Construct validation of the hospital anxiety and depression scale with clinical populations

Abstract

Objective: the hospital anxiety and depression scale (HADS) attempts to measure anxiety and depression without confounding by somatic symptoms of physical disorder, and is widely used for this purpose. This paper addresses three questions about the validity of the HADS concerning its independence of physical symptoms, the extent to which its items robustly measure the identified constructs with varying clinical populations and situations, and its capacity to differentiate anxiety and depression.Methods: data from patients with breast disease, myocardial infarction (MI), and stroke were examined using factor analytic and psychometric analyses.Results: using exploratory factor analysis in patients with breast disease, 13 of the 14 HADS items fell on a psychological factor and loadings on the psychological factor were higher than loadings on the somatic factor for all items. The HADS showed high levels of internal consistency and there was little evidence that removing items would improve it. Confirmatory factor analyses (CFA) in MI and stroke groups confirmed the separation of anxiety and depression. Analyses indicated items, which were performing poorly for these clinical groups.Conclusions: there was support for the validity of the HADS for all three questions. However, there were some evidences of individual items performing poorly. Given the ease of administration and the acceptability of this measure to ill or weak respondents, the HADS continues to perform satisfactorily.

Introduction

The hospital anxiety and depression scale (HADS) [1] is widely used as a measure of mood, emotional distress, anxiety, depression and emotional disorder in clinical populations with symptoms of physical disease. It is useful because it is quick to use and easily acceptable to patients who may be quite unwell. With only 14 items each answered on a four-point verbal rating scale, it can be used to give measures of anxiety (7 items), depression (7 items) or emotional distress (all 14 items). Alternatively, it can be used to screen for emotional disorder using cut-offs on both the anxiety and depression scales. The major attraction of the HADS is that it was designed to be valid with these clinical populations, avoiding items that might be endorsed due to physical rather than psychological states.

Previous psychometric investigations have shown that the HADS achieve good internal consistency and test–retest reliabilities, are sensitive to change and give valid assessments [2]. Factor analytic studies have found support for the two factors, anxiety and depression [2], [3], [4] with items largely loading on the appropriate scales.

This paper addresses three further questions about the validity of the HADS. First, how well does it succeed in separating symptoms of physical conditions from somatic indices of psychological states? If it is successful, then it should be possible to demonstrate distinguishable factor structures. On the other hand, one might expect a more general ‘quality of life’ (QoL) construct where a conglomerate of the individual's experience results in a single coherent state. If symptoms of disease affect mood, then it might prove impossible to distinguish two factors. Then it would be meaningless to use a measure such as the HADS to assess the impact of disease on mood states. Further, Watson and Pennebaker [5] have pointed to the ‘nuisance factor in health research’ of negative affectivity (NA), the individual's predisposition to see and report experiences in a negative way. The effect of NA would be to produce one factor dominated by the tendency to report both physical and psychological states with a more or less negative outlook. Thus, it may be impossible to achieve Zigmond and Snaith's [1] goal because of the pervasive effects of NA or because all symptoms combine in a single state of QoL.

Second, are all the items on the HADS valid for patients having different clinical conditions and management? Researchers frequently comment informally or in oral presentations on the possible inappropriateness of some of the items for particular patients. For example, ‘I feel as if I am slowed down’ might simply reflect the effect of the disorder in stroke, or “I get a sort of frightened feeling like ‘butterflies’ in the stomach” might be confused with cardiac symptoms. Similarly, ‘I can enjoy a good book or radio or TV programme’ may not be valid in an acute medical ward. Psychometric studies have noted items performing poorly in their analyses [3] and this may be due to invalidity of items for specific populations. While any questionnaire will have items that are not a coherent part of the dimension being measured for individual patients, the problems noted would result in systematic patterns of invalidity for whole populations. So for example, stroke, cardiac, and acute medical patients might each have a higher score in the above examples than they would have if the scale did not confound their mood with other aspects of their condition and management. It would then be difficult to study mixed populations or to compare different clinical populations without some adjustments to the scale. Indices of internal consistency allow examination of the extent to which the scale has aberrant items of this kind for a particular population.

Third, different scores are derived from the HADS and one can ask if they are all valid. Zigmond and Snaith [1] demonstrated that the scale was sensitive in detecting emotional disorder and published validated cut-offs for possible and probable disorder for both the anxiety and depression scales. However, the distinction between the anxiety and depression scales needs to be demonstrated, as otherwise the HADS might more appropriately be used as a measure of general distress. Theoretically, the distinction is important as threatening situations such as anticipating a surgical procedure or receiving results of an HIV test are postulated to raise anxiety, while loss situations such as bereavement or loss of function are seen as causing depressed mood. There is evidence of distinct patterns of cognitive performance for anxiety and depressive disorder [6] and it is therefore valuable to be able to discriminate the two. Exploratory analyses have largely found support for the two scales and the allocation of items to scales [2], [3], [4], but it has yet to be demonstrated that the HADS achieves validity for this separation of the two subscales in a confirmatory analysis.

The aims of this paper were to examine the following three questions using psychometric and factor analytic techniques:

• Do HADS items load on a separate factor from items dealing with symptoms of physical disorder?

• Do HADS items achieve satisfactory internal consistency in populations with different clinical conditions and at different stages of management?

• Is there support for the separation of the anxiety and depression items of the HADS?