GAD-7

GAD-7 in Mental Health Trials, Services and Economic Evaluation

The GAD-7 was developed by Spitzer, Kroenke, Williams and Löwe in United States primary care as a brief way to identify probable cases of generalised anxiety disorder and to grade how severe the symptoms are. It is now used as an anxiety outcome in treatment trials and in routine services such as NHS Talking Therapies. For a health economist the GAD-7 matters in two places. In England's NHS Talking Therapies programme the GAD-7 helps decide who counts as a clinical case at the start of treatment, and therefore who is counted as recovered at the end, so GAD-7 scores sit underneath the recovery rates that NHS England reports for the programme. In trials, the GAD-7 may be the anxiety outcome that an economic evaluation has to translate into something comparable across conditions, and because its scores carry no preference weights that translation needs either a separate preference-based measure or a mapping function. Here GAD-7 names the questionnaire, while generalised anxiety disorder is the diagnosis it screens for.

Items, response scale and scoring of the GAD-7

The GAD-7 is a short self-report patient-reported outcome measure. Its development and scoring rules come from the original validation study, which Spitzer and colleagues ran in 15 primary care clinics in the United States between November 2004 and June 2005.

The developers began with a pool of 13 items. Nine reflected the DSM-IV symptom criteria for generalised anxiety disorder and four were drawn from existing anxiety scales. The published GAD-7 keeps the seven items with the highest correlation with the total score on that 13-item scale. Each item asks how often, over the previous two weeks, the respondent has been bothered by a symptom. The four response options are "not at all", "several days", "more than half the days" and "nearly every day", scored 0, 1, 2 and 3.

The total is the simple sum of the seven item scores.

$$S = \sum_{i=1}^{7} x_i$$

where $S$ is the GAD-7 total score and $x_i$ is the score from 0 to 3 on item $i$. The total therefore runs from 0 to 21, with higher scores meaning more frequent anxiety symptoms. No item is weighted more than another, and no value set or tariff converts the total into a utility. The GAD-7 is a non-preference-based measure of symptom frequency, much like most disease-specific instruments used as trial outcomes.

Cut points and accuracy against a diagnostic interview

In the validation study, 2,740 adult patients completed the questionnaire and 965 of them had a telephone interview with a mental health professional within one week. The interview diagnoses served as the reference standard for judging the scale's sensitivity and specificity at different thresholds.

Spitzer and colleagues reported that at a cut point of 10 or greater both sensitivity and specificity exceed 0.80, and that sensitivity is nearly maximised there. At a score of 10 the scale had a sensitivity of 89 per cent, a specificity of 82 per cent and a positive predictive value of 29 per cent for generalised anxiety disorder. The low positive predictive value follows from prevalence. In a companion analysis of the same 965 patients, Kroenke and colleagues found that 7.6 per cent had generalised anxiety disorder, so even an accurate screen produces many false positives among those who score 10 or more.

The authors also suggested that cut points of 5, 10 and 15 might be interpreted as representing mild, moderate and severe levels of anxiety. These bands were offered as an interpretation of severity, not as separately validated diagnostic thresholds.

Later pooled evidence puts sensitivity lower and specificity higher than the original study did. The evidence review for the US Preventive Services Task Force on anxiety screening reported a pooled sensitivity of 0.79 and a pooled specificity of 0.89 for a GAD-7 threshold of 10 or more in detecting generalised anxiety disorder. For a model of a screening or case-finding programme, pooled estimates of this kind, with their uncertainty, are a better input than the single validation study, and a ROC curve shows how the trade-off shifts at other thresholds.

Reliability and general population norms

Reliability limits how much of a change in GAD-7 score can be read as real change. The original study reported excellent internal consistency (Cronbach's alpha of 0.92) and good test-retest reliability (intraclass correlation of 0.83).

Löwe and colleagues then tested the scale in a nationally representative face-to-face household survey of 5,030 people in Germany in May and June 2006. Confirmatory factor analysis supported a single underlying dimension, with factorial invariance across gender and age, and internal consistency was 0.89. GAD-7 scores correlated with the PHQ-2 depression screen at 0.64 and with self-esteem at minus 0.43. About 5 per cent of the general population scored 10 or more and about 1 per cent scored 15 or more.

These norms help with interpretation, since a trial population with a mean baseline score of 12 is above the score of 10 that only about 5 per cent of the German general population reached.

The GAD-7 in NHS Talking Therapies outcome reporting

NHS Talking Therapies for anxiety and depression, the programme formerly called Improving Access to Psychological Therapies (IAPT), is the setting where GAD-7 scores have the most direct link to how services are judged. The programme's manual, published by NHS England, states that patients complete measures of both depression and anxiety at every session.

Table 9 of the manual sets the clinical cut-off and reliable change index for each measure. For the GAD-7 the cut-off for caseness is 8 and the reliable change index is 4. For the PHQ-9 the corresponding figures are 10 and 6. NHS England's service standards describe reliable improvement as a change between pre-treatment and post-treatment scores that exceeds the measurement error of the scales, which is what the reliable change index captures.

NHS England Digital's monthly statistics define recovery in terms of caseness. A referral has recovered when it was a clinical case at the start of treatment and is not a clinical case at the end, and it has reliably recovered when it has also shown a significant improvement. Recovery and reliable change are calculated from a pair of measures, one for depression and one for anxiety. The manual's table marks in bold the pair to use for each presenting problem, and when the measure in bold is missing the calculation falls back on the PHQ-9 and GAD-7 combination.

When recovery rates from Talking Therapies data are used as effectiveness inputs in mental health economics models, the analyst inherits these definitions. A recovery rate is a binary summary built on two thresholds and a pairing rule, and changing any of them changes the proportion recovered without any change in patients' health.

Turning GAD-7 results into QALYs for cost-utility analysis

A cost-utility analysis needs health states valued on a scale anchored at full health and death, so that time in each state can be summed into QALYs. The GAD-7 total is a count of symptom frequency with no such anchor, and nothing guarantees that a three-point fall has the same value at every point on the scale. Drummond and colleagues, and Brazier and colleagues, describe two routes for a measure of this kind.

The first route is to collect a preference-based measure such as the EQ-5D alongside the GAD-7 in the same trial or service, and to compute QALYs from that measure. The GAD-7 then serves as the clinical outcome and the preference-based measure as the economic one. The second route, used when only GAD-7 data exist, is to estimate utility values from GAD-7 scores with a mapping function fitted in a dataset where both measures were collected. Mapping depends on the two instruments overlapping in content. The EQ-5D has a single anxiety or depression dimension, so it registers only part of what the GAD-7 records and adds mobility, self-care, usual activities and pain, which the GAD-7 does not cover.

Several choices change the result when GAD-7 data are mapped. The estimation dataset may differ in severity or setting from the population being modelled, the target instrument and its value set may differ from the one a decision maker prefers, and a model that predicts mean utility well can still shrink the spread of individual values towards the mean, which understates uncertainty. A mapping function should therefore be chosen for its fit in a comparable population, and the uncertainty in its predictions should be carried into the probabilistic analysis.

Illustrative worked example of recovery and cost per recovery

All numbers here are illustrative. A service treats 200 patients who all start at or above the GAD-7 caseness cut-off of 8. For simplicity the example classifies outcomes on the GAD-7 alone, whereas NHS Talking Therapies uses a depression and an anxiety measure together.

1. Classifying three patients. Patient A falls from 15 to 6. The change is 15 − 6 = 9 points, above the reliable change index of 4, and the final score is below 8, so patient A has reliably recovered. Patient B falls from 13 to 9. The change is 13 − 9 = 4 points, which meets the reliable change index, but 9 is still at caseness, so patient B has reliably improved without recovering. Patient C falls from 9 to 7. The change is 9 − 7 = 2 points, less than 4, so although the final score is below 8 and patient C counts as recovered, the recovery is not reliable.

2. Service rates. Suppose 110 of the 200 patients recover and 100 reliably recover.

$$r = \frac{R}{N} = \frac{100}{200} = 0.50$$

where $r$ is the reliable recovery rate, $R$ is the number of patients who reliably recovered and $N$ is the number who started at caseness. The recovery rate is 110 ÷ 200 = 0.55.

3. Cost per reliable recovery. At an illustrative GBP 650 per course of treatment, the service costs 200 × 650 = 130,000, so GBP 130,000 in total, and the cost per reliable recovery is 130,000 ÷ 100 = 1,300, or GBP 1,300.

4. Cost per QALY from a separate preference-based measure. Suppose the same patients also completed a preference-based measure, and against an untreated comparator with no cost the mean utility gain is 0.05 sustained for one year. The QALY gain is 200 × 0.05 = 10 QALYs.

$$\text{ICER} = \frac{\Delta C}{\Delta Q} = \frac{130000}{10} = 13000$$

where $\Delta C$ is the incremental cost in GBP and $\Delta Q$ is the incremental QALY gain, so the ICER is GBP 13,000 per QALY.

The two summaries answer different questions. GBP 1,300 per reliable recovery cannot be set against a cost-effectiveness threshold, because a recovery has no value in QALYs and a patient like B, who improves without crossing the cut-off, contributes nothing to it. The ICER in step 4 uses all of the change in every patient but depends entirely on the preference-based measure, not on the GAD-7.

How the GAD-7 differs from nearby measures

Anxiety symptoms are the clinical outcome, and the GAD-7 is one instrument for measuring them. Other anxiety measures use different items and scales, so their scores cannot be pooled with GAD-7 scores without conversion.

The PHQ-9 is the depression measure that NHS Talking Therapies pairs with the GAD-7 when recovery calculations fall back on a standard combination. The two have different caseness cut-offs and reliable change indices, so a fall of 4 points means a reliable change on the GAD-7 but not on the PHQ-9.

A preference-based measure such as the EQ-5D produces utilities directly from a value set, while the GAD-7 describes anxiety in more detail without valuing it. Whether a GAD-7 change is large enough to matter to patients is a separate question of minimal clinically important difference, which the reliable change index does not answer.

Limitations and common misreadings of GAD-7 scores

Several errors recur when GAD-7 results are used in economic work. Most come from treating a threshold or a total score as more than the instrument was built to provide.

• Treating a screening score as a diagnosis. A score of 10 or more identifies probable generalised anxiety disorder, and at a positive predictive value of 29 per cent in the validation sample most people above the cut point did not have the disorder.
• Mixing thresholds. The screening cut point of 10 and the NHS Talking Therapies caseness cut-off of 8 serve different purposes, and recovery rates calculated with one cannot be compared with rates calculated with the other.
• Reading recovery as health gain. A recovery rate counts crossings of a threshold. Two services with the same recovery rate can deliver very different average changes in score, and a service that treats milder cases will find recovery easier to achieve.
• Using the total as a utility. The 0 to 21 total is a sum of ordinal item scores and has no anchor at full health or death, so it cannot be multiplied by time to produce QALYs.
• Ignoring missing data. In its 2015 to 2016 reporting, the IAPT annual report kept referrals with unknown starting caseness in the denominator of the recovery rate, a rule intended to encourage complete recording of questionnaire scores, so the handling of missing scores affects the rate reported.

Sources

• Brazier J, Ratcliffe J, Salomon JA, Tsuchiya A. Measuring and Valuing Health Benefits for Economic Evaluation. 2nd ed. Oxford University Press; 2017.
• Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes. 4th ed. Oxford University Press; 2015.
• NHS Digital. Psychological Therapies: Annual report on the use of IAPT services, England 2015-16. https://files.digital.nhs.uk/pdf/1/0/psyc-ther-ann-rep-2015-16_v2.pdf
• Kroenke K, Spitzer RL, Williams JBW, Monahan PO, Löwe B. Anxiety disorders in primary care: prevalence, impairment, comorbidity, and detection. Annals of Internal Medicine. 2007;146(5):317-325. https://pubmed.ncbi.nlm.nih.gov/17339617/
• Löwe B, Decker O, Müller S, Brähler E, Schellberg D, Herzog W, Herzberg PY. Validation and standardization of the Generalized Anxiety Disorder Screener (GAD-7) in the general population. Medical Care. 2008;46(3):266-274. https://pubmed.ncbi.nlm.nih.gov/18388841/
• NHS England. NHS Talking Therapies for anxiety and depression manual, version 7. First published June 2018, updated March 2024. https://www.england.nhs.uk/wp-content/uploads/2018/06/nhs-talking-therapies-manual-v7.1-updated.pdf
• NHS England. NHS Talking Therapies service standards. Web page, accessed 17 October 2026. https://www.england.nhs.uk/mental-health/adults/nhs-talking-therapies/service-standards/
• NHS England Digital. NHS Talking Therapies Monthly Statistics Including Employment Advisors, Performance April 2025, Outcomes. https://digital.nhs.uk/data-and-information/publications/statistical/nhs-talking-therapies-monthly-statistics-including-employment-advisors/performance-april-2025/outcomes
• Spitzer RL, Kroenke K, Williams JBW, Löwe B. A brief measure for assessing generalized anxiety disorder: the GAD-7. Archives of Internal Medicine. 2006;166(10):1092-1097. doi:10.1001/archinte.166.10.1092
• O'Connor E, Henninger M, et al. Anxiety screening: evidence report and systematic review for the US Preventive Services Task Force. JAMA. 2023. https://jamanetwork.com/journals/jama/fullarticle/2806251