Paroxetine: Independent Evidence on Uses, Side Effects & Withdrawal

Table of contents

• Evidence summary
• What paroxetine is
• How it works
• What it is prescribed for
• What works and what does not
• Benefits by claim
• Risks and all side effects
• All interactions
• Who should avoid paroxetine
• Dosage and how to take it
• Follow the money: who makes it and who funded the evidence
• Related research
• Frequently asked questions
• Sources and funding notes

Evidence summary

Independent evidence and credibility scorecard

Independence tier: 1 means no money from paroxetine's sellers, 4 means seller-funded. Credibility grade runs from A (highest) to D.

What paroxetine is

Paroxetine is a selective serotonin reuptake inhibitor (SSRI) antidepressant. It is available only on prescription in both the UK and the US (NHS). The UK brand is Seroxat. Its UK marketing authorisation holder is SmithKline Beecham Limited, trading as GlaxoSmithKline UK, in London. It was first authorised on 11 December 1990, and the current text was revised on 4 April 2025 (Seroxat SmPC).

In the US, the brand is Paxil. FDA first approved it (NDA 020031) on 29 December 1992 as a new molecular entity, and the application is now held by Apotex (Drugs@FDA). The current US labels for Paxil and the extended-release Paxil CR list Apotex Corp as the labeller (DailyMed Paxil, DailyMed Paxil CR). A low-dose 7.5 mg paroxetine product, Brisdelle, was approved in 2013 for menopausal hot flushes only (Brisdelle label).

The molecule was first developed in the 1970s by the Danish company Ferrosan. Beecham, now part of GSK, bought the rights in 1980 and developed the hydrochloride salt used in Paxil and Seroxat (Bučar, Lancaster & Bernstein, Angew Chem 2015). Paroxetine has long been off patent. Openfda lists 14 generic (ANDA) approvals in the US and dozens of generic and repackager labels on DailyMed (openFDA Drugs@FDA).

How it works

Paroxetine is a potent, selective blocker of serotonin (5-HT) reuptake into nerve cells. The US label says its mechanism in depression and anxiety "is unknown, but is presumed to be linked to potentiation of serotonergic activity" (FDA label, 12.1). That wording matters because the idea that depression is simply "low serotonin" is itself disputed. An umbrella review found no consistent evidence for it, and 35 researchers rebutted that review. Neither side's paper tests whether the drugs work (Moncrieff et al. 2022, Jauhar et al. 2023).

Three features of paroxetine's chemistry explain most of what makes it different from other SSRIs:

• Short half-life and non-linear kinetics. The mean elimination half-life is about 21 hours. One of the enzymes that clears paroxetine (CYP2D6) becomes saturated at clinical doses, so blood levels rise more than in proportion to the dose. At steady state, drug exposure was about 8 times what single-dose data predicted (FDA label, 12.3). The reverse is thought to happen when the dose is cut, which fits with withdrawal being worse than with long-acting fluoxetine. A randomised interruption study found more discontinuation symptoms with paroxetine than with fluoxetine (Rosenbaum et al., Biol Psychiatry 1998).
• Strong, irreversible CYP2D6 inhibition. Paroxetine raises blood levels of other drugs that CYP2D6 clears, such as pimozide, thioridazine, metoprolol and some tricyclics. It also blocks the conversion of tamoxifen into its active form, endoxifen (SmPC 4.5).
• Anticholinergic activity. The UK SmPC describes paroxetine as having "low affinity for muscarinic cholinergic receptors" and "only weak anticholinergic properties". The American Geriatrics Society Beers Criteria nevertheless lists it alongside tricyclics as an antidepressant with "strong anticholinergic activity" that older adults should avoid (AGS Beers 2023). The two sources disagree, and we report both.

Steady state is reached in 7 to 14 days. Paroxetine is about 95% protein-bound and is cleared almost entirely by the liver (SmPC 5.2).

What it is prescribed for

What works and what does not

Benefits by claim

Depression: real but modest, and better than some SSRIs in head-to-head trials

The largest independent synthesis is Cipriani and colleagues' 2018 network meta-analysis: 522 double-blind trials, 116,477 adults, 21 antidepressants. Every drug was more effective than placebo, with response odds ratios ranging from 2.13 (95% CrI 1.89–2.41) for amitriptyline down to 1.37 (1.16–1.63) for reboxetine. In head-to-head trials, "agomelatine, amitriptyline, escitalopram, mirtazapine, paroxetine, venlafaxine, and vortioxetine were more effective than other antidepressants (range of ORs 1·19–1·96)". The authors also note that escitalopram, mirtazapine, paroxetine, agomelatine and sertraline "had a relatively higher response and lower dropout rate than the other antidepressants". Paroxetine was one of the drugs for which they obtained a "considerable amount of unpublished data", which lowers the risk of publication bias for this drug in particular (Cipriani et al., Lancet 2018, full text). The certainty of evidence was "moderate to very low". We could not extract paroxetine's own odds ratio against placebo from the text, because it appears only in a figure.

How big is the benefit in practice? Across antidepressants in general, FDA reviewers analysing patient-level data from 232 trials (73,388 participants) found a mean difference of 1.75 points on the 52-point HAM-D scale (95% CI 1.63–1.86). They estimated that "about 15% of participants have a substantial antidepressant effect beyond a placebo effect" (Stone et al., BMJ 2022). An independent reanalysis of Cipriani's data put the difference at 1.97 points and judged the certainty "very low" (Munkholm et al., BMJ Open 2019). Using FDA files, Turner and colleagues showed that published antidepressant literature overstated effect sizes by 32% overall, because negative trials went unpublished or were written up as positive (Turner et al., NEJM 2008). So paroxetine's advantage is real on average, but modest, and many people improve on placebo too.

Relapse prevention

In a randomised withdrawal study, adults who had responded to paroxetine were switched either to more paroxetine or to placebo. Relapse was 15% on paroxetine vs 39% on placebo (FDA label 14.1). The UK SmPC describes a 52-week study with relapse of 12% vs 28% (SmPC 5.1). These designs cannot fully tell relapse apart from withdrawal after abrupt switching to placebo. For paroxetine, which has the most withdrawal symptoms of the SSRIs, that limitation is especially relevant (Cochrane, Van Leeuwen 2021).

Anxiety disorders

For panic disorder, social anxiety disorder, OCD, GAD and PTSD, the efficacy data come almost entirely from the manufacturer's 8- to 12-week registration trials, as summarised in the regulator-approved label. In the panic trials, 76% of patients on 40 mg were panic-free at the end vs 44% on placebo. In a flexible-dose trial the figures were 51% vs 32%. In social anxiety disorder, the proportion rated much or very much improved was 69% vs 29% and 77% vs 42% in two trials, with "no indication… of any additional benefit for doses higher than 20 mg daily". In OCD, 40 mg and 60 mg cut YBOCS scores by about 6–7 points vs about 3 on placebo, but 20 mg did not. One of three GAD trials and one of three PTSD trials failed on at least one primary outcome (FDA label, section 14).

The independent check for GAD is Slee and colleagues' 89-trial network meta-analysis (25,441 patients, no funding). It found duloxetine, pregabalin, venlafaxine and escitalopram effective with relatively good acceptability, while "paroxetine and benzodiazepines were effective but also poorly tolerated when compared with placebo" (Slee et al., Lancet 2019). This matches why NICE puts sertraline, escitalopram or psychological therapy ahead of paroxetine for anxiety.

Placebo response and what "works" means

Most of paroxetine's licences rest on showing statistical superiority over placebo on rating scales in short trials. That is evidence that it works on average, not a promise for any one person. Placebo response in these trials is large; in the social anxiety trials, 29–42% of people on placebo were rated much improved. NICE's GAD guideline states there is "no evidence that either mode of treatment (individual high-intensity psychological intervention or drug treatment) is better" (NICE CG113).

Risks and all side effects

Boxed warning (US)

The FDA boxed warning says: antidepressants "increased the risk of suicidal thoughts and behaviors in pediatric and young adult patients in short-term studies". In pooled trials, per 1,000 patients treated compared with placebo, there were 14 additional cases in under-18s and 5 additional in 18–24-year-olds, 1 fewer case at ages 25–64 and 6 fewer at 65 and over. "PAXIL is not approved for use in pediatric patients" (FDA label, 5.1).

A paroxetine-specific adult analysis in the UK SmPC found suicidal behaviour in 2.19% of 18–24-year-olds on paroxetine vs 0.92% on placebo. In adults of all ages with major depression, the figures were 0.32% vs 0.05%. All of these events were suicide attempts, and 8 of the 11 on paroxetine were in younger adults (SmPC 5.1). The SmPC also warns about akathisia, an inner restlessness that makes it hard to sit still, which is most likely in the first few weeks, and adds that "increasing the dose may be detrimental".

Common side effects

In 6-week US depression trials, the effects seen more often on paroxetine than on placebo included:

• nausea: 26% vs 9%
• somnolence (drowsiness): 23% vs 9%
• dry mouth: 18% vs 12%
• asthenia (weakness): 15% vs 6%
• constipation: 14% vs 9%
• dizziness and insomnia: 13% each vs 6%
• diarrhoea: 12% vs 8%
• sweating: 11% vs 2%
• tremor: 8% vs 2%
• ejaculatory disturbance: 13% vs 0% (corrected for gender)
• other male genital disorders: 10% vs 0%

Twenty per cent of paroxetine-treated patients in the depression trials stopped because of an adverse reaction (FDA label, 6.1). The NHS lists nausea and vomiting, sexual problems, sleepiness, dizziness, weakness or agitation, sleep problems, headaches, weight changes, diarrhoea or constipation, and blurred vision as common. It says not to drive or cycle if affected (NHS).

Withdrawal: why paroxetine stands out

Withdrawal symptoms are not the same as addiction; the SmPC says they are "not the same as the drug being addictive or dependence producing". They are nonetheless common and can be disabling. In paroxetine trials, adverse events on stopping occurred in 30% of patients vs 20% on placebo. Typical symptoms include dizziness, sensory disturbances such as electric-shock sensations and tinnitus, intense dreams, agitation, nausea, tremor, sweating, headache, palpitations and emotional instability. They usually start within days, are "generally… self-limiting and usually resolve within two weeks, though in some individuals they may be prolonged (two-three months or more)" (SmPC 4.4).

Across all antidepressants, the independent Henssler meta-analysis (79 studies, 21,002 patients) found at least one discontinuation symptom in 31% of people stopping an antidepressant vs 17% stopping placebo. That gives a placebo-adjusted incidence of about 15%, "one in six to seven patients". Severe symptoms affected 2.8% vs 0.6%. Imipramine, paroxetine and desvenlafaxine or venlafaxine "were associated with a higher severity of symptoms" (Henssler et al. 2024). A survey-heavy review by withdrawal researchers estimated a much higher average incidence of 56% (Davies & Read 2019). The true figure depends on how symptoms are measured. In a randomised 5–8-day interruption study, patients stopping paroxetine had more new symptoms than those on sertraline and far more than those on fluoxetine (Rosenbaum et al. 1998). The SmPC even notes "very rare reports" of symptoms after a single missed dose.

All interactions

Who should avoid paroxetine

• Anyone taking MAOIs, thioridazine or pimozide, or with known hypersensitivity to paroxetine. These are formal contraindications (SmPC 4.3).
• Children and adolescents under 18. UK: "should not be used", because trials showed more suicidal behaviour and hostility without adequate efficacy. US: not approved (SmPC 4.2).
• People planning pregnancy or in early pregnancy. The US label says paroxetine "should be initiated only after consideration of the other available treatment options" in women who intend to become pregnant or are in their first trimester. In 2005 the FDA changed paroxetine's pregnancy category from C to D after two unpublished registry analyses. One found about a 2% cardiac-defect risk with paroxetine vs 1% in the whole registry (FDA Public Health Advisory, 8 Dec 2005).
  • A 2016 meta-analysis of 23 studies found pooled ORs of 1.28 (95% CI 1.11–1.47) for major cardiac malformations and 2.38 (1.14–4.97) for atrial septal defects. Its lead author declares that she consults for plaintiffs in antidepressant birth-defect litigation (Bérard et al. 2016, full text).
  • A publicly funded US Medicaid cohort of 949,504 pregnancies found no significant association between paroxetine and right ventricular outflow tract obstruction (RR 1.07, 95% CI 0.59–1.93) once depression was taken into account (Huybrechts et al. 2014).
  • Stopping suddenly in pregnancy should be avoided. Untreated depression carries its own risks: the US label cites a study in which women who stopped antidepressants were more likely to relapse. Late-pregnancy SSRI exposure is linked to neonatal adaptation problems and persistent pulmonary hypertension of the newborn. The NHS says your doctor will discuss risks and benefits and may advise giving birth in hospital (NHS).
• Breastfeeding: generally considered compatible. Relative infant doses are 0.4–2.2%. Watch the baby for agitation, poor feeding or poor weight gain (FDA label 8.2, NHS).
• Women taking tamoxifen, including those who might otherwise use paroxetine for hot flushes (Kelly 2010).
• Older adults (65+). The Beers Criteria say to avoid it as a strongly anticholinergic antidepressant, with high-quality evidence and a strong recommendation (AGS Beers 2023). If it is used, the maximum is 40 mg, and the risks of hyponatraemia, bleeding and falls are higher (SmPC).
• Bipolar disorder or a history of mania, unless a specialist is managing treatment. Epilepsy, narrow-angle glaucoma, bleeding disorders, diabetes, and heart disease or long QT all need caution (NHS, SmPC 4.4).
• Liver or kidney impairment: blood levels rise. In the US, severe impairment means starting at 10 mg with a maximum of 40 mg. In the UK, doses are kept at the lower end of the range (FDA label 2.5).

Dosage and how to take it

Sources: Seroxat SmPC 4.2, FDA Paxil label, section 2. Paxil CR (extended release) uses different strengths: 25–62.5 mg for depression and 12.5–25 mg for PMDD. Brisdelle is 7.5 mg at bedtime for hot flushes only. These products are not interchangeable milligram for milligram (Paxil CR label).

• How to take it: once a day in the morning. The UK advises taking it with food to reduce nausea, and swallowing the tablet rather than chewing it (NHS, SmPC). If you miss a dose, take it when you remember unless your next dose is nearly due. Never double up (NHS).
• How long before it works: the SmPC says improvement "starts after one week but may only become evident from the second week". The NHS says it "can take several weeks". The dose is usually reviewed within 3–4 weeks (SmPC).
• How long to take it: for depression, at least 6 months after symptoms have gone (SmPC, NICE NG222). For GAD, NICE advises continuing for at least a year if the drug is working, because relapse risk is high (CG113, 1.2.33).
• How to stop: only with your prescriber, and gradually. The SmPC describes the trial regimen of reducing by 10 mg a week, then returning to the previous dose and tapering more slowly if symptoms are intolerable. NICE NG222 (2022) recommends step-wise reductions of a proportion of the previous dose, for example 50%, then smaller steps such as 25% as the dose gets lower. It suggests liquid preparations if available for very small doses, a pace "led by and agreed with the person", and it recognises that "withdrawal may take weeks or months" (NICE NG222, 1.4.16). The US oral suspension is listed as "not currently marketed" (FDA label).
• Overdose: taking more than prescribed can be dangerous. In the UK, call NHS 111 (NHS). The US label lists delayed seizures and QRS/QTc prolongation among the effects of overdose.

Follow the money: who makes it and who funded the evidence

Who makes and sells it

• Originator: Ferrosan, a Danish company, developed paroxetine in the 1970s. Beecham bought the rights in 1980 and later became part of GlaxoSmithKline (GSK, London). The source for this history is an academic chemistry review whose co-author Joel Bernstein is quoted in it answering questions in the paroxetine patent litigation (Bučar et al. 2015).
• UK: Seroxat's marketing authorisation holder is SmithKline Beecham Limited / GlaxoSmithKline UK, London (SmPC).
• US: the Paxil and Paxil CR NDAs and labels are now held by Apotex (Drugs@FDA). We could not verify when or on what terms the US brand passed from GSK to Apotex. Brisdelle's current US labeller is Legacy Pharma USA.
• Generics: paroxetine is a widely genericised, low-cost medicine. OpenFDA lists 14 US ANDA approvals, and many manufacturers and repackagers appear on DailyMed, among them Aurobindo, Zydus, Mylan, Lupin, Solco and Alembic. We found no current, primary-source brand revenue figure.

Who funded the evidence

• The licensing trials for every indication were sponsored by the manufacturer (SmithKline Beecham, later GSK). Their results reach patients mainly through the label, which the manufacturer writes and the regulator approves. The FDA's human-drug review process is itself about 77% funded by industry user fees (FDA PDUFA FY2025 report).
• The independent syntheses this article leads with had no seller funding. Cipriani 2018 was funded by the NIHR and JSPS, although some co-authors declared lecture or consulting fees from antidepressant makers. Henssler 2024 and Slee 2019 had no funding. Huybrechts 2014 was funded by AHRQ and NIH, Bérard 2016 by Quebec public funds, and Le Noury 2015 had no grant.
• Conflicts that point against the drug are disclosed too. Two Le Noury authors worked for plaintiffs in paroxetine litigation, and Bérard consults for plaintiffs in antidepressant birth-defect litigation. We report these for the same reason we report industry fees: they are incentives, not proof of error.

Documented integrity events

• Study 329. This SmithKline Beecham trial of paroxetine and imipramine in 275 adolescents (1994–1998) was published in 2001 by Keller and colleagues as supporting paroxetine. The independent RIAT team reanalysed the full clinical study report. They describe the 2001 article as "largely ghostwritten" and say it "claimed efficacy and safety for paroxetine that was at odds with the data". Their reanalysis found no statistically or clinically significant benefit on any prespecified primary or secondary outcome. Suicidal and self-injurious behaviour was higher on paroxetine: 11 of 93 patients vs 2 of 87 on placebo. The 2001 paper had reported 5 vs 1. In correspondence, GSK stated that it did "not agree that the article is false, fraudulent or misleading" (Le Noury et al., BMJ 2015).
• 2012 US guilty plea. GSK agreed to plead guilty to three criminal counts, two of them for misbranding Paxil and Wellbutrin, and to pay US$3 billion in total. That comprised US$1 billion criminal (of which US$757,387,200 related to the Paxil and Wellbutrin misbranding offences) and US$2 billion civil. The government alleged that from April 1998 to August 2003 GSK promoted Paxil for depression in under-18s. It said GSK helped prepare and distribute "a misleading medical journal article that misreported that a clinical trial of Paxil demonstrated efficacy" in under-18s, and did not make available data from two other failed paediatric studies. GSK also entered a corporate integrity agreement (DOJ press release, 2 July 2012). The DOJ release does not name the article; the civil claims were settled, not adjudicated.
• Publication bias across antidepressants. Using FDA files, Turner and colleagues found that the published literature made 94% of trials look positive, whereas the FDA judged 51% positive (Turner et al., NEJM 2008).

Taken together: the evidence that paroxetine helps adults in the short term survives independent reanalysis. The evidence that it helped adolescents did not. Its distinctive harms, namely withdrawal, the pregnancy signal, the tamoxifen interaction and anticholinergic burden in older people, are documented mostly by regulators and publicly funded researchers rather than by the manufacturer's own publications.

Related research

• Stress, anxiety and depression: prevention and management guide
• Supplements for stress, anxiety and depression: the evidence
• St John's wort (do not combine with paroxetine)
• Saffron for depression
• Ashwagandha: evidence and safety
• L-theanine and magnesium
• Sleep: prevention and management guide and melatonin
• Other antidepressants: sertraline, fluoxetine, escitalopram, citalopram, venlafaxine, mirtazapine

Frequently asked questions

Sources and funding notes

• NHS, Paroxetine (page last reviewed 23 July 2026). UK government health service; no manufacturer funding.
• Seroxat 20 mg tablets, Summary of Product Characteristics (revised 4 April 2025). Written by the licence holder, SmithKline Beecham Ltd / GSK UK (London), and approved by the MHRA. Manufacturer document, regulator-controlled.
• FDA prescribing information, Paxil (Apotex Corp), via DailyMed. Manufacturer-written, FDA-approved. The FDA's drug review is about 77% funded by industry user fees.
• FDA prescribing information, Paxil CR (Apotex Corp). As above.
• FDA prescribing information, Brisdelle (Legacy Pharma USA). As above.
• Drugs@FDA, NDA 020031 (Paxil), approved 29 Dec 1992. US regulator record; generic counts from openFDA.
• FDA Public Health Advisory: Paroxetine, 8 December 2005 (archived). US regulator; based on then-unpublished manufacturer and registry analyses.
• NICE NG222, Depression in adults (2022). UK government (DHSC) funded; committee declarations not checked.
• NICE CG113, Generalised anxiety disorder and panic disorder in adults (2011, updated 2020). UK government funded.
• NICE CG159, Social anxiety disorder (2013). UK government funded.
• NICE CG31, OCD and body dysmorphic disorder (2005). UK government funded; older guideline.
• NICE NG116, Post-traumatic stress disorder (2018). UK government funded.
• WHO mhGAP guideline, 3rd edition (2023). WHO; panel interests declared, mostly none.
• Lam et al., CANMAT 2023 update, Can J Psychiatry 2024. Canada; internal funds, no industry funding to the organisation in 2019–23; many authors declare industry fees.
• Cipriani et al., Lancet 2018 (full text). UK/Japan; funded by NIHR Oxford Health BRC and JSPS; 78% of included trials industry-funded; some co-authors had industry fees.
• Stone et al., BMJ 2022. USA; FDA staff analysis; no external funding; one author reported unrelated industry fees.
• Munkholm, Paludan-Müller & Boesen, BMJ Open 2019. Denmark; none declared.
• Turner et al., NEJM 2008. USA (VA Portland / OHSU); funding and conflicts not verified.
• Slee et al., Lancet 2019. UK (UCL); "No funding was received"; author conflicts not verified.
• Le Noury et al., BMJ 2015 (RIAT reanalysis of Study 329). UK/Australia/Canada/USA; no specific grant; two authors declared paid work for plaintiffs in paroxetine litigation.
• US Department of Justice, GSK settlement press release, 2 July 2012 (archived). US government.
• Henssler et al., Lancet Psychiatry 2024. Germany; funding "None"; no competing interests.
• Davies & Read, Addictive Behaviors 2019. UK; funding not shown in the abstract; the authors are linked to prescribed-drug-dependence advocacy groups.
• Rosenbaum et al., Biol Psychiatry 1998. USA (Massachusetts General Hospital); funding not shown in the PubMed record, not verified.
• Van Leeuwen et al., Cochrane 2021 (approaches to stopping antidepressants). Cochrane; no commercial sponsorship permitted.
• Fava et al., J Clin Psychiatry 2000. USA; funding not shown in the PubMed record, not verified.
• Kelly et al., BMJ 2010. Canada; academic population cohort; one author disclosed consulting for multiple companies including GSK.
• Bérard et al., Br J Clin Pharmacol 2016 (full text). Canada; funded by FRQ-S and RQRM; lead author is a consultant for plaintiffs in antidepressant birth-defect litigation.
• Huybrechts et al., NEJM 2014. USA; funded by AHRQ and NIH; author disclosures not fetched.
• American Geriatrics Society 2023 Beers Criteria. USA; professional society; panel conflicts were minor and disclosed.
• EMA PRAC, new product information wording, May 2019. EU regulator; about 91.5% fee-funded.
• Moncrieff et al., Molecular Psychiatry 2022 and Jauhar et al., Molecular Psychiatry 2023. The serotonin-hypothesis debate; authors on both sides declare interests pointing in opposite directions.
• Bučar, Lancaster & Bernstein, Angewandte Chemie 2015. Academic chemistry review (history of paroxetine's development); a co-author appears as a witness in the patent litigation it describes.
• FDA PDUFA financial report FY2025. US regulator; source for the share of FDA drug-review costs paid by industry fees.

Last reviewed: 26 September 2026. Educational information from Pure City Research, not medical advice. Medicine decisions belong with a qualified prescriber.