GAD-7 (Generalised Anxiety Disorder-7)

Structural validity Highcontestedevidence form: canonical

direct unidimensional evidence includes UK-inclusive nationally representative samples (Shevlin 2022), though much replication is in non-UK, clinical and disease-specific populations

GAD-7 was designed as, and is most often confirmed as, a single-factor (unidimensional) scale. The original development study confirmed anxiety and depression as distinct dimensions (Spitzer 2006), and a nationally representative German general-population confirmatory factor analysis (N=5030) substantiated a one-dimensional structure with factorial invariance for gender and age (Lowe 2008). Unidimensional fit has been replicated across many settings: Cypriot perinatal women (Vogazianos 2022), an Italian coronary heart disease sample (Bolgeo 2023), four nationally representative European samples including the UK (Shevlin 2022), Canadian young adults where a one-factor model fit best (Riglea 2025), and 20 Czech samples (N=5529) supporting a unidimensional structure (Cigler 2026). However, a competing two-factor structure separating cognitive-emotional from somatic items recurs: a network analysis of Spanish primary-care patients revealed a two-factor solution (Moriana 2021), a US college-student study fit both one- and two-factor models (White 2025), a large psychotherapy sample found the scale technically multidimensional though sum scores remained justifiable (Stochl 2020), and a Malaysian study preferred a six-item second-order model (Pheh 2023). The one-factor model dominates practice and fits well in most samples, but the dimensionality question is genuinely unsettled.

Sub-grades (evidence differs by subgroup):

• {"subgroup": "one-factor (unidimensional) model", "grade": "High", "note": "Well replicated across languages and settings; the dominant and recommended scoring model"}
• {"subgroup": "cognitive-emotional vs somatic two-factor model", "grade": "Moderate", "note": "Recurs in network analyses and some CFAs; keeps dimensionality contested rather than settled"}

Convergent and discriminant validity Moderatewell-establishedevidence form: canonical

indirect convergent/discriminant evidence is mostly from general-population, student and clinical non-workplace samples (German general population, US students, disease-specific cohorts)

Convergent validity is consistently supported. In the German general population GAD-7 correlated r=0.64 with the PHQ-2 depression module and r=-0.43 with the Rosenberg Self-Esteem Scale (Lowe 2008). In US college students the total score correlated r=0.70 with the trait scale of the State-Trait Anxiety Inventory (convergent) and only r=-0.04 with a behavioural activation reward subscale (discriminant) (White 2025). Increasing scores were strongly associated with multiple domains of functional impairment in the original study (Spitzer 2006). The recurring discriminant concern is the strong overlap with depression: GAD-7 and PHQ-9 anxiety and depression factors are highly correlated and frequently co-occur (Stochl 2020; Bolgeo 2023), so the scale distinguishes anxiety from unrelated constructs well but discriminates anxiety from depression less cleanly.

Criterion validity: reference standard Highwell-establishedevidence form: canonical

indirect validated against diagnostic interview in primary-care and disease-specific clinical samples, not in workplace populations, and UK-specific diagnostic-accuracy data are a minority of the pooled evidence

Against a structured or semi-structured clinical interview, GAD-7 has been extensively evaluated. The original primary-care study identified a cut-off of 10 with sensitivity 89% and specificity 82% for GAD (Spitzer 2006). An early diagnostic meta-analysis (12 samples, 5223 participants) found pooled sensitivity 0.83 and specificity 0.84 at a cut-off of 8, with cut-offs 7 to 10 performing similarly (Plummer 2016). The most comprehensive synthesis, a 2025 Cochrane review of 48 studies (19,228 participants, 27 countries, 24 languages), reported that at the recommended cut-off of 10 or higher the GAD-7 summary sensitivity was 0.64 (95% CI 0.56 to 0.72) and specificity 0.91 (95% CI 0.87 to 0.93) for detecting GAD, with an area under the curve of 0.86; for detecting any anxiety disorder sensitivity fell to 0.48 (specificity 0.91) (Akturk 2025). The pooled sensitivity at cut-off 10 is therefore markedly lower than the original single-study estimate, with pronounced heterogeneity, and the Cochrane authors caution that the summary estimates are rough averages that may deviate substantially in specific situations. Setting-specific validations against interview (for example Taiwanese epilepsy patients, optimal cut-off 7) add further threshold variability (Shih 2022).

Criterion validity: organisational Absent (a finding about the literature)untestedevidence form: mixed

indirect the closest evidence uses the derivative GAD-2 and self-reported work outcomes, and no study links the full GAD-7 to objective organisational records

No study validating the full GAD-7 against objective organisational outcomes (recorded sickness absence, turnover, or measured job performance) was located in the latest review pass. The nearest evidence is adjacent rather than direct: a study of 4953 working Australians linked probable anxiety to worse self-reported presenteeism and absenteeism on the WHO Health and Work Performance Questionnaire, but it used the 2-item GAD-2, not the full GAD-7, and relied on self-reported rather than employer-recorded outcomes (Deady 2021). A Polish validation among employees related GAD-7 to professional burnout and psychological distress, again self-reported constructs rather than organisational records (Basinska 2023). The original study related GAD-7 to self-reported disability days and functional impairment, not to work-context criterion outcomes (Spitzer 2006). Organisational criterion validity for the fielded GAD-7 is therefore essentially untested.

Internal consistency Highwell-establishedevidence form: canonical

direct UK-inclusive nationally representative samples contribute (Shevlin 2022, Saunders 2023), although most individual alpha estimates come from non-UK or clinical samples

Internal consistency is uniformly high across populations and languages. Cronbach's alpha was 0.89, identical across all gender and age subgroups, in the German general population (Lowe 2008); 0.91 in US college students (White 2025); alpha 0.907 with McDonald's omega 0.909 in Cypriot perinatal women (Vogazianos 2022); alpha 0.89 with composite reliability 0.90 in an Italian cardiac sample (Bolgeo 2023); and 0.928 in Taiwanese epilepsy patients (Shih 2022). Lower but still acceptable values appear in some translations, for example alpha 0.81 in a Swahili HIV sample (Nyongesa 2020) and a median alpha of 0.86 across 20 Czech samples (Cigler 2026). Values consistently sit in the 0.81 to 0.93 range.

Test-retest reliability Lowthin

indirect the original ICC of 0.83 is reported second-hand in retrieved sources without interval or sample size, and the two dedicated coefficients located are from a Kenyan HIV sample and a Czech life-events design, none in a UK working population

Dedicated short-interval test-retest studies of GAD-7 are sparse. The original validation reportedly gave ICC 0.83, but its interval and sample size were not captured in sources retrieved in the latest review pass (the value is cited second-hand in Nyongesa 2020, doi:10.1186/s12991-020-00312-4). A 2-week ICC of 0.59 was found in a Swahili HIV sample (Nyongesa 2020), and the Czech multi-sample study found moderate stability of r=0.46 to 0.53, but over intervals spanning major life events rather than a fixed short retest window (Cigler 2026, doi:10.1016/j.janxdis.2026.103149). Test-retest reliability for a UK working-adult population is untested. The absence of a consistent, short-interval, well-sampled retest estimate is itself the finding.

Measurement invariance Highwell-establishedevidence form: canonical

direct UK treatment-seeking and working-age-versus-older samples are included (Saunders 2023, Delamain 2024), though occupational-group invariance specifically is untested

Measurement invariance is one of the most heavily studied properties of the GAD-7, and it is largely supported across groups but contested over time. Across sex/gender, invariance or an absence of differential item functioning is repeatedly demonstrated, including in very large UK treatment-seeking samples (N=165,872) (Saunders 2023), Canadian adolescents where strict invariance held by sex and grade (Romano 2021), and four European nationally representative samples with no DIF across sex, age or country (Shevlin 2022). Across age, invariance held between UK working-age and older adults with only limited DIF (Delamain 2024). Across language and country, invariance is supported across English and French (Riglea 2025), across 18 countries after traumatic brain injury (Teymoori 2020), and with full scalar invariance in rural India (De Man 2021). Longitudinal invariance is where the literature conflicts: strict temporal invariance was established across 10 psychotherapy sessions (Stochl 2020) and across time in Czech and Canadian samples (Cigler 2026; Riglea 2025), yet longitudinal invariance was NOT established in a partial-hospital sample, implying that raw pre-post change scores may be unreliable in that setting (Ong 2021).

Sub-grades (evidence differs by subgroup):

• {"subgroup": "sex / gender", "grade": "High", "note": "Strongly supported across many samples including large UK datasets"}
• {"subgroup": "age", "grade": "High", "note": "Supported including UK working-age versus older adults (Delamain 2024)"}
• {"subgroup": "language / country", "grade": "Moderate", "note": "Supported across several languages and a UK-inclusive four-country study, but by translation rather than exhaustive"}
• {"subgroup": "longitudinal / over time", "grade": "Low", "note": "Contested: strict temporal invariance in some large samples but not established in a partial-hospital sample, complicating change-score use (Ong 2021)"}
• {"subgroup": "occupation / industry", "grade": "Absent", "note": "No study testing invariance across occupational groups located in the latest review pass"}

Responsiveness and MIC Moderatethinevidence form: canonical

indirect responsiveness evidence comes from a chronic-depression trial and Czech psychotherapy samples, not from UK or workplace populations, and MIC anchors vary by setting

Responsiveness (sensitivity to change) is demonstrated, but formal minimal important change estimates are sparse. In a multisite chronic-depression trial (N=261), GAD-7 scores fell significantly in patients who improved on the Hamilton depression rating (effect size -0.51 at 12 weeks, -1.0 at 48 weeks) and rose in those who worsened, supporting sensitivity to change (Toussaint 2020). The Czech multi-sample study found clear sensitivity to change during psychotherapy and derived a reliable-change threshold of about plus or minus 5.2 points (Cigler 2026). A single, well-anchored minimal important change value validated for a working-adult population was not located in the latest review pass; responsiveness is established while the MIC anchor remains setting-dependent.