Open questions — generalized anxiety disorder¶
Last curated: 2026-09-02 (full build). This file replaces the seed set written at scoping. Stable IDs are continuous with the seed: OQ-1 to OQ-5 are the seed questions, re-grounded against live evidence; OQ-6 onward are new.
Tier 1 = answering it would change practice, and a study is designable today. Tier 2 = blocked on tools, numbers, definitions, or a Tier-1 answer. Tier 3 = foundational questions whose answers would reorganise the field rather than change a decision.
Every asserted absence was re-searched on 2026-09-02 and is dated.
Dots not yet connected¶
Cross-domain junctions where two bodies of evidence in this knowledge base both exist and no study has joined them.
| # | Dot A | Dot B | The missing junction | Powers |
|---|---|---|---|---|
| D1 | Cross-national DSM-5 epidemiology rests on the excessiveness criterion (Ruscio 2017, PMID 28297020) | The same group has published the case for deleting that criterion (Ruscio 2024, PMID 39364896) | Prevalence and comorbidity have been re-derived without excessiveness; treatment response has not. No trial dataset has been re-analysed under the alternative definition | OQ-1 |
| D2 | Network meta-analyses rank treatments (Slee 2019, PMID 30712879; Papola 2024, PMID 37851421) | The Cochrane antidepressant review excluded serious comorbidity while the psychological/pharmacological network allowed all comorbidities (Kopcalic 2025, PMID 39880377; Chen 2019, PMID 31494377) | No treatment estimate exists for GAD stratified by presence or absence of comorbid depression, in either direction | OQ-2 |
| D3 | GAD-7 pooled accuracy is estimated across 48 studies, with context-specific variation (Aktürk 2025, PMID 40130828) | USPSTF recommends screening adults on instrument accuracy plus treatment efficacy, with no screening trial showing benefit (USPSTF 2023, PMID 37338866; O'Connor 2023, PMID 37338868) | A 2026-09-02 search found no screening-plus-pathway trial with treatment initiation and symptom outcomes | OQ-3 |
| D4 | GAD is chronic: 0.38 five-year remission, high recurrence after recovery (Yonkers 2000, PMID 10974960; Bruce 2005, PMID 15930067) | Longest double-blind acute phase in the Cochrane antidepressant review is 28 weeks; relapse reported in 6 of 69 psychotherapy RCTs (Kopcalic 2025, PMID 39880377; van Dis 2020, PMID 31758858) | The disorder's defining temporal feature is the one its evidence base least addresses | OQ-4 |
| D5 | Digital interventions show g=0.62–0.79 in GAD with guided ≈ unguided (Eilert 2021, PMID 33225589; Pauley 2023, PMID 34047264) | The GAD-specific delivery-format network finds individual CBT superior to remote CBT (SMD 0.96) and remote CBT no better than TAU (Liu 2025, PMID 40506439) | No study has reconciled the two by matching comparator type, guidance intensity and delivery modality | OQ-5 |
| D6 | Placebo pre-post effect in GAD is d_av 1.23, second only to depression (Bschor 2024, PMID 38809560) | Drug–placebo HAM-A differences are 2.5–3.6 points, and antidepressant response NNTB is 7 (Slee 2019, PMID 30712879; Kopcalic 2025, PMID 39880377) | No analysis has asked whether GAD's placebo magnitude is a property of the disorder or of trial design, or whether design changes could shrink it | OQ-6 |
| D7 | The anxiety GWAS highlights GABAergic signalling across 58 loci (Strom 2026, PMID 41634414) | Benzodiazepines and pregabalin — GABA-system drugs — were developed decades before that signal and are restricted on tolerability and dependence grounds (Fernandes 2025, PMID 40544830; Slee 2019, PMID 30712879) | Genetics points at a mechanism the field is actively de-prescribing; no work links the genetic signal to differential response to GABAergic agents | OQ-7 |
| D8 | CALM collaborative care produced an NNT of 5.27 for its response outcome (Roy-Byrne 2010, PMID 20483968) | Patients describe valuing continuity, validation and participation (Parker 2020, PMID 32059636; Hurtado 2020, PMID 31908140) | A 2026-09-02 search found no dismantling trial separating care coordination, treatment choice and measurement-based follow-up | OQ-8 |
| D9 | Preference was strong enough to terminate a randomised trial: three-quarters of eligible patients declined, mostly refusing randomisation to medication (Kalpakidou 2019, PMID 31126337) | A quasi-experimental GAD study later allowed choice among e-CBT, medication and combination (Stephenson 2023, PMID 38125282) | Preference-based delivery is feasible, but nonrandom allocation leaves comparative causal effects unresolved | OQ-9 |
| D10 | GAD in stable coronary heart disease carries a 62% higher cardiovascular event rate after full adjustment (Martens 2010, PMID 20603456), and SSRI use in CAD+GAD associates with lower MACE (HR 0.77; Wu 2025, PMID 40466339) | Every GAD treatment trial uses symptom scales as its primary endpoint | No trial has tested whether treating GAD changes cardiovascular outcomes | OQ-10 |
| D11 | Intolerance of uncertainty is the best-supported mechanism and IU-targeting CBT outperforms general CBT at post-treatment (Wilson 2023, PMID 37271039; Dai 2025, PMID 40367584) | The advantage disappears by follow-up, and IU is definitively transdiagnostic (r=0.51 across 335 effect sizes; McEvoy 2019, PMID 31678816) | Nobody has tested whether a longer or maintained IU-targeting dose preserves the advantage, or whether IU-targeting works equally in depression | OQ-11 |
| D12 | GAD has a disorder-specific structural signature — reduced insula and lateral/medial prefrontal volume, increased right putamen — relative to fear-based disorders (Liu 2022, PMID 36151073) | GAD resembles experimentally induced anxiety least of the anxiety disorders (Chavanne 2021, PMID 33054384) | If induced anxiety is the translational model used in anxiolytic drug development, GAD may be the disorder that model fits worst — and nobody has drawn that inference in a drug-development context | OQ-12 |
| D13 | Perinatal GAD prevalence is 4.1% antenatal and 5.7% postnatal (Dennis 2017, PMID 28302701) | A later cluster-RCT secondary analysis measured GAD symptoms after systems-level perinatal interventions (Zimmermann 2024, PMID 38992743) | A targeted 2026-09-02 PubMed search still located no perinatal GAD pharmacotherapy RCT | OQ-24 |
| D14 | Muscle tension is the only DSM-IV GAD symptom that distinguished GAD from non-GAD in a community sample; fatigue, restlessness and concentration difficulty were more common in depression (Faravelli 2012, PMID 22578985) | The criteria weight all six associated symptoms equally | No revision has proposed weighting or reordering the symptom list by discriminative value | OQ-14 |
| D15 | Four independent literatures converge on uncontrollability of worry as GAD's discriminating feature: incremental validity (Hallion 2013, PMID 23713499), symptom specificity (Faravelli 2012, PMID 22578985), network centrality without bridge status (Cai 2024, PMID 38238548) and metacognitive specificity (Sun 2017, PMID 28763680) | Every published revision proposal targets excessiveness, not uncontrollability (Ruscio 2024, PMID 39364896; Andrews 2010, PMID 20058241) | No proposal has been built on the convergence — i.e. keep uncontrollability, drop excessiveness, and reweight the associated symptoms by discriminative value | OQ-14, OQ-32, OQ-36 |
| D16 | Guidelines agree on first-line treatment across every major body (literature/guidelines/REGISTRY.md) | Guideline-concordant care in a 721-patient primary-care cohort produced equal 12-month improvement at higher cost, and concordance was patterned by education level rather than clinical need (Prins 2010, PMID 20049547; Prins 2011, PMID 22099636; Prins 2011, PMID 20586845) | Nobody has tested whether the recommendations, the measurement of concordance, or confounding by indication explains the null | OQ-37 |
| D17 | In paediatric anxiety, acute response predicted long-term remission while assigned treatment did not predict remission (Ginsburg 2014, PMID 24477837; Ginsburg 2018, PMID 29960692); CBT nevertheless predicted some long-term functional trajectories (Swan 2018, PMID 30138013) | Guidelines recommend modalities, not response-contingent strategies (Walter 2020, PMID 32439401) | The missing analysis is how early response and assigned modality jointly shape remission and function | OQ-38 |
| D18 | 70% of treated patients find their treatment helpful, and virtually all would eventually obtain helpful treatment by persisting with up to 10 professionals (Stein 2021, PMID 34372811) | Only 29.7% would persist that long, and patients describe being offered medication first without preference elicitation (Hurtado 2020, PMID 31908140) | No intervention has been designed to increase persistence through unhelpful encounters — the modelled binding constraint on getting effective care | OQ-39 |
Tier 1 — designable today, would change practice¶
OQ-1. Should excessive worry remain a required criterion, and what happens to the treatment evidence if it goes?¶
Removing it raises global lifetime prevalence 2.6% → 4.0%, and non-excessive cases match diagnosed cases on family history, secondary comorbidity and suicidality while reporting comparable impairment (Ruscio 2024, PMID 39364896). Reliability rises from κ=0.53 to κ=0.78 when the criterion is dropped (Wittchen 1995, PMID 7666382). Prevalence has been re-derived; treatment response has not. Re-analysing existing trial datasets by excessiveness status is designable now. → the-diagnostic-boundary.md, epidemiology-and-burden.md
OQ-2. What works for GAD without comorbid depression — and what works for the comorbid state?¶
The Cochrane antidepressant review excluded serious comorbidity and included few participants with secondary psychiatric comorbidity; the psychological/pharmacological network allowed all comorbidities (Kopcalic 2025, PMID 39880377; Chen 2019, PMID 31494377). Emotion regulation therapy is the only GAD protocol deliberately trialled across both (43% comorbid MDD; Mennin 2018, PMID 29504794). An IPD meta-analysis with comorbidity recorded would answer it. → comorbidity-and-primary-care.md, ssri-and-snri-pharmacotherapy.md
OQ-3. Does screening with the GAD-7 change outcomes, or only detection?¶
Accuracy is settled (sensitivity 0.64, specificity 0.91 at ≥10; Aktürk 2025, PMID 40130828). Benefit is not: only two studies evaluated screening itself and neither found benefit, yet a B recommendation was issued (O'Connor 2023, PMID 37338868; USPSTF 2023, PMID 37338866). → screening-and-measurement.md
OQ-6. Can GAD's placebo response be reduced by trial design?¶
d_av 1.23 (1.06–1.41), second highest of nine disorders (Bschor 2024, PMID 38809560). Single-dose designs produce larger drug–placebo differences (MM120 −6.0 HAM-A points; Robison 2025, PMID 40906494), but confound with functional unblinding in 46.2% of dosed participants. Design manipulations — run-in periods, blinded raters, dose-response modelling — are testable. → clinical-trials-landscape.md
OQ-8. Which component of collaborative care does the work?¶
CALM bundled treatment choice, care management and web-based outcome monitoring, giving NNT 5.27 for response (Roy-Byrne 2010, PMID 20483968). No dismantling trial has been located as of 2026-09-02. → comorbidity-and-primary-care.md
OQ-9. How should drug-versus-therapy comparative effectiveness be established when preference undermines randomisation?¶
Three-quarters of eligible patients declined a commissioned UK RCT, mostly because they would not risk medication allocation (Kalpakidou 2019, PMID 31126337). A later quasi-experimental study let patients choose e-CBT, medication or combination, but its 41/41/33 groups remain vulnerable to selection and confounding (Stephenson 2023, PMID 38125282). A randomised-preference or stronger causal design is still needed. → patient-experience-and-advocacy.md
OQ-15. Does metacognitive therapy's advantage over CBT survive independent replication?¶
Recovery 65% vs 38% post-treatment and 57% vs 38% at nine years, with GAD re-diagnosis 9.5% vs 23.1% (Nordahl 2018, PMID 30294448; Solem 2021, PMID 34520637). But 81 randomised of 246 assessed, single centre, and both protocol authors were investigators with declared royalty interests. A pre-registered multi-centre replication with allegiance controls is designable now. → other-psychological-therapies.md
OQ-16. Is pregabalin first-line?¶
A 14-study, 4,822-patient meta-analysis argues yes on efficacy, tolerability and cost-effectiveness (Cardoner 2025, PMID 39989902); it ranked third on HAM-A with good acceptability in the reference network (Slee 2019, PMID 30712879); no guideline in literature/guidelines/REGISTRY.md places it first-line, and misuse risk is real but concentrated in people with substance use disorders (Bonnet 2017, PMID 28988943). A non-inferiority head-to-head against an SSRI would settle it. → pregabalin-benzodiazepines-and-others.md
OQ-17. Does treating GAD change cardiovascular outcomes?¶
GAD independently predicts a 62% higher cardiovascular event rate in stable CHD after adjustment including depression (Martens 2010, PMID 20603456), and SSRI use in CAD+GAD is associated with lower one-year MACE in a propensity-matched cohort of 109,052 (HR 0.77, 0.74–0.81; Wu 2025, PMID 40466339). Both are observational. → course-relapse-and-long-term-outcome.md
OQ-18. Should CYP2C19 phenotype guide escitalopram dosing in adolescents with GAD?¶
Increasing CYP2C19 metabolism was associated with lower escitalopram exposure (AUC₀₋₂₄, p<0.05) in an adolescent GAD RCT, and the authors raise genotype-guided dosing explicitly (Strawn 2020, PMID 32857933, NCT02818751). A genotype-stratified dosing trial is designable. → special-populations.md
OQ-36. Should a criterion revision keep uncontrollability and drop excessiveness?¶
Four independent methods converge on uncontrollability as the discriminating feature and on the arousal/somatic criteria as the shared ones (Hallion 2013, PMID 23713499; Faravelli 2012, PMID 22578985; Cai 2024, PMID 38238548; Sun 2017, PMID 28763680). Every revision proposal on the table targets excessiveness instead. Assembling these literatures into a testable criterion set is designable today. → the-diagnostic-boundary.md, diagnosis-and-classification.md
OQ-37. Does guideline-concordant care improve outcomes in GAD?¶
39% concordance in 721 Dutch primary-care patients; concordant patients were more severe at baseline and improved equally at 12 months, at higher cost, with concordance predicted by education level and perceived need rather than clinical need (Prins 2010, PMID 20049547; Prins 2011, PMID 22099636; Prins 2011, PMID 20586845). Confounding by indication, over-estimated trial effects and mismeasured concordance are all untested. → guidelines.md
OQ-38. If acute response rather than treatment type predicts long-term remission, what secures early response?¶
CAMELS: 46.5% remission at ~6 years, 21.7% stable remission across four follow-up years, and acute responders more likely to remit; treatment assignment did not predict remission (Ginsburg 2014, PMID 24477837; Ginsburg 2018, PMID 29960692). Treatment did predict some functional trajectories (Swan 2018, PMID 30138013). A PubMed search rerun on 2026-09-02 located no equivalent adult joint analysis of acute response, modality, remission and function. → special-populations.md
OQ-39. What increases persistence through unhelpful treatment encounters?¶
The modelled ceiling on helpful treatment is reached only by persisting with up to 10 professionals; 29.7% would (Stein 2021, PMID 34372811). No intervention targets persistence directly; collaborative care may do so incidentally (Roy-Byrne 2010, PMID 20483968). → patient-experience-and-advocacy.md, comorbidity-and-primary-care.md
OQ-40. Should adult GAD adopt a core outcome set?¶
Endpoint choice changes the answer: mean HAM-A change vs remission gives different drug rankings (Slee 2019, PMID 30712879 vs Kong 2020, PMID 33343351); LOCF vs observed cases changes whether escitalopram separates from placebo (Bose 2008, PMID 18050245); censoring vs ITT changes whether it works in older adults (Lenze 2009, PMID 19155456); worry improves where GAD severity does not (Stanley 2009, PMID 19351943). A core outcome set exists in development for paediatric anxiety (Monga 2023, PMID 37244653) and not for adult GAD. → clinical-trials-landscape.md, screening-and-measurement.md
Tier 2 — blocked on definitions, tools, or a Tier-1 answer¶
OQ-4. What maintains remission over years rather than weeks?¶
Continuation treatment works over 26–76 weeks (relapse 19% vs 56% on escitalopram; Allgulander 2006, PMID 16316482), and guidelines stop at 6–12 months post-remission (Bandelow 2022, PMID 34609587). A PubMed search rerun on 2026-09-02 located no trial comparing fixed-duration against indefinite continuation with a specified taper. The gap is one of trial duration and funding, not design. → course-relapse-and-long-term-outcome.md
OQ-5. Is guided digital CBT equivalent to face-to-face CBT in GAD specifically?¶
Blocked on GAD-specific rather than pooled-anxiety analysis: the equivalence estimate (g=0.14, 9 comparisons, 683 participants) pools anxiety disorders (Pauley 2023, PMID 34047264), while the GAD-specific format network finds individual CBT superior (Liu 2025, PMID 40506439). → digital-and-remote-delivery.md
OQ-7. Does the GABAergic genetic signal predict response to GABAergic drugs?¶
Blocked on GAD-specific genotyped treatment cohorts. The 58-locus finding is for anxiety disorders pooled (Strom 2026, PMID 41634414), and the dedicated fear-versus-GAD GWAS finds four genome-wide loci for GAD (Ter Kuile 2026, PMID 42374129). → mechanism-and-models.md
OQ-11. Does targeting intolerance of uncertainty produce durable rather than transient advantage?¶
CBT-IU beats general CBT on IU and worry at post-treatment (p<0.01 each) with effect size rising with time spent on IU, but the advantage is not maintained at follow-up (Wilson 2023, PMID 37271039). Blocked on dose-and-duration trials. → mechanism-and-models.md
OQ-19. Will the treatment-resistance consensus be validated and adopted in GAD?¶
A 2024 Delphi exercise with 36 international experts produced 14 recommendations and a potential staging model across anxiety disorders (Domschke 2024, PMID 38214637). It postdates most GAD trials and is not yet a GAD-specific validated instrument, so historical populations remain heterogeneous. → treatment-resistant-gad.md
OQ-20. Does quetiapine augmentation work?¶
Two systematic reviews of overlapping literature reach opposite conclusions: no significant mean HAM-A reduction (Samuel 2011, PMID 21088608) versus RCT-supported augmentation (Schiele 2026, PMID 40946318). Blocked on an adequately powered trial and on OQ-19. → treatment-resistant-gad.md
OQ-21. Is the rTMS effect in GAD real?¶
SMD −1.857 (−2.219 to −1.494) from 6 studies and 152 patients (Parikh 2022, PMID 34791241) would make rTMS the most effective GAD treatment ever measured; a network reports high-frequency rTMS response OR 291.40 with a 95% CI of 13.08–6490.21 (Duan 2025, PMID 40203547). Blocked on adequately powered sham-controlled trials, which the authors themselves call urgent. → treatment-resistant-gad.md
OQ-22. Does escitalopram work in older adults with GAD?¶
The answer differs by analysis convention inside a single trial: cumulative response 69% vs 51% (p=0.03) censoring at dropout, 57% vs 45% (p=0.11) on conservative ITT (Lenze 2009, PMID 19155456). A PubMed search rerun on 2026-09-02 located no adequately powered conservative-ITT replication. → special-populations.md
OQ-23. Would age-adapted CBT close the older-adult effect-size gap?¶
CBT effect is g=0.55 (0.22–0.88) in older adults versus 0.94 (0.52–1.36) in working-age adults; content analysis found protocols were robust CBT but not adapted to gerontological evidence, and no older-adult study used ITT (Kishita 2017, PMID 28119196). Directly testable once an adapted protocol exists. → special-populations.md
OQ-24. What is safe and effective pharmacotherapy for GAD in pregnancy?¶
WPSI identified no treatment trials through its 2020 review (Nelson 2020, PMID 32510989), while a later systems-level cluster-RCT secondary analysis measured GAD symptoms among depression-screen-positive perinatal participants (Zimmermann 2024, PMID 38992743). CANMAT prioritised large observational medication-safety data (Vigod 2025, PMID 39936923). A targeted PubMed search rerun on 2026-09-02 located no perinatal GAD pharmacotherapy RCT; that narrower gap remains. → special-populations.md
OQ-25. Does the GAD-7's cultural differential item functioning appear in the languages where it is deployed?¶
Item-level racial DIF was found in one US undergraduate sample (Parkerson 2015, PMID 25725310). Sex invariance has also been examined: partial strong invariance in young adults across sex, time and language (Af Winklerfelt Hammarberg 2025, PMID 39880312) and invariance in 165,872 treatment seekers (Saunders 2023, PMID 37118684). Cross-cultural and diagnosed-GAD invariance remain incompletely mapped. → screening-and-measurement.md
OQ-26. What is the minimal clinically important difference for the HAM-A in GAD?¶
Severity bands exist (mild 8–14, moderate 15–23, severe ≥24; Matza 2010, PMID 20718076) and one trial pre-specified 2.5 points as its MCID (Robison 2025, PMID 40906494), but no GAD-specific anchor-based MCID study was located on 2026-09-02. Without it, a 2.5–3.6-point network difference cannot be interpreted. → screening-and-measurement.md
OQ-27. Why are completed GAD phase 2/3 programmes unpublished?¶
Troriluzole (NCT03829241, n=881), ENX-102 (NCT05749055, n=252), SEP-363856 (NCT05729373, n=434) and extended-release lorazepam (NCT02305797, n=495) are registered as completed, but NCT-ID PubMed searches on 2026-09-02 located no journal report. NCT03829241 and NCT02305797 have posted registry results. A same-day PubMed search located no systematic audit confined to GAD trial reporting, so the field-wide reporting rate remains unknown. → clinical-trials-landscape.md
Tier 3 — foundational¶
OQ-28. Is GAD a category at all, and if not, what justifies a threshold?¶
Taxometric analyses in a 2,061-adult community sample and a 1,175-patient clinical sample both find dimensional structure, with the authors of the former stating that any threshold is "likely to be arbitrary" (Marcus 2014, PMID 24377439; Kertz 2014, PMID 24334160). Dimensionality does not by itself invalidate a threshold — hypertension is dimensional and useful — but a defensible threshold needs an anchor: impairment, treatment cost-effectiveness, or predictive validity for secondary disorders. None has been proposed. → the-diagnostic-boundary.md
OQ-29. Why do genetic correlations between GAD and depression approach unity while impairment profiles remain separable?¶
rg = +1.00 in women, +0.74 in men (Kendler 2007, PMID 17121688); yet pure GAD impairs comparably to pure MDD (Hoffman 2008, PMID 17146763) and GAD/depression symptoms load as distinct factors with independent effects on disability (Spitzer 2006, PMID 16717171). Shared etiology and separable phenotype are compatible in principle; nobody has specified the mechanism that makes them compatible in fact. → the-diagnostic-boundary.md
OQ-30. Is any psychological mechanism GAD-specific?¶
Intolerance of uncertainty is definitively transdiagnostic (r=0.51 across 181 studies and 52,402 participants; McEvoy 2019, PMID 31678816) and its correlation with GAD (0.57) barely exceeds that with depression (0.53) or OCD (0.50) (Gentes 2011, PMID 21664339). Contrast avoidance appears across GAD, depression and social anxiety (Newman 2023, PMID 36565682). If no mechanism is specific, the models describe the internalizing spectrum rather than a disorder. → mechanism-and-models.md
OQ-31. Why is GAD-specific qualitative evidence sparse, and why is there no current APA guideline?¶
The live audit found several GAD-specific qualitative studies rather than one (PMIDs: 31908140, 33218311, 35716017, 21644188, 41587136), but they remain a very small, treatment- and life-stage-concentrated literature. A PubMed search on 2026-09-02 still found no current APA practice guideline for GAD; the 1998 anxiety guideline covers panic disorder (PMID 9585731). → patient-experience-and-advocacy.md, guidelines.md
OQ-32. Should the associated-symptom criteria be weighted rather than counted?¶
Only excessive worry and muscle tension were specific to GAD in a community sample; restlessness, concentration problems and fatigue were more prevalent in major depression (Faravelli 2012, PMID 22578985). A weighted or reordered criterion set has never been proposed, let alone tested. → diagnosis-and-classification.md, the-diagnostic-boundary.md
Note to the next sweep¶
Re-run every asserted absence and re-date it. After the audit, the standing gaps are: no treatment-response reanalysis by excessiveness status (OQ-1); no comorbidity-stratified treatment estimate (OQ-2); no screening trial with clinical outcome endpoints (OQ-3); no fixed-versus-indefinite continuation trial (OQ-4); no collaborative-care dismantling trial (OQ-8); no preference design supporting causal drug-versus-therapy inference (OQ-9); no GAD-specific anchor-based HAM-A MCID (OQ-26); no systematic audit of GAD trial reporting (OQ-27); no current APA GAD guideline (OQ-31); no general-adult GAD core outcome set (OQ-40); no adult equivalent of the CAMELS joint response/modality analysis (OQ-38); and no intervention specifically targeting persistence (OQ-39). A perinatal GAD pharmacotherapy RCT also remained absent in the 2026-09-02 search (OQ-24). Where a source pools anxiety disorders, say so—a pooled anxiety effect size is not a GAD effect size.