Review
The validity of the Hospital Anxiety and Depression Scale
An updated literature review
Ingvar Bjelland a,*,Alv A.Dahl b ,Tone Tangen Haug c ,Dag Neckelmann d
a
Department of Public Health and Primary Health Care,Section for Preventive Medicine,Haukeland Hospital,
Armauer Hann Building,University of Bergen,N-5021Bergen,Norway b
女性秃顶Department of Psychiatry,Aker Hospital,University of Oslo,Oslo,Norway
c
Institute of Psychiatry,University of Bergen,Bergen,Norway
d
Department of Psychiatry,Haukeland Hospital,University of Bergen,Bergen,Norway
Received 3June 2001;accepted 9October 2001
Abstract
Objective:To review the literature of the validity of the Hospital Anxiety and Depression Scale (HADS).Method:A review of the 747identified papers that ud HADS was performed to address the following questions:(I)How are the factor structure,discriminant validity and the internal consistency of HADS?(II)How does HADS perform as a ca finder for anxiety disorders and depression?(III)How does HADS agree with other lf-rating instruments ud to rate anxiety and depression?Results:Most factor analys demonstrated a two-factor solution in good accordance with the HADS subscales for Anxiety (HADS-A)and Depression (HADS-D),respectively.The correlations between the two subscales varied from .40to .74(mean .56).Cronbach’s alpha for HADS-A varied from .68to
.93(mean .83)and for HADS-D from .67to .90(mean .82).In most studies an optimal balance between nsitivity and specificity was achieved when caness was defined by a score of 8or above on both HADS-A and HADS-D.The nsitivity and specificity for both HADS-A and HADS-D of approximately 0.80were very similar to the nsitivity and specificity achieved by the General Health Questionnaire
(GHQ).Correlations between HADS and other commonly ud questionnaires were in the range .49to .83.Conclusions:HADS was found to perform well in asssing the symptom verity and caness of anxiety disorders and depression in both somatic,psychiatric and primary care patients and in the general population.D 2002Elvier Science Inc.All rights rerved.
Keywords:Anxiety;Depression;Psychiatric Status Rating Scales;Psychometrics;Reproducibility of results;Sensitivity and specificity
Introduction
To make cost-effective screening of mental disorders feasible,veral brief questionnaires asssing a limited t of symptoms have been developed.The Hospital Anxiety and Depression Scale (HADS)[1]was developed by Zigmond and Snaith in 1983to identify caness (possible and probable)of anxiety disorders and depression among patients in nonpsychiatric hospital clinics.It was divided into an Anxiety subscale (HADS-A)and a Depression subscale (HADS-D)both containing ven intermingled items.To prevent ‘noi’from somatic disorders on the scores,all symptoms of anxiety or depression relating also
to physical disorder,such as dizziness,headaches,insom-nia,anergia and fatigue,were excluded.Sym
ptoms relating to rious mental disorders were also excluded,since such symptoms were less common in patients attending a non-psychiatric hospital clinic.The authors [1]also intended to ‘‘define carefully and distinguish between the concepts of anxiety and depression.’’
HADS has been ud extensively,and we identified 747papers that referred to HADS in Medline,ISI and PsycINFO indexed journals by May 2000.
The evaluation of psychometric properties and dia-gnostic efficacy of questionnaires is often inadequate [2].To our knowledge,there has been only one review of the literature addressing the issues in HADS [3].Bad on approximately 200papers on HADS in approximately 35,000individuals in various patient populations,Herr-mann concluded in 1996that ‘‘HADS is a reliable and
0022-3999/02/$–e front matter D 2002Elvier Science Inc.All rights rerved.PII:S 0022-3999(01)00296-3
*Corresponding author.Tel.:+47-5597-4610;fax:+47-5597-5896.E-mail address :ingvar. (I.Bjelland).
Journal of Psychosomatic Rearch 52(2002)69–
77
valid instrument for asssing anxiety and depression in medical patients.’’
Since Herrmann’s review the number of‘HADS-papers’that have been published has incread almost fourfold. The papers also include samples from the general popu-lation,which Herrmann’s review did not.Another reason for conducting an updated review of HADS-related papers was to achieve more information about the following issues:(I) How is the factor structure,discriminant validity and the internal consistency of HADS?(II)How does HADS perform as a ca finder for anxiety disorders and depres-sion?(III)To what extent does HADS agree with other lf-rating instruments(concurrent validity)?
色播图片
Method
A arch in the Medline,ISI and PsycINFO databas was performed in May2000.All papers containing the terms‘Hospital’and‘Anxiety’and‘Depression’or‘HAD’or‘HADS’in the title or abstract were identified.The abstracts of the studies(n=1403)were then inspected to ascertain whether they contained information about the HADS.The authors then reviewed747studies using the HADS for information regarding issues(I),(II)and(III), and71relevant papers were identified.
Factor structure,discriminant validity and
形容美味的成语internal consistency
The following information was gathered:the number of factors in HADS identified by factor analys,the correla-tion between the subscales of HADS,and the internal consistency of the subscales(Cronbach’s alpha).
HADS as a ca finder for anxiety disorders and depression Sensitivity and specificity of HADS in the different studies were chon according to the cut-off value deter-mined by a receiver operating characteristic(ROC)curve giving a maximal diagnostic contribution[4,5].In studies without ROC curve
s,but with at least four cut-off values with given nsitivities and specificities,we plotted the ROC curves ourlves.The area under the ROC curve(AUC)is a measure of the information value inherent in a test to determine caness over the whole range of possible thresh-old values[6].An AUC value of0.50reflects a test that is unable to discriminate between cas and noncas,while a value of1.00means perfect nsitivity and specificity at all cut-off values.In the studies where the ROC curves were plotted by us,approximations of AUC were calculated using the trapezium rule[7].(AUC between two cut-off points on the curve is a trapezium.All the trapeziums are summarized.) Reported positive and negative predictive values were not regarded as appropriate measures for review becau of their nsitivity to varying prevalence of‘true cas.’
Only studies where the diagnos were made by a structured or mistructured diagnostic interview were con-sidered for nsitivity and specificity measures.
Concurrent validity
The performance of HADS relative to other commonly ud questionnaires and rating scales of anxiety and depression was bad on correlation coefficients for instru-ments with a continuous scale,and nsitivity and specifi-city for instruments categorising individuals as having a disorder or not.
Results
Most studies using HADS have been done on lected samples of patients with cancer or other somatic illness. The psychometric properties of HADS were ldom the main issue in the studies,the sample sizes were mostly relatively small(n<250),and the results were frequently given without further discussion.From general population samples,psychometric properties of HADS were only reported in three papers.Spinhoven et al.[8]reported from three different Dutch samples(total N=5393), Lisspers et al.[9]from a sample of624Swedish subjects and Jimenez et al.[10]from a sample of207elderly Spanish subjects.
Factor structure,discriminant validity and
毕业于internal consistency
Among the19studies reporting factor analysis of HADS (Table1),11studies(total N=14,588)achieved a two-factor structure,5studies(total N=3459)a three-factor structure and2studies(total N=235)a four-factor structure.Most studies ud principal component analysis.The studies of Spinhoven et al.[8]and Lisspers et al.[9]bad on data from the general population both reported a two-factor structure(total N=6017).Spinhoven et al.found that the two-factor solution was stable across different
age groups from the general population and in different clinical samples (general practice,medical outpatients with unexplained somatic symptoms and psychiatric outpatients).Lisspers et al.found the same two-factor structure for both males and females.Dunbar et al.[11]tested different factor models using a confirmatory factor analysis on samples of three different age groups(aged approximately18,39and 58years)from the general population(n=2547).A three-factor model derived from the tripartite theory of anxiety and depression[12]produced the clost fit to the data across all the age groups,though testing the two-factor model achieved by Moorey et al.[13]showed measures of goodness of fit relatively clo to the three-factor model (comparative fit index0.93vs.0.97and root mean square error of approximation0.06vs.0.04).
I.Bjelland et al./Journal of Psychosomatic Rearch52(2002)69–77 70
Bad on the studies HADS performed as a bidimen-sional test,although the factors were not absolutely consist-ent with the subscales of Anxiety and Depression.The most consistent finding was that the HADS-A4item(‘‘I can sit at ea and feel relaxed’’)showed relatively low loadings (<0.60)on the anxiety factor and some loadings on the depression factor(>0.45)[3,9,13–17].
Twenty-one studies reported the Pearson correlation coef-ficient between HADS-A and HADS-D(me
an.56).In ven studies of nonpatient samples[10,17–22]the correlations varied between.49and.74(mean.59).In12studies of somatic patient samples[14,20,23–32]the correlations var-ied between.40and.64(mean.55).The last two studies of psychiatric patients both achieved a correlation of.56[8,33].
Cronbach’s alpha coefficient of internal consistency was reported in15studies(Table1)and varied for HADS-A from.68to.93(mean.83),and for HADS-D from.67to .90(mean.82)[3,9,13–16,21,30,34–40].
HADS as a ca finder for anxiety disorders and depression Optimal balance between nsitivity and specificity for HADS as a screening instrument was achieved most fre-quently at a cut-off score of8+for both HADS-A and HADS-D giving nsitivities and specificities for both subscales of approximately0.80.
The findings from the24papers reporting nsitivity and specificity are summarid according to the popula-tions studied.More details are given in Table 2.Only one community survey(n=330)was found[41]and ROC curves identified8+to be an optimal cut-off score for caness for both anxiety disorders and depression bad on ICD-9.Sensitivity and specificity for both anxiety and depression were approximately0.90.The author reported similar results in samples from medical inpatient populations.
HADS was tested in three studies of primary care populations.Wilkinson and Barczak[42](n=100)found an excellent ability of HADS to detect DSM-III-defined psychiatric morbidity,and the ROC curves showed that a score of8+was the optimal threshold.The AUC was found to be0.96.el Rufaie and Absood[35]studied patients(n=217)attending a primary health care centre. The ROC curves(calculated by us)showed that the optimal cut-off scores for caness were9+for HADS-A (nsitivity0.66,specificity0.93)and7+for HADS-D (nsitivity0.66,specificity0.97),when using DSM-III diagnos obtained by the Clinical Interview Schedule as gold standard.AUC(calculated by us)was0.86for both anxiety and depression.Lam et al.[43],however,identified (by ROC curves)a lower optimal cut-off in their sample from a general practice(n=100),3+for HADS-A and 6+for HADS-D giving the nsitivities0.67and0.78and specificities0.83and0.91,respectively.Their gold stand-ard was not reported,but the Clinical Interview Schedule was ud,presumably giving DSM-III diagnos.
Table1
Factor analysis and internal consistency of the HADS
Version of Method of factor Number of Cronbach’s a
Reference HADS n analysis factors HADS-A HADS-D Anderson[75]Swedish163PCA4
Bedford et al.[16]English132PCA2.83.86 Brandberg et al.[39]Swedish273PCA3.85.81 Costantini et al.[38]Italian197PCA2.85.89 Dagnan et al.[15]English341PCA2.84.83 Dunbar et al.[11]English2547CFA3
Hammerlid et al.[36]Norwegian Swedish351PCA2.89.82 Herrmann et al.[3]German5338PCA?a2.80.81 Leung et al.[21]English100PCA3
Chine100PCA3.81.74 Lewis[29]English117PCA3
Lisspers et al.[9]Swedish624PCA2.84.82 Martin and Thompson[40]English72MLA4.82.78 Martin and Thompson[30]English194MLA3.76.72 Moorey et al.[13]English568PCA2.93.90 Razavi et al.[31]French228PCA3
Savard et al.[14]French Canadian162PCA2.89.89 Sigurdardottir et al.[72]Swedish89PCA2
Soriano and Salavert[17]Spanish621PCA2
Spinhoven et al.[8]Dutch6165PCA2
Botega et al.[34]Portugue78.68.67
el Rufaie et al.[35]Arabic217.78.88 Wettergren et al.[37]Swedish20.88.86 CFA:confirmatory factor analysis;HADS:Hospital Anxiety and Depression Scale;HADS-A:Anxiety subscale of HADS;HADS-D:Depression subscale of HADS;MLA:maximum likelihood factor analysis;PCA:principal component analysis.
a Not reported.
I.Bjelland et al./Journal of Psychosomatic Rearch52(2002)69–7771
We identified12studies that addresd optimal cut-off scores for caness in noncancer medical patients(total N=2109).For HADS-A the mean optimal cut-off score was approximately8+(7.5),with resulting mean nsitivity 0.90,and mean specificity0.78.Similarly,for HADS-D the mean optimal cut-off score also was approximately 8+(8.1),with mean nsitivity0.83,and mean specificity 0.79.Johnson et al.[44]studied poststroke patients(n=93) and we estimated their optimal cut-off scores to be5+for HADS-A and4+for HADS-D,giving significantly lower specificity for both anxiety and depression(0.46and0.44, respectively)than in studies of other medical samples. Using the highest score of either HADS-A or HADS-D as an indicator of psychiatric morbidity,Morriss and Wearden [45]found that a cut-off score for caness of10+resulted in nsitivity0.92and specificity0.71in a sample of chronic fatigue syndrome patients(n=136).Hamer et al.
[46]prented findings from a sample of100lf-harming patients with an ROC curve,which showed8+to be the optimal cut-off score of caness of HADS-D giving ns-itivity0.88and specificity0.78.
In the10studies of cancer patients(total N=1803),the mean optimal cut-off score for caness on HADS-A was approximately9+(8.8),with mean nsitivity0.72,and mean specificity0.81.For HADS-D the mean optimal cut-off score of caness was approximately8+(8.3),with mean nsitivity0.66,and
最小的国家是哪个
mean specificity0.83.
Concurrent validity
Six studies reported the correlations between Beck’s Depression Inventory(BDI)and HADS.The correlations between BDI and HADS-D were.62to.73,BDI and HADS-A.61to.83and BDI and HADS-total score (HADS-T).73[9,14,47–50](Table3).Two studies demon-strated that the correlations between the General Health Questionnaire(GHQ-28)and HADS-D were.50and.66, and between GHQ-28and HADS-A.50and.68[18,24].The correlations between the Clinical Anxiety Scale[51]and HADS-A were.69and.75in two studies[52,53].The correlations between Spielberger’s State-Trait Anxiety Inventory(STAI)and HADS were examined in five studies [9,14,20,54,55].Between STAI and HADS-A the correla-tions were in the range of.64to.81,between STAI and HADS-D.52to.65and between STAI and HADS-T.68 to.71.Two studies examined the relationship between the
河北气象局Table3
Correlation coefficients between the HADS and other questionnaires and interview-bad measures
Compared Correlation coefficients between HADS and the other questionnaire Reference questionnaire HADS-A HADS-D HADS-T Beck et al.[47]BDI-PC.62
Lisspers et al.[9]BDI.64.71.73
Sua`rez-Mendosa et al.[48]BDI.83
Savard et al.[14]BDI.68.70
Tedman et al.[49]BDI.61.73
Watson et al.[50]BDI.69
Lewis and Wesly[60]GHQ-12.75 Caplan[18]GHQ-28.68.66
霄壤之别Chandarana et al.[24]GHQ-28.50.50
Elliot[54]STAI.64.52
Herrmann et al.[20]STAI.66.59
Lisspers et al.[9]STAI-S.64.68
STAI-T.66.64.71 Millar et al.[55]STAI-S.81
Savard et al.[14]STAI-S.78.65
Lepine et al.[59]MADRS.62
描写小狗外貌的句子Snaith and Taylor[52]MADRS.37.81
Upadhyaya and Stanley[53]MADRS.80
Aylard et al.[58]MADRS(item3).77
CAS.67 Snaith and Taylor[52]CAS.69.44
Upadhyaya and Stanley[53]CAS.75
Spinhoven and van der Does[56]SCL-90,Anxiety,Depression.49.69
Watson et al.[50]SCL-90,Anxiety,Depression.73.67
Lepine et al.[59]HAMA-S,HAMA-P,HAMA-T.34.40.44
Millar et al.[55]V AS.74
BDI:Beck Depression Inventory;BDI-PC:Beck Depression Inventory for Primary Care;CAS:Clinical Anxiety Scale;HADS-A:Anxiety subscale of HADS; HADS-D:Depression subscale of HADS;HADS-T:Total score of HADS;HAMA-S:Hamilton Anxiety Scale—Somatic Items;HAMA-P:Hamilton Anxiety Scale—Psychic Items;HAMA-T:Hamilton Anxiety Scale—Total Scale;MADRS:Montgomery–Asberg Depression Rating Scale;SCL-90:Symptom Checklist90Scale;STAI-S:Spielberger State-Trait Anxiety Inventory—State Form;STAI-T:Spielberger State-Trait Anxiety Inventory—Trait Form;V AS: Visual Analogue Scale.
I.Bjelland et al./Journal of Psychosomatic Rearch52(2002)69–7773
SCL-90subscales of Anxiety and Depression and HADS [50,56].The correlations between SCL-90Anxiety and HADS-A were.49and.73,while the correlations between SCL-90Depression and HADS-D were.69in both studies. Finally,in four studies the correlations between the interview-bad Montgomery Asberg Depression Rating Scale[57]and HADS-D were in the range.62to.81,while the correlation with HADS-T was.77[52,53,58,59].Low correlations(.34 to.44)were found between Hamilton Anxiety Rating Scale and HADS-A[59].
Three studies[42,60,61]compared the nsitivity and specificity of HADS to that of various editions of
GHQ. HADS and GHQ had clo to identical nsitivities and specificities,both at the level of0.80for HADS-A, HADS-D as well as for HADS-T.Clarke et al.[62] compared HADS,GHQ and BDI(against DSM-III-R dia-gnos)by using Quality ROC curves.Here the GHQ performed marginally better than HADS and BDI. Discussion
Bidimensionality
The results of our review support the two-factor struc-ture of HADS.In most studies where empirically bad exploratory factor analys were ud HADS revealed two relatively independent dimensions of anxiety and de-pression cloly identical to the Anxiety and Depression subscales.The three-factor model supported by the theory-driven confirmatory factor analysis of Dunbar et al.[11], however,challenge the bidimensionality of HADS.Never-theless,the fit measures of the two-factor model propod by Moorey et al.[13]were relatively clo to the three-factor model.In addition,Dunbar et al.did not test more than one two-factor model,while four three-factor models were tested,among whom one showed a much wor fit than the two-factor model.
Recognising the extensive comorbidity between anxiety and depression[63–65],the moderate to strong correlations between HADS-A and HADS-D subscales reported were to be expected.Burns a
nd Eidelson[66]argued that the correlation between any valid and reliable measure of depression and anxiety should be at the.70level,not becau of shared symptoms between anxiety and depres-sion,but becau of a common causal factor.However, other authors have claimed that a low correlation between the two measures of anxiety and depression is a hallmark of good discriminant validity of a bidimensional test[12]. Watson et al.[50]stated that:‘‘Phenomenologically,anxi-ety and depression are clearly distinct from each other. Anxiety is centered on the emotion of fear and involves feelings of worry,apprehension,and dread;in contrast, depression is dominated by the emotion of sadness and is associated with feelings of sorrow,hopelessness,and gloom.Nevertheless,despite their eming distinctiveness,it has proven difficult to distinguish the constructs empirically.Many studies have shown that lf-report measures are highly correlated,with coefficients typically in the.45to.75range.’’Some authors have recommended not only the u of correlations between subscales to asss their divergent validity,but also a multitrait–multimethod approach[67].In our arch,however,no papers reported such a comprehensive asssment.
Internal consistency
It has been recommends that Cronbach’s coefficient alpha should be at least.60for a lf-report instrument to be reliable[68].This demand was fulfilled in all studies of HADS in various translations t
hat report data on internal consistency.Similar findings of internal consistency from different translations of HADS supported the robustness of the instrument.
HADS as a ca finder for anxiety disorders
and depression
In this review the threshold values identified for optimal balance between nsitivity and specificity showed rel-atively little variability,and they were very clo to8+, defined as the cut-off for‘possible cas’suggested by Zigmond and Snaith in their original paper on HADS[1]. This threshold was found for HADS-A and HADS-D in the general population as well as in somatic patients samples. Two papers reported some deviating cut-off values;Lam et al.[43]found an optimal cut-off value of HADS-A at 3+and of HADS-D at6+,while Johnson et al.[44]found the optimal cut-off values of both HADS-A and HADS-D at 4+.An explanation may be that in both studies HADS was administered completely or partly as an interview,possibly biasing the respons to the items.
The nsitivity and specificity of HADS-A and HADS-D with a threshold of8+were most often found to be in the range of0.70to0.90.The variation in both optimal cut-off values and nsitivity and specificity might be due to differences in diagnostic systems,‘gold standard’instru-ments,HADS transl
ations ud[21,69,70],as well as to differences in samples and procedures in administration of HADS[71](such an explanation may also be applied to the varying results of the other psychometric properties of HADS).Among three studies of general practice patients AUCs were found to be0.84–0.96.The results indicate excellent ca finding abilities of HADS in unlected samples of patients eking a general practitioner. Concurrent validity
This review revealed that HADS,despite of its brevity, exhibited similar nsitivity and specificity as longer ver-sions of GHQ.When compared to other questionnaires for anxiety and depression in common u such as BDI,STAI,
I.Bjelland et al./Journal of Psychosomatic Rearch52(2002)69–77 74
