About the CDGAD© Scale

Scale Properties

We identified six desirable properties of the CDGAD© scale that make it an effective GAD scale:

  1. First, we wanted a self-rating format, which is more easily and economically utilized than an observer-based measure.
  2. Second, a low cognitive burden was considered critical, i.e. the scale would be easy to understand and quick to complete, regardless of the subject’s level of education or age. To achieve this goal, we designed the CDGAD© items to be answered based on a “Yes” or “No” choice rather than a multiple-choice construction, which is the basis of most comparable scales.
  3. Third, since 50% of the US adult population have functional literacy skills below the 6th grade level, we aimed at easy readability.
  4. Fourth, brevity was a required feature of the scale.
  5. Fifthly, we needed to include sufficient number of items to provide a score that meaningfully captured varying levels of severity from no symptoms to severe GAD symptoms.
  6. Sixthly, we intended for the scale to serve as a diagnostic screener of possible GAD, and therefore included all the GAD symptoms that made up the criteria for this diagnosis in the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition (DSM-IV). Because the current criteria for GAD in the 5th edition of the DSM (DSM-5) are identical to those of DSM-IV, apart from a re-ordering of Criteria D, E and F, we see no reason why the performance and usefulness of the CD-GAD relative to DSM-5 should be any different from the results of testing the scale with reference to DSM-1V.

Scale Development

The project was initiated and led by Dr. Bernard J. Carroll, formerly Chair of the Department of Psychiatry at Duke University, and with assistance of Dr. George Parkerson, former Chair of Family and Community Medicine, and other colleagues in the Department of Psychiatry Drs. William Wilson, Kathryn Connor and Jonathan Davidson. Drs. John Beyer, Leslie Forman and Kishore Gadde provided critical review of the scale. The study was supported by an unrestricted grant from Wyeth.

The process of development and results have been described by Carroll et al (2002, unpublished report) and are given here. Over a four-month period, our group met to propose initial symptom statements, which were then given to three experienced psychiatrists for their consideration. Approval was then obtained from the Duke University Institutional Review Board (IRB) to approach a group of patients with the diagnosis of GAD, who then provided feedback on the suitability of statements and proposed alternative wording from their experience of the disorder. Readability of the scale was assessed with Microsoft Office 2000 software. After this review, the final scale was submitted to field testing, and a Spanish translation, verified by back-translation was also prepared (Bobes et al, 2006).

Scale Construction

The final scale contains 12 symptom statements covering all DSM-IV and DSM-5 criteria (Carroll et al, 2002 unpublished report). Two alternative statements are provided for the mandatory single symptoms of Criterion A (excessive worry/anxiety), Criterion B (inability to stop worrying), and E (DSM-IV) or D (DSM-5) (significant distress or impairment). One statement is provided for each of the 6 symptoms of Criterion C, at least three of which are required for the diagnosis of GAD. The DSM requirement of six months symptom duration was specified in an explanatory sentence before the printed symptoms. The total score is determined as the sum of affirmative responses in the “YES” column.

Field Testing

Clinical and population norms were determined by field testing in 400 subjects comprising the primary care medical setting (PC, n=198), psychiatric outpatient clinic (POP, n=102), and healthy volunteers (HV, n=100). All subjects provided informed consent following approval of this phase of the project by the Duke (DUMC) IRB.

Sample Identification

DUMC primary care clinic subjects (n=198) with at least minimal worry or nervousness at a screening contact by phone or face-to-face were identified and scheduled for a more in-depth interview. A convenience sample of 108 psychiatric outpatients were identified from physician referral, and 102 healthy volunteers were identified by word of mouth in the community.

Psychometric Assessment

Face validity was established by the scale’s concordance with DSM inclusion criteria for GAD.

Convergent validity was tested by comparing the CD-GAD scale score and diagnostic index (DI) relative to an independent GAD diagnosis from use of the MINI Structured Interview of Sheehan et al, 1998). Convergent validity was further tested against the Duke Health Profile (Parkerson et al, 1990).

Discriminant validity was evaluated against the Brief Carroll Depression Scale (BCDS) (Carroll, 1998).

Internal reliability was assessed by Cronbach’s alpha.

Test-retest reliability was assessed in a subgroup of 44 patients whose clinical state had not changed based on another GAD scale, the Clinical Anxiety Scale (CAS) (Snaith et al, 1982) between times 1 and 2 seven days later. Test-retest intra-class correlations were determined on the total score and the DI was evaluated by kappa coefficient.

Other Evaluations

Associations between the scale score and demographics, and sample type were assessed using ANOVA and Chi-Squared testing.

Conditional probability analysis was used to compare the scale’s DI by crosswalk to DSM-IV MINI diagnoses of GAD. We computed positive predictive value (PPV), negative predictive value (NPV), sensitivity, specificity, positive likelihood ratio (PLR), odds ratio of posterior to prior probabilities and Cohen’s kappa coefficient of concordance. These measures were examined, in addition, for the total score and the individual statements of the scale. The same measures were computed for each of Criteria A, B, C and E of DSM-IV (or for Criteria A, B, C and D as they are in DSM-5). Receiver Operating Characteristic (ROC) analysis was applied to the total CDGAD© score in relation to MINI diagnoses of GAD to explore the optimal cut-point based on the number of positive responses, independently of the DSM-IV algorithm.


Sample profile: 400 subjects were assessed, mean age 42.5 years, with variation by site. Sex distribution was 27% male and 73% female. Race proportions were 74% white, 23% black, 3% other. Fifty-two percent were married, 32% single, 8% divorced/separated and 8% other. Less than HS education was 6%, 13% HS graduates, 14% some college, 44% college graduates and 23% postgraduate experience.

Readability: Brevity was achieved with a scale length of 123 words. The Flesch Reading Ease level is high at 87% and the Flesch-Kincaid Reading Grade Level is 2.7.

Scale Reliability: Cronbach’s overall alpha coefficient was 0.92. By site it was 0.59 in HV, 0.91 in PC and 0.90 in POP samples. The test-retest reliability intraclass correlation coefficient was 0.94. Test-retest kappa coefficients for the individual items ranged from 0.50 – 0.77. The median test-retest kappa was 0.66. The test-retest kappa for the DI by crosswalk to DSM-IV was 0.74. The scale-derived GAD diagnosis was unchanged in 89% of the 44 cases.

MINI Diagnoses: All HVs were free of the diagnosis, 13% of the PC sample and 34% of the POP were positive for a MINI GAD diagnosis. In MINI +ve cases compared to GAD -ve, the mean CDGAD© score was 8.97 and 2.50 (p<0.0001). By site, CDGAD scores differed significantly as expected: all (n=400) mean = 3.5, for HVs the mean = 0.57, for PC = 3.7 and for POP = 6.0. This range of score was matched by the performance of the GAD standard comparator, the CAS.

The CAS correlated with the CDGAD© in all subjects (r = 0.774) as did the CDGAD© and the BCDS (r = 0.734), which reflects the fact that those with GAD exhibit a considerable degree of depression and that there is overlap between constructs.

Diagnostic Agreement: For the full sample and both clinical samples, Table 1 shows the sensitivity, specificity, PLR, PPV%, NPV%, prior probability and posterior probability of GAD, odds ratio, kappa and diagnostic efficiency. These figures are shown for all sites combined and separately for the primary care (PC), psychiatric outpatient (POP), PC + POP groups.

Table 1: GAD Scale Diagnostic Index in Relation to MINI Diagnoses of GAD

Sensitivity 0.64 0.58 0.69 0.64
Specificity 0.9 0.89 0.78 0.86
PLR 6.3 5.2 3.1 4.5
PPV % 53.4 44.1 61.5 53.4
NPV % 93.2 93.2 82.5 93.5
Prior P of GAD 0.15 0.13 0.34 0.21
Posterior P of GAD 0.53 0.44 0.61 0.53
Odds Ratio 6.34 5.19 3.06 4.44
Kappa 0.5 0.41 0.45 0.46
Dx Efficiency 0.86 0.85 0.75 0.81


Dx Efficiency – Diagnostic Efficiency

NPV – Negative Predictive Value

P – Probability

PC – Primary Care

PLR – Positive Likelihood Ratio

POP – Psychiatry Outpatient

PPV – Positive Predictive Value

The sensitivity and specificity are acceptable, and the PLR associated with a positive DI (i.e. the ratio of true positive: false positive cases) in the full sample = 6.3, and in the PC sample = 5.2. PLR represents the quotient of sensitivity divided by 1-specificity and, like sensitivity and specificity, are prevalence-free measures of performance. The remaining measures of PPV, NPV, odds ratio and kappa are influenced by base rate of the disorder in the sample, which was 0.13 (13%) in PC group.

The findings also suggest that the CDGAD© is suitable to screen for GAD in conditions where the base rate is low. This is shown by the fact that if the HV and PC samples are pooled, the PLR and odds ratio were both = 8.2. By contrast, and as would be expected, in the POP sample, with higher rates of GAD and kindred conditions, the lower specificity resulted in a lower PLR and OR each = 3.1.

Performance of Individual GAD Scale Item Groups: As shown in Table 2, the items were studied for their agreement with the MINI-derived diagnosis of GAD. Best concordance was attained for all items together, which yielded sensitivity of 0.64 and specificity of 0.90. Items 3 and 4, i.e. DSM-IV/5 Cluster B, came a close second, which suggests that the symptom of uncontrollable worry is a key defining feature of GAD, with considerably better specificity than other symptoms.

Table 2: GAD Scale Item Groups in Relation to MINI Diagnoses of GAD (Total Sample)

Items 1-2 for
Criterion A
0.84 0.74 37 96.2 3.2 0.38
Items 3-4 for
Criterion B
0.77 0.86 49.5 95.4 5.4 0.51
Items 5-10 for
Criterion C
0.9 0.73 37.7 98 3.3 0.4
Items 11-12 for
Criterion E
0.85 0.72 35.9 96.4 3.1 0.37
by crosswalk
0.64 0.9 53.4 93.2 6.3 0.5


NPV – Negative Predictive Value

PLR – Positive Likelihood Ratio

PPV – Positive Predictive Value

Receiver Operating Characteristic (ROC) Analysis in Relation to MINI Diagnosis: ROC analysis was used to estimate concordance of the new scale with MINI diagnosis (Table 3). The optimal CDGAD© cut-point was a score of 6 or greater suggesting the presence of GAD, and less than 6 its absence. Here a score of 6 could be obtained from “YES” responses to any of the 12 items. The ROC curve showed that the scale powerfully discriminates GAD +ve from GAD -ve cases on the MINI, with an area-under-the-curve (AUC) = 0.91.

Table 3: Comparison of ROC Solution and Diagnostic Index by Crosswalk in Relation to MINI Diagnoses of GAD (Total Sample)

6 and
0.84 0.83 47.7 96.6 5 5.05 0.5
0.64 0.9 53.4 93.2 6.3 6.34 0.5


NPV – Negative Predictive Value

PLR – Positive Likelihood Ratio

PPV – Positive Predictive Value

Diagnostic Index by Crosswalk in Relation to MINI GAD Diagnosis: Table 3 shows a sensitivity = 0.64, specificity = 0.90 for the CDGAD “YES” responses matching the DSM requirements, i.e. to derive from all the four DSM GAD criteria.

Of note for primary care, our ROC and Diagnostic Crosswalk findings demonstrated that the CDGAD© scale was superior to the mental health subscale of the Duke Health Profile, used in the study. Thus, the new scale does not duplicate existing instruments for screening in primary medical care.


The following pages on this site provide more information about CDGAD©:

To learn more about GAD, visit the National Institute of Mental Health’s GAD page.