Skip to content

Cognitive behavioural therapy

TL;DR — CBT is the best-evidenced psychological treatment for GAD, but its headline effect size depends almost entirely on the comparator. Against waitlist, pooled g≈0.84 across 41 studies and 2,132 GAD patients (Cuijpers 2014, PMID 24487344); against treatment as usual in the definitive network meta-analysis of 65 RCTs and 5,048 patients, SMD −0.74 (95% CI −1.09 to −0.38, moderate certainty), with third-wave CBTs statistically indistinguishable at −0.76 (−1.15 to −0.36) and relaxation therapy at −0.59 (−1.07 to −0.11, low certainty) — and relaxation lost superiority entirely when high-risk-of-bias studies were excluded (Papola 2024, PMID 37851421). Against placebo the effect is moderate: Hedges' g=0.56 across anxiety disorders pooled, with GAD in the large-effect stratum (Carpenter 2018, PMID 29451967) — a pooled source whose GAD figure must be taken from its own stratum. Long-term, the picture is weakest exactly where GAD needs it most: at ≥12 months CBT remained significantly better than controls for GAD, but at only Hedges' g=0.22 across 10 studies (van Dis 2020, PMID 31758858). Two comparative facts recur and are rarely stated together: applied relaxation performs about as well as CBT in head-to-head trials (Borkovec 1993, PMID 8370856; Arntz 2003, PMID 12732372; Flückiger 2022, PMID 35311299), and remote CBT underperforms individual face-to-face CBT in the only delivery-format network meta-analysis (Liu 2025, PMID 40506439).

The effect-size ladder

Every number below is for GAD unless flagged as pooled.

Comparison Effect Studies / n Source
Psychotherapy vs control (mostly waitlist) g=0.84 (95% CI 0.71–0.97); on depression g=0.71 (0.59–0.82) 38 comparisons from 28 studies; 41 studies, 2,132 patients total Cuijpers 2014, PMID 24487344
CBT vs treatment as usual SMD −0.74 (−1.09 to −0.38), moderate certainty 65 RCTs, 5,048 participants Papola 2024, PMID 37851421
Third-wave CBT vs treatment as usual SMD −0.76 (−1.15 to −0.36), moderate certainty same network Papola 2024, PMID 37851421
Relaxation therapy vs treatment as usual SMD −0.59 (−1.07 to −0.11), low certainty; −0.47 (−1.18 to 0.23), non-significant, after excluding high-RoB studies same network Papola 2024, PMID 37851421
CBT vs placebo (pooled anxiety disorders) Hedges' g=0.56 target symptoms; response OR 2.97; GAD in the large-effect subgroup with OCD and acute stress disorder 41 studies, N=2,843 Carpenter 2018, PMID 29451967 — pooled source
CBT vs placebo (pooled anxiety disorders, earlier) g=0.73 completer continuous; response OR 4.06 (2.78–5.92); no attrition difference 27 studies Hofmann 2008, PMID 18363421 — pooled source
CBT vs inactive controls (unified series) GAD falls in the g=0.5–1.0 band, alongside panic, social anxiety, depression, OCD and the eating disorders; waitlist-controlled trials give g>0.94 375 trials, 32,968 patients across disorders Cuijpers 2025, PMID 40238104 — cross-disorder source
CBT vs controls at ≥12 months (GAD) Hedges' g=0.22 (k=10) 69 RCTs, 4,118 outpatients across disorders van Dis 2020, PMID 31758858
CBT vs controls at 0–12 months (GAD) Hedges' g range 0.07–0.40 — the smallest range of any disorder in that analysis same van Dis 2020, PMID 31758858
Acceptability (all-cause discontinuation) No psychotherapy differed from treatment as usual; e.g. CBT RR 1.04 (0.64–1.67) 65 RCTs Papola 2024, PMID 37851421

Two observations that the summary sentences of these papers do not make:

  1. GAD has the smallest CBT-vs-control effects of the anxiety disorders in the long-term analysis (g 0.07–0.40 across follow-up windows, vs 0.59–0.72 for PTSD and 0.70–0.85 for OCD) (van Dis 2020, PMID 31758858). The disorder most often described as "responsive to CBT" is the one where CBT's durability evidence is thinnest.
  2. Control-condition choice dominates. Waitlist gives g>0.94 in the unified series; care-as-usual gives 0.22–1.13 (Cuijpers 2025, PMID 40238104). Bandelow's cross-treatment meta-analysis makes the same point from the other side: pre-post effect sizes for placebo pills (d=1.29) exceed those for waitlists (d=0.20) and for several psychotherapies (Bandelow 2015, PMID 25932596).

The applied-relaxation problem

If CBT's active ingredient were cognitive restructuring, a purely somatic treatment should underperform it. It does not, and this has been replicated for three decades.

Study Design Result
Borkovec 1993 (PMID 8370856) Nondirective vs applied relaxation vs CBT AR and CBT equivalent at post-treatment and both superior to nondirective therapy; at follow-up, gains maintained in AR and CBT (especially CBT), with highest end-state functioning in CBT
Arntz 2003 (PMID 12732372) 45 community mental-health patients, cognitive therapy vs applied relaxation, 12 sessions Both effective (ES 0.53–1.14); AR tended to do better at 1 month, CT caught up by 6 months; 55% (CT) vs 53.3% (AR) recovered on STAI-trait at 6 months
Flückiger 2022 (PMID 35311299) Longitudinal multilevel meta-analysis, 54 effect sizes in 23 studies No significant omnibus differences among bona fide psychotherapies on primary outcomes; negligible AR-vs-CBT differences at each assessment; small advantage for augmented integrative CBT over well-established CBT
Papola 2024 (PMID 37851421) Network meta-analysis, 65 RCTs Relaxation −0.59 vs TAU, but non-significant after excluding high-RoB studies — the one analysis where relaxation separates from CBT, and it does so on risk-of-bias grounds

The honest reading is that GAD responds to structured, credible, active psychological treatment, and the specific cognitive machinery is not clearly doing the work — with the single exception of protocols that explicitly target intolerance of uncertainty (mechanism and models), which outperform general CBT on IU and worry at post-treatment though not at follow-up (Wilson 2023, PMID 37271039).

Which protocol, and which format

Protocol. A Bayesian network meta-analysis of 56 studies and 4,388 participants found Dugas's CBT protocol — self-monitoring, education about intolerance of uncertainty, examination of worry beliefs, problem-orientation training, and work with core fears — superior in both short- and long-term reduction of GAD severity among CBT protocols; among third-wave approaches, acceptance-based therapy was notable. Meta-regression found clinician guidance significantly moderated effect size (Dai 2025, PMID 40367584). This is the clearest published protocol-level recommendation in the GAD literature, and it converges with the IU mechanism evidence.

Format. 52 trials, 4,361 patients (mean age 43, 69.7% women); 75% at low or unclear risk of bias:

Format comparison SMD (95% CI) Source
Individual CBT vs remote CBT 0.96 (0.13–1.79) — individual superior Liu 2025, PMID 40506439
Individual CBT vs treatment as usual 1.12 (0.24–2.00) Liu 2025, PMID 40506439
Individual CBT vs waiting list 1.62 (1.03–2.22) Liu 2025, PMID 40506439
Group CBT vs waiting list 1.65 (0.47–2.84) Liu 2025, PMID 40506439
Remote CBT vs TAU or waiting list Not superior Liu 2025, PMID 40506439
Acceptability across formats No significant differences Liu 2025, PMID 40506439

A third estimate sits between the two: a GAD-specific meta-analysis of 10 studies and 1,071 participants in remote CBT found within-group pre-post g=1.30 (95% CI 1.03–1.58) and between-group g=0.76 (0.47–1.06) at post-treatment, with no significant difference between low- and high-intensity remote formats (Trenoska Basile 2022, PMID 35403706). The conclusion drawn from the format network — group CBT is a reasonable alternative to individual CBT, remote CBT may be less effective — sits in direct tension with the digital-delivery literature, which reports large effects for guided internet CBT (digital and remote delivery). The tension is unresolved and is one of this condition's open questions.

Low-intensity CBT — the bottom of the stepped-care ladder

NICE's stepped-care model puts low-intensity CBT (LICBT) first for mild-to-moderate GAD, and that step now has its own meta-analysis: 12 RCTs identified from 1,205 papers over 2000–2023, covering the three NICE-recommended LICBT forms — non-facilitated self-help, guided self-help and psychoeducational groups (PROSPERO CRD42021285590). Adjusted effect sizes were medium and consistent across outcomes: anxiety g=−0.63 (12 effect sizes, n=1,201), depression g=−0.48 (11, n=1,164), worry g=−0.64 (9, n=908), all favouring LICBT over control. Between-study heterogeneity was significant for anxiety and worry with no moderator identified by meta-regression (Powell 2024, PMID 38166836).

LICBT's g≈0.63 and high-intensity CBT's SMD −0.74 come from different syntheses and comparator structures and cannot establish equivalence. The LICBT estimate nevertheless supports its use as a stepped-care option (Powell 2024, PMID 38166836; Papola 2024, PMID 37851421).

Mechanisms of change: what actually moves

Unusually for a psychotherapy literature, GAD has session-by-session mediation studies with defensible designs.

Study Design Finding
Donegan 2012 (PMID 22506794) 57 GAD patients in an RCT; daily self-report of worry and somatic anxiety during CBT or applied relaxation Change was bidirectional in both treatments, but change in worry accounted for subsequent change in somatic anxiety to a greater extent in CBT than in AR — evidence that two equally efficacious treatments work through different mechanisms, each consistent with its own theoretical rationale
Gómez Penedo 2021 (PMID 34881910) 80 GAD patients in CBT within an RCT; dynamic structural equation modelling of session-by-session data Greater change in interpersonal cognitions and in self-related cognitions in a session predicted lower worry at the start of the next session; the interpersonal effect survived adjustment for self-related cognitions
Coyne 2019 (PMID 30829505) 85 GAD patients randomised to CBT alone or CBT integrated with motivational interviewing; repeated alliance, interpersonal-problem and outcome ratings Both within-patient and between-patient alliance related to subsequent worry reduction; change in interpersonal problems mediated the association only at the within-patient level
Costa 2023 (PMID 36447332) RCT of mindfulness-based intervention vs fluoxetine vs an active comparison, 8 weeks; mediation and moderated mediation Self-compassion (C=0.11, 95% CI 0.01–0.28) and non-judgement of inner experience (C=0.10, 0.004–0.21) mediated the baseline-worry-to-endpoint-anxiety path; mindfulness state did not (C=0.06, −0.05 to 0.20). Intervention modality did not moderate
Jacoby 2024 (PMID 37577773) Sleep outcomes in the 226-patient CBT vs Kundalini yoga vs stress-education RCT Sleep improved pre-to-post in all three arms (all d>0.97) with no significant advantage for CBT or yoga over the attention control; worry, mindfulness and perceived stress each mediated the time-to-sleep effect, and the degree of sleep change attributable to worry was ordered CBT > KY > SEdu
Wilson 2023 (PMID 37271039) 26 studies, 1,199 GAD patients CBT explicitly targeting intolerance of uncertainty outperformed general CBT on IU and worry at post-treatment (p<0.01 each), with effect size rising with time spent on IU — but not at follow-up
Picó-Pérez 2023 (PMID 35916600) Meta-analysis of 17 whole-brain task fMRI studies, n=442, anxiety-related disorders pooled Activity in a salience/interoception network — right inferior frontal gyrus/anterior insula and dorsomedial prefrontal/dorsal anterior cingulate cortex — was strongly associated with positive CBT outcome

Two conclusions follow. First, worry reduction repeatedly appears as a statistical mediator of somatic-anxiety change (Donegan 2012, PMID 22506794), sleep improvement (Jacoby 2024, PMID 37577773) and functioning in a 670-participant worry-focused trial (Andersson 2024, PMID 37706604); mediation alone does not establish causation. Second, candidate pathways are not limited to cognitive content: interpersonal change and self-compassion also carry mediation signals, motivating interpersonally oriented and acceptance-based variants (other psychological and non-drug therapies).

Comparators other than CBT

  • Yoga. In a 3-arm, single-blind RCT (n=226 with primary GAD; Kundalini yoga 93, CBT 90, stress education 43; twelve 120-min group sessions), response rates were 54.2% (yoga) vs 33.0% (stress education), OR 2.46 (1.12–5.42), NNT 4.59; and 70.8% (CBT) vs 33.0%, OR 5.00 (2.12–11.82), NNT 2.62. The pre-specified noninferiority test of yoga vs CBT was not met (Simon 2021, PMID 32805013). Yoga works; it is not equivalent to CBT.
  • Transdiagnostic protocol. The Unified Protocol was statistically equivalent to diagnosis-specific protocols across panic, GAD, OCD and social anxiety (n=223), with better treatment completion (OR 3.11, 1.44–6.74); both beat waitlist (UP d=−0.93; SDP d=−1.08) (Barlow 2017, PMID 28768327). Pooled across four disorders, so not a GAD-specific effect.
  • Drugs vs therapy. Bandelow's 234-study, 37,333-patient meta-analysis found higher average pre-post effect sizes for medications (d=2.02) than psychotherapies (d=1.22), and noted psychotherapy pre-post effects did not differ from pill placebo — while cautioning that pre-post comparison across differently-controlled literatures is the weak link, and recommending patient preference as the deciding factor (Bandelow 2015, PMID 25932596). Pooled across panic, GAD and social phobia. The GAD-specific head-to-head evidence remains thin; see SSRI and SNRI pharmacotherapy and Chen 2019 (PMID 31494377), which found most pharmacological interventions had larger effect sizes than psychological ones — while allowing all comorbidities.

Older adults

CBT for late-life GAD has its own meta-analysis: 14 RCTs, N=985, all required to use the PSWQ or PSWQ-A. CBT beat waitlist or treatment as usual at post-treatment and at 6-month follow-up, but against active controls the advantage was small and non-significant at post-treatment and equivalent at follow-up; effect size was associated with attrition rates and depression outcomes (Hall 2016, PMID 27687212). The authors state plainly that durability and superiority over other available treatments such as supportive psychotherapy "remain to be tested" (special populations).

Quality caveats that apply to this whole literature

  • Erratum on the reference network meta-analysis. Papola 2024 carries a published correction (PMID 38265780); figures quoted here are from the article record as retrieved 2026-09-02.
  • Waitlist dominance. The majority of studies in Cuijpers 2014 used waiting lists as the control condition, and the authors report indications of publication bias (Cuijpers 2014, PMID 24487344).
  • Risk of bias. 75% of the delivery-format trials were at low or unclear risk of bias (Liu 2025, PMID 40506439); most trials in the long-term analysis were "mainly of low quality" (van Dis 2020, PMID 31758858).
  • Relapse is barely measured. Relapse rates of 0–14% at 3–12 months were reported in only 6 of 69 RCTs, predominantly for panic disorder rather than GAD (van Dis 2020, PMID 31758858) — see course, relapse and long-term outcome.

Open questions

  • Why is CBT's long-term effect in GAD (g=0.22 at ≥12 months) so much smaller than in PTSD or social anxiety (van Dis 2020, PMID 31758858), when its acute effect is comparable?
  • Is anything in CBT doing disorder-specific work beyond structure and credibility, given that applied relaxation matches it across four decades of head-to-head trials (Borkovec 1993, PMID 8370856; Flückiger 2022, PMID 35311299)?
  • Why does remote CBT underperform individual CBT in the format network (Liu 2025, PMID 40506439) while the internet-CBT literature reports large effects (digital and remote delivery)? Comparator choice and guidance intensity are the leading candidates.
  • Does Dugas's IU-targeting protocol's superiority (Dai 2025, PMID 40367584) replicate in an adequately powered head-to-head trial against standard CBT? The evidence is currently network-indirect.
  • What is CBT's effect in GAD without comorbid depression? A PubMed search rerun on 2026-09-02 located no meta-analytic estimate stratified by depressive comorbidity; this remains a dated effect-modification gap.

References

  1. Papola D, et al. Psychotherapies for Generalized Anxiety Disorder in Adults: A Systematic Review and Network Meta-Analysis of Randomized Clinical Trials. JAMA Psychiatry. 2024;81:250-259. PMID 37851421
  2. Papola D, et al. Errors in Network Meta-Analysis of Generalized Anxiety Disorder Psychotherapies. JAMA Psychiatry. 2024;81:320. PMID 38265780
  3. Cuijpers P, et al. Psychological treatment of generalized anxiety disorder: a meta-analysis. Clin Psychol Rev. 2014;34:130-40. PMID 24487344
  4. Cuijpers P, et al. Cognitive Behavior Therapy for Mental Disorders in Adults: A Unified Series of Meta-Analyses. JAMA Psychiatry. 2025;82:563-571. PMID 40238104
  5. van Dis EAM, et al. Long-term Outcomes of Cognitive Behavioral Therapy for Anxiety-Related Disorders: A Systematic Review and Meta-analysis. JAMA Psychiatry. 2020;77:265-273. PMID 31758858
  6. Carpenter JK, et al. Cognitive behavioral therapy for anxiety and related disorders: A meta-analysis of randomized placebo-controlled trials. Depress Anxiety. 2018;35:502-514. PMID 29451967
  7. Hofmann SG, Smits JA. Cognitive-behavioral therapy for adult anxiety disorders: a meta-analysis of randomized placebo-controlled trials. J Clin Psychiatry. 2008;69:621-32. PMID 18363421
  8. Borkovec TD, Costello E. Efficacy of applied relaxation and cognitive-behavioral therapy in the treatment of generalized anxiety disorder. J Consult Clin Psychol. 1993;61:611-9. PMID 8370856
  9. Arntz A. Cognitive therapy versus applied relaxation as treatment of generalized anxiety disorder. Behav Res Ther. 2003;41:633-46. PMID 12732372
  10. Flückiger C, et al. The relative efficacy of bona fide cognitive behavioral therapy and applied relaxation for generalized anxiety disorder at follow-up: A longitudinal multilevel meta-analysis. J Consult Clin Psychol. 2022;90:339-352. PMID 35311299
  11. Liu S, et al. CBT treatment delivery formats for generalized anxiety disorder: a systematic review and network meta-analysis of randomized controlled trials. Transl Psychiatry. 2025;15:197. PMID 40506439
  12. Dai X, et al. (Third-wave) cognitive behavioral therapy for generalized anxiety disorder in adults: A systematic review and Bayesian network meta-analysis. J Psychiatr Res. 2025;187:134-143. PMID 40367584
  13. Simon NM, et al. Efficacy of Yoga vs Cognitive Behavioral Therapy vs Stress Education for the Treatment of Generalized Anxiety Disorder: A Randomized Clinical Trial. JAMA Psychiatry. 2021;78:13-20. PMID 32805013
  14. Barlow DH, et al. The Unified Protocol for Transdiagnostic Treatment of Emotional Disorders Compared With Diagnosis-Specific Protocols for Anxiety Disorders: A Randomized Clinical Trial. JAMA Psychiatry. 2017;74:875-884. PMID 28768327
  15. Bandelow B, et al. Efficacy of treatments for anxiety disorders: a meta-analysis. Int Clin Psychopharmacol. 2015;30:183-92. PMID 25932596
  16. 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
  17. Hall J, et al. Efficacy of Cognitive Behavioral Therapy for Generalized Anxiety Disorder in Older Adults: Systematic Review, Meta-Analysis, and Meta-Regression. Am J Geriatr Psychiatry. 2016;24:1063-1073. PMID 27687212
  18. Wilson EJ, et al. The impact of psychological treatment on intolerance of uncertainty in generalized anxiety disorder: A systematic review and meta-analysis. J Anxiety Disord. 2023;97:102729. PMID 37271039
  19. Powell CLYM, et al. A meta-analysis on the efficacy of low-intensity cognitive behavioural therapy for generalised anxiety disorder. BMC Psychiatry. 2024;24:10. PMID 38166836
  20. Donegan E, Dugas MJ. Generalized anxiety disorder: a comparison of symptom change in adults receiving cognitive-behavioral therapy or applied relaxation. J Consult Clin Psychol. 2012;80:490-6. PMID 22506794
  21. Gómez Penedo JM, et al. Interpersonal cognitions as a mechanism of change in cognitive behavioral therapy for generalized anxiety disorder? J Consult Clin Psychol. 2021;89:898-908. PMID 34881910
  22. Coyne AE, et al. Interpersonal change as a mediator of the within- and between-patient alliance-outcome association in two treatments for generalized anxiety disorder. J Consult Clin Psychol. 2019;87:472-483. PMID 30829505
  23. Costa MA, et al. Mechanisms of improvement in generalized anxiety disorder: A mediation and moderated mediation analysis from a randomized controlled trial. Br J Clin Psychol. 2023;62:196-208. PMID 36447332
  24. Jacoby RJ, et al. Effect of cognitive behavioural therapy and yoga for generalised anxiety disorder on sleep quality in a randomised controlled trial. J Sleep Res. 2024;33:e13992. PMID 37577773
  25. Andersson E, et al. The impact of reduced worry on general functioning: A mediation analysis from a randomized trial. Stress Health. 2024;40:e3319. PMID 37706604
  26. Picó-Pérez M, et al. Neural predictors of cognitive-behavior therapy outcome in anxiety-related disorders: a meta-analysis of task-based fMRI studies. Psychol Med. 2023;53:3387-3395. PMID 35916600
  27. Trenoska Basile V, et al. Remote cognitive-behavioral therapy for generalized anxiety disorder: A preliminary meta-analysis. J Clin Psychol. 2022;78:2381-2395. PMID 35403706