The Patient Health Questionnaire Anxiety and Depression Scale (PHQ-ADS): Initial Validation in Three Clinical Trials

Abstract

Objective

We examine the reliability and validity of the Patient Health Questionnaire Anxiety-Depression Scale (PHQ-ADS) – which combines the PHQ-9 and GAD-7 scales – as a composite measure of depression and anxiety.

Methods

Baseline data from 896 patients enrolled in 2 primary-care based trials of chronic pain and 1 oncology-practice based trial of depression and pain were analyzed. The internal reliability, standard error of measurement (SEM), and convergent, construct, and factor structure validity, as well as sensitivity to change of the PHQ-ADS were examined.

Results

The PHQ-ADS demonstrated high internal reliability (Cronbach's alpha of 0.8 to 0.9) in all 3 trials. PHQ-ADS scores can range from 0 to 48 (with higher scores indicating more severe depression/anxiety), and the estimated SEM was approximately 3 to 4 points. The PHQ-ADS showed strong convergent (most correlations 0.7-0.8 range) and construct (most correlations 0.4-0.6 range) validity when examining its association with other mental health, quality of life and disability measures. PHQ-ADS cutpoints of 10, 20, and 30 indicated mild, moderate, and severe levels of depression/anxiety, respectively. Bi-factor analysis showed sufficient unidimensionality of the PHQ-ADS score. PHQ-ADS change scores at 3 months differentiated (P < .0001) between individuals classified as worse, stable, or improved by a reference measure, providing preliminary evidence for sensitivity to change.

Conclusions

The PHQ-ADS may be a reliable and valid composite measure of depression and anxiety which, if validated in other populations, could be useful as a single measure for jointly assessing two of the most common psychological conditions in clinical practice and research.

Trial Registration

clinicaltrials.gov Identifier: NCT00926588 (SCOPE); NCT00386243 (ESCAPE); NCT00313573 (INCPAD);

Introduction

Depression and anxiety are the two most common mental health conditions in the general population as well as in clinical practice.^1-6 Depression and anxiety also result in substantial disability, representing the 2^nd and 5^th leading causes of years lived with disability in the United States⁷ and accounting for enormous losses in work productivity as well as high direct and indirect health care costs^8;9

There are a number of well-validated measures that assess depression and anxiety as separate domains. However, a measure that provides a single composite score for depression and anxiety also has several potential advantages. First, depression and anxiety frequently co-occur.^3-5;10-16 Indeed, the Diagnostic and Statistical Manual for Mental Disorders, 5^th Edition (DSM 5) acknowledges this comorbidity by including a specifier “with anxious distress” to for depressive disorders accompanied by significant levels of anxiety.¹⁷ Thus, a single score that summarizes the collective effect of depression and anxiety may be useful. Second, some interventions (e.g., cognitive-behavioral therapy; certain classes of antidepressants) are effective for both depression and anxiety. Consequently, selecting a composite score as the primary outcome for interventional studies targeting both depression and anxiety would allow for a smaller sample size than using depression and anxiety as separate co-primary outcomes. As a corollary, a single score that captures both depression and anxiety severity may be attractive to practitioners who are monitoring response to treatment of patients with comorbid depression and anxiety in clinical practice. Third, theoretical and empiric evidence supports an overarching psychological construct that encompasses distinct but related dimensions of depression and anxiety.^18;19 Fourth, the moderately strong intercorrelation between depression and anxiety makes a composite score attractive as a covariate in multivariate modeling and other types of adjusted analyses.

The Patient Health Questionnaire 9-item depression scale (PHQ-9) and 7-item Generalized Anxiety Disorder scale (GAD-7) are among the best validated and most commonly used depression and anxiety measures, respectively.^20-25 They have been used in hundreds of research studies, incorporated into numerous clinical practice guidelines, and adopted by a variety of medical and mental health care practice settings. Importantly, the PHQ-9 and GAD-7 are public domain measures available in more than 80 translations, many of which can be freely downloaded at www.phqscreeners.com. This paper uses data from 3 clinical trials to examine the reliability and convergent, construct, and factor structure validity as well as sensitivity to change of the Patient Health Questionnaire Anxiety-Depression Scale (PHQ-ADS) – a 16-item scale comprising the PHQ-9 and GAD-7 – as a composite measure of depression and anxiety.

Methods

Patient Sample

Data was drawn from 3 clinical trials enrolling a total of 896 patients (Table 1). Two trials enrolled primary care patients with chronic musculoskeletal pain, and one trial enrolled oncology patients who had depression and/or cancer-related pain. The Stepped Care to Optimize Pain care Effectiveness (SCOPE) trial enrolled 250 patients with chronic musculoskeletal pain from 5 primary care clinics in a single Veterans Affairs (VA) Medical Center, randomizing participants to a telecare collaborative management intervention arm optimizing analgesic therapy (n = 124) or a usual care arm (n = 126).^26,27 The Evaluation of Stepped Care for Chronic Pain (ESCAPE) trial enrolled 241 Operation Enduring Freedom/Operation Iraqi Freedom veterans, randomizing them to an intervention (n = 120) or usual care (n = 121) group.²⁸ The intervention involved 12 weeks of optimized analgesic therapy coupled with pain self-management strategies (Step 1) followed by 12 weeks of brief cognitive behavioral therapy (Step 2). The Indiana Cancer Pain and Depression (INCPAD) trial enrolled 405 patients with depression and/or cancer-related pain from 16 community-based oncology practices, randomizing them to a telecare intervention arm optimizing analgesic and antidepressant therapy (n = 202) or a usual care arm (n = 203).^29;30 Data collection occurred from March 2006 through August 2009 in INCPAD, from December 2007 through April 2012 in ESCAPE, and from June 2010 through May 2013 in SCOPE,

Table 1. Characteristics of Patient Samples in the Three Trials.

Variable	SCOPE (n = 250)	ESCAPE (n = 241)	INCPAD (n = 405)
Clinical sites	Primary care	Primary care	Oncology
Primary eligibility condition	Chronic musculo-skeletal pain	Chronic musculo-skeletal pain	Pain and/or Depression
Veterans, %	100.0	100.0%	7.7%
Age, mean (range) yr.	55.1 (28-65)	36.7 (21-73)	58.8 (23-86)
Men, %	82.8	88.4	32.1
Race, %
White	76.8	77.7	79.5
Black	19.2	12.8	18.0
Other	4.0	9.5	2.5
Education, %
Some college	74.0	75.9	39.0
High school or less	26.0	24.1	61.0
Major depression, %	24.0	32.0	69.9

Measures

PHQ-9 and GAD-7

The PHQ-9 consists of 9 items representing the criterion symptoms for DSM 5 major depressive disorder.³¹ Respondents are asked how much each symptom has bothered them over the past 2 weeks, with response options of “not at all”, “several days”, “more than half the days”, and “nearly every day”, scored as 0, 1, 2, and 3, respectively. The PHQ-9 can be scored as either a continuous variable from 0 to 27 (with higher scores representing more severe depression) or categorically using a diagnostic algorithm for major depressive or other depressive disorder. The GAD-7 has 7 items with response options identical to the PHQ-9 and therefore can be scored as a continuous variable from 0 to 21 (with higher scores representing more severe anxiety). Although originally developed as a measure to detect generalized anxiety disorder³², the operating characteristics of the GAD-7 are nearly as good for the other common anxiety disorders in clinical practice – panic disorder, social anxiety disorder, and posttraumatic stress disorder.²³ The PHQ-9 and GAD-7 have strong internal and test-retest reliability as well as construct and factor-structure validity.²⁰ Moreover, both measures have proven sensitive to change when monitoring treatment response.^20;33-36 The PHQ-ADS is the sum of the PHQ-9 and GAD-7 scores and thus can range from 0 to 48, with higher scores indicating higher levels of depression and anxiety symptomatology.

Other Mental Health Measures for Assessing Convergent Validity

The 5-item Mental Health Inventory (MHI-5) is one of eight scales that constitute the widely-used 36-item Medical Outcomes Study Short Form health survey (SF-36).³⁷ Scores on the MHI-5 range from 0 to 100, with lower scores representing worse mental health. The MHI-5 has been found to have reasonable sensitivity and specificity in screening for DSM-IV depressive and anxiety disorders^38;39 The Mental Component Summary (MCS) score of the SF-12 was administered, which serves as a measure of impairment related to mental disorders; the MCS is scored from 0 to 100 with higher scores representing better mental functioning and is one of the most widely-used measures of mental health functioning and quality of life.⁴⁰ Finally, participants in the SCOPE trial completed the 4-item depression and 4-item anxiety scales from the PROMIS-29 profile; scores for each scale range from 4 to 20 with higher scores representing worse symptoms (www.nihpromis.org).^41-43 A composite PROMIS anxiety-depression score was also calculated (i.e., the sum of the depression and anxiety scores), which could range from 8 to 40.

Quality of Life and Disability Measures for Assessing Construct Validity

Two quality of life domains that have shown moderate associations with depression and anxiety are vitality and social functioning which were assessed with the SF-36 vitality and social functioning scales; these, like other SF-36 scales, have scores that range from 0 to 100, with lower scales representing worse impairment. Disability days were assessed In two trials (SCOPE and INCPAD) with a single item that asked participants to indicate the number of days during the preceding 4 weeks that they were either in bed or had to reduce work or usual activities by 50% or more due to physical health or emotional problems?.^26;44 Another measure of disability used in the INCPAD trial was the Sheehan Disability Scale (SDS) which consists of three items asking how much the participant's health condition has interfered with his/her family life, social life, and work over the past month on a scale of 0 (not at all) to 10 (unable to carry on any activities).^44;45 The SDS score is a mean of these three items with higher scores reflecting greater disability. In the SCOPE and ESCAPE trials, work effectiveness was assessed with a single item asking how effective the respondent was on his or her job during the past 4 weeks on a scale of 0% (not at all effective) to 100% (completely effective).²⁶

Statistical Analysis

Because of substantial differences in the patient samples and study interventions, we analyzed data for each trial separately rather than pooling the data. For a number of analyses, results are reported for both the PHQ-ADS as well as its component scales, the PHQ-9 and GAD-7. The mean, standard deviation, and internal reliability (Cronbach's alpha) was calculated for each of the 3 scales. The standard error of measurement (SEM) was calculated as the standard deviation of the baseline score for a measure multiplied by the square root of one minus the Cronbach's alpha.^43;46 The SEM can be regarded as the standard deviation of an individual score, and either 1 or 2 SEMs have been considered one approach to estimating the minimal clinically important difference (MCID) for a scale.^33;47 Pearson's correlation coefficients of the PHQ-ADS, PHQ-9 and GAD-7 with other mental health measures and quality of life/disability measures were calculated to assess convergent and construct validity, respectively.

Cutpoints of 10, 20, and 30 on the PHQ-ADS were examined as thresholds of mild, moderate, and severe depression/anxiety symptoms, respectively. This resulted in 4 ordinal PHQ-ADS categories of 0-9, 10-19, 20-29, and 30-48, representing, minimal, mild, moderate, and severe levels of depressive-anxiety symptomatology. The rationale for these cutpoints was three-fold: 1) Because 5, 10, and 15 represent mild, moderate, and severe cutpoints on the PHQ-9 and GAD-7, it seemed logical to select 10, 20, and 30 on a composite scale that is the simple sum of the two scales; 2) Examination of the frequency distribution of the PHQ-ADS scores in the 3 trials suggested a reasonable distribution of scores using these predefined cutpoints; 3) 10, 20, and 30 are easy-to-remember cutpoints, a pragmatic consideration that may increase clinical uptake.⁴⁸ The convergent and construct validity of PHQ-ADS ordinal categories were evaluated by comparing the four groups on mental health and quality of life/disability measures using analysis of variance models.

The structural validity of a single summed PHQ-ADS score was evaluated using confirmatory one-factor, two-factor, and bi-factor models.^49,50 The one-factor models represent the set of items as being explained by a strictly unidimensional single trait and indicate the measurement validity of a single score when the model fits the data. Bi-factor models represent the set of items as a sufficiently unidimensional trait – one which has some construct-relevant multidimensionality that does not interfere with the interpretation of a single general trait score. Sufficient unidimensionality is indicated when analyses demonstrate that the preponderance of the variance is attributable to the general trait despite the presence of secondary relationships between clusters of items.

Strict unidimensional model fit was evaluated using absolute (i.e., chi square), parsimony-adjusted RMSEA (i.e., root mean square error of approximation; cutoff ≤ .06) and WRMR (i.e., weighted root mean square residual; cutoff ≤ 1.0), and incremental CFA fit indices (i.e., comparative fit index; cutoff ≥ .95). Sufficient unidimensionality in the bi-factor model was evidenced by: explained common variance (ECV) ≥ .60, omega hierarchical index ≥ .70, and a high correlation (e.g. r >.90) between the factor loadings of the unidimensional model and the general factor of the bi-factor model. All factor analyses were performed by modeling the items as ordinal categorical with the non-linear logistic link function between items and factors. This non-linear factor analytic model is identical, within a transformation, to an item response theory (IRT) model.⁵¹ We performed factor analysis instead of IRT modeling because our focus was more on dimensionality assessment than item characteristics.

Sensitivity of the PHQ-ADS scores to change was assessed.^52,53 Specifically, because the MHI-5 and PHQ-ADS were both administered at baseline and at 3 months in two of the trials (SCOPE and INCPAD), and because the MHI-5 is essentially a composite depression-anxiety score (consisting of 3 depression and 2 anxiety items), three MHI-5 change groups (worse, same, improved) were computed for each patient by determining whether the MHI-5 declined or improved by more than 1.0 standard error of measurement (SEM) from baseline to follow-up at 3 months. The SEM for the MHI-5 was 8 in SCOPE and 9 in INCPAD, so we classified those with an MHI-5 decrease or increase of 10 or greater as worse or improved, respectively, with the remainder of patients classified as same. Sensitivity to change of the PHQ-ADS was assessed by computing the standardized response mean (SRM) for each MHI-5 change group, and comparing the SRMs using analysis of variance, with pairwise Tukey-Kramer post hoc tests controlling the overall Type I error rate at 0.05.

Analyses were performed using SAS Version 9.3 (SAS Institute, Cary, North Carolina) and MPlus Version 7.2 (Muthen and Muthen).

Results

Psychometric Characteristics of PHQ-9, GAD-7 and PHQ-ADS in the 3 Trials

As shown in Table 2, the mean PHQ-9 and GAD-7 scores in the 3 trials represent moderate levels of depression and mild levels of anxiety, respectively. The INCPAD trial enrolled patients with depression as well as pain and therefore, not surprisingly, had the highest mean depression scores, whereas the SCOPE trial had the lowest depression and anxiety scores. All 3 scale scores demonstrated good internal reliability, with Cronbach's alphas in the 0.8 to 0.9 range. PHQ-ADS item means (SD) and item-total correlations are summarized in Table S1, Supplemental Digital Content 1; all item-total correlations were good (0.42 to 0.69). Correlations of the 16 PHQ-ADS items with one another are shown in Table S2, Supplemental Digital Content 1.

Table 2. Selected Characteristics of PHQ-9, GAD-7, and PHQ-ADS in Three Trials.

Variable	SCOPE (n = 250)	ESCAPE (n = 241)	INCPAD (n = 405)
Scale scores, mean (SD)
PHQ-9,	9.1 (6.3)	11.2 (5.9)	13.0 (6.7)
GAD-7	5.9 (5.6)	8.8 (5.3)	7.9 (5.8)
PHQ-ADS	14.9 (11.2)	20.0 (10.4)	20.8 (11.0)
Cronbach's alpha
PHQ-9	0.842	0.846	0.816
GAD-7	0.882	0.853	0.855
PHQ-ADS	0.917	0.908	0.878
Standard error of measurement
PHQ-9	2.51	2.29	2.91
GAD-7	1.97	2.04	2.94
PHQ-ADS	3.18	3.13	3.81
PHQ-ADS Categories, n %
Minimal (0-9)	96 (38.4)	53 (22.0)	65 (16.1)
Mild (10-19)	78 (31.2)	66 (27.4)	122 (30.1)
Moderate (20-29)	42 (16.8)	68 (28.2)	122 (30.1)
Severe (30-39)	34 (13.6)	54 (22.4)	96 (23.7)

Using a 1-SEM change to estimate a minimal clinically important difference (MCID), the MCID estimated from these 3 trials would be approximately 2 to 3 points for the PHQ-9 and GAD-7 and 3 to 4 points for the PHQ-ADS. Using a more conservative estimate of a 2-SEM change, the MCID would be approximately 4 to 6 points for the PHQ-9 and GAD-7 and 6 to 8 points for the PHQ-ADS. The distribution of the PHQ-ADS ordinal categories indicated more than a third (38.4%) of patients in the SCOPE trial had minimal depression/anxiety symptoms, approximately a third had mild symptoms (31.2%), and close to a third (30.4%) had moderate to severe symptoms. In the ESCAPE trial, about a quarter (22%-28%) of patients fell into each of the 4 categories, whereas in the INCPAD trial which targeted depressed patients, the majority of patients had some level of depression/anxiety symptoms.

The most commonly used cutpoint on both the PHQ-9 and GAD-7 to screen for depressive and anxiety disorders, respectively, is 10 or greater.²⁰ The number of patients in the 3 trials that achieved this cutpoint on both the PHQ-9 and GAD-7 was 286 (31.9%); on the PHQ-9 only, 266 (29.7%); on the GAD-7 only, 21 (2.3%); and on neither measure, 323 (36.1%). Thus, if only the PHQ-9 had been used in these trials, 307 (34.3%) of patients with chronic pain who had anxiety only or, more commonly, combined anxiety and depression, would not have been detected. This supports joint use of the PHQ-9 and GAD-7 to increase the detection of comorbid anxiety

Convergent and Construct Validity of the PHQ-ADS, PHQ-9 and GAD-7

As shown in Table 3, the PHQ-ADS had the strongest correlations with the PHQ-9 and GAD-7 (its two component scales), and the PHQ-9 and GAD-7 had moderately strong correlations with one another. The 3 scales also showed moderately strong convergent validity with the 3 composite psychological measures (PROMIS-ADS, MHI-5, and MCS) with the PHQ-ADS having slightly higher correlations than the PHQ-9 and GAD-7. As expected, the highest correlations were with the two scales measuring exclusively depression and anxiety symptoms (PROMIS-ADS and MHI-5). Construct validity was supported by moderate correlations of each of the 3 scales with quality of life and disability measures.

Table 3. Correlations of PHQ-ADS, PHQ-9, and GAD-7 with Mental Health (Convergent Validity) and Quality of Life and Disability (Construct Validity) Measures ^*.

Variable	PHQ-ADS	PHQ-9	GAD-7
Convergent Validity
PHQ-9
SCOPE	.95	--	--
ESCAPE	.94	--	--
INCPAD	.89	--	--
GAD-7
SCOPE	.94	.77	--
ESCAPE	.93	.75	--
INCPAD	.86	.54	--
PROMIS-ADS
SCOPE	.83	.76	.80
SF Mental (MHI-5)
SCOPE	.83	.78	.78
ESCAPE	.81	.79	.72
INCPAD	.76	.65	.69
SF MCS
SCOPE	.79	.75	.74
ESCAPE	.82	.81	.73
INCPAD	.67	.60	.57
Construct Validity
SF Vitality
SCOPE	.69	.63	.50
ESCAPE	.57	.60	.45
INCPAD	.46	.45	.36
SF Social
SCOPE	.62	.60	.57
ESCAPE	.66	.65	.58
Disability Days
SCOPE	.48	.46	.44
INCPAD	.35	.31	.30
Sheehan Disability Scale
INCPAD	.45	.41	.38
Work Effectiveness
SCOPE	-.46	-.47	-.39
ESCAPE	-.41	-.34	-.43

^*Values shown are Pearson's correlation coefficients

Convergent and Construct Validity of the PHQ-ADS Ordinal Categories

Data in Table 4 demonstrate the convergent and construct validity of the PHQ-ADS ordinal categories. There is a large incremental increase in depression (PHQ-9), anxiety (GAD-7), and psychological composite (PROMIS-ADS, MHI-5, and MCS) scores as one goes from minimal to mild to moderate to severe levels of depression/anxiety as classified by the four PHQ-ADS ordinal categories. A similar incremental “dose-response” effect is seen on all quality of life and disability domains.

Table 4. Convergent and Construct Validity of PHQ-ADS Ordinal Categories.

Measure	PHQ-ADS Category (Score Range)				P-value*
	Minimal	Mild	Moderate	Severe
	(0-9)	(10-19)	(20-29)	(30-48)
Convergent Validity	Mean (SD)
PHQ-9
SCOPE	3.3 (2.3)	8.8 (2.3)	14.2 (3.2)	19.5 (3.7)	< .001
ESCAPE	3.8 (1.9)	8.6 (2.3)	13.5 (3.0)	18.7 (3.3)	< .001
INCPAD	1.9 (2.5)	11.1 (3.3)	15.5 (3.5)	19.8 (3.6)	< .001
GAD-7
SCOPE	1.3 (1.6)	4.9 (2.5)	9.8 (2.9)	16.3 (3.2)	< .001
ESCAPE	2.4 (1.4)	6.5 (2.2)	10.5 2.6)	15.8 (2.8)	<.001
INCPAD	1.9 (2.1)	4.1 (2.9)	8.6 (3.6)	15.8 (3.2)	< .001
PROMIS-ADS
SCOPE	9.4 (2.3)	12.0 (4.2)	18.8 (5.8)	26.6 (7.1)	< .001
SF Mental (MHI-5)
SCOPE	85.5 (7.7)	75.0 (13.2)	52.5 (16.9)	36.5 (18.2)	< .001
ESCAPE	81.0 (13.5)	67.3 (13.7)	50.5 (14.4)	34.1 (15.1)	< .001
INCPAD	82.2 (10.8)	64.5 (15.3)	49.7 (15.0)	35.1 (17.0)	< .001
SF MCS
SCOPE	56.8 (5.4)	50.7 (8.3)	39.6 (9.7)	29.4 (9.8)	< .001
ESCAPE	55.4 (7.2)	46.9 (8.8)	37.0 (7.9)	27.8 (7.3)	< .001
INCPAD	54.0 (8.7)	44.8 (8.7)	36.8 (9.9)	30.5 (10.8)	< .001
Construct Validity	Mean (SD)
SF Vitality
SCOPE	56.1 (19.4)	38.3 (16.1)	25.3 (19.5)	21.0 (16.9)	< .001
ESCAPE	50.0 (20.0)	40.8 (16.0)	31.2 (15.2)	21.5 (13.4)	< .001
INCPAD	46.7 (18.7)	30.6 (18.4)	23.3 (16.0)	19.1 (14.7)	< .001
SF Social
SCOPE	82.8 (19.8)	69.2 (21.6)	48.8 (24.8)	37.9 (23.9)	< .001
ESCAPE	75.9 (21.1)	61.0 (21.0)	46.0 (22.1)	29.9 (17.1)	< .001
Disability Days
SCOPE	4.7 (6.5)	9.1 (8.2)	13.0 (7.4)	16.2 (8.6)	< .001
INCPAD	10.4 (9.9)	15.3 (10.4)	18.9 (9.8)	20.5 (8.3)	< .001
Sheehan Disability Scale
INCPAD	3.3 (2.6)	4.8 (2.7)	6.1 (2.6)	6.9 (2.4)	< .001
Work Effectiveness
SCOPE	82.0 (18.8)	73.6 (20.5)	61.8 (22.9)	52.9 (26.3)	<.001
ESCAPE	84.0 (17.8)	79.9 (18.2)	74.3 (20.0)	59.1 (24.5)	< .001

^*Analysis of variance was used to compare mean scores across the four categories.

Structural Validity of the PHQ-ADS

Table 5 includes the fit statistics for the1-factor, 2-factor, and bi-factor models. Although the chi-square test was significant in all 3 trials (suggesting some deviation from good fit), this fit index yields high power in larger samples to detect minor deviations. Therefore, consistent with tradition in confirmatory latent variable modeling, we will emphasize the fit indices (CFI, RMSEA, WRMR) which are less dependent on sample sizes. There was generally a small improvement in fit when comparing the 2-factor to 1-factor model, and a greater improvement when comparing the bi-factor to either the 1-factor or 2-factor models. The CFI threshold of ≥ .95 was achieved for all 3 models in the SCOPE and ESCAPE trials but only for the bifactor model in the INCPAD trial. The RMSEA threshold of ≤ .06 was achieved for the bifactor model in two of the trials but in none of the trials for the 1-factor and 2-factor models. Finally, the WRMR threshold of ≤ 1.0 was achieved for the bi-factor model in all 3 trials, the 2-factor model in only 1 trial, and the 1-factor model in none of the trials. As shown in Table S3, Supplemental Digital Content 1, most of the factor loadings were substantially higher than the acceptable threshold of 0.40, and were only slightly higher for the 2-factor compared to the 1-factor model. Moreover, the general factor loadings from the bi-factor model were generally in the range of loadings from the 1-factor model.

Table 5. Confirmatory One-Factor, Two-Factor, and Bi-factor Model Statistics for the PHQ-ADS ^*.

Fit Index	SCOPE Trial (n = 250)			ESCAPE Trial (n = 241)			INCPAD Trial (n = 405)
	1-factor	2-factor	Bi-factor	1-factor	2-factor	Bi-factor	1-factor	2-factor	Bi-factor
Number of parameters	64	65	80	64	65	80	64	65	80
Chi-square (df)	318.0 (104)	290.6 (103)	250.39 (88)	278.7 (104)	228.0 (103)	167.1 (88)	817.7 (104)	407.1 (103)	161.6 (88)
RMSEA	.091	.085	.086	.083	.071	.061	.130	.085	.045
CFI	0.956	0.962	0.967	0.954	0.967	0.979	0.862	.941	.986
WRMR	1.179	1.110	0.949	1.114	0.981	0.755	2.040	1.384	0.740
Estimated factor correlations	n/a	0.912	†	n/a	0.865	†	n/a	0.653	†
Explained common variance			0.854			0.792			0.634
Omega hierarchical index			0.906			0.891			0.743
Correlation between 1-factor model loadings and general factor loadings from bi-factor model			0.97			0.73			0.79

^*Strict unidimensional model fit was evaluated using absolute (i.e., chi square), parsimony-adjusted RMSEA (i.e., root mean square error of approximation; cutoff ≤ .06), incremental CFA fit indices (i.e., comparative fit index; cutoff ≥ .95), and WRMR fit indices (i.e., weighted root mean square residual, cutoff ≤ 1.0) Sufficient unidimensionality in the bi-factor model was evidenced by: explained common variance greater than 0.60, omega hierarchical index greater than 0.70, and a high correlation (e.g. r >.90) between the factor loadings of the unidimensional model and the general factor of the bi-factor model.

^†Each pair constrained to zero

In the bi-factor model (Table 5), the general factor strength indices (i.e., ECV, omega hierarchical) and the correlation between factor loadings of the unidimensional model and the general factor of the bi-factor model each exceeded cutoffs (0.60, 0.70, and 0.90, respectively), further suggesting sufficient unidimensionality and supporting the structural validity of a single PHQ-ADS composite score. Finally, the scree plots (Figure S1, Supplemental Digital Content 1) of the eigenvalues indicated that there was one dominant factor, because the eigenvalues dropped greatly from the first to the second factor, after which eigenvalues leveled off with much smaller drops between the second and remaining factors. Taken together, the fit indices and the factor loadings point to the validity of the traditional scoring of the PHQ-9 and GAD-7 as depression and anxiety scale scores as well as the sufficient unidimensionality of scoring the PHQ-ADS as a composite score.

Sensitivity to Change of the PHQ-ADS

According to the MHI-5 change scores at 3 months, there were 56 patients in the SCOPE trial who were classified as worse, 113 as unchanged, and 75 as improved. The mean PHQ-ADS score increased 3.63 points in the worse group, declined 3.12 points in the stable group, and declined 7.96 points in the improved group, resulting in SRMs of -0.45, --0.51, and --0.98, respectively. In the INCPAD trial, there were 73 patients classified as worse, 115 as unchanged, and 147 as improved. The mean PHQ-ADS score decreased in all 3 groups (--5.10 points in the worse group, --9.72 points in the unchanged group, and --16.40 in the improved group, resulting in SRMs of --0.57, --1.28, and -1.89, respectively. The PHQ-ADS change scores among categories were significantly different (P < .0001) by analysis of variance, and pairwise comparisons between the worse, unchanged, and improved categories also differed (p < .001) in both trials. Thus, although the direction of PHQ-ADS change for the worse group in the INCPAD trial was unexpected, the PHQ-ADS change scores significantly differentiated between the worse, unchanged, and improved groups in both trials.

Discussion

In this validation study of the PHQ-ADS, several important findings emerge. First, the PHQ-ADS demonstrated good internal reliability as well as strong convergent and construct validity in 3 separate trials. Second, cutpoints of 10, 20, and 30 on the PHQ-ADS indicate mild, moderate, and severe levels of depression/anxiety symptoms, respectively. Third, factor analysis confirmed sufficient unidimensionality of the PHQ-ADS to support its use as a composite depression/anxiety measure. Fourth, there is preliminary evidence for sensitivity to change of the PHQ-ADS in that it significantly differed between groups that were categorized as worse, unchanged, or improved at 3 months post-randomization.

The PHQ-ADS cutpoints of 10, 20, and 30 are easy for clinicians to remember and, interestingly, are double the cutpoints of the individual PHQ-9 and GAD-7 scales for which scores of 5, 10, and 15 represent thresholds for mild, moderate, and severe depressive and anxiety symptoms, respectively. Since the PHQ-9 and GAD-7 ordinal cutpoints have proven useful in patient care as well as in practice guidelines for stratifying treatment decisions²⁰, future investigations should examine the utility of ordinal severity categories for the PHQ-ADS. The statistically-determined SEM suggests that a 3 to 4 point change on the PHQ-ADS may represent a clinically important difference. Also, the comparison of PHQ-ADS change scores among worse, stable, and improved groups as defined by the MHI-5 suggest the PHQ-ADS is sensitive to change over time. However, it will also be important to assess responsiveness in treatment trials that jointly target depression and anxiety to further examine what amount of change in PHQ-ADS scores is clinically meaningful.

The high comorbidity of depression and anxiety is one reason a composite measure may be useful. A WHO study involving the administration of a structured psychiatric interview to 5438 primary care patients from 15 international primary sites found that 39% of patients with current depression also had an anxiety disorder, and 44% with a current anxiety disorder also had comorbid depression.^54;55 A U.S. study of 2091 patients from 15 primary care clinics found that 30% of patients with depression and/or anxiety (defined as PHQ-8 and GAD-7 scores ≥ 15, respectively) had both conditions.¹² A Dutch psychiatric cohort study of 1783 patients found that of those with a DSM-IV depressive disorder, 67% had a current and 75% had a lifetime comorbid anxiety disorder, and of persons with a current anxiety disorder, 63% had a current and 81% had a lifetime depressive disorder.⁵⁶ Similarly, numerous other studies have confirmed 30-50% or higher co-occurrence rates of depression and anxiety ^3-5;12-16;57

The number of composite depression-anxiety scales is limited. One well-validated composite measure is the 14-item Hospital Depression and Anxiety Scale (HADS) which provides both a single composite score as well as separate depression and anxiety scores.^58-60 Notably, a systematic review of studies examining the latent structure of the HADS tend to support both an overarching unidimensional structure as well as two underlying factors, which can vary with both the sample and the analytic strategies used.⁵⁹ Another measure is the Mental Health Inventory (the 5-item mental health scale of the SF-36) which provides a composite score^38;39 as well as depression (3 items) and anxiety (2 items) subscores; however, the latter are calculated differently than the composite score and have only occasionally been used in research.^43;58-60 and seldom in clinical practice. Both the HADS and MHI are proprietary measures and thus require a user fee to the practice or researcher for their administration. In contrast, the PHQ-9 and GAD-7 are public domain measures. Another set of public domain measures developed with NIH funding are the PROMIS scales⁶⁴, which include depression and anxiety scales of varying lengths (4 to 8 items) as well as computer-adapted testing (CAT) administration that draws upon larger item banks. One study demonstrated good correspondence between PROMIS depression and anxiety scores and PHQ-9 and GAD-7 scores.⁴³ Also, the PHQ-ADS was strongly associated with scores on the PROMIS Anxiety-Depression composite score (Tables 3 and 4). Thus, future research could compare the PHQ-ADS and PROMIS composite anxiety-depression scores in terms of validity and responsiveness.

Our study has several limitations. First, all 3 trials focused on patients with pain, rather than individuals with depression (except INCPAD) or anxiety. However, previous studies have supported the utility of the PHQ-9^65-70 and GAD-7⁷¹ in individuals with pain, and one would expect similar performance from a composite score of the two measures. Also, there was a substantial number of patients who met clinical cutpoints for depression and combined depression/anxiety in the 3 trials, but only a small proportion with anxiety only. Thus, the PHQ-ADS should be further evaluated in populations without pain as well as those with a more representative distribution of anxiety and depression, including patient samples where a structured diagnostic interview is used rather than cutpoints on a scale. Moreover, it is important to evaluate the PHQ-ADS in patients seen in mental health settings where the types and severity of psychiatric disorders may vary substantially compared to medical populations. For example, although the PHQ-9 has proven useful in psychiatric patients using similar cutpoints as those used in medical settings^72,73, the operating characteristics may be somewhat different in psychiatric populations (i.e., similar specificity but lower sensitivity).⁷⁴ Second, patient samples in two of the trials were exclusively Veterans and predominantly men; thus, data on the PHQ-ADS in non-Veteran samples including more women is warranted. Third, we did not test responsiveness to treatment of the PHQ-ADS since none of the 3 trials were specifically treating anxiety and only one was targeting depression. Thus, evaluating responsiveness to treatment (e.g., intervention groups versus control group) of the PHQ-ADS in interventional studies targeting depression and anxiety (ideally in the same trial) is needed. Fourth, the results in the INCPAD trial of oncology patients were, though generally comparable to the two primary care trials, weaker on a few of the psychometric analyses. This suggests that further study of the PHQ-ADS in patients with cancer as well as other specialty populations is warranted. Fifth, we did not use a structured criterion standard diagnostic interview in these 3 trials to determine which patients met criteria for depressive or anxiety disorders, and thus were unable to compare the sensitivity and specificity of the PHQ-ADS with the PHQ-9 and GAD-7. Certainly, a PHQ-ADS screening cutpoint would be higher than that of the PHQ-9 or GAD-7 (which are ≥ 10) since its score range is greater; for example, 10 represents a cutpoint for moderate depressive symptoms on the PHQ-9 and moderate anxiety symptoms on the GAD-7, whereas 20 represented a cutpoint for moderate depressive/anxiety symptoms on the PHQ-ADS in our sample. However, the PHQ-ADS is not intended to replace its constituent subscales in screening for depressive and anxiety disorders, since the operating characteristics of the PHQ-9 and GAD-7 are already well-established.^20-25 Sixth, our assessment of construct validity relied on relatively brief PROMIS and SF mental health measures; future studies should compare the PHQ-ADS to more detailed depressive and anxiety scales, both in terms of construct validity as well as responsiveness to treatment.

The PHQ-ADS composite score does not override the value of the individual PHQ-9 depression and GAD-7 anxiety scores but instead complements them as a measure of overall psychological symptomatology when the latter is manifested principally by varying levels of depressive and anxiety symptoms. Our findings in terms of reliability and convergent, construct, and structural validity (both fit indices and factorial loadings) support the established value of the PHQ-9 and GAD-7 as measures of depression and anxiety, respectively, while at the same time demonstrating sufficient unidimensionality of the PHQ-ADS as a composite measure. There are conceptual and clinical reasons in support of distinct depression and anxiety scores as well as a single summative score. Despite their comorbidity, depression and anxiety represent different groups of disorders in psychiatric classification; and while responding to several common treatments, depression and anxiety also have some specific treatments that differ. The PHQ-ADS score may be useful in studies for which a single depression/anxiety score is desirable as either an outcome variable or as a covariate to adjust for in multivariable analyses. The PHQ-ADS may also be useful in monitoring the concomitant treatment of depression and anxiety, especially since some treatments work across both conditions.

Abbreviations

PHQ-9

9-item Patient Health Questionnaire depression scale

GAD-7

7-item Generalized Anxiety Disorder anxiety scale

PHQ-ADS

Patient Health Questionnaire Anxiety-Depression Scale

SCOPE

Stepped Care to Optimized Pain care Effectiveness trial

ESCAPE

Evaluation of Stepped Care for Chronic Pain trial

INCPAD

Indiana Cancer Pain and Depression trial

MHI-5

5-item Mental Health Inventory

SF-36

36-item Short Form Health Survey

SF-12

12-item Short Form Health Survey

MCS

Mental Component Summary

PCS

Physical Component Summary

PROMIS

Patient Reported Outcomes Measurement Information System

SDS

Sheehan Disability Scale

SEM

standard error of measurement

MCID

minimal clinically important difference

CFA

comparative fit index

ECV

explained common variance