Skip to content

SSRI and SNRI pharmacotherapy

TL;DR — Antidepressants beat placebo in GAD with high-certainty evidence and a modest effect: response (≥50% HAM-A reduction) RR 1.41 (95% CI 1.29–1.55), NNTB 7 (5–9) across 20 studies and 7,267 participants; acceptability equals placebo (dropout RR 1.03, 0.93–1.14); fewer dropouts for lack of efficacy (RR 0.41, NNTB 27) and more for adverse effects (RR 2.18, 1.81–2.61, NNTH 17) (Kopcalic 2025, PMID 39880377). The reference ranking across 89 trials and 25,441 patients puts duloxetine (HAM-A MD −3.13, 95% CrI −4.13 to −2.13), pregabalin (−2.79), venlafaxine (−2.69) and escitalopram (−2.45) as efficacious with relatively good acceptability; quetiapine has the largest effect (−3.60) but poor tolerability (discontinuation OR 1.44, 1.16–1.80), as do paroxetine and benzodiazepines (Slee 2019, PMID 30712879). Class-level effects from the psychological/pharmacological network: SSRIs SMD −0.67 (95% CrI −0.90 to −0.43), SNRIs −0.54 (−0.79 to −0.30) (Chen 2019, PMID 31494377) — a network that allowed all comorbidities. Continuation treatment prevents relapse convincingly: 56% vs 19% relapse on placebo vs escitalopram over 24–76 weeks (Allgulander 2006, PMID 16316482); 41.8% vs 13.7% for duloxetine over 26 weeks (Davidson 2008, PMID 18559291). The unresolved issue is not whether these drugs work but whether a 2.5–3.1-point HAM-A difference is the right unit of clinical meaning (screening and measurement).

What the class-level evidence says

Question Answer Certainty Source
Do antidepressants beat placebo for response? RR 1.41 (1.29–1.55), NNTB 7 (5–9); 20 studies, 7,267 participants High Kopcalic 2025, PMID 39880377
Are they acceptable? All-cause dropout RR 1.03 (0.93–1.14); 33 studies, 11,294 participants High Kopcalic 2025, PMID 39880377
Do they prevent efficacy-related dropout? RR 0.41 (0.33–0.50), NNTB 27 (24–32) High Kopcalic 2025, PMID 39880377
Do they cause harm-related dropout? RR 2.18 (1.81–2.61), NNTH 17 (13–112) High Kopcalic 2025, PMID 39880377
Does the class matter? "Similar findings when classes of antidepressants were compared with placebo", certainty high High Kopcalic 2025, PMID 39880377

Trial duration in that review ranged from 4 to 28 weeks, participants had moderate-to-severe GAD without serious medical comorbidity, studies with regular benzodiazepine use were excluded, and few studies included secondary psychiatric comorbidity — the review's own listed gap (Kopcalic 2025, PMID 39880377). That exclusion profile is the reason "what works in GAD with depression" remains open (comorbidity and primary care).

Drug-level ranking

From the 89-trial, 25,441-patient network meta-analysis; primary efficacy outcome is mean difference in HAM-A change, acceptability is all-cause discontinuation (Slee 2019, PMID 30712879):

Drug HAM-A mean difference vs placebo (95% CrI) Acceptability Note
Quetiapine −3.60 (−4.83 to −2.39) — largest effect Poor (discontinuation OR 1.44, 1.16–1.80) See pregabalin, benzodiazepines and others
Duloxetine −3.13 (−4.13 to −2.13) Relatively good SNRI
Pregabalin −2.79 (−3.69 to −1.91) Relatively good Not an antidepressant
Venlafaxine −2.69 (−3.50 to −1.89) Relatively good SNRI
Escitalopram −2.45 (−3.27 to −1.63) Relatively good SSRI
Mirtazapine, sertraline, fluoxetine, buspirone, agomelatine Efficacious and well tolerated — Findings limited by small sample sizes
Paroxetine Effective Poorly tolerated SSRI
Benzodiazepines Effective Poorly tolerated See pregabalin, benzodiazepines and others

The authors' own conclusion is often overlooked and is clinically load-bearing: "The failure of initial pharmacological therapy might not be a reason to abandon a pharmacological treatment strategy" (Slee 2019, PMID 30712879) — i.e. the network supports sequential trials within class before switching modality (treatment-resistant GAD).

Two later networks broadly agree while adding comparators. The pharmacological-plus-psychological network ranked classes: NDRIs SMD −1.84 (95% CrI −3.05 to −0.62, single-drug class with wide interval), NaSSAs −0.91, melatonergic agonists −0.68, SSRIs −0.67 (−0.90 to −0.43), azapirones −0.58, anticonvulsants −0.56, SNRIs −0.54 (−0.79 to −0.30), benzodiazepines −0.40; serotonin modulators and second-generation antipsychotics did not beat placebo in that model; overall, most pharmacological interventions had larger effect sizes than psychological ones (Chen 2019, PMID 31494377). A 100-trial, 28,637-participant anxiolytic network placed clomipramine highest for efficacy and vortioxetine lowest, with clomipramine also producing the most discontinuations; only four treatments had fewer adverse events than placebo — diazepam, agomelatine, clobazam and silexan (Müller 2026, PMID 40788541). That last network mixes anxiety disorders and HAM-A-reporting populations, so its rankings are not GAD-specific.

The commentary layer disagrees with the enthusiasm. An accompanying Lancet comment on the Slee network cautioned about how such rankings should be read (Bandelow 2019, PMID 30712877), and an evidence-based nursing appraisal framed the same review as consolidating existing knowledge rather than changing it (Roles 2020, PMID 31138563).

Individual agents with GAD-specific trial evidence

Agent Design Result Source
Escitalopram 20 mg 491 open-label → 375 responders randomised to escitalopram (187) or placebo (188), 24–76 weeks Relapse 19% vs 56%; risk of relapse 4.04× higher on placebo (p<0.001); AE withdrawal 7% vs 8%; discontinuation symptoms on taper mainly dizziness (10–12%) Allgulander 2006, PMID 16316482
Duloxetine 60–120 mg 887 open-label 26 weeks → responders randomised, 26-week double-blind continuation Relapse 13.7% (n=204) vs 41.8% (n=201), p≤0.001, and worse on every outcome measure Davidson 2008, PMID 18559291
Duloxetine, responder analysis Continuous responder analysis from the long-term trial Informs randomised-withdrawal design assumptions Llorca 2009, PMID 19142122
Agomelatine 25–50 mg Responders to 16 weeks randomised to agomelatine (113) or placebo (114) for 26 weeks; 31 centres, 6 countries Relapse 19.5% vs 30.7%; log-rank p=0.046; superiority retained in the more severe subgroup (HARS ≥25, CGI-S ≥5); no discontinuation-symptom difference on withdrawal Stein 2012, PMID 22901350 (ISRCTN38094599)
Sertraline Multicentre placebo-controlled RCT in GAD GAD-specific SSRI trial Brawman-Mintzer 2006, PMID 16848646
Venlafaxine XR Randomised, double-blind, placebo-controlled, Japanese patients with GAD Regional replication Otsubo 2025, PMID 41123332
Vortioxetine Systematic review and meta-analysis for GAD; separate evidence synthesis Mixed/negative — vortioxetine ranked least efficacious in the 100-trial anxiolytic network Pae 2015, PMID 25851751; Meza 2021, PMID 34038400; Müller 2026, PMID 40788541
Duloxetine, safety Systematic review of efficacy, safety and tolerability Cross-indication safety profile Rodrigues-Amorim 2020, PMID 33192668

Dose, time course and the shape of response

Question Answer Source
Does more drug mean more benefit? Across 57 trials and 16,056 patients with anxiety disorders, higher SSRI doses within the therapeutic range give greater symptom improvement and higher response likelihood; higher SNRI doses do not. Higher doses of both classes increase dropout due to side effects Jakubovski 2019, PMID 30479005 — pooled anxiety disorders
What is the trajectory? For SNRIs a logarithmic model fits best (greatest incremental gain early); for SSRIs a linear model fits best (similar improvement across the acute phase) Jakubovski 2019, PMID 30479005
Does agent choice matter within class? No significant efficacy differences between agents within either class in that analysis Jakubovski 2019, PMID 30479005
Does the GAD-specific dose evidence agree? In a 681-patient fixed-dose trial, escitalopram 10 mg and 20 mg both beat placebo but 20 mg was not significantly superior to 10 mg, while more patients withdrew for adverse events at 20 mg Baldwin 2006, PMID 16946363

The escitalopram fixed-dose result is the cleanest GAD-specific dose-response datum available and it cuts against the pooled finding: doubling the dose bought tolerability cost without measurable efficacy gain (Baldwin 2006, PMID 16946363).

Trial-level detail for the first-line agents

Trial Design Result
Davidson 2004 (PMID 15274172) 315 outpatients, HAM-A ≥18, escitalopram 10–20 mg flexible vs placebo, 8 weeks after a 1-week placebo lead-in HAM-A change −11.3 vs −7.4 (LOCF, p<0.001); separation from week 1 onward; response 68% vs 41% (completers), 58% vs 38% (LOCF); AE discontinuation 8.9% vs 5.1% (p=0.27)
Baldwin 2006 (PMID 16946363) 681 patients; placebo vs escitalopram 5/10/20 mg vs paroxetine 20 mg, 12 weeks, fixed dose Escitalopram 10 and 20 mg beat placebo; escitalopram 10 mg was more efficacious than paroxetine 20 mg; more AE withdrawals on escitalopram 20 mg and on paroxetine
Bose 2008 (PMID 18050245) 392 randomised; escitalopram 10–20 mg vs venlafaxine XR 75–225 mg vs placebo, 8 weeks Venlafaxine XR separated from placebo on the primary LOCF analysis (−2.27, p=0.01); escitalopram did not (−1.52, p=0.09) — both separated on observed-cases analysis and on every secondary measure. AE discontinuation: escitalopram 7% vs placebo 5% (p=0.61); venlafaxine XR 13% vs placebo (p=0.03)
Rynn 2008 (PMID 17311303) Flexible-dose, progressive-titration, placebo-controlled duloxetine trial GAD registration-programme evidence for duloxetine
Strawn 2015 (PMID 25791145) 272 youth aged 7–17 with primary GAD; duloxetine 30–120 mg vs placebo, 10 weeks, then 18 weeks open label PARS-GAD improvement −9.7 vs −7.1 (p≤0.001, Cohen's d 0.5); response 59% vs 42%, remission 50% vs 34%, functional remission 37% vs 24% (all p≤0.05); pulse +6.5 vs +2.0 bpm and weight −0.1 vs +1.1 kg (p≤0.01)

Bose 2008 is worth dwelling on: the same trial reaches opposite conclusions about escitalopram depending on whether LOCF or observed-cases analysis is used (PMID 18050245). This is the acute-phase analogue of the older-adult problem in Lenze 2009 (PMID 19155456) — in a disorder with a placebo pre-post effect of d_av 1.23 (Bschor 2024, PMID 38809560), analysis convention decides drug–placebo separation more often than is comfortable.

Independent syntheses that predate and corroborate the reference network

Source Method Finding
Baldwin 2011 (PMID 21398351) Bayesian mixed-treatment meta-analysis, 46 RCTs identified / 27 analysed, nine drugs; outcomes response (≥50% HAM-A reduction), remission (HAM-A ≤7) and withdrawal for adverse events The first probabilistic ranking of GAD drugs, eight years before Slee 2019 (PMID 30712879), and using responder and remitter outcomes rather than mean HAM-A change
Li 2018 (PMID 29558528) 8 RCTs of duloxetine in adult GAD HADS-anxiety MD 2.32 (1.77–2.88) and HAM-A psychic anxiety MD 2.15 (1.61–2.68) favouring duloxetine; AE discontinuation OR 2.62 (1.35–5.06); any treatment-emergent AE OR 1.76 (1.36–2.28); serious AEs not different (OR 1.13, 0.52–2.47)
Zhang 2016 (PMID 26238298) Meta-analysis of published duloxetine RCTs in GAD Independent replication of the duloxetine signal
Mavranezouli 2013 (PMID 23512146) Decision-analytic model with two network meta-analyses (efficacy and tolerability), UK NHS/PSS perspective, 2011 prices, six first-line drugs plus no treatment Sertraline was best at limiting discontinuation for side effects, second best for response, lowest cost and highest QALYs; 75% probability of being most cost-effective at £20,000/QALY

Two things follow. The duloxetine effect is concentrated in psychic rather than somatic anxiety (Li 2018, PMID 29558528), which matters because GAD presents somatically in primary care (comorbidity and primary care). And the most cost-effective drug in a formal economic model — sertraline — is not the most efficacious in the network ranking (Mavranezouli 2013, PMID 23512146; Slee 2019, PMID 30712879), because tolerability and price dominate the model. Efficacy ranking, cost-effectiveness ranking and guideline position are three different orderings.

Relapse prevention: the strongest pharmacological result in GAD

Three independent randomised-withdrawal trials, three drug classes, one consistent answer:

Trial Drug Continuation period Relapse on drug Relapse on placebo
Allgulander 2006 (PMID 16316482) Escitalopram 20 mg 24–76 weeks 19% 56%
Davidson 2008 (PMID 18559291) Duloxetine 60–120 mg 26 weeks 13.7% 41.8%
Stein 2012 (PMID 22901350) Agomelatine 25–50 mg 26 weeks 19.5% 30.7%

The effect sizes here (risk of relapse 4× higher off drug in the escitalopram trial) are far larger than the acute-phase HAM-A differences, and this asymmetry is under-discussed: GAD's pharmacology looks unimpressive in 8-week acute trials and impressive in 6–18-month withdrawal designs. The design caveat is that randomised-withdrawal enriches for responders by construction, and part of the placebo-arm "relapse" may be discontinuation effects rather than illness return (course, relapse and long-term outcome).

Discontinuation

Two 2024–2025 meta-analyses reframed the discontinuation debate quantitatively, both in mixed populations rather than GAD alone:

  • Incidence. Across 79 studies and 21,002 patients, ≥1 discontinuation symptom occurred in 31% (95% CI 27–35) after stopping an antidepressant vs 17% (14–21) after stopping placebo; the RCT-based between-group difference was 8% (4–12), i.e. roughly one in six to seven patients experiences a drug-attributable discontinuation symptom. Severe symptoms: 2.8% (1.4–5.7) vs 0.6% (0.2–1.3). Desvenlafaxine, venlafaxine, imipramine and escitalopram had higher frequencies; imipramine, paroxetine and desvenlafaxine/venlafaxine higher severity (Henssler 2024, PMID 38851198).
  • Nature and severity. Across 50 studies and 17,828 participants, DESS scores at week 1 were higher after stopping antidepressants (SMD 0.31, 0.23–0.39; k=11, n=3,915)—about one additional symptom and below the review's clinically significant threshold. Dizziness dominated (OR 5.52, 3.81–8.01; risk difference 6.24%), followed by nausea, vertigo and nervousness. The meta-analysis found no association between discontinuation and mood worsening and interpreted later depressive presentation as more consistent with relapse; individual clinical attribution still requires a differential assessment (Kalfas 2025, PMID 40632531).

In GAD specifically, the escitalopram relapse-prevention trial reported low discontinuation-symptom incidence on tapered withdrawal, mainly dizziness 10–12% (Allgulander 2006, PMID 16316482), and the agomelatine trial reported no discontinuation-symptom difference versus continued treatment (Stein 2012, PMID 22901350). These are the only GAD-specific discontinuation data located in this build; the larger meta-analyses are dominated by depression trials.

Where this evidence is thin

  1. Comorbidity is excluded by design. The Cochrane review excluded serious medical comorbidity and regular benzodiazepine use and included few participants with secondary psychiatric comorbidity (Kopcalic 2025, PMID 39880377). The 91-trial psychological/pharmacological network went the opposite way and allowed all comorbidities (Chen 2019, PMID 31494377). Neither design answers the question clinicians actually face.
  2. Head-to-head drug-versus-therapy trials are rare. Rankings come from networks, not direct comparisons (cognitive behavioural therapy).
  3. Duration. Longest double-blind acute phase in the Cochrane review: 28 weeks (Kopcalic 2025, PMID 39880377) — against a disorder measured in decades (course, relapse and long-term outcome).
  4. Placebo response. The acute-phase drug–placebo gap of 2.5–3.1 HAM-A points is small relative to within-arm change; the placebo-response problem is treated on clinical trials landscape.

Open questions

  • Is NNTB 7 for response (Kopcalic 2025, PMID 39880377) compatible with an average HAM-A difference of 2.5–3.1 points (Slee 2019, PMID 30712879)? Responder and continuous analyses of the same literature imply different clinical stories, and no reconciliation has been published.
  • How much apparent relapse after discontinuation reflects recurrence versus withdrawal? Discontinuation symptoms are usually mild and short in pooled antidepressant evidence (Kalfas 2025, PMID 40632531). A targeted PubMed search rerun on 2026-09-02 located no GAD relapse-prevention reanalysis that adjudicated withdrawal symptoms separately; the proportion remains unknown.
  • Does drug choice matter, given that class comparisons versus placebo look "similar" (Kopcalic 2025, PMID 39880377) while the network separates duloxetine from paroxetine on tolerability (Slee 2019, PMID 30712879)?
  • What is the effect in GAD without comorbid depression, and in GAD with it? Neither stratum has a dedicated estimate (OPEN-QUESTIONS.md, OQ-2).

References

  1. Kopcalic K, et al. Antidepressants versus placebo for generalised anxiety disorder (GAD). Cochrane Database Syst Rev. 2025;1:CD012942. PMID 39880377
  2. Slee A, et al. Pharmacological treatments for generalised anxiety disorder: a systematic review and network meta-analysis. Lancet. 2019;393:768-777. PMID 30712879
  3. Bandelow B, et al. Network analyses to rank pharmacological treatments for generalised anxiety disorder. Lancet. 2019;393:721-722. PMID 30712877
  4. Roles S. The largest meta-analysis of pharmaceutical treatments of Generalised Anxiety Disorder consolidates current knowledge and reveals convincing effectiveness of Venlafaxine. Evid Based Nurs. 2020;23:74. PMID 31138563
  5. Chen TR, et al. Pharmacological and psychological interventions for generalized anxiety disorder in adults: A network meta-analysis. J Psychiatr Res. 2019;118:73-83. PMID 31494377
  6. Müller TJ, et al. Comparative efficacy and acceptability of anxiolytic drugs for the treatment of anxiety disorders: a systematic review and network meta-analysis. Eur Arch Psychiatry Clin Neurosci. 2026;276:1879-1894. PMID 40788541
  7. Allgulander C, et al. Prevention of relapse in generalized anxiety disorder by escitalopram treatment. Int J Neuropsychopharmacol. 2006;9:495-505. PMID 16316482
  8. Davidson JR, et al. Duloxetine treatment for relapse prevention in adults with generalized anxiety disorder: a double-blind placebo-controlled trial. Eur Neuropsychopharmacol. 2008;18:673-81. PMID 18559291
  9. Llorca PM, et al. A continuous responder analysis from a long-term trial of duloxetine treatment for generalized anxiety disorder: implications for the randomized withdrawal relapse prevention study design. J Clin Psychopharmacol. 2009;29:96-7. PMID 19142122
  10. Stein DJ, et al. Agomelatine prevents relapse in generalized anxiety disorder: a 6-month randomized, double-blind, placebo-controlled discontinuation study. J Clin Psychiatry. 2012;73:1002-8. PMID 22901350
  11. Brawman-Mintzer O, et al. Sertraline treatment for generalized anxiety disorder: a randomized, double-blind, placebo-controlled study. J Clin Psychiatry. 2006;67:874-81. PMID 16848646
  12. Otsubo T, et al. A randomized, double-blinded, placebo-controlled study to evaluate the efficacy and safety of venlafaxine extended release in Japanese patients with generalized anxiety disorder. Psychiatry Clin Neurosci. 2025;79:849-858. PMID 41123332
  13. Pae CU, et al. Vortioxetine, a multimodal antidepressant for generalized anxiety disorder: a systematic review and meta-analysis. J Psychiatr Res. 2015;64:88-98. PMID 25851751
  14. Meza N, et al. Vortioxetine for generalised anxiety disorder in adults. Medwave. 2021;21:e8172. PMID 34038400
  15. Rodrigues-Amorim D, et al. A Systematic Review of Efficacy, Safety, and Tolerability of Duloxetine. Front Psychiatry. 2020;11:554899. PMID 33192668
  16. Henssler J, et al. Incidence of antidepressant discontinuation symptoms: a systematic review and meta-analysis. Lancet Psychiatry. 2024;11:526-535. PMID 38851198
  17. Kalfas M, et al. Incidence and Nature of Antidepressant Discontinuation Symptoms: A Systematic Review and Meta-Analysis. JAMA Psychiatry. 2025;82:896-904. PMID 40632531
  18. Kapczinski F, et al. Antidepressants for generalized anxiety disorder. Cochrane Database Syst Rev. 2003:CD003592. PMID 12804478
  19. Jakubovski E, et al. Systematic review and meta-analysis: Dose-response curve of SSRIs and SNRIs in anxiety disorders. Depress Anxiety. 2019;36:198-212. PMID 30479005
  20. Davidson JR, et al. Escitalopram in the treatment of generalized anxiety disorder: double-blind, placebo controlled, flexible-dose study. Depress Anxiety. 2004;19:234-40. PMID 15274172
  21. Baldwin DS, et al. Escitalopram and paroxetine in the treatment of generalised anxiety disorder: randomised, placebo-controlled, double-blind study. Br J Psychiatry. 2006;189:264-72. PMID 16946363
  22. Bose A, et al. Randomized placebo-controlled trial of escitalopram and venlafaxine XR in the treatment of generalized anxiety disorder. Depress Anxiety. 2008;25:854-61. PMID 18050245
  23. Rynn M, et al. Efficacy and safety of duloxetine in the treatment of generalized anxiety disorder: a flexible-dose, progressive-titration, placebo-controlled trial. Depress Anxiety. 2008;25:182-9. PMID 17311303
  24. Strawn JR, et al. A randomized, placebo-controlled study of duloxetine for the treatment of children and adolescents with generalized anxiety disorder. J Am Acad Child Adolesc Psychiatry. 2015;54:283-93. PMID 25791145
  25. Baldwin D, et al. Efficacy of drug treatments for generalised anxiety disorder: systematic review and meta-analysis. BMJ. 2011;342:d1199. PMID 21398351
  26. Li X, et al. Efficacy and tolerability of short-term duloxetine treatment in adults with generalized anxiety disorder: A meta-analysis. PLoS One. 2018;13:e0194501. PMID 29558528
  27. Zhang Y, et al. Duloxetine in treating generalized anxiety disorder in adults: A meta-analysis of published randomized, double-blind, placebo-controlled trials. Asia Pac Psychiatry. 2016;8:215-25. PMID 26238298
  28. Mavranezouli I, et al. The cost effectiveness of pharmacological treatments for generalized anxiety disorder. Pharmacoeconomics. 2013;31:317-33. PMID 23512146
  29. Bschor T, et al. Differential Outcomes of Placebo Treatment Across 9 Psychiatric Disorders. JAMA Psychiatry. 2024;81:757-768. PMID 38809560
  30. Lenze EJ, et al. Escitalopram for older adults with generalized anxiety disorder: a randomized controlled trial. JAMA. 2009;301:295-303. PMID 19155456