg AS, Hua X, et al. Selective serotonin reuptake inhibitors (SSRIs) for stroke recovery. Cochrane Database Syst Rev. 2021.
- EFFECTS Trial Collaboration. Safety and efficacy of fluoxetine on functional recovery after acute stroke (EFFECTS): a randomised, double-blind, placebo-controlled trial. Lancet Neurol. 2020.
- American Geriatrics Society. 2023 AGS Beers Criteria for potentially inappropriate medication use in older adults. J Am Geriatr Soc. 2023.
- O'Mahony D, Cherubini A, Guiteras AR, et al. STOPP/START criteria for potentially inappropriate prescribing in older people: version 3. Eur Geriatr Med. 2023.
- Bixby AL, VandenBerg A, Bostwick JR. Clinical management of bleeding risk with antidepressants. Ann Pharmacother. 2019. (Narrative review.)
- McFarland D, Merchant D, Khandai A, et al. Selective serotonin reuptake inhibitor (SSRI) bleeding risk: considerations for the consult-liaison psychiatrist. Curr Psychiatry Rep. 2023. (Narrative review.)
- Na KS, Jung HY, Cho SJ, Cho SE. Can we recommend mirtazapine and bupropion for patients at risk for bleeding? A systematic review and meta-analysis. J Affect Disord. 2018;225:221–226.
- Laine L, Barkun AN, Saltzman JR, Martel M, Leontiadis GI. ACG clinical guideline: upper gastrointestinal and ulcer bleeding. Am J Gastroenterol. 2021. Together with the ACG-CAG guideline on anticoagulants and antiplatelets during acute gastrointestinal bleeding (2022) and the ACG guideline for prevention of NSAID-related ulcer complications (2009).
- Gutiérrez-Barbero M, et al. European Society of Gastrointestinal Endoscopy guideline: endoscopic diagnosis and management of nonvariceal upper gastrointestinal hemorrhage, update 2021. Endoscopy. 2021.
- Siau K, Hearnshaw S, Stanley AJ, et al. British Society of Gastroenterology-led multisociety consensus care bundle for the early clinical management of acute upper gastrointestinal bleeding. Frontline Gastroenterol. 2020.
- Pisters R, Lane DA, Nieuwlaat R, de Vos CB, Crijns HJ, Lip GY. A novel user-friendly score (HAS-BLED) to assess 1-year risk of major bleeding in patients with atrial fibrillation. Chest. 2010.
- Comparative review of risk scores for upper gastrointestinal bleeding, covering the Glasgow-Blatchford, AIMS-65, ABC, MAP(ASH), CANUKA, Rockall, CSMCPI, PNED and Forrest instruments. 2023.
- Cheng YL, Hu HY, Lin XH, et al. Use of SSRI, but not SNRI, increased upper and lower gastrointestinal bleeding: a nationwide population-based cohort study in Taiwan. Medicine. 2015.
- Rahman AA, He N, Rej S, Platt RW, Renoux C. Concomitant use of selective serotonin reuptake inhibitors and oral anticoagulants and risk of major bleeding: a systematic review and meta-analysis. Thromb Haemost. 2023.
- Montastruc JL, Bura-Rivière A. Association of antidepressants plus antithrombotics and bleeding risk: a pharmacovigilance study. Eur J Clin Pharmacol. 2023. (Spontaneous reporting database; reporting odds ratios are not risk estimates.)
- Spina E, Barbieri MA, Cicala G, de Leon J. Clinically relevant drug interactions between newer antidepressants and oral anticoagulants. Expert Opin Drug Metab Toxicol. 2020.
- Ghafarian AM, Nancoo N, Ghafarian H, et al. Perioperative selective serotonin reuptake inhibitor and serotonin-norepinephrine reuptake inhibitor use and 30-day bleeding after elective colectomy. Am J Surg. 2026. (Propensity-matched cohort.)
- Singh I, Achuthan S, Chakrabarti A, et al. Influence of pre-operative use of serotonergic antidepressants on the risk of bleeding in patients undergoing different surgical interventions: a meta-analysis. Pharmacoepidemiol Drug Saf. 2015.
- Gomez-Lumbreras A, Tawfik AG, Del Fiol G, et al. Risk of gastrointestinal bleeding by specific SSRIs and SNRIs: a systematic review and meta-analysis. Br J Clin Pharmacol. 2026.
- Kaye AD, Cooper HD, Mashaw SA, et al. Clinical implications of antidepressants and associated risk of bleeding: a narrative review. Curr Pain Headache Rep. 2025. (Source of the serotonin-transporter affinity tiers discussed in section 7.)
- Coupland C, Hill T, Morriss R, et al. Antidepressant use and risk of adverse outcomes in people aged 20–64 years: cohort study using a primary care database. BMC Med. 2018.
- Zhou X, Xiang S, Xu B, et al. A scientific approach to hemorrhage risk assessment of SSRIs/SNRIs utilizing the FAERS database. Psychiatry Res. 2025. (Pharmacovigilance disproportionality analysis.)
- Vakil N. Editorial: selective serotonin receptor inhibitors and gastrointestinal bleeding — managing the risk. Aliment Pharmacol Ther. 2017. (Editorial, reporting a cohort study by Laursen and colleagues.)
- Liu L, Fuller M, Behymer TP, et al. Selective serotonin reuptake inhibitors and intracerebral hemorrhage risk and outcome. Stroke. 2020. (Post hoc analysis of the ERICH case-control study.)
- Robbins-Welty GA, Fiedorowicz JG, Gensler L, et al. Antidepressants and bleeding risk: expert consensus from the Association of Medicine and Psychiatry. J Psychosom Res. 2026. (Expert consensus statement with a clinical decision algorithm. Cited here for its existence and scope; its algorithm is not reproduced.)
- Engler J, Filliter C, Montastruc F, et al. Risk of abnormal uterine bleeding associated with high-affinity compared with low-affinity serotonin and norepinephrine reuptake inhibitors. J Affect Disord. 2024.
- Zhuo C, Chen G, Lin C, et al. Risk-to-benefit ratios of consecutive antidepressants for heavy menstrual bleeding in young women with bipolar disorder or major depressive disorder. Front Psychiatry. 2022.
- Janjua AU, Dhingra AL, Greenberg R, McDonald WM. The efficacy and safety of concomitant psychotropic medication and electroconvulsive therapy. CNS Drugs. 2020. (Narrative review.)
- Bakker S, Burggraaf JLI, Kruip MJHA, et al. Selective serotonin reuptake inhibitor use and risk of major bleeding during treatment with vitamin K antagonists: results of a cohort study. Thromb Haemost. 2023.
- Kubiszewski P, Sugita L, Kourkoulis C, et al. Association of selective serotonin reuptake inhibitor use after intracerebral hemorrhage with hemorrhage recurrence and depression severity. JAMA Neurol. 2020. (Single-centre observational cohort, 1,279 survivors, competing-risks analysis.)
- Schmidt LB, Goertz S, Wohlfahrt J, Melbye M, Munch TN. Recurrent intracerebral hemorrhage: associations with comorbidities and medicine with antithrombotic effects. PLoS One. 2016;11(11):e0166223. (Danish nationwide administrative cohort; the contrasting null result.)
- Greenberg SM, Ziai WC, Cordonnier C, et al. 2022 guideline for the management of patients with spontaneous intracerebral hemorrhage: a guideline from the American Heart Association/American Stroke Association. Stroke. 2022. (Quoted here only for its take-home statement on fluoxetine and functional recovery, which was reproduced verbatim from the society's own published document.)
Responses