Skip to content

Screening and measurement

TL;DR — The GAD-7 is the most widely deployed mental-health instrument in primary care after the PHQ-9, and the 2025 Cochrane review of its accuracy is sobering: across 48 studies, 19,228 participants, 27 countries and 24 languages, at the recommended cut-off ≥10 the GAD-7 detects generalised anxiety disorder with sensitivity 0.64 (95% CI 0.56–0.72) and specificity 0.91 (0.87–0.93) — i.e., it misses roughly a third of cases — and for any anxiety disorder sensitivity falls to 0.48 (0.40–0.57) (Aktürk 2025, PMID 40130828). The two-item GAD-2 performs statistically indistinguishably (sens 0.68, spec 0.86 for GAD). The original validation reported sensitivity 89% / specificity 82% at ≥10 in 965 US primary-care patients (Spitzer 2006, PMID 16717171); the gap between 89% and 64% is the difference between a development sample and 35 heterogeneous replications. The USPSTF nonetheless recommends screening adults, including pregnant and postpartum people (B recommendation, moderate certainty of moderate net benefit) while finding evidence insufficient in adults ≥65 (USPSTF 2023, PMID 37338866) — a recommendation made with no trial showing that screening improves outcomes (O'Connor 2023, PMID 37338868). Two other measurement facts matter: the GAD-7 shows cultural bias (Black/African American participants with high symptoms scored lower than others at equivalent symptom levels; Parkerson 2015, PMID 25725310), and the minimal clinically important difference is about 4 points (Toussaint 2020, PMID 32090765).

The instrument landscape

Instrument Items Role Key performance Source
GAD-7 7 Screen + severity ≥10: sens 0.64 (0.56–0.72), spec 0.91 (0.87–0.93) for GAD; sens 0.48 for any anxiety disorder; AUC 0.86 (0.84–0.88) for GAD Aktürk 2025, PMID 40130828
GAD-2 2 Ultra-brief screen ≥3: sens 0.68 (0.59–0.75), spec 0.86 (0.82–0.89) for GAD; AUC 0.82; no statistically significant difference from GAD-7 Aktürk 2025, PMID 40130828
HADS-A 7 Screen in medical settings ≥8: sens 0.82 (0.76–0.87), spec 0.74 (0.70–0.77) for GAD; AUC 0.82. In 1,000 people at 17% prevalence: 325 test positive, of whom 199 are false positives Fomenko 2025, PMID 40600405
PSWQ 16 Trait worry, the GAD-specific construct Discriminates college samples meeting all / some / no DSM-III-R GAD criteria, and GAD from PTSD; in 34 GAD patients it did not correlate with other anxiety or depression measures — tapping an independent construct Meyer 1990, PMID 2076086
PSWQ-A / abbreviated 8/3 Brief worry Psychometrically evaluated for abbreviated and ultra-brief forms Kertz 2014, PMID 24932640
HAM-A 14 Clinician-rated severity; the primary endpoint of nearly every GAD drug trial Proposed severity bands: none/minimal ≤7, mild 8–14, moderate 15–23, severe ≥24 (n=144; correspond closely to CGI-S) Matza 2010, PMID 20718076
PHQ-15 / PHQ-9 companions 15 / 9 Somatic and depressive symptoms alongside GAD-7 Optimal cut-points ≥10 on parent scales, ≥3 on ultra-brief versions; 5/10/15 = mild/moderate/severe Kroenke 2010, PMID 20633738
IAQ — ICD-11-aligned self-report GAD measure Unidimensional by IRT; 7.1% met the ICD-11 algorithm in a UK-representative sample of 2,058 Shevlin 2023, PMID 36215152

Note what the top two rows imply jointly: adding five items to the GAD-2 buys no measurable accuracy. The GAD-7's advantage over the GAD-2 is severity grading and change-sensitivity, not detection.

The accuracy story, in order

  1. Development (2006). 2,740 patients across 15 US primary-care clinics, 965 with a blinded telephone interview by a mental health professional: a cut-point optimising sensitivity (89%) and specificity (82%); increasing scores strongly associated with all six SF-20 functional scales and with disability days; factor analysis separated GAD and depression symptoms as distinct dimensions with independent effects on impairment (Spitzer 2006, PMID 16717171).
  2. Extension to other anxiety disorders (2007). In the same programme, GAD-7 and GAD-2 performed well (AUC 0.80–0.91) as screens for all four of GAD, panic, social anxiety and PTSD; 19.5% of primary-care patients had ≥1 anxiety disorder and 41% of those reported no current treatment (Kroenke 2007, PMID 17339617).
  3. Rational Clinical Examination (2014). Across 10 studies, the GAD-7 was the best-performing screen for GAD (LR+ 5.1, 95% CI 4.3–6.0; LR− 0.13, 0.07–0.25) — but the review's own caveat was that neither instrument had been replicated in more than one primary-care population (Herr 2014, PMID 25058220).
  4. Single-country validations. In 178 Spanish primary-care patients against the CIDI, a computerised GAD-7 at ≥10 gave sensitivity 0.87, specificity 0.78, PPV 0.93, NPV 0.64 (Muñoz-Navarro 2017, PMID 28666201) — much closer to the development sample than to the pooled estimate.
  5. Cochrane (2025). 48 studies, 19,228 participants: the pooled figures above, with a 95% prediction region for GAD-7/GAD wider than for the other three analyses, i.e. pronounced heterogeneity. Sensitivity tended to be higher and specificity lower in populations with specific medical conditions. The authors describe their own summary estimates as "rough averages" whose performance "may deviate substantially" in specific situations (Aktürk 2025, PMID 40130828).

The trajectory — 89% → 87% in a single validation → 64% pooled — is the standard optimism-decay curve of a diagnostic instrument, and it matters because service systems set thresholds using the development figure.

What screening does and does not buy

The USPSTF evidence report is explicit: only two studies evaluated screening itself, and neither found benefit; the recommendation rests on the accuracy of the instruments plus the separate, larger evidence that treatment works (O'Connor 2023, PMID 37338868). The 2020 Women's Preventive Services Initiative review reached the same structural conclusion — "no studies evaluated the overall effectiveness or harms of screening" (Nelson 2020, PMID 32510989).

Body Population Recommendation Basis
USPSTF 2023 (PMID 37338866) Adults ≥19, including pregnant and postpartum Screen (B) — moderate certainty of moderate net benefit Instrument accuracy + treatment efficacy
USPSTF 2023 (PMID 37338866) Adults ≥65 I statement — evidence insufficient Older adults under-represented in accuracy studies
USPSTF 2022 (PMID 36219403) Children/adolescents 8–18 Screen (B) 7.8% current anxiety-disorder prevalence in 3–17-year-olds (2018–2019 NSCH)
USPSTF 2022 (PMID 36219403) Children ≤7 I statement —

The late-life I statement drew a direct published objection: anxiety in older adults is argued to be under-studied rather than absent, and the absence of evidence is itself the problem the recommendation entrenches (Andreescu 2023, PMID 36652241). See special populations.

The unanswered question is not accuracy but consequence: whether deploying a 64%-sensitive screen at population scale changes who receives treatment and what happens to them. That is a trial-shaped question and no trial has answered it (OPEN-QUESTIONS.md, OQ-3).

Measurement properties that change how results should be read

  • Change sensitivity and MCID. In 261 patients from a multisite chronic-depression trial, GAD-7 scores fell significantly in HRSD-24 improvers (12 wk ES −0.51, SRM −0.57; 48 wk ES −1.0, SRM −1.7) and rose in worseners; the estimated MCID is 4 points (Toussaint 2020, PMID 32090765). Caveat stated by the authors: this was a depression trial, so replication in anxiety-specific treatment is needed — an example of the pooling hazard reaching even the measurement layer.
  • Factor structure and invariance. A one-factor model fits best, with full strong invariance across language (English/French), partial strong invariance across sex, and strong invariance across time (ages 30→35) in 799 Canadian young adults (Riglea 2025, PMID 39880312).
  • Cultural bias. In White/Caucasian, Hispanic and Black/African American undergraduates, a modified one-factor model fitted across groups, but Black/African American participants with high GAD symptoms scored lower on the GAD-7 than others at equivalent symptom levels (Parkerson 2015, PMID 25725310). Differential item functioning of this direction inflates apparent between-group prevalence differences in the opposite direction to the usual assumption.
  • Age-related item functioning. Older adults show DIF for distress/interference (higher) and fatigue (lower) at equal latent GAD severity (Correa 2019, PMID 31938010).
  • Screening-scale prevalence ≠ disorder prevalence. 18.2% of US adults reported any past-two-week anxiety symptoms on the GAD-7 in 2022 (Terlizzi 2024, PMID 39591466), against 12-month DSM-5 GAD of 1.8% in the WMH surveys (Ruscio 2017, PMID 28297020). Both are correct; conflating them is the single most common error in secondary reporting of GAD statistics.

Performance in specific populations

Pooled accuracy hides substantial variation, and the Cochrane review explicitly warns that performance "may deviate substantially" from the summary in specific situations (Aktürk 2025, PMID 40130828). Where GAD-specific population studies exist, they show exactly that.

Population Finding Source
Pregnant and postpartum women 240 perinatal women (155 pregnant, 85 postpartum) referred for psychiatric consultation, against psychiatrist DSM-IV diagnosis: GAD-7 sensitivity 61.3%, specificity 72.7% at an optimal cut-off of 13 — higher than the standard ≥10 — and the GAD-7 outperformed the Edinburgh Postnatal Depression Scale and its EPDS-3A anxiety subscale across a wider range of cut-offs, identifying GAD more accurately in patients with comorbid depression Simpson 2014, PMID 25161068
Low-income African American and Caucasian older adults Pooled from three community-based late-life worry/anxiety trials: PSWQ-A and GAD-7 both showed adequate one-factor fit, internal consistency in the total sample and in both racial subgroups, and good convergent, discriminant and predictive validity; the Geriatric Anxiety Inventory–short form was acceptable only in the African American subgroup Shrestha 2020, PMID 30810345
Young urban South African women 6,028 women aged 18–28 in Soweto: GAD-7 α=0.84 with reasonable fit for one- and two-factor (somatic/cognitive) models; the authors' own conclusion is that the scales identify those at risk but are not a substitute for diagnostic evaluation Hart 2025, PMID 39321979
Populations with specific medical conditions Across the Cochrane analyses, sensitivity tended to be higher and specificity lower in participants with specific conditions than in the other two settings; the most-studied groups were epilepsy (9 studies), cancer (5), cardiovascular disease (5) and general primary care (5) Aktürk 2025, PMID 40130828

The perinatal cut-off finding matters practically: the standard threshold of ≥10, applied to pregnant and postpartum women, is not the threshold that optimised performance in the one perinatal validation located (Simpson 2014, PMID 25161068) — and USPSTF recommends screening this exact population (USPSTF 2023, PMID 37338866).

The ultra-brief layer

Instrument Composition Evidence
PHQ-4 PHQ-2 (depression) + GAD-2 (anxiety) 2,149 patients across 15 US primary-care clinics: factor analysis confirmed two discrete factors explaining 84% of total variance; increasing scores strongly associated with functional impairment, disability days and health-care use; anxiety had a substantial effect on functional status independent of depression (Kroenke 2009, PMID 19996233)
PHQ-4, general population Same Nationally representative German household survey, 5,030 participants (response 72.9%): two-factor solution with very good fit (RMSEA 0.027, 90% CI 0.023–0.032), structurally invariant across age and gender; GAD-2 score of 3 = 95.2nd percentile, score of 5 = 99.2nd percentile. Limitation stated by the authors: no criterion-standard diagnostic interview (Löwe 2010, PMID 19616305)

The percentile normative data are the useful and rarely-used part: a GAD-2 of 3 — the recommended screening cut-off — places a person in the top 5% of the general population, and a GAD-2 of 5 in the top 1% (Löwe 2010, PMID 19616305). That is a different framing of the same threshold from sensitivity and specificity, and it makes the base-rate problem visible.

Measurement-based care: measuring is not the same as acting

Measuring symptoms repeatedly is only useful if the measurement changes what is done. That link has been tested mostly outside GAD:

  • A systematic review of 103 studies of ecological momentary assessment and remote measurement-based care, of which 15 used RMBC and 9 RCTs were meta-analysed, is the current synthesis (Machleid 2026, PMID 41603797).
  • In depression, a multicentre randomised trial has tested measurement-based care to enhance antidepressant outcomes directly (Husain 2025, PMID 40892412).
  • The GAD-7's change sensitivity supports this use — ES −0.51 at 12 weeks and −1.0 at 48 weeks in HRSD-24 improvers, with an MCID of 4 points (Toussaint 2020, PMID 32090765) — but that evidence too comes from a chronic-depression trial.

So the instrument most used for monitoring GAD has had its responsiveness established in depression, and the care model that would justify monitoring has been tested in depression. Both gaps are GAD-specific and both are closable (OPEN-QUESTIONS.md).

The endpoint problem in trials

Nearly every GAD pharmacotherapy trial uses the HAM-A as its primary endpoint, and the largest network meta-analysis reports its results as mean differences in HAM-A change — duloxetine −3.13 (95% CrI −4.13 to −2.13), quetiapine −3.60 (−4.83 to −2.39) (Slee 2019, PMID 30712879). Against Matza's severity bands (mild 8–14, moderate 15–23, severe ≥24; Matza 2010, PMID 20718076), a 3-point HAM-A change is well under the width of a single severity band. Whether a 2.5–3.6-point HAM-A difference is clinically meaningful is therefore a live measurement question, not merely a statistical one (SSRI and SNRI pharmacotherapy, clinical trials landscape).

Psychological-treatment trials predominantly use the PSWQ instead, which measures the GAD-specific construct rather than general anxiety — the CBT-in-older-adults meta-analysis required PSWQ or PSWQ-A as an inclusion criterion (Hall 2016, PMID 27687212). Drug trials and therapy trials therefore do not share a primary outcome, which is one reason head-to-head comparison rests on network meta-analysis with all its assumptions (cognitive behavioural therapy).

Open questions

  • Does screening change outcomes, or only detection? Two studies, no benefit found, and a B recommendation issued anyway (O'Connor 2023, PMID 37338868; USPSTF 2023, PMID 37338866).
  • Why is pooled GAD-7 sensitivity 0.64 when single-population validations repeatedly reach 0.87–0.89 (Spitzer 2006, PMID 16717171; Muñoz-Navarro 2017, PMID 28666201)? Case-mix, reference standard and prevalence are all candidates, and the Cochrane heterogeneity analysis could not resolve it (Aktürk 2025, PMID 40130828).
  • If GAD-2 ≈ GAD-7 for detection (Aktürk 2025, PMID 40130828), what justifies the extra five items outside severity monitoring?
  • How broadly does GAD-7 measurement invariance generalise? One young-adult study found partial strong invariance across sex, time and language (Af Winklerfelt Hammarberg 2025, PMID 39880312), and a clinical sample of 165,872 treatment seekers supported invariance between males and females (Saunders 2023, PMID 37118684). Parkerson 2015 found item-level racial DIF in a US undergraduate sample (PMID 25725310). Cross-cultural and diagnosed-GAD invariance therefore remain open rather than wholly unstudied.
  • Should the perinatal GAD-7 threshold be 13 rather than 10 (Simpson 2014, PMID 25161068), given that USPSTF recommends screening pregnant and postpartum people at the standard cut-off (USPSTF 2023, PMID 37338866)?
  • Does measurement-based care improve GAD outcomes? The model has been tested in depression, not GAD (Husain 2025, PMID 40892412; Machleid 2026, PMID 41603797).
  • What is the anchor-based MCID for the HAM-A in GAD, and how does it compare with the 2.5–3.6-point differences that drive treatment rankings (Slee 2019, PMID 30712879)? A targeted PubMed search rerun on 2026-09-02 located no GAD-specific anchor-based validation; the 2.5-point value prespecified in the MM120 trial is a design assumption, not a validation study (Robison 2025, PMID 40906494).

References

  1. Aktürk Z, et al. Generalized Anxiety Disorder 7-item (GAD-7) and 2-item (GAD-2) scales for detecting anxiety disorders in adults. Cochrane Database Syst Rev. 2025;3:CD015455. PMID 40130828
  2. Spitzer RL, et al. A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch Intern Med. 2006;166:1092-7. PMID 16717171
  3. Kroenke K, et al. Anxiety disorders in primary care: prevalence, impairment, comorbidity, and detection. Ann Intern Med. 2007;146:317-25. PMID 17339617
  4. Kroenke K, et al. The Patient Health Questionnaire Somatic, Anxiety, and Depressive Symptom Scales: a systematic review. Gen Hosp Psychiatry. 2010;32:345-59. PMID 20633738
  5. Herr NR, et al. Does this patient have generalized anxiety or panic disorder? The Rational Clinical Examination systematic review. JAMA. 2014;312:78-84. PMID 25058220
  6. Muñoz-Navarro R, et al. Screening for generalized anxiety disorder in Spanish primary care centers with the GAD-7. Psychiatry Res. 2017;256:312-317. PMID 28666201
  7. Fomenko A, et al. Hospital Anxiety and Depression Scale Anxiety subscale (HADS-A) for detecting anxiety disorders in adults. Cochrane Database Syst Rev. 2025;7:CD015456. PMID 40600405
  8. Meyer TJ, et al. Development and validation of the Penn State Worry Questionnaire. Behav Res Ther. 1990;28:487-95. PMID 2076086
  9. Kertz SJ, et al. Psychometric properties of abbreviated and ultra-brief versions of the Penn State Worry Questionnaire. Psychol Assess. 2014. PMID 24932640
  10. Matza LS, et al. Identifying HAM-A cutoffs for mild, moderate, and severe generalized anxiety disorder. Int J Methods Psychiatr Res. 2010;19:223-32. PMID 20718076
  11. Shevlin M, et al. The development and initial validation of self-report measures of ICD-11 depressive episode and generalized anxiety disorder. J Clin Psychol. 2023;79:854-870. PMID 36215152
  12. US Preventive Services Task Force. Screening for Anxiety Disorders in Adults: US Preventive Services Task Force Recommendation Statement. JAMA. 2023;329:2163-2170. PMID 37338866
  13. US Preventive Services Task Force. Screening for Anxiety in Children and Adolescents: US Preventive Services Task Force Recommendation Statement. JAMA. 2022;328:1438-1444. PMID 36219403
  14. O'Connor EA, et al. Anxiety Screening: Evidence Report and Systematic Review for the US Preventive Services Task Force. JAMA. 2023;329:2171-2184. PMID 37338868
  15. Nelson HD, et al. Screening for Anxiety in Adolescent and Adult Women: A Systematic Review for the Women's Preventive Services Initiative. Ann Intern Med. 2020;173:29-41. PMID 32510989
  16. Andreescu C, et al. Is Anxiety in Late Life an Uncharted Territory? — Questioning the USPSTF Draft Recommendation Statement for Anxiety Screening in Older Adults. JAMA Psychiatry. 2023;80:197-198. PMID 36652241
  17. Toussaint A, et al. Sensitivity to change and minimal clinically important difference of the 7-item Generalized Anxiety Disorder Questionnaire (GAD-7). J Affect Disord. 2020;265:395-401. PMID 32090765
  18. Riglea T, et al. Factor structure and measurement invariance of the GAD-7 across time, sex, and language in young adults. J Affect Disord. 2025;375:359-365. PMID 39880312
  19. Parkerson HA, et al. Cultural-based biases of the GAD-7. J Anxiety Disord. 2015;31:38-42. PMID 25725310
  20. Correa JK, Brown TA. Expression of Generalized Anxiety Disorder Across the Lifespan. J Psychopathol Behav Assess. 2019;41:53-59. PMID 31938010
  21. Terlizzi EP, Zablotsky B. Symptoms of Anxiety and Depression Among Adults: United States, 2019 and 2022. Natl Health Stat Report. 2024. PMID 39591466
  22. Ruscio AM, et al. Cross-sectional Comparison of the Epidemiology of DSM-5 Generalized Anxiety Disorder Across the Globe. JAMA Psychiatry. 2017;74:465-475. PMID 28297020
  23. Slee A, et al. Pharmacological treatments for generalised anxiety disorder: a systematic review and network meta-analysis. Lancet. 2019;393:768-777. PMID 30712879
  24. 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
  25. Simpson W, et al. Comparative efficacy of the generalized anxiety disorder 7-item scale and the Edinburgh Postnatal Depression Scale as screening tools for generalized anxiety disorder in pregnancy and the postpartum period. Can J Psychiatry. 2014;59:434-40. PMID 25161068
  26. Shrestha S, et al. Psychometric properties of worry and anxiety measures in a sample of african american and caucasian older adults. Aging Ment Health. 2020;24:315-321. PMID 30810345
  27. Hart C, et al. Examining the psychometric properties of the Patient Health Questionnaire-9 and Generalized Anxiety Disorder-7 among young urban South African women. J Affect Disord. 2025;369:61-70. PMID 39321979
  28. Kroenke K, et al. An ultra-brief screening scale for anxiety and depression: the PHQ-4. Psychosomatics. 2009;50:613-21. PMID 19996233
  29. Löwe B, et al. A 4-item measure of depression and anxiety: validation and standardization of the Patient Health Questionnaire-4 (PHQ-4) in the general population. J Affect Disord. 2010;122:86-95. PMID 19616305
  30. Machleid F, et al. Remote Measurement-Based Care Interventions for Mental Health: Systematic Review and Meta-Analysis. JMIR Ment Health. 2026;13:e63088. PMID 41603797
  31. Husain MI, et al. Measurement-Based Care to Enhance Antidepressant Treatment Outcomes in Major Depressive Disorder: A Randomized Clinical Trial. JAMA Netw Open. 2025;8:e2529427. PMID 40892412
  32. Saunders R, et al. Measurement invariance of the PHQ-9 and GAD-7 across males and females seeking treatment for common mental health disorders. BMC Psychiatry. 2023;23:298. PMID 37118684
  33. Robison R, et al. Single Treatment With MM120 (Lysergide) in Generalized Anxiety Disorder: A Randomized Clinical Trial. JAMA. 2025;334:1358-1372. PMID 40906494