Case identification of depression in patients with chronic physical health problems: a diagnostic accuracy meta-analysis of 113 studies

Abstract

Background

Depression is more likely in patients with chronic physical illness, and is associated with increased rates of disability and mortality. Effective treatment of depression may reduce morbidity and mortality. The use of two stem questions for case finding in diabetes and coronary heart disease is advocated in the Quality and Outcomes Framework, and has become normalised into primary care.

Aim

To define the most effective tool for use in consultations to detect depression in people with chronic physical illness.

Design

Meta-analysis.

Method

The following data sources were searched: CENTRAL, CINAHL, Embase, HMIC, MEDLINE, PsycINFO, Web of Knowledge, from inception to July 2009. Three authors selected studies that examined identification tools and used an interview-based ICD (International Classification of Diseases) or DSM (Diagnostic and statistical Manual of Mental Disorders) diagnosis of depression as reference standard. At least two authors independently extracted study characteristics and outcome data and assessed methodological quality.

Results

A total of 113 studies met the eligibility criteria, providing data on 20 826 participants. It was found that two stem questions, PHQ-9 (Patient Health Questionnaire), the Zung, and GHQ-28 (General Health Questionnaire) were the optimal measures for case identification, but no method was sufficiently accurate to recommend as a definitive case-finding tool. Limitations were the moderate-to-high heterogeneity for most scales and the facts that few studies used ICD diagnoses as the reference standard, and that a variety of methods were used to determine DSM diagnoses.

Conclusion

Assessing both validity and ease of use, the two stem questions are the preferred method. However, clinicians should not rely on the two-questions approach alone, but should be confident to engage in a more detailed clinical assessment of patients who score positively.

INTRODUCTION

Depression is one of the leading causes of disability and disease burden.¹ It is associated with the most years lost to disability of all diseases worldwide. Identifying depression in patients with chronic physical health problems is important for several reasons. First, a number of studies suggest depression is approximately two to three times as prevalent in such populations, including patients with cancer,² chronic heart disease,^3,4 and chronic obstructive pulmonary disease (COPD).⁵ Secondly, there appears to be greater disease burden, in terms of healthcare use and functional disability, in people with comorbid depression compared with those with physical health problems alone.^6,7 Thirdly, mortality is greater in several medical conditions when depression is present — heart disease,⁸ COPD,⁹ stroke,¹⁰ cancer¹¹ — and in medically ill older adults.¹² Furthermore, morbidity and mortality may diminish with effective treatment of depression.^13,14

There is convincing evidence that many cases of depression go unrecognised in the general population and in primary care.^15–17 Reasons for under-recognition include a low rate of mood problems as the presenting complaint, infrequent specific enquiry from clinicians, and uncertainty about diagnostic criteria.^18,19 Identifying depression in people with chronic physical health problems may be even more complex, and primary care physicians may be less likely to diagnose depression in this population.^20,21 Reasons for difficulties in raising the issue of depression in consultations are complex.²² In addition, depressed individuals presenting with somatic complaints are less likely to be detected.^23–26

Improving case identification for depression has received much attention. For example, the US Preventive Services Task Force recommended screening for depression for all people in primary care (whether they had a physical illness or not), along with the necessary treatment resources for those subsequently identified.²⁷ In the UK, through the Quality and Outcomes Framework (QOF), GPs are incentivised to ask the case-identification questions of people with diabetes and coronary heart disease.²⁸ This approach is also advocated in the National Institute for Health and Clinical Excellence (NICE) guidelines.²⁹ However, there is much debate in the literature concerning the effectiveness of screening and case identification.³⁰ Gilbody and colleagues have shown untargeted screening was not effective in improving the recognition of depression in primary care and general hospital settings.³⁰ There is also much debate concerning the terminology used in the field. The present study proposes to separate overall accuracy (case identification) into more clinically understandable rule-in and rule-out performance. Rule-in accuracy (positive predictive value) is the ability to correctly identify those with the disorder with minimal false positives, whereas rule-out accuracy (negative predictive value) is the ability to correctly identify those without the disorder with minimal false negatives (missed cases). In order to differentiate from untargeted screening approaches, which appear to be ineffective, this data synthesis will focus on case identification in a population at higher risk of depression (that is, people with chronic physical health problems). This is vital before further case finding is advocated by the QOF for patients with other physical problems.

How this fits in

There is strong evidence that the prevalence of depression is raised among patients with long-term conditions and that this comorbidity is associated with adverse outcomes. Inadequate and inaccurate identification of depression has been documented in both primary care and general medical settings. This meta-analysis provides evidence that several brief and feasible depression case-finding approaches can be used as a first assessment for patients with chronic physical health problems, and that two stem questions referring to core depression features appear the most efficient initial approach.

There are a large number of scales used both in clinical practice and in research studies, few of which have been originally developed for the physically ill. In addition, there are no existing definitive meta-analyses across a comprehensive range of measures. Therefore, a diagnostic accuracy meta-analysis was conducted to assess the sensitivity and specificity of the most widely used case-identification instruments in people who are physically ill.

METHOD

Data sources and searches

The full review protocol can be found in the guideline on depression in people with chronic physical health problems, which was commissioned by NICE.³¹ Briefly, a search for studies assessing the validity of case-identification instruments was made using seven electronic bibliographic databases (CENTRAL, CINAHL, Embase, HMIC, MEDLINE, PsycINFO, Web of Knowledge). Each database was searched from inception to October 2009. Additional papers were found by searching the references of retrieved articles, tables of contents of relevant journals, previous systematic reviews and meta-analyses of case identification for depression, written requests to experts, and suggestions made by the members of the Guideline Development Group (comprising clinicians, academics, and service users with expertise in depression and chronic physical health problems).

Study selection

The study included validation studies of mood questionnaires agreed by the authors (see Appendix 1 for further details). The reference standard was diagnoses according to the Diagnostic and Statistical Manual of Mental Disorders (DSM) of the American Psychiatric Association (for example DSM-IV)³² or International Classification of Diseases (ICD) (for example ICD-10)³³ of the World Health Organization criteria. Studies that did not clearly state the comparator to be DSM or ICD diagnosis of depression, or that did not provide sufficient data to be extracted in the meta-analysis were excluded.

Data extraction and quality assessment

All published studies that met the eligibility criteria were assessed for methodological quality using the Scottish Intercollegiate Guidelines Network (SIGN) checklist for diagnostic studies.²⁹ Data were extracted independently by at least two authors, and 2×2 tables were constructed, from which the primary outcomes were calculated: that is sensitivity, specificity, and likelihood ratios.

To maximise the available data, the most consistently reported and recommended cut-off points were extracted for each of the scales. There are limitations to this approach, as noted by Furukawa and colleagues,³⁴ ;who found that the optimal cut-offs for the General Health Questionnaire (GHQ)-12 and GHQ-28 differed according to the prevalence of depression, and it is likely there are similar problems for most other scales. However, a Bayesian approach makes allowance for variations according to prevalence (see below), therefore seeking to take into account this potential limitation.

Data synthesis and analysis

A bivariate diagnostic accuracy meta-analysis was conducted using Stata (version 10) with the metandi³⁵ commands, to obtain pooled estimates of sensitivity, specificity, and likelihood ratios. This method was originally developed as a mixed effects regression model for meta-analysis of trials, and modified more recently for studies of diagnostic accuracy.^36,37 Between-study heterogeneity was assessed using the I² statistic.³⁸ In addition, publication bias was assessed by visual inspection of funnel plots, and formal use of Egger's test.³⁹

A Bayesian curve analysis was also undertaken; this plots post-test conditional probabilities from all possible pre-test probabilities (prevalence). The area under the Bayesian curve (AUC) for positive results can be used as a statistical comparison of rule-in success and 1 — AUC for negatives results used as an indicator of rule-out success. An area of more than 0.75 can be interpreted as ‘satisfactory’ and more than 0.80 interpreted as ‘good’. If a test achieved more than 0.90 in a rule-in capacity, this was considered sufficient for a recommendation that this tool could be used on its own for case finding.

Additional meta-regression analyses were planned to assess differences in diagnostic accuracy for disease groups. Such analyses were conducted on a scale when there were a minimum of four studies for at least two disease groups.

RESULTS

A total of 113 studies on 20 826 participants met the eligibility criteria of the review (see Figure 1 for full details on study flow information). These studies were both on populations specifically targeted for a chronic physical health problem (such as cancer, heart disease, and stroke), and in general medical settings where all were physically ill and a substantial proportion had a chronic physical health problem. In total, 83 studies specifically targeted people with chronic physical health problems in any setting (Appendix 2). The mean prevalence of depression was 0.25 (95% confidence interval [CI] = 0.05 to 0.61). A further 30 studies were on people in general medical settings, with a mean prevalence of depression of 0.24 (95% CI = 0.04 to 0.52).

Figure 1.

Study flow diagram.

Studies recruiting for chronic physical health problem

Sensitivity and specificity

Table 1 provides an evidence summary for the various scales on people recruited for specific chronic physical health problems. There was moderate to high sensitivity for most scales. The tools with the highest sensitivity were the two stem questions (0.98; 95% CI = 0.85 to 0.99), followed by the GHQ-28, Patient Health Questionnaire (PHQ)-9, Beck Depression Inventory (BDI), and BDI non-somatic (Table 1). Sensitivity was lowest for the one-item measure.

Table 1.

Evidence summary of scales in studies recruiting for chronic physical illness

Instrument	Total sample size (studies)	Sensitivity (95% CI)	Specificity (95% CI)	Positive LR (95% CI)	Negative LR (95% CI)	I2, %	Prevalence
Patient Health Questionnaire-9	1617 (6)	0.84 (0.69 to 0.92)	0.88 (0.83 to 0.91)	6.77 (4.96 to 9.24)	0.19 (0.10 to 0.37)	93	0.26
Two stem questions (low mood and loss of interest)	1860 (6)	0.98 (0.85 to 0.99)	0.86 (0.70 to 0.94)	6.81 (2.98 to 15.57)	0.02 (0.002 to 0.21)	0	0.20
Beck Depression Inventory (BDI)	3486 (22)	0.83 (0.79 to 0.87)	0.79 (0.74 to 0.84)	3.96 (3.12 to 5.03)	0.21 (0.16 to 0.27)	90	0.20
BDI-non somatic items	946 (8)	0.83 (0.68 to 0.92)	0.79 (0.70 to 0.85)	3.89 (2.66 to 5.69)	0.22 (0.11 to 0.44)	87	0.22
Center of Epidemiological Studies — Depression	1812 (11)	0.77 (0.71 to 0.85)	0.85 (0.80 to 0.90)	5.25 (3.60 to 7.65)	0.27 (0.21 to 0.36)	0	0.23
Geriatric Depression Scale-30	687 (13)	0.79 (0.73 to 0.84)	0.73 (0.67 to 0.79)	2.95 (2.37 to 3.68)	0.29 (0.22 to 0.38)	0	0.28
Geriatric Depression Scale-15	823 (8)	0.84 (0.78 to 0.88)	0.77 (0.73 to 0.81)	3.70 (3.03 to 4.52)	0.21 (0.16 to 0.28)	0	0.29
One item	1940 (11)	0.73 (0.60 to 0.83)	0.77 (0.62 to 0.88)	3.21 (1.98 to 5.21)	0.35 (0.24 to 0.49)	98	0.26
General Health Questionnaire-12	517 (5)	0.81 (0.70 to 0.89)	0.64 (0.52 to 0.75)	2.27 (1.68 to 3.09)	0.29 (0.19 to 0.47)	53	0.16
General Health Questionnaire-28	465 (5)	0.90 (0.79 to 0.96)	0.80 (0.62,0.90)	4.39 (2.31 to 8.35)	0.13 (0.06 to 0.25)	94	0.33
Hospital Anxiety and Depression Scale — Depression	5087 (29)	0.75 (0.67 to 0.81)	0.81 (0.74 to 0.86)	3.90 (2.94 to 5.17)	0.31 (0.24 to 0.41)	99	0.26
Hamilton Depression Rating Scale	985 (11)	0.81 (0.75 to 0.86)	0.85 (0.76 to 0.91)	5.32 (3.24 to 8.71)	0.22 (0.16 to 0.31)	52	0.26
Zung Self Rating Depression Scale	190 (4)	0.78 (0.56 to 0.91)	0.92 (0.68 to 0.98)	9.82 (2.31 to 41.63)	0.24 (0.11 to 0.50)	83	0.30

LR = likelihood ratio.

The Zung Self Rating Depression Scale had the highest specificity 0.92 (95% CI = 0.68 to 0.98). This was followed by the two stem questions, the Hamilton Depression Rating Scale (HDRS), PHQ-9 and the Centre for Epidemiologic Studies Depression Scale (CES-D); all had high specificity. The lowest specificity was found for the one-item measure and the GHQ-12.

Rule-in (positive predictive value) and rule-out accuracy (negative predictive value)

Using Bayesian plots of conditional probabilities to examine rule-in and rule-out performance, only three tools had less than satisfactory rule-in performance, namely the single question: the Geriatric Depression Scale (GDS-30) and GHQ-12. The optimal single tool was the Zung, although it did not reach the a priori standard for recommendation when applied alone. For rule-out performance, four methods were not satisfactory. These were the single queston, the Hospital Anxiety and Depression Scale (HADS), GDS-30, and GHQ-12. The optimal tools were the two stem questions and GHQ-28. Overall accuracy was best for the two stem questions, Zung, PHQ-9, and GHQ-28. However, it should be noted that data for the Zung scale were based on just four studies and a relatively small total sample size (n = 190).

Meta-regression comparing the diagnostic accuracy for different disease groups was only possible for the BDI and HADS-D. There was no evidence of difference in sensitivity (beta = 0.93, P = 0.34) and specificity (beta = 1.56, P = 0.35) of the HADS between stroke and cancer patients. There was no evidence of difference in sensitivity (beta = 1.49, P = 0.60), but some evidence for differences in specificity (beta = 1.20, P = 0.02) of the BDI between heart disease and cancer patients.

Studies in general medical settings

Table 2 summarises the results for general medical settings. There were only three scales that provided sufficient data for analyses. All these scales performed equally well in this setting as compared to populations specifically targeted for chronic physical health problems with a large overlap in confidence intervals.

Table 2.

Evidence summary of scales in general medical settings

Instrument	Total sample size (studies)	Sensitivity (95% CI)	Specificity (95% CI)	Positive LR (95% CI)	Negative LR (95% CI)	I2, %	Prevalence
Geriatric Depression Scale-30	1255 (9)	0.85 (0.78 to 0.90)	0.76 (0.68 to 0.83)	3.55 (2.73 to 4.60)	0.20 (0.15 to 0.27)	98	0.36
Geriatric Depression Scale-15	1108 (8)	0.89 (0.84 to 0.92)	0.74 (0.67 to 0.80)	3.38 (2.65 to 4.31)	0.15 (0.11 to 0.22)	63	0.24
Hospital Anxiety and Depression Scale — Depression	2506 (6)	0.81 (0.73 to 0.87)	0.75 (0.70 to 0.79)	3.22 (2.79 to 3.72)	0.25 (0.18 to 0.36)	89	0.15

LR = likelihood ratio.

Sensitivity and specificity

Sensitivity was relatively high in all measures but particularly high in the GDS-15 (0.89; 95% CI = 0.84 to 0.92). Specificity was very similar for the GDS-30, GDS-15, and HADS when used in general medical settings (Table 2).

Rule-in and rule-out accuracy

Using the same methodology for each measure in general medical settings and correcting for prevalence using a Bayesian analysis, the GDS-15 was most successful and the HADS least successful. No method came close to the a priori; standard for rule-in performance when applied alone. For rule-out accuracy, the HADS was significantly less accurate than the GDS-15 (Area HADS = 0.71, 95% CI = 0.68 to 0.74 versus Area GDS-15 = 0.78, 95% CI = 0.75 to 0.82).

DISCUSSION

Most of the scales performed adequately as case-identification measures for depression, with modest differences in validity coefficients. Most studies targeted chronic physically ill populations rather than general medical settings such as primary care. In order to detect depression in those with chronic physically ill health, the most sensitive instruments appear to be two stem questions, PHQ-9, and GHQ-28. The most specific measure was the Zung. Overall, optimal accuracy was achieved by the two stem questions, Zung, PHQ-9, and GHQ-28. However, it should be noted that estimates on the Zung and GHQ-28 analysis were based on a relatively small sample size; therefore, it is possible that conclusions regarding these scales may change with further data. No method came close to the a priori standard for case-finding recommendation when applied alone.

Another important factor to consider when comparing the different measures is the ease of implementation. The Zung is a 20-item scale and therefore is more resource intensive and less likely to be implemented in primary care compared to shorter measures. Taking into account both the psychometric properties and ease of implementation, it would appear the two stem questions may be the preferred measure for case identification in patients with chronic physical health problems. From these data, the authors do not recommend relying upon a single question alone, and recommend two questions as a minimum initial enquiry. This is consistent with previous pooled data in primary care40 and cancer settings.⁴¹

In general medical settings, there were fewer studies, and analysable data were only available for the GDS, GDS-15, and HADS-D. Specificity was similar for all three scales but sensitivity was highest in the GDS-15. Further research is needed to confirm whether the optimal tools in the chronically ill (two stem questions, PHQ-9, and the Zung) perform equally well in general medical samples.

There are several limitations to the results of this systematic review. First, there was moderate to high heterogeneity for most measures. Secondly, there is a paucity of validity studies using the ICD-10 as the criterion standard compared with the DSM-IV, which may favour tools using DSM items, and therefore the authors recommend future examination using this outcome. Thirdly, there were widely used or potentially useful scales that had few or no studies in the physically ill; these include the Montgomery–Asberg Depression Scale (MADRS),⁴² and the Clinically Useful Outcome Depression Scale (CUDOS).⁴³ Further research is needed on these scales for people with chronic physical health problems. Fourthly, there were a number of different semi-structured methods used to determine the interview-based diagnosis, including the Schedules for Clinical Assessment in Neuropsychiatry (SCAN),⁴⁴ the Composite International Diagnostic Interview (CIDI),⁴⁵ the Structured and Clinical Interview for DSM-III-R (SCID),⁴⁶ and the Diagnostic Interview schedule (DIS),⁴⁷ all of which may vary in diagnostic accuracy. A further limitation is the lack of cost-effectiveness analyses assessing the cost impact of false positives associated with the use of case-identification measures. However, it should be noted that the cost-effectiveness of case identification is very complex to model and requires a number of assumptions concerning probabilities assigned to events in the depression treatment care pathway, and explicit values of treatment outcomes.⁴⁸ Therefore, such issues were considered beyond the scope of this paper.

It should also be acknowledged that the use of case-identification tools may not be translated into real benefit in clinical practice. Case identification may bring limited benefit if there are no effective assessment and treatment services in place, as professionals may be reluctant to make a diagnosis of depression if they have limited resources on which to call.⁴⁹ The aim of the NICE guideline for which this review was conducted,³¹ is to promote the commissioning of such services. The impact of case finding on the individual consultation may be important, since the use of the PHQ-9 severity questionnaire can cause a tension within the consultation, with GPs struggling to manage formal assessment versus personal enquiry.⁵⁰

From this data synthesis, it appears that there are a number of instruments for the case identification of depression in the medically ill that have similar accuracy. A consideration of both accuracy and acceptability suggests that the two stem questions may be the most efficient initial method, although further validation is needed. We do not recommend the use of a single question used alone. GPs and practice nurses should not rely on the case-finding questions alone; they should be confident to complete an assessment of the patient's mental state and risk, and a pathway within the practice should be in place (particularly when it is the practice nurse who has done the case finding). Resources within the practice should be available to support patients who have depression and a chronic physical health problem, and primary care practitioners should have well-defined links with local primary care mental health services, which should offer appropriate interventions for such patients, including a collaborative care approach as recommended by NICE.³¹

Appendix 1. Full list of instruments considered

1. Beck Depression Inventory (BDI):

   • BDI-II51

   • BDI Cognitive-Affective scale⁵¹

   • BDI Fast Screen⁵²

2. Patient Health Questionnaire (PHQ):

   • PHQ-953

   • PHQ-254

3. Two stem questions:⁵⁵

   • These are similar to the PHQ-2 except that the scoring system is dichotomous (‘yes’ or ‘no’) rather than the Likert scale used for both PHQ-9 and PHQ-2, and the period of reported low mood or loss of interest is 1 month rather than 2 weeks

4. General Health Questionnaire:⁵⁶

   • GHQ-12

   • GHQ-28

5. Centre of Epidemiological Studies - Depression (CES-D)⁵⁷

6. Geriatric Depression Scale (GDS)

   • GDS-30⁵⁸

   • GDS-15⁵⁹

7. Zung Self Rating Depression Scale⁶⁰

8. Hospital Anxiety and Depression Scale - Depression⁶¹

9. Hamilton Depression Rating Scale HDRS⁶²

   • Both 17– and 21-item versions

10. Montgomery-Asberg Depression Rating Scale (MADRS)⁶³

11. Clinically Useful Depression Outcome Scale⁶⁴

12. One-item measures of depression

13. Edinburgh Postnatal Depression Scale¹⁷⁸

Appendix 2.

Summary characteristics of included studies

Study	Country	Patients	Female	Mean age, years	Physical health condition	Instrument
Aben et al, 200265	Netherlands	171	91	68	Stroke	BDI
						HADS-D
						HDRS
Agrell and Dehlin, 198966	Sweden	40	22	80	Stroke	CES-D
						GDS
						HDRS
						Zung
Akechi et al, 200667	Japan	205	68	61	Cancer	1-item
						Two stem questions
Akizuki et al, 200368	Japan	275	164	52	Cancer	HADS-D
						1-item
Aydin and Ulusahin, 200169	Turkey	100	Not reported	Not reported	Tuberculosis and COPD	GHQ-12
Berard et al, 199870	South Africa	100	87	50	Cancer	HADS-D
						BDI
Berg et al, 200971	Finland	100	32	55	Stroke	BDI, HDRS
Blank et al, 200472	US	125	Not reported	77	Various (general medical)	GDS
						GDS-15
						CES-D
						Two stem questions
Burke et al, 199273	US	67	44	77	Various (general medical)	GDS
Chilcot et al, 200874	UK	41	16	53	Dialysis	BDI
Chochinov et al, 199775	US	197	103	Not reported	Cancer	Two stem questions
Costantini et al, 199976	Italy	132	132	53	Breast cancer	HADS-D
Craven et al, 198877	Canada	99	36	51	Renal dialysis	BDI
Cullum et al, 200678	UK	618	371	80	Medically ill (general medical)	GDS-15
Diez-Quevedo et al, 200179	Spain	1003	451	43	Medically ill (general medical)	PHQ-9
Ertan et al, 200580	Turkey	109	36	67	Parkinson's Disease	GDS
Forkman et al, 200981	Germany	126	37	51	Heart disease	BDI
						BDI-non somatic
Freedland et al, 200382	US	682	327	66	Congestive heart failure	BDI
Furlanetto et al, 200583	US	155	98	50	Medically ill (general medical)	BDI non-somatic items
Galaria et al, 200084	US	70	41	77	Visual impairment	GDS
						GDS-15
Gilley and Wilson, 199785	US	93	42	70	Medically ill (general medical)	GDS
Golden et al, 200786	US	88	23	Not reported	Hepatitis C	BDI
						BDI non somatic items
						HADS-D
Grassi et al, 200987	Italy	109	83	55	Cancer	HADS-D
Hahn et al, 200688	Germany	204	98	50	Medically ill (general medical)	GHQ-12
						HADS-D
Hall et al, 199989	UK	266	266	Not reported	Breast cancer	HADS-D
Hammer et al, 200890	US	39	19	58	Amyotrophic lateral sclerosis	BDI
Harter et al, 200691	Germany	206	103	48	Musculoskeletal diseases	GHQ-12
						HADS-D
Harter et al, 200192	Germany	569	285	54	Medically ill (general medical)	GHQ-12
						HADS-D
Haughey et al, 200593	US	226	226	40	Medically ill (general medical)	Two stem questions
Haworth et al, 200794	US	88	15	70	Heart failure	GDS-15
						HADS-D
Healey et al, 200895	UK	49	28	79	Stroke	BDI non-somatic
Hedayati et al, 200696	US	98	44	57	Haemodialysis	CES-D
						BDI
Hermanns et al, 200697	Germany	376	148	52	Diabetes	BDI
						CES-D
Herrero et al, 200398	Spain	385	181	38	Medically ill (general medical)	HADS-D
Hickie et al, 198799	US	39	Not reported	Not reported	Medically ill (general medical)	GDS
Hopko et al, 2007100	US	33	25	54	Cancer	HAM-D
						BDI
						CES-D
Hoyl et al, 1999101	US	74	2	74	Medically ill (general medical)	GDS-15
Hughson et al, 1988102	UK	75	75	51	Cancer	GHQ-28
Ibbotson et al, 1994103	UK	513	282	Not reported	Cancer	HADS-D
						GHQ-28
Jackson and Baldwin, 1993104	US	59	Not reported	77	Medically ill (general medical)	GDS
						GDS-15
Jefford et al, 2004105	US	100	Not reported	Not reported	Cancer	1-item
Johnson et al, 1995106	Australia	204	Not reported	71	Stroke	GDS
						HADS-D
						GHQ-28
Katz et al, 2004107	Canada	60	13	61	Cancer	BDI
						HADS-D
						CES-D
Kawase et al, 2006108	Japan	305	Not reported	62	Cancer	1-item
Koenig et al, 1992109	US	109	0	74	Medically ill (general medical)	GDS
						GDS-15
Kugaya et al, 1998110	Japan	128	48	61	Cancer	HADS-D
Lam et al, 2004111	Hong Kong	100	56	69	Medically ill (general medical)	HADS
Lamers et al, 2008112	Netherlands	713	350	71	Medically ill (general medical)	PHQ-9
Laska et al, 2007113	Sweden	89	40	74	Stroke	MADRS
LeFevre et al, 1999114	UK	79	35	70	Cancer	HADS-D
Lee et al, 2008115	China	253	94	Not reported	Stroke	GDS-15
Leentjens et al, 2000116	Netherlands	53	Not reported	67	Parkinson's disease	BDI
						HDRS
						MADRS
Leung et al, 1998117	Taiwan	50	Not reported	54	Chronic medical disorders	Zung
Lightbody et al, 2007118	UK	28	Not reported	72	Stroke	MADRS
Lincoln et al, 2003119	UK	143	70	66	Stroke	BDI
						GHQ-28
Lloyd-Williams et al, 2000120	UK	100	56	57	Cancer	EPDS
						1-item
Lloyd-Williams et al, 2001121	UK	100	56	57	Cancer	HADS-D
Lloyd-Williams et al, 2004122	UK	74	37	68	Cancer	1-item
						EPDS
Love et al, 2002123	Australia	303	303	Not reported	Cancer	HADS-D
Love et al, 2004124	Australia	227	227	52	Breast cancer	HADS
						BDI non-somatic
Low et al, 2007125	Canada	119	30	63	Myocardial infarction or angina	BDI
						GDS
Lowe et al, 2004126	Germany	501	167	42	Medically ill (general medical)	HADS-D
						PHQ-9
						PHQ-2
Lustman et al, 1997127	US	172	83	48	Diabetes	BDI
Lykouras et al, 1996128	Greece	107	57	43	Neurological disorder	GHQ-28
Magni et al,1986129	Italy	220	109	76	Medically ill (general medical)	GDS
McManus et al, 2005130	US	1024	184	67	Chronic heart disease	PHQ-9
						PHQ-2
						Two stem questions
						CES-D
McQuillan et al,2003131	US	415	344	58	Rheumatoid Arthritis	CES-D
Meyer et al, 2003132	US	45	Not reported	Not reported	Cancer	1-item
Mitchell et al, 2008133	UK	129	Not reported	58	Cancer	PHQ-2
						1-item
Mohr et al, 2007134	US	260	190	51	Multiple sclerosis	Two stem questions
						1-item
Narding et al, 2002135	Netherlands	44	16	70	Stroke	HDRS
Neal et al, 1994136	UK	45	28	77	Medically ill (general medical)	GDS
						GDS-15
O'Rourke et al, 1998137	UK	105	Not reported	68	Stroke	HADS-D
Okimoto et al, 1982138	Japan	55	Not reported	Not reported	Medically ill (general medical)	Zung
Olden et al, 2009139	US	439	239	66	Cancer	HDRS
Ozalp et al, 2008140	Turkey	208	208	51	Cancer	HADS-D
Parikh et al, 1988141	US	80	40	58	Stroke	CES-D
Parker et al, 2002142	Australia	302	175	47	Medically ill (general medical)	HADS-D
						BDI non-somatic
Passik et al, 2001143	US	60	58	31	Cancer	Zung
Patterson et al, 2006144	US	310	37	40	HIV	BDI non-somatic
Payne et al, 2007145	US	167	Not reported	Not reported	Cancer	Two stem questions
						1-item
Persoons et al, 2003146	Netherlands	97	64	48	Otolaryngology	PHQ-9
Picardi et al, 2005147	Italy	141	79	38	Dermatology	GHQ-12
						PHQ-9
Pomeroy et al, 2001148	US	87	52	78	Medically ill (general medical)	GDS
						GDS-15
						1-item
Poole et al, 2006149	US	115	47	43	Heart disease	HADS-D
Rapp et al, 1988150	US	150	48	69	Medically ill (general medical)	GDS
						BDI
Razavi et al, 1990151	Belgium	210	140	55	Cancer	HADS-D
Reuter and Harter, 2000152	Germany	188	51	54	Cancer	HADS-D
						GHQ-12
Rinaldi et al, 2003153	US	181	Not reported	79	Medically ill (general medical)	GDS-15
Roger et al, 2009154	US	67	35	71	Stroke	CES-D
						HDRS
						GDS-15
Rovner et al, 1997155	US	70	41	77	Visual impairment	GDS
Sagen et al, 2009156	Norway	104	43	65	Stroke	HADS-D
						MADRS
Scheinthal et al, 2001157	US	75	42	74	Medically ill (general medical)	GDS-15
						BDI non-somatic
Schein et al, 1997158	US	76	35	70	Medically ill (general medical)	CES-D
Serrano-Duenas et al, 2008159	Ecuador	115	33	70	Parkinson's disease	HDRS
Shinar et al, 1986160	Israel	27	16	56	Stroke	CES-D
Silberman et al, 2006161	Brazil	46	19	68	Parkinson's disease	MADRS
						BDI
Singer et al, 2008162	Germany	250	23	Not reported	Cancer	HADS-D
Sivrioglu et al, 2009163	Turkey	85	53	59	Stroke	GDS
Stafford et al, 2007164	Australia	193	39	64	Heart disease	HADS-D
						PHQ-9
Strik et al, 2001165	US	206	49	59	Myocardial infarction	HADS-D
						HDRS
						BDI
Tang et al, 2004166	China	100	45	74	Stroke	GDS
						HADS-D
						GDS-15
Tang et al, 2004167	China	60	Not reported	Not reported	Stroke	GDS
						HADS-D
Thekkumpurath et al, 2009168	UK	150	86	70	Cancer	GHQ-12
Turner et al, 1984169	US	40	20	47	Chronic pain	Zung
						BDI
						BDI-non somatic
Upadhyaya et al, 1997170	UK	72	35	71	Medically ill (general medical)	HADS-D
Vahter et al, 2007171	Estonia	134	Not reported	44	Multiple sclerosis	1-item
Vargas et al, 2007172	Portugal	484	276	70	Medically ill (general medical)	GDS
Walker et al, 2007173	UK	361	238	Not reported	Cancer	HADS-D
Watnick et al, 2005174	US	62	20	63	Dialysis	PHQ-9
						BDI
Weintraub et al, 2006175	US	148	Not reported	71	Parkinson's disease	GDS-15
						HDRS
Wilhelm et al, 2004176	Australia	212	117	Not reported	Medically ill (general medical)	BDI non-somatic
Williams et al, 2005177	US	296	219	59	Stroke	PHQ-2
						PHQ-9

Funding

Stephen Pilling received financial support from the National Institute for Health and Clinical Excellence.

Additional information

The Bayesian plots of conditional probabilities of scale are available on request fromthe authors.

Provenance

Freely submitted; externally peer reviewed.

Competing interests

David Goldberg developed the General Health Questionnaire. The other authors have declared no competing interests.

Discuss this article

Contribute and read comments about this article on the Discussion Forum: http://www.rcgp.org.uk/bjgp-discuss