- Email: [email protected]
The structure of the hospital anxiety and depression scale (HAD): An appraisal with normal, psychiatric and medical patient subjects
Pergamon PII:
Person. indiuid. D/jf: Vol. 23, No. 3, pp. 473478, 1997 a 1997 Elsevier Science Ltd. All rights reserved Printed in Great Britain 0191-8869/97 $17.00+0.00 SOlSl-8869(97)OOOSS-5
THE STRUCTURE OF THE HOSPITAL ANXIETY AND DEPRESSION SCALE (HAD): AN APPRAISAL WITH NORMAL, PSYCHIATRIC AND MEDICAL PATIENT SUBJECTS Alan Bedford,‘,’ Karel de Pauw3 and Eryk Grant4 ‘York Clinical Psychology Services, Monkgate Health Centre, 31-35 Monkgate, York Y03 7PB, ‘Department of Psychology, University of York, York YOl 5DD, ‘Department of Psychiatry, St James’s University Hospital, Beckett Street, Leeds LS9 7TF, and 4Evaluation and Research Support Unit, Verandah Cottages, Bluebeck Drive, Clifton Hospital, Shipton Road, York Y03 6RA, England (Received 20 August 1996; receivedfor publication 10 April 1997)
Summary-Four published studies which include factor analyses of the items of the Hospital Anxiety and Depression Scale (HAD) are reviewed and new data on psychiatric out-patients are presented. By contrast with the rest, the two studies in English with British subjects clearly support the interpretation of the HAD as a bidimensional measure. However, one anxiety item appears sex-related and misaligned. The problem of item meaning in other cultures is raised. Overall, despite the HAD’s use internationally, there has been a lack of systematic structural evaluation. 0 1997 Elsevier Science Ltd
INTRODUCTION
The Hospital Anxiety and Depression Scale (Zigmond & Snaith, 1983) is a brief, self-administered questionnaire specifically constructed for use with physically ill patients. Hence, its 14 items, seven each for anxiety and depression, attempt to avoid somatic symptoms, whilst five of the depression items are intended to assess anhedonia. The scale’s acceptance is witnessed by its wide clinical use in British hospitals, including psychiatric units, as well as around twenty publications annually in the professional literature and over thirty translations. However, little attention has been given to the item structure of the HAD, and no such report has involved patients undergoing psychiatric treatment. Hence, the aims of this report are to review the current findings on scale structure and relate them to a first report of psychiatric patient data. In subsequent sections the HAD item numbers cited are those in the British test booklet, and as in the Appendix.
STUDIES
INVOLVING
SCALE
STRUCTURE
ANALYSIS
In the first study, with an HAD translation, Razavi, Delvaux, Farvacques and Robaye (1989) reported the responses of 228 French cancer in-patients. Their mean age was 55 yr, SD 14.5, and 67.1% of Ss were female. The employment of a principal components analysis (‘la methode d’Hotelling’) and Kaiser’s eigenvalues-greater-than-one criterion resulted in a three-component solution. When only loadings greater than 0.40 were considered by them, their first component was of six depression items and four anxiety items. Component two had loadings from two depression items and one anxiety item, whilst component three comprised a single anxiety item. The data were then subjected to a factor analysis (‘l’analyse de Thurstone’), the only change being that a depression item now loaded on factor one instead of two. No information on type of rotated solutions, eigenvalues or percentages of variance accounted for was given. The only tables provided were the ‘matrice d’intercorrelations’ of the items and one of item content. From the former it is apparent that their item A6 (Item 11 in the British test booklet) had a curious pattern of correlates. With two other anxiety items it correlated 0.79 and 0.75, but with the rest of the anxiety subscale items the correlations ranged only from 0.11 to 0.25. Likewise, this same item (‘I feel restless as if I have to be on the move’) correlated 0.71 and 0.51 with two items of the depression subscale and between 0.10 and -0.08 with the remaining items. Subsequently, these authors, referring to this their earlier article, said that “factorial analysis ?AlD 23-M
473
474
Alan Bedford et
al.
using Hotelling and Thurstone methods failed to show the bidimensionality” (Razavi, Delvaux, Farvacques & Robaye, 1990). This second paper contained HAD, but not subscale, means scores of 9.3 (SD 4.9), 16.1 (SD 7.3) and 22.4 (SD 7.1) for the cancer patients divided into groups of ‘no psychiatric diagnosis’, ‘adjustment disorder’ and ‘major depressive disorder’, respectively. Next chronologically, Moorey, Greer, Watson, Gorman, Rowden, Tunmore, Robertson and Bliss (199 1) examined both the factor structure and stability of the HAD in 568 cancer out-patients. The mean age of the sample (70.8% men, 29.2% women) was 55.1 yr (SD 12.8). These authors performed (a) principal components analyses, with both orthogonal and oblique rotations, on the 14 items, (b) principal components analyses on the two subscales separately, and (c) oblique principal components analyses on the whole scale, first separated by sex and then with the sample split in half. In brief, these analyses produced unidimensional solutions for both the separate subscales of anxiety (57% of the variance and Cronbach’s alpha 0.93) and depression (47% of the variance and Cronbach’s alpha 0.90). The HAD as a whole was regarded as bidimensional, tapping these two separate but related concepts which together accounted for 53% of the variance. The only weakness was for Item 7 of the anxiety subscale (‘sit at ease and feel relaxed’), which consistently loaded higher on the depression component (around 0.50) than on the anxiety component (about 0.30). The mean scores on the HAD were 5.4 for anxiety (SD 4.1; range O-19) and 3.0 for depression (SD 3.0; range O-1 5). Their total mean score was therefore about 8.4. This study will be considered in greater detail subsequently in relation to our own findings. The next study involved two non-clinical groups using a Swedish translation of the HAD (Andersson, 1993). One sample consisted of 40 men and 47 women, parents of mentally handicapped children, whilst the other was a matched group of 29 men and 47 women, the parents of nonhandicapped children. The data were analyzed using the principal components method with varimax rotation and orthogonal transformation. Andersson’s interpretation of the two-component solution was the conventional one that “an item ought to be represented. . . where it is most powerful, that is where the item has its highest loading”. On this basis, which ignores both the absolute size of loading (four of the ‘powerful’ items were below 0.40) and the differential size between the loadings (one, Item 6 (‘I feel cheerful’), was 0.495 vs 0.472, i.e. equally split). Using the author’s criteria, of the eight items constituting component one (anxiety) only four were from the original anxiety scale. The second component was equally divided in having three items from each subscale. The two components correlated 0.29 and, with eigenvalues of 3.68 and 1.47, accounted for only 36.8% of the variance. The author concluded that, “there is no statistical evidence for the original split of the questions into two parts, Anxiety and a Depression subscale”. He then went on to compute threeand four-factor solutions. Only the latter, which he regarded as superior, was reported, and with additional eigenvalues of 1.45 and 1.03 accounted for 54.5% of the variance. Component one consisted of three high loading (above 0.70) depression items and was named ‘Well-being’. The second component, ‘Momentary anxiety’, comprised five anxiety and one depression item, whilst component three, ‘Power to relax’, was made up of two depression and one anxiety item. The final and fourth component, which the author chose not to interpret, had just one item from each subscale. Andersson’s item intercorrelation matrix shows, unsurprisingly, that these latter two items (8, ‘I feel as if I am slowed down’, and 11, ‘I feel restless as if I have to be on the move’) showed a low correlation with the rest of the scale items, ranging from 0.01 to 0.12 and from 0.04 to 0.33, respectively (Andersson, 1993). His suggestion that there may be a different interpretation for Item 7 (‘sit and feel relaxed’) between general population members and those under treatment may have broader relevance than that specific item. The mean score for answers on the anxiety scale was 6.6 and on the depression scale 4.1. The total mean was therefore around 10.7. The most recent report (Leung, Ho, Kan, Hung & Chen, 1993) involved 100 first- and fourthyear Hong Kong medical students completing both the English and Chinese versions of the HAD in one sitting, with the Eysenck Personality Questionnaire (Eysenck & Eysenck, 1975) used as a filler. In seven preliminary drafts of translations and back-translations most difficulties and disagreement arose from expressions such as Item 4 (‘I can laugh and see the funny side of things’), Item 9 (‘butterflies in the stomach’) and choices for the question in Item 10 (‘lost interest in appearance’).
The structure of HAD
415
To assess the linguistic equivalence of the versions r-tests were performed, with statistically significant differences being found for Anxiety Items 9 and 11 (‘feel restless as if I have to be on the move’) and Depression Item 2 (‘still enjoy the things I used to’). When the item correlation matrices for the English version were examined “all anxiety items except Item 9 showed high correlation with the depression subscores while all depression items except Item 10 showed high correlation with the anxiety subscores”. For the Chinese version “all depression items except Item 10 showed high correlation with the anxiety subscores while all anxiety items showed high correlation with the depression subscore without exception”. The correlation between the subscores was given as “Pearson’s coefficient 0.56 for the English version and 0.61 for the Chinese version”. Cronbach’s alpha for the anxiety and depression subscales, Chinese version only reported, was found to be 0.81 and 0.74, respectively. When “the data were further factor-analyzed. . . oblique rotation was performed. . . “. It is unclear whether this is a factor or a principal component analysis. For both language versions this resulted in three factor/component solutions. For the Chinese version these “accounted for 55.9% of the total variance: 37.5%, 9.6% and 8.8% for factors 1, 2 and 3, respectively”. Their Table 5 only presents the highest loading for each of the 14 items, although from this it is obvious that factor one is a depression factor. It comprises items from that a priori scale (5 in the English version and 6 in the Chinese version) plus Anxiety Item 7 (‘sit at ease and feel relaxed’). Factor two is an anxiety factor of four and five items, respectively. Of the four items (English) and two items (Chinese), versions of factor three Depression Item 10 and Anxiety Item 11 are ever-present. With regard to descriptive statistics, the English version had an Anxiety mean of 7.3 (SD 3.4) and a Depression mean of 5.0 (SD 3.1). The total HAD mean score was therefore 12.3. The Chinese translation gave an Anxiety mean of 7.3 (SD 3.4) and a Depression mean of 4.9 (SD 3.1). This resulted in a highly comparable total HAD mean score of 12.2. PRESENT
STUDY
Method and subjects
The 132 Ss were consecutive referrals to an adult psychiatry out-patient clinic in a district general hospital. Eighty-nine per cent of these patients were from general practitioners. There were 62 men and 70 women, with mean age of 37.6 yr (SD 12.0). Clinical and diagnostic formulations were conducted by a consultant psychiatrist (K.P.) and based upon a battery of assessment techniques, including the HAD. Only this measure at first interview is the concern of the present article, although it is anticipated that other results will be presented in subsequent reports. RESULTS For reasons of comparability we followed the sequence of procedures as carried out by Moorey et al. (1991). First, a principal components analysis of the total group was performed on the 14 items with rotation to a specified two-component orthogonal solution. The first component had an eigenvalue of 5.82 and accounted for 41.6% of the variance, whilst the second had an eigenvalue of 1.70 with 12.1% of the variance. When their criterion of taking loadings of 0.45 as the cut-off point was employed, the anxiety and depression items loaded satisfactorily onto separate factors. The weakest was Item 7, from the anxiety subscale (‘can sit at ease and feel relaxed’), which, whilst loading 0.50 on the anxiety component, loaded 0.43 on the depression component. For the anxiety subscale a mean of 13.9 (SD 4.4) was obtained, whilst for depression the mean was 9.9 (SD 4.9). The HAD mean was 23.7 (SD 8.1). Next an oblique rotation was performed. Table 1 shows the comparison between the component loadings found by Moorey et al. (1991) and those of the current study. Again all the items unequivocally loaded on the appropriate component except Anxiety Item 7, which loaded higher on depression in the Moorey et al. (199 1) study, and was almost equally divided in our data. For our psychiatric group the highest anxiety loading was 0.81 for Item 3 (‘frightened feeling as if something awful is about to happen’) and the lowest was 0.48 for Item 7. Among the depression subscale items a highest loading of 0.82 was found for Item 2 (‘still enjoy the things I
476
Alan Bedford et al. Table 1. Component loadings (oblique rotation) of the anxiety and depression subscale items of the HAD as found by Moorey et al. (1991) and in the current studv
HAD items
I
Depression Moorey Current
3 5 7 9 11 13
0.13 0.09 0.17 0.51 -0.10 -0.07 -0.01
2 4 6 8 IO 12 14
0.78 0.74 0.76 0.51 0.49 0.79 0.65
Anxiety subscale 0.44 0.15 0.25 0.45 0.25 0.03 0.24 Depression subscale 0.82 0.76 0.72 0.69 0.63 0.72 0.65
Anxiety Moorey Current
0.71 0.77 0.72 0.30 0.86 0.66 0.83
0.62 0.81 0.63 0.48 0.73 0.50 0.77
-0.02 -0.06 -0.01 0.14 0.07 0.01 -0.08
0.08 0.27 0.28 0.12 - 0.05 0.27 0.21
used to enjoy’) whilst the lowest was Item 10 (‘lost interest in my appearance’) with 0.63. The correlation between the two oblique components was 0.53, which is closely similar to the 0.50 found by Moorey et al. (1991). Like them we went on using “this method of rotation.. . in the rest of the study to test the stability of the factor structure”. Hence, the next step was to carry out principal components analyses with oblique solution on the two subscales of the HAD separately. Analysis of the anxiety items revealed a single component with eigenvalue 3.54 accounting for 50.5% of the variance. Item 3 had the highest loading (0.80) and Item 11 (‘feel restless as if I have to be on the move’) the lowest (0.45). Cronbach’s alpha was 0.825 for the anxiety scale. Item 11 had the lowest item-total scale correlation of 0.34; when this item was removed the resulting 6-item scale had an increased Cronbach’s alpha of 0.843. With the depression items an eigenvalue of 3.82 was obtained for the only component, which accounted for 54.6% of the variance. Items 2 and 4 (‘can laugh and see the funny side of things’) had joint highest loadings at 0.81, whilst Item 10, with 0.58, was lowest. Cronbach’s alpha was 0.856 for the depression scale. Item 10 had the lowest item-total scale correlation of 0.47; the deletion of this item resulted in a trivial change of Cronbach’s alpha to 0.857. For completeness the full HAD Cronbach was computed and found to be 0.883. Item 11 was again the weakest with an item-total correlation of only 0.28; dropping this item raised Cronbach’s alpha to 0.888. To test the stability of the component structure obtained, the sample was split in half by taking alternate cases and again each was subjected to a principal component oblique solution with two-component specification. As can be seen in Table 2, each item’s highest loading is on the same component in both samples. Next the data were divided by sex into 62 men and 70 women. Table 3 compares their descriptive statistics. No mean sex differences were obtained, as the largest t-test value, for the total scale scores, was only 1.04. Finally, the male and female subscale scores were subjected separately to a principal components oblique solution with two-component specification. A correlation of 0.62 was found between the components in men; for women the correlation was 0.47. Table 4 shows the loadings by sex for each anxiety and depression item. Whereas Moorey et al. (1991) found that for women Depression Items 8 and 10 just failed their 0.45 criterion, in our data they had loadings of 0.65 and 0.74, respectively. The ‘weakest’ item in that subscale was Item 10 in men, with 0.47 against a -0.09 loading on the anxiety component. For the anxiety subscale our data repeated their finding that Item 7 met the criterion in both sexes on the ‘wrong’ component, i.e. depression. However, for our males there was an even higher loading on the anxiety component (0.64) but not for our females (0.23). Generally, the cross-loadings in our data were larger than those of Moorey et al. (1991), particularly for males and the anxiety items.
477
The structure of HAD Table 2. Component
Depression Sample 2 Sample I
HAD items
Anxiety Depression Total
3
SD
13.5 9.4 23.0
4.7 5.2 8.9
Anxiety Sample 2 Sample 1
3 5 7 9 11 13
38 17 35 42 19 -4 28
Anxiety subscale 45 -7 IO 45 29 -1 18
66 83 68 42 71 50 72
64 81 61 59 74 48 81
2 4 6 8 10 I2 14
83 73 75 56 12 69 64
Depression subscale 80 74 67 81 55 72 64
-6 27 28 28 -7 18 -6
17 35 34 -2 -1 42 32
Table 3. Male and female means, standard
HAD
loadings of split halves of total sample (oblique rotation)
deviations,
ranges and Cronbach
Males (N = 62) Cronbach’s Range I-21 o-19 l-39
0.850 0.869 0.900
Table 4. Component
HAD items
alpha
loadings
alphas for the HAD anxiety, depression
R
SD
14.2 10.3 24.4
4.1 4.6 7.4
2-2 I 1-21 542
alpha
0.800 0.840 0.860
for men and women (oblique rotation)
Depression Men Women
Anxiety Men
Women
3 5 7 9 I1 13
49 27 46 45 43 -6 43
Anxiety subscale 44 16 IO 53 18 19 15
61 76 52 64 57 61 71
63 81 70 23 78 40 75
2 4 6 8 10 12 14
86 79 81 13 47 80 61
Depression subscale 78 76 65 65 74 66 66
-9 28 23 12 -9 23 -2
-4 16 25 -1 -3 22 31
1
and total scales
Females (N = 70) Cronbach’s Range
DISCUSSION Of the four previously published studies reported here, only Moorey et al. (1991) appear to find a two factor-component solution. However, Leung et al. (1993) after clearly identified anxiety and depression dimensions in both their English- and Chinese-language versions, end up with a small third mixed factor in one case of only two items. This is of no practical or theoretical use, and therefore their study probably belongs more properly to the Moorey et al. (1991) bidimensionality camp. As our own results are in line with Moorey et al. (1991), this means that all three analyses using an English-language version of the HAD are interpretable in terms of a two oblique component solution. By contrast, Andersson’s Swedish version (Andersson, 1993) produces two orthogonal components of mixed items, and the French version of Razavi et al. (1989) found that 11 and 10 of the items loaded on the first factor or component, respectively. Until other reports appear using the
Alan Bedford et
478
al.
English and translated versions, the generalisability of these structural findings can only be regarded as tentative. However, within these limitations the two studies in English with British Ss both find Anxiety Item 7 to be somewhat misaligned in having high loadings on the depression component. Our data suggested that it was a more pure measure of depression for females (loadings of 0.53 vs 0.23) than for males (0.45 vs 0.64). It may be apparent from our earlier description that the majority of reports fail to give a full account of the statistical (and sometimes other) procedures carried out. Also, these omissions are not consistent across studies. There are, of course, other important differences between these reports other than those concerned with data analysis and reportage. The most obvious ones relate to the status of the respondents, i.e. patients or non-hospital Ss, their ages, and the representativeness and balance of each sex. These add to the imponderables when attempting to evaluate the HAD’s structure. Additionally, we have already mentioned that Leung et al. (1993) experienced difficulties in translations and backtranslations of items for cultural reasons. Similarly, Cheng and Hamid (1995), with Rosenberg’s Self-Esteem Scale, noted that problems in syntax may occur when a scale is translated and its psychometric properties may then be adversely affected. These points have been specifically addressed by El-Rufaie, Absood and Abou-Saleh (1997) in producing a culture-oriented screening scale for use in Arabic primary health care settings. The 12 anxiety-depression items were mainly derived, and modified, from the HAD scales. It is to be regretted that a measure used internationally has been subjected to so little evaluation, and that the few structural researches carried out make little contribution to the advancement of practical knowledge. REFERENCES Andersson, E. (1993). The Hospital Anxiety and Depression Scale. Homogeneity of the subscales. Social Eehaoiour and Personality, 21, 197-204. Cheng, S. T. & Hamid, P. N. (1995). An error in the use of translated scales. The Rosenberg Self-Esteem Scale for Chinese. Perceptual and Motor Skills, 81,431&434. El-Rufaie, 0. E. F., Absood, G. H. & Abou-Saleh, M. T. (1997). The Primary Care Anxiety and Depression (PCAD) Scale. A culture-oriented screening scale. Acta Psychiatrica Scandinauica, 95, 119-124. Eysenck, H. J. & Eysenck, S. B. G. (1975). The Eysenckpersonality questionnaire. Sevenoaks: Hodder and Stoughton. Leung, C. M., Ho, S., Kan, C. S., Hung, C. H. & Chen, C. N. (1993). Evaluation of the Chinese version of the Hospital Anxiety and Depression Scale. A cross-cultural perspective. International Journal of Psychosomatics, 40, 29-34. Moorey, S., Greer, S., Watson, M., Gorman, C., Rowden, L., Tunmore, R., Robertson, B. & Bliss, J. (1991). The factor structure and factor stability of the Hospital Anxiety and Depression Scale in patients with cancer. British Journal of Psychiatry, 158, 255-259. Razavi, D., Delvaux, N., Farvacques, C. & Robaye, E. (1989). Validation de la version Francaise du HADS dans une population de patients cancereux hospitalises. Reoue de Psychologie Appliquee, 39(4), 295-308. Razavi, D., Delvaux, N., Farvacques, C. & Robaye, E. (1990). Screening for adjustment disorders and major depressive disorders in cancer in-patients. British Journal of Psychiatry, 156, 79-83. Zigmond, A. B. & Snaith, R. P. (1983). The Hospital Anxiety and Depression Scale. Acta Psychiatrica Scandinauica, 67, 361-370.
APPENDIX Items of the Hospital IA. 2D. 3A. 4D. 5A. 6D. 7A. 8D. 9A. 10D. 1 IA. 12D. 13A. 14D.
Anxiety
and Depression
Scale (HAD)
I feel tense or ‘wound up’. I still enjoy the things I used to enjoy. I get a sort of frightened feeling as if something awful is about to happen. I can laugh and see the funny side of things. Worrying thoughts go through my mind. I feel cheerful. I can sit at ease and feel relaxed. I feel as if I am slowed down. I get a sort of frightened feeling like ‘butterflies’ in the stomach. I have lost interest in my appearance. I feel restless as if I have to be on the move. I look forward with enjoyment to things. I get sudden feelings of panic. I can enjoy a good book or radio or TV programme.
Item numbers
are followed
by ‘A’ to denote an anxiety
subscale
item and ‘D’ for a depression
item.
Person. indiuid. D/jf: Vol. 23, No. 3, pp. 473478, 1997 a 1997 Elsevier Science Ltd. All rights reserved Printed in Great Britain 0191-8869/97 $17.00+0.00 SOlSl-8869(97)OOOSS-5
THE STRUCTURE OF THE HOSPITAL ANXIETY AND DEPRESSION SCALE (HAD): AN APPRAISAL WITH NORMAL, PSYCHIATRIC AND MEDICAL PATIENT SUBJECTS Alan Bedford,‘,’ Karel de Pauw3 and Eryk Grant4 ‘York Clinical Psychology Services, Monkgate Health Centre, 31-35 Monkgate, York Y03 7PB, ‘Department of Psychology, University of York, York YOl 5DD, ‘Department of Psychiatry, St James’s University Hospital, Beckett Street, Leeds LS9 7TF, and 4Evaluation and Research Support Unit, Verandah Cottages, Bluebeck Drive, Clifton Hospital, Shipton Road, York Y03 6RA, England (Received 20 August 1996; receivedfor publication 10 April 1997)
Summary-Four published studies which include factor analyses of the items of the Hospital Anxiety and Depression Scale (HAD) are reviewed and new data on psychiatric out-patients are presented. By contrast with the rest, the two studies in English with British subjects clearly support the interpretation of the HAD as a bidimensional measure. However, one anxiety item appears sex-related and misaligned. The problem of item meaning in other cultures is raised. Overall, despite the HAD’s use internationally, there has been a lack of systematic structural evaluation. 0 1997 Elsevier Science Ltd
INTRODUCTION
The Hospital Anxiety and Depression Scale (Zigmond & Snaith, 1983) is a brief, self-administered questionnaire specifically constructed for use with physically ill patients. Hence, its 14 items, seven each for anxiety and depression, attempt to avoid somatic symptoms, whilst five of the depression items are intended to assess anhedonia. The scale’s acceptance is witnessed by its wide clinical use in British hospitals, including psychiatric units, as well as around twenty publications annually in the professional literature and over thirty translations. However, little attention has been given to the item structure of the HAD, and no such report has involved patients undergoing psychiatric treatment. Hence, the aims of this report are to review the current findings on scale structure and relate them to a first report of psychiatric patient data. In subsequent sections the HAD item numbers cited are those in the British test booklet, and as in the Appendix.
STUDIES
INVOLVING
SCALE
STRUCTURE
ANALYSIS
In the first study, with an HAD translation, Razavi, Delvaux, Farvacques and Robaye (1989) reported the responses of 228 French cancer in-patients. Their mean age was 55 yr, SD 14.5, and 67.1% of Ss were female. The employment of a principal components analysis (‘la methode d’Hotelling’) and Kaiser’s eigenvalues-greater-than-one criterion resulted in a three-component solution. When only loadings greater than 0.40 were considered by them, their first component was of six depression items and four anxiety items. Component two had loadings from two depression items and one anxiety item, whilst component three comprised a single anxiety item. The data were then subjected to a factor analysis (‘l’analyse de Thurstone’), the only change being that a depression item now loaded on factor one instead of two. No information on type of rotated solutions, eigenvalues or percentages of variance accounted for was given. The only tables provided were the ‘matrice d’intercorrelations’ of the items and one of item content. From the former it is apparent that their item A6 (Item 11 in the British test booklet) had a curious pattern of correlates. With two other anxiety items it correlated 0.79 and 0.75, but with the rest of the anxiety subscale items the correlations ranged only from 0.11 to 0.25. Likewise, this same item (‘I feel restless as if I have to be on the move’) correlated 0.71 and 0.51 with two items of the depression subscale and between 0.10 and -0.08 with the remaining items. Subsequently, these authors, referring to this their earlier article, said that “factorial analysis ?AlD 23-M
473
474
Alan Bedford et
al.
using Hotelling and Thurstone methods failed to show the bidimensionality” (Razavi, Delvaux, Farvacques & Robaye, 1990). This second paper contained HAD, but not subscale, means scores of 9.3 (SD 4.9), 16.1 (SD 7.3) and 22.4 (SD 7.1) for the cancer patients divided into groups of ‘no psychiatric diagnosis’, ‘adjustment disorder’ and ‘major depressive disorder’, respectively. Next chronologically, Moorey, Greer, Watson, Gorman, Rowden, Tunmore, Robertson and Bliss (199 1) examined both the factor structure and stability of the HAD in 568 cancer out-patients. The mean age of the sample (70.8% men, 29.2% women) was 55.1 yr (SD 12.8). These authors performed (a) principal components analyses, with both orthogonal and oblique rotations, on the 14 items, (b) principal components analyses on the two subscales separately, and (c) oblique principal components analyses on the whole scale, first separated by sex and then with the sample split in half. In brief, these analyses produced unidimensional solutions for both the separate subscales of anxiety (57% of the variance and Cronbach’s alpha 0.93) and depression (47% of the variance and Cronbach’s alpha 0.90). The HAD as a whole was regarded as bidimensional, tapping these two separate but related concepts which together accounted for 53% of the variance. The only weakness was for Item 7 of the anxiety subscale (‘sit at ease and feel relaxed’), which consistently loaded higher on the depression component (around 0.50) than on the anxiety component (about 0.30). The mean scores on the HAD were 5.4 for anxiety (SD 4.1; range O-19) and 3.0 for depression (SD 3.0; range O-1 5). Their total mean score was therefore about 8.4. This study will be considered in greater detail subsequently in relation to our own findings. The next study involved two non-clinical groups using a Swedish translation of the HAD (Andersson, 1993). One sample consisted of 40 men and 47 women, parents of mentally handicapped children, whilst the other was a matched group of 29 men and 47 women, the parents of nonhandicapped children. The data were analyzed using the principal components method with varimax rotation and orthogonal transformation. Andersson’s interpretation of the two-component solution was the conventional one that “an item ought to be represented. . . where it is most powerful, that is where the item has its highest loading”. On this basis, which ignores both the absolute size of loading (four of the ‘powerful’ items were below 0.40) and the differential size between the loadings (one, Item 6 (‘I feel cheerful’), was 0.495 vs 0.472, i.e. equally split). Using the author’s criteria, of the eight items constituting component one (anxiety) only four were from the original anxiety scale. The second component was equally divided in having three items from each subscale. The two components correlated 0.29 and, with eigenvalues of 3.68 and 1.47, accounted for only 36.8% of the variance. The author concluded that, “there is no statistical evidence for the original split of the questions into two parts, Anxiety and a Depression subscale”. He then went on to compute threeand four-factor solutions. Only the latter, which he regarded as superior, was reported, and with additional eigenvalues of 1.45 and 1.03 accounted for 54.5% of the variance. Component one consisted of three high loading (above 0.70) depression items and was named ‘Well-being’. The second component, ‘Momentary anxiety’, comprised five anxiety and one depression item, whilst component three, ‘Power to relax’, was made up of two depression and one anxiety item. The final and fourth component, which the author chose not to interpret, had just one item from each subscale. Andersson’s item intercorrelation matrix shows, unsurprisingly, that these latter two items (8, ‘I feel as if I am slowed down’, and 11, ‘I feel restless as if I have to be on the move’) showed a low correlation with the rest of the scale items, ranging from 0.01 to 0.12 and from 0.04 to 0.33, respectively (Andersson, 1993). His suggestion that there may be a different interpretation for Item 7 (‘sit and feel relaxed’) between general population members and those under treatment may have broader relevance than that specific item. The mean score for answers on the anxiety scale was 6.6 and on the depression scale 4.1. The total mean was therefore around 10.7. The most recent report (Leung, Ho, Kan, Hung & Chen, 1993) involved 100 first- and fourthyear Hong Kong medical students completing both the English and Chinese versions of the HAD in one sitting, with the Eysenck Personality Questionnaire (Eysenck & Eysenck, 1975) used as a filler. In seven preliminary drafts of translations and back-translations most difficulties and disagreement arose from expressions such as Item 4 (‘I can laugh and see the funny side of things’), Item 9 (‘butterflies in the stomach’) and choices for the question in Item 10 (‘lost interest in appearance’).
The structure of HAD
415
To assess the linguistic equivalence of the versions r-tests were performed, with statistically significant differences being found for Anxiety Items 9 and 11 (‘feel restless as if I have to be on the move’) and Depression Item 2 (‘still enjoy the things I used to’). When the item correlation matrices for the English version were examined “all anxiety items except Item 9 showed high correlation with the depression subscores while all depression items except Item 10 showed high correlation with the anxiety subscores”. For the Chinese version “all depression items except Item 10 showed high correlation with the anxiety subscores while all anxiety items showed high correlation with the depression subscore without exception”. The correlation between the subscores was given as “Pearson’s coefficient 0.56 for the English version and 0.61 for the Chinese version”. Cronbach’s alpha for the anxiety and depression subscales, Chinese version only reported, was found to be 0.81 and 0.74, respectively. When “the data were further factor-analyzed. . . oblique rotation was performed. . . “. It is unclear whether this is a factor or a principal component analysis. For both language versions this resulted in three factor/component solutions. For the Chinese version these “accounted for 55.9% of the total variance: 37.5%, 9.6% and 8.8% for factors 1, 2 and 3, respectively”. Their Table 5 only presents the highest loading for each of the 14 items, although from this it is obvious that factor one is a depression factor. It comprises items from that a priori scale (5 in the English version and 6 in the Chinese version) plus Anxiety Item 7 (‘sit at ease and feel relaxed’). Factor two is an anxiety factor of four and five items, respectively. Of the four items (English) and two items (Chinese), versions of factor three Depression Item 10 and Anxiety Item 11 are ever-present. With regard to descriptive statistics, the English version had an Anxiety mean of 7.3 (SD 3.4) and a Depression mean of 5.0 (SD 3.1). The total HAD mean score was therefore 12.3. The Chinese translation gave an Anxiety mean of 7.3 (SD 3.4) and a Depression mean of 4.9 (SD 3.1). This resulted in a highly comparable total HAD mean score of 12.2. PRESENT
STUDY
Method and subjects
The 132 Ss were consecutive referrals to an adult psychiatry out-patient clinic in a district general hospital. Eighty-nine per cent of these patients were from general practitioners. There were 62 men and 70 women, with mean age of 37.6 yr (SD 12.0). Clinical and diagnostic formulations were conducted by a consultant psychiatrist (K.P.) and based upon a battery of assessment techniques, including the HAD. Only this measure at first interview is the concern of the present article, although it is anticipated that other results will be presented in subsequent reports. RESULTS For reasons of comparability we followed the sequence of procedures as carried out by Moorey et al. (1991). First, a principal components analysis of the total group was performed on the 14 items with rotation to a specified two-component orthogonal solution. The first component had an eigenvalue of 5.82 and accounted for 41.6% of the variance, whilst the second had an eigenvalue of 1.70 with 12.1% of the variance. When their criterion of taking loadings of 0.45 as the cut-off point was employed, the anxiety and depression items loaded satisfactorily onto separate factors. The weakest was Item 7, from the anxiety subscale (‘can sit at ease and feel relaxed’), which, whilst loading 0.50 on the anxiety component, loaded 0.43 on the depression component. For the anxiety subscale a mean of 13.9 (SD 4.4) was obtained, whilst for depression the mean was 9.9 (SD 4.9). The HAD mean was 23.7 (SD 8.1). Next an oblique rotation was performed. Table 1 shows the comparison between the component loadings found by Moorey et al. (1991) and those of the current study. Again all the items unequivocally loaded on the appropriate component except Anxiety Item 7, which loaded higher on depression in the Moorey et al. (199 1) study, and was almost equally divided in our data. For our psychiatric group the highest anxiety loading was 0.81 for Item 3 (‘frightened feeling as if something awful is about to happen’) and the lowest was 0.48 for Item 7. Among the depression subscale items a highest loading of 0.82 was found for Item 2 (‘still enjoy the things I
476
Alan Bedford et al. Table 1. Component loadings (oblique rotation) of the anxiety and depression subscale items of the HAD as found by Moorey et al. (1991) and in the current studv
HAD items
I
Depression Moorey Current
3 5 7 9 11 13
0.13 0.09 0.17 0.51 -0.10 -0.07 -0.01
2 4 6 8 IO 12 14
0.78 0.74 0.76 0.51 0.49 0.79 0.65
Anxiety subscale 0.44 0.15 0.25 0.45 0.25 0.03 0.24 Depression subscale 0.82 0.76 0.72 0.69 0.63 0.72 0.65
Anxiety Moorey Current
0.71 0.77 0.72 0.30 0.86 0.66 0.83
0.62 0.81 0.63 0.48 0.73 0.50 0.77
-0.02 -0.06 -0.01 0.14 0.07 0.01 -0.08
0.08 0.27 0.28 0.12 - 0.05 0.27 0.21
used to enjoy’) whilst the lowest was Item 10 (‘lost interest in my appearance’) with 0.63. The correlation between the two oblique components was 0.53, which is closely similar to the 0.50 found by Moorey et al. (1991). Like them we went on using “this method of rotation.. . in the rest of the study to test the stability of the factor structure”. Hence, the next step was to carry out principal components analyses with oblique solution on the two subscales of the HAD separately. Analysis of the anxiety items revealed a single component with eigenvalue 3.54 accounting for 50.5% of the variance. Item 3 had the highest loading (0.80) and Item 11 (‘feel restless as if I have to be on the move’) the lowest (0.45). Cronbach’s alpha was 0.825 for the anxiety scale. Item 11 had the lowest item-total scale correlation of 0.34; when this item was removed the resulting 6-item scale had an increased Cronbach’s alpha of 0.843. With the depression items an eigenvalue of 3.82 was obtained for the only component, which accounted for 54.6% of the variance. Items 2 and 4 (‘can laugh and see the funny side of things’) had joint highest loadings at 0.81, whilst Item 10, with 0.58, was lowest. Cronbach’s alpha was 0.856 for the depression scale. Item 10 had the lowest item-total scale correlation of 0.47; the deletion of this item resulted in a trivial change of Cronbach’s alpha to 0.857. For completeness the full HAD Cronbach was computed and found to be 0.883. Item 11 was again the weakest with an item-total correlation of only 0.28; dropping this item raised Cronbach’s alpha to 0.888. To test the stability of the component structure obtained, the sample was split in half by taking alternate cases and again each was subjected to a principal component oblique solution with two-component specification. As can be seen in Table 2, each item’s highest loading is on the same component in both samples. Next the data were divided by sex into 62 men and 70 women. Table 3 compares their descriptive statistics. No mean sex differences were obtained, as the largest t-test value, for the total scale scores, was only 1.04. Finally, the male and female subscale scores were subjected separately to a principal components oblique solution with two-component specification. A correlation of 0.62 was found between the components in men; for women the correlation was 0.47. Table 4 shows the loadings by sex for each anxiety and depression item. Whereas Moorey et al. (1991) found that for women Depression Items 8 and 10 just failed their 0.45 criterion, in our data they had loadings of 0.65 and 0.74, respectively. The ‘weakest’ item in that subscale was Item 10 in men, with 0.47 against a -0.09 loading on the anxiety component. For the anxiety subscale our data repeated their finding that Item 7 met the criterion in both sexes on the ‘wrong’ component, i.e. depression. However, for our males there was an even higher loading on the anxiety component (0.64) but not for our females (0.23). Generally, the cross-loadings in our data were larger than those of Moorey et al. (1991), particularly for males and the anxiety items.
477
The structure of HAD Table 2. Component
Depression Sample 2 Sample I
HAD items
Anxiety Depression Total
3
SD
13.5 9.4 23.0
4.7 5.2 8.9
Anxiety Sample 2 Sample 1
3 5 7 9 11 13
38 17 35 42 19 -4 28
Anxiety subscale 45 -7 IO 45 29 -1 18
66 83 68 42 71 50 72
64 81 61 59 74 48 81
2 4 6 8 10 I2 14
83 73 75 56 12 69 64
Depression subscale 80 74 67 81 55 72 64
-6 27 28 28 -7 18 -6
17 35 34 -2 -1 42 32
Table 3. Male and female means, standard
HAD
loadings of split halves of total sample (oblique rotation)
deviations,
ranges and Cronbach
Males (N = 62) Cronbach’s Range I-21 o-19 l-39
0.850 0.869 0.900
Table 4. Component
HAD items
alpha
loadings
alphas for the HAD anxiety, depression
R
SD
14.2 10.3 24.4
4.1 4.6 7.4
2-2 I 1-21 542
alpha
0.800 0.840 0.860
for men and women (oblique rotation)
Depression Men Women
Anxiety Men
Women
3 5 7 9 I1 13
49 27 46 45 43 -6 43
Anxiety subscale 44 16 IO 53 18 19 15
61 76 52 64 57 61 71
63 81 70 23 78 40 75
2 4 6 8 10 12 14
86 79 81 13 47 80 61
Depression subscale 78 76 65 65 74 66 66
-9 28 23 12 -9 23 -2
-4 16 25 -1 -3 22 31
1
and total scales
Females (N = 70) Cronbach’s Range
DISCUSSION Of the four previously published studies reported here, only Moorey et al. (1991) appear to find a two factor-component solution. However, Leung et al. (1993) after clearly identified anxiety and depression dimensions in both their English- and Chinese-language versions, end up with a small third mixed factor in one case of only two items. This is of no practical or theoretical use, and therefore their study probably belongs more properly to the Moorey et al. (1991) bidimensionality camp. As our own results are in line with Moorey et al. (1991), this means that all three analyses using an English-language version of the HAD are interpretable in terms of a two oblique component solution. By contrast, Andersson’s Swedish version (Andersson, 1993) produces two orthogonal components of mixed items, and the French version of Razavi et al. (1989) found that 11 and 10 of the items loaded on the first factor or component, respectively. Until other reports appear using the
Alan Bedford et
478
al.
English and translated versions, the generalisability of these structural findings can only be regarded as tentative. However, within these limitations the two studies in English with British Ss both find Anxiety Item 7 to be somewhat misaligned in having high loadings on the depression component. Our data suggested that it was a more pure measure of depression for females (loadings of 0.53 vs 0.23) than for males (0.45 vs 0.64). It may be apparent from our earlier description that the majority of reports fail to give a full account of the statistical (and sometimes other) procedures carried out. Also, these omissions are not consistent across studies. There are, of course, other important differences between these reports other than those concerned with data analysis and reportage. The most obvious ones relate to the status of the respondents, i.e. patients or non-hospital Ss, their ages, and the representativeness and balance of each sex. These add to the imponderables when attempting to evaluate the HAD’s structure. Additionally, we have already mentioned that Leung et al. (1993) experienced difficulties in translations and backtranslations of items for cultural reasons. Similarly, Cheng and Hamid (1995), with Rosenberg’s Self-Esteem Scale, noted that problems in syntax may occur when a scale is translated and its psychometric properties may then be adversely affected. These points have been specifically addressed by El-Rufaie, Absood and Abou-Saleh (1997) in producing a culture-oriented screening scale for use in Arabic primary health care settings. The 12 anxiety-depression items were mainly derived, and modified, from the HAD scales. It is to be regretted that a measure used internationally has been subjected to so little evaluation, and that the few structural researches carried out make little contribution to the advancement of practical knowledge. REFERENCES Andersson, E. (1993). The Hospital Anxiety and Depression Scale. Homogeneity of the subscales. Social Eehaoiour and Personality, 21, 197-204. Cheng, S. T. & Hamid, P. N. (1995). An error in the use of translated scales. The Rosenberg Self-Esteem Scale for Chinese. Perceptual and Motor Skills, 81,431&434. El-Rufaie, 0. E. F., Absood, G. H. & Abou-Saleh, M. T. (1997). The Primary Care Anxiety and Depression (PCAD) Scale. A culture-oriented screening scale. Acta Psychiatrica Scandinauica, 95, 119-124. Eysenck, H. J. & Eysenck, S. B. G. (1975). The Eysenckpersonality questionnaire. Sevenoaks: Hodder and Stoughton. Leung, C. M., Ho, S., Kan, C. S., Hung, C. H. & Chen, C. N. (1993). Evaluation of the Chinese version of the Hospital Anxiety and Depression Scale. A cross-cultural perspective. International Journal of Psychosomatics, 40, 29-34. Moorey, S., Greer, S., Watson, M., Gorman, C., Rowden, L., Tunmore, R., Robertson, B. & Bliss, J. (1991). The factor structure and factor stability of the Hospital Anxiety and Depression Scale in patients with cancer. British Journal of Psychiatry, 158, 255-259. Razavi, D., Delvaux, N., Farvacques, C. & Robaye, E. (1989). Validation de la version Francaise du HADS dans une population de patients cancereux hospitalises. Reoue de Psychologie Appliquee, 39(4), 295-308. Razavi, D., Delvaux, N., Farvacques, C. & Robaye, E. (1990). Screening for adjustment disorders and major depressive disorders in cancer in-patients. British Journal of Psychiatry, 156, 79-83. Zigmond, A. B. & Snaith, R. P. (1983). The Hospital Anxiety and Depression Scale. Acta Psychiatrica Scandinauica, 67, 361-370.
APPENDIX Items of the Hospital IA. 2D. 3A. 4D. 5A. 6D. 7A. 8D. 9A. 10D. 1 IA. 12D. 13A. 14D.
Anxiety
and Depression
Scale (HAD)
I feel tense or ‘wound up’. I still enjoy the things I used to enjoy. I get a sort of frightened feeling as if something awful is about to happen. I can laugh and see the funny side of things. Worrying thoughts go through my mind. I feel cheerful. I can sit at ease and feel relaxed. I feel as if I am slowed down. I get a sort of frightened feeling like ‘butterflies’ in the stomach. I have lost interest in my appearance. I feel restless as if I have to be on the move. I look forward with enjoyment to things. I get sudden feelings of panic. I can enjoy a good book or radio or TV programme.
Item numbers
are followed
by ‘A’ to denote an anxiety
subscale
item and ‘D’ for a depression
item.