A validation study of the Hospital Anxiety and Depression Scale (HADS) in a large sample of French employees

The publication of the Hospital Anxiety and Depression Scale (HADS) in 1983 [1] triggered off a large number of significant research studies. There is therefore an abundant literature on the subject. This tool has been translated and validated in many countries and its capacity to detect anxiety and depressive disorders is widely recognized. Nonetheless, as soon as it was published, researchers rightly questioned the structure of the questionnaire, its links with the other tools for determining anxiety and depression and its psychometric properties in general. In the first instance, we shall present the points on which there seems to be a consensus, followed by the one for which this is much less the case.

HADS comprises 14 items, seven of which relate to anxiety symptoms and seven to depressive symptoms. Each item is coded from 0 to 3. The scores for anxiety and depression can therefore vary from 0 to 21, depending on the presence and severity of the symptoms. The authors [1] have proposed cut-off points or thresholds: a score between 0 and 7 does not indicate the presence of the symptoms of anxiety or depression; a score between 8 and 10 indicates the presence of the symptomology but to a moderate degree, therefore doubtful cases; a score greater than or equal to 11 indicates a significant number of symptoms of anxiety or depression corresponding to confirmed cases. The studies concerning the accuracy of these thresholds all showed them to be reliable: the review of the literature by Bjelland et al. [2], covering 747 papers, concluded that the threshold of 8 is decisive in order to not exclude people suffering from anxiety or depressive disorders, which would be the case if the threshold of 10 alone was taken into account. Since 2002, further studies have measured the sensitivity and specificity of the HADS: This tool was proposed in Norway to a general population in the doctor's consulting room [3], in England with people suffering from heart disease [4] and to a representative sample of the German population [5], the threshold of 8 is highly recommended for each of the two scales. More recently, Roberge et al. [6] calculated different thresholds relating to each scale: 10 or more for the anxiety scale, 7 or more for the depression scale. When the HADS is considered on one scale only, a possibility envisaged but not recommended by Zigmond and Snaith [1] as a measure of the intensity of “emotional distress”, the recommended threshold is 13 [5], even 16 [6].

The concurrent validity of each of the two HADS scales, HADS Depression subscale (HADS-D) and HADS Anxiety subscale (HADS-A), has been demonstrated with populations suffering from various pathologies, in different countries and with a variety of methodologies: Clinical interviews and/or standardized tools. Thus, high correlations were found between the scores on the Beck Depression Inventory and the HAD-D scores in diabetic patients [7], in elderly people in hospital, people attending hospital consultations and people in the general population [8] or in patients suffering from cancer [9], to quote only the most recent studies. The Depression Screening Questionnaire has also been used to check the concurrent validity of the HADS-D scale in the patients of general practices [3], and the Quality of Life Questionnaire in Iranian patients suffering from different stages of cancer [10] or people suffering from musculoskeletal disease [11]. Regarding anxiety, the two principal questionnaires used are the State-Trait Anxiety Inventory and the Generalized Anxiety Questionnaire [3],[8].

However, the HADS-D and HADS-A are not independent: according to Bjelland et al. [2], the average correlation would be .56, which is not astonishing, given the comorbidity between anxiety and depressive pathologies. The main symptom measured by the HADS-D relates to an anhedonic depression. So, the correlation between sub-score is estimated, in the general population, at between .5 and .6 [12] and is clearly higher in the populations suffering from pathologies, which the vast majority of the studies involved. For this reason, some authors recommend the use of HADS-A to assess both anxious and depressive symptoms [7],[13]. It should, however, be remembered that the meta analysis produced by Mitchell, Meader and Symonds [13] involved only patients suffering from cancer and in palliative care, for whom it can easily be understood that anxiety symptoms are closely intertwined with depressive symptoms; the same argument can be advanced for the study by Sultan, Luminet and Hartemann [7] which concerns diabetics, for whom the depressive experience is highly marked by anxiety. Others, that we have already mentioned, recommend the use of the global scale to assess the intensity of emotional distress (e.g. [5],[7],[13]-[16]).

The few divergences we have just mentioned are not, however, subject to significant debate. On the other hand, they are revealing about a point on which the authors are not in agreement, that is to say the factorial structure of the HADS. This has been under discussion for almost 30 years and the debate is still ongoing.

Zigmond and Snaith [1] drew up the HADS with two objectives: i) retain only items “based solely on the psychic symptoms of neurosis” (p. 362), ii) “to define carefully and distinguish between the concepts of anxiety and depression” (id.). This is in order to evaluate the disorders more related to the patients' psychological symptoms rather than physical ones. Thus, the depressive dimension is strongly marked by the measure of anhedonia (five items out of seven), a characteristic symptom of depression, and the anxiety dimension by the measure of feelings of tension, worry, fear and panic (five items out of seven). Nonetheless, some of these ten items can give an account of both an emotional state and a physical state as well as the two supplementary items that complete each of the scales. We can estimate that these items do not fully meet the authors' first objective and we shall see that these are the items that are subject to discussion in the majority of cases.

Although some research studies point to a structure with four factors [17],[18] or a single factor [5],[7],[13]-[16], the main debate concerns two or three factors (cf. literature review of Cosco et al. [19]). Martin [20] deems that we have to take the negative linguistic form of some items into consideration either as a general factor influencing the factors of anxiety and depression, or as a supplementary factor. Others take the tripartite model of Clark and Watson [21],[22] as a basis and obtain a more satisfactory factorial solution by considering three factors: An anxiety factor, a depression factor and a negative emotional or psychomotor agitation factor [14],[23]-[28]. Lastly, a large number of research studies point to the initial two-factor structure of HADS [2],[3],[7]-[9],[12],[13],[29]-[40]. Nonetheless, it is difficult, a priori, to adopt a position in favor of one or the other, given the very diverse populations involved in the analyses (witness the debate between Friedmann et al. and Dahl et al. [41] and the different statistical methods used (EFA - with orthogonal or oblique rotation - or CFA). In addition, the study by Straat, Andries van der Ark and Sijtsma [42] shows that the differences obtained by the authors in respect of the dimensional structure of the HADS are essentially due to methodological factors. However, Norton et al. [43] present a meta confirmatory factor analysis by which they tested 10 models; although they prefer the solution with a general distress factor, they conclude that a bifactorial model is acceptable.

Our aim is to present some psychometric properties of the HADS on a large population of French employees and to see whether we obtain the same main results as in the literature. The population in our research is original, since it concerns working employees; it is assumed to be in better health than all the other populations. There is no study relating to this type of population. Also, this questionnaire is very often used by companies in France (we specify the context in the next paragraph) without there being any previous psychometric analyses done on such data.

We shall compare our results with the studies involving a general population only [2],[5],[23],[31],[33],[34] and with a small number of existing French studies [7],[14],[26],[44]-[46].

After carrying out EFAs on all the people in our sample, as well as the sub-groups defined according to gender and age, we used CFA to test the different possible theoretical models. It turns out that the original model, as defined by Zigmond and Snaith [1] is the one that demonstrates the best fit. This is therefore the one we selected at least for the population we are interested in, that is to say, French employees.

In EFAs, two types of items are observed: items which substantially load one of the two factors and two items with smaller loadings that load the two factors. Thus, HAD2-D1, HAD4-D2, HAD6-D3, HAD10-D5, HAD12-D6 and HAD14-D7 items have a loading that varies from .57 to .80 on factor 1, which we identify as accounting for the scale of depression. Factor 2, corresponding to the anxiety scale, comprises HAD1-A1, HAD3-A2, HAD5-A3, HAD9-A5, HAD11-A6 and HAD13-A7 items whose loadings vary from -.50 to -.80. Two items HAD8-D4 and HAD7-A4 do not clearly load on one of the two factors.

We propose to test four models of CFA: the original model and the three others which take into account all the possible combinations of items HAD8-D4 and HAD7-A4 belonging to one or other factor (cf. Table 2). The chosen model is the original two-factor model but, to obtain a good adequacy of the model with the data, we have of to add a link between HAD8-D4 and the anxiety factor as well as a link between HAD7-A4 and depression factor; what was expected according to the results of the EFAs. These items are often quoted in the literature as having no weight satisfying on the factor to which they are supposed to belong (e.g. [8]-[10],[38],[39]). Roberge et al. [6] list the various interpretations mentioned by authors in literature (p. 176–177): the item HAD7-A4, in its formulation, refers to the anxious symptomatology “I cannot sit at ease” and to depressive symptoms related to anhedonia ‘I cannot feel relaxed”. As for item HAD8-D4, if questions are asked about the psychic and motor slowing, which may be a symptom of depression, (“I feel as if I am slowed down”), it can also be reminiscent of the anxious inhibition sometimes present in the anxious spectrum [57].

Nevertheless, even if we did not find the “true” original model, we consider that we can continue to base ourselves on the calculation of the scores according to Zigmond and Snaith [1] with seven items relative to the depression and seven items relative to the anxiety. Indeed, there have been a large number of studies concerning the thresholds calculated from the initial anxiety and depression scales. These studies [2]-[5] have largely validated the use of thresholds (one threshold at 8 and another at 11) and a calling into question of the calculation of anxiety and depression scores necessarily calls into question the value of the thresholds. The last reason is related to our sample: the HADS was devised a priori for detecting the presence of anxiety and depressive disorders in hospital patients. Although the two-factor structure was also found in general populations [3],[5],[8],[33],[34], our sample is not fully comparable to a general population because it only concerns people at work. However, when we compare groups (according to gender, age, occupational status and especially other variables specific to each company, such as business line or department), using initial anxiety and depression scores or scores based only on the five items that most load the anxiety (HAD1-A1, HAD3-A2, HAD5-A3, HAD9-A5 and HAD13-A7) and depression factors (HAD2-D1, HAD4-D2, HAD6-D3, HAD10-D5 and HAD12-D6), the global results are unchanged: We find exactly the same groups with the highest scores. As the important aspect of our psychosocial risk prevention approach is to detect "at risk" groups, the scores calculated on the initial scales remain valid and we may, in addition, use the thresholds.

We presented the averages of the scores of anxiety and depression by subgroups (by crossing gender, age group and status). We cannot exactly compare these results in detail with the results published in the literature, with the exception of a few studies that validated the questionnaire with a general population [5],[34] or included a control group taken from the general population [8],[15]. The overall mean values and the mean values by gender (when these are given) are either comparable to those of this study, or slightly higher, as in the case of the German validation study [5] for the depression score; in the latter case, the population was older than the one in our study, which accounts for the differences. Nevertheless, we can note that women have a higher score of anxiety on average than men (6.55 vs 5.42) and that the scores of anxiety and especially depression increase with the age (HADS-A: 5.39 for the 30-year-old employees less vs 6.09 for those of 50 and more years old; HADS-D: 2.64 for the 30-year-old employees less vs 3.84 for those of 50 and more years old). Occupational status also impacts scores: in every age group, blue- or white-collar workers and technicians have anxiety and depression scores higher than those of engineers or junior executives and upper Management. For example, anxiety (depression) mean score for employees from 30 to 39 years old is 6.14 (3.62) for blue- or white-collar workers and technicians vs 5.55 (2.83) for engineers or junior executives and upper Management. The influence of the gender can be illustrated by the difference between anxiety scores of women and the men who are more than 50 years old: if we mention these differences with cut-off defined by Zigmond and Snaith [1], 18.4% of women have a score upper to 11 versus 9% of men. Occupational status influence can be illustrated in its turn by the difference between depression scores of the employees who are more than 50 years old: 6.8% of blue- or white-collar workers and technicians have a score upper to 11 versus 3.7% of managers. These results are consistent with those found in literature; we cite in particular Cohidon [57] who reports the results of the SMPG "Mental health in the general population" survey conducted between 1999 and 2003 on a sample of 36,000 people in France. The questionnaire used was the Mini [58]. The author reports that mood disorders affect 11% of men and 16% of women; anxiety disorders 17% of men and 25% of women. Analysis of the data shows equal and systematic differences according to socio-professional category: the least qualified persons are those most affected by disorders. The hypothesis which we are able to put forward (which concerns our results and also Cohidon's) is that persons on a lower socio-economic level have a more restricted access to care, whether in financial terms and/or on a personal level [59],[60]. Actually, the general practitioner remains the preferred point of contact for these people, whereas executives and managers are less reluctant to approach a psychiatrist or psychologist (the latter is never covered by health insurance in France).

The findings from this study are consistent with the analyses conducted with other populations in other countries: the HADS has good reliability and discriminant validity; its bi-dimensional structure allows one score to be calculated for anxiety symptoms and one score for depression symptoms. However, as Straat, Andries van der Ark and Sijtsma [42] suggest, and given the recurring results for the problem items, perhaps this questionnaire should be reviewed.

However, the HADS remains a useful instrument for detecting anxiety and depression symptoms, both at an individual and a collective level. As we have explained, the individual level is not within our remit but that of the company doctor. On the other hand, we are more particularly interested at a collective level in the context of preventing psychosocial risks. The population covered by this study is obviously not representative of all French employees, since our data practically only concerns major companies. Nevertheless, we update our data every year and we can compare a given company with all the other companies on "our panel", obviously taking into consideration the characteristics of each of them (gender, age, etc.). We therefore see a wide variability between the average scores for anxiety and depression in companies: for example, for 2011 data, the average anxiety score varies between 10.3 (SD = 4.4) and 5.11 (SD = 3.07); the average depression score varies between 6.37 (SD = 4.07) and 2.51 (SD = 2.45). Most of the time, this great variability is due to the economic situation in that business sector; the variability within a company (between different jobs or different departments) results more from internal characteristics (management policy, training, etc.).

The last report drafted by experts and asked by the French Ministry of Employment and Health shows that the anxious and depressive disorders are the ones which are the most present in the population of employees, that they recover from psychosocial risk factors and that it is thus important to measure them [61]. Even if the HADS is criticized [42], it remains a quick and relevant tool for use in companies.