Factor structure of the Hospital Anxiety and Depression Scale (HADS) in German coronary heart disease patients

Coronary heart disease (CHD) is of profound interest to health and clinical psychology due to the high levels of anxiety and depression observed in patients following the occurrence of a coronary event [1‐5]. CHD was the most leading diagnosis for treatment in hospitals in Germany for men (320,000 patients per year), and, after childbirth and breast cancer, the third reason for in-patient treatment for women (150,000 patients per year) [6]. Most in-patient rehabilitation hospitalizations in Germany for men (about 60,000 per year) were caused by CHD [7, 8]. Recent research showed, that at least one in five patients in cardiac rehabilitation suffer from a psychological disorder [9]. Accurate identification of significant anxiety and depression as soon as possible following a cardiac event is essential in order to facilitate delivery of an effective and comprehensive treatment package which takes into account psychological as well as coronary disease symptoms [10]. This is particularly relevant since anxiety and especially depression have been demonstrated to be predictors of mortality in this clinical group [11]. The availability of easy to administer, reliable and valid screening tools would logically be a critical component of a clinical protocol seeking to identify CHD patients with psychological disturbance. A suitable measure would readily identify those patients for whom additional referral to a clinical psychologist or to a liaison psychiatry service would be more appropriate.

A candidate screening tool that has been widely and increasingly used with CHD patients is the Hospital Anxiety and Depression Scale (HADS: [12]), an easily administered 14-item self-report measure comprising 7 anxiety items and 7 depression items from which separate anxiety and depression sub-scale scores are calculated [13]. The HADS was designed to exclude symptoms that might arise from the somatic aspects of illness such as insomnia, anergia, and fatigue, therefore the instrument has been designed for use within the clinical context of general medicine. The HADS has been used for screening purposes in a diverse and broad range of clinical groups [14‐24]. A number of investigations have suggested that the HADS is a suitable instrument to accurately assess anxiety and depression in CHD patients [10, 17, 24‐27]. A fundamental assumption underpinning the clinical usefulness of the HADS across a broad range of clinical groups, including CHD, is that the instrument reliably assesses anxiety and depression as two distinct and separable dimensions [28].

On the other hand, recent psychometric evaluations of the HADS in a range of clinical populations have suggested that the proposed factor structure of the instrument may indeed be compromised by the physiological aspects of the disease or by changes in health status [23, 29, 30]. Conversely, there is accumulating evidence that the fundamental factor structure of the HADS comprises three factors instead of two [24, 26, 31‐33]. The finding that the three-factor structure offers a superior fit to clinical data than the two-factor (anxiety and depression) model formulated as part of the original instrument development by Zigmond and Snaith [12] has implications in terms of the use, scoring and future development of this assessment tool.

Dunbar et al. [32] found a three-factor structure of the HADS in a non-clinical population (for an overview see table 1) and interpreted their findings in relation to the conceptually rich 'tripartite' model proposed by Clark and Watson [34]. Extending these observations to a clinical population, Friedman et al. [33] found a three-factor structure to the HADS in a patient group being treated for major depression, which incidentally, was similar to that observed by Dunbar et al. [32]. Martin and Newell [35] found that Friedman et al.'s [33] three-factor model offered the best-fit to their data examining individuals with significant facial disfigurement compared to competing two-factor models. Martin and Thompson [22] observed a three-factor structure to the HADS in myocardial infarction patients and, in a later study, Martin et al. [26] extended further the findings of both Dunbar et al. [32] and Friedman et al. [33] to myocardial infarction patients finding additional support for the three-factor structure suggested by these researchers to underlie the HADS. A recent study [24] of the psychometric properties of the HADS in Chinese acute coronary syndrome (ACS) patients has established further support for the three-factor structure of the HADS furnishing evidence that the three-dimensional structure of the instrument appears to be consistent across diverse cultures. Caci et al. [31] suggested a three-factor underlying structure to the HADS that represents a modification to the three-factor model identified by Friedman et al. [33] and replicated by Martin et al. [26]. However, Caci et al.'s [31] model was based on a student population and it should be remembered that the presence of significant pathology or physiological change states does impact on the underlying factor structure of this instrument [23, 30, 36]. The most consistent contemporary observation in terms of the underlying factor structure of this instrument in cardiac patients is strongly indicative that the HADS comprises three underlying dimensions. However, these cardiac studies have used either clinical populations from the United Kingdom [22, 26] or from the Far East [24].

There has been little systematic investigation of the factor structure of the translated HADS in German cardiac patients, though interestingly a German language version of the instrument has been developed in Germany using cardiac patients within the context of the original assumed two-dimensional (anxiety and depression) structure [37]. To date, the HADS has been found to comprise a two-factor structure consistent with the anxiety and depression sub-scales proposed by Zigmond and Snaith [12] in cardiac [38] and non-clinical [39] populations in Germany. However, these large studies did not investigate the possibility that alternative factor models may provide a better explanation of the data.

Identification of a coherent three-factor underlying structure of the HADS has a number of significant implications in terms of the validity of the tool as a screening instrument. Firstly, referral to mental health services could be undermined based on a two-dimensional (anxiety and depression) interpretation of HADS scores in cardiac patients. Secondly, further replication of a three-factor structure of the HADS in a German cardiac population would be valuable in determining if the HADS should be more effectively used as a screening instrument when comprised of three sub-scales in this group. Thirdly, replication of a consistent three-factor structure in the German-translated version of the HADS would provide strong evidence that the three-dimensional structure is implicit to the instrument and not a language-based artifact. Finally, the widespread international use of the HADS provides a compelling rationale to establish the psychometric properties of the instrument not only in broad diagnostic categories, but also across culturally-diverse groups.

The purpose of the present study was to determine whether the three-factor structure of the HADS identified by Martin and colleagues [26] in myocardial infarction patients in the UK and Martin et al. [24] in Chinese ACS patients has the same psychometric properties as that of the German-translated version of the HADS in a cohort of German patients presenting with CHD. The present study addresses two research questions:

The findings of the current study offer a further important contribution to the evidence base regarding the underlying factor structure of the widely used HADS. It is worthy of note that high levels of HADS assessed anxiety and depression were observed in the study. This finding is consistent with investigators using this instrument in cardiac populations in other parts of the world [22, 24, 26] and verification of the need to screen for psychological disturbance in patients presenting with CHD.

The findings from the factor analyses conducted on the HADS data are of pertinent methodological as well as clinical interest. EFA of the HADS revealed two factors, the loadings of individual items being consistent with the anxiety and depression sub-scale domains. However, it was also observed that the HADS-A item-7 'I can sit at ease and feel relaxed' was jointly loading on both anxiety and depression factors, this split-loading slightly in favour of the depression factor. A recent EFA of the HADS conducted with patients with significant facial disfigurement [35] also revealed item-7 to be split-loaded between anxiety and depression latent domains. Martin and Newell [35] suggest that in circumstances of split-loading such as those observed in the current study, a two-factor solution may offer the most parsimonious solution in EFA, but may not provide the best identification of factors in terms of model fit. Martin and Newell's [35] rationale for this is that EFA is not a model evaluation technique, therefore identification of factors based on arbitrary cut-points such as eigenvalues and scree plots is likely to produce a psychometrically reductionist account of sophisticated relationships between observed and latent variables. Martin and Newell [35] proposed that the lack of apriori model specification in EFA provides a convincing psychometrically plausible explanation of inconsistencies between EFA and CFA in extracted factors and interpretation of data. Indeed, Martin and Newell [35] found a similar finding to that of the current investigation, EFA support for a two-factor model and CFA support for the superiority of three-factor compared to two-factor models. Interestingly, Dagnan et al. [55] and Mykletun et al. [56] identified three-factor initial solutions within the HADS but chose to dismise the third-factor, without a sound psychometric rationale. It is likely that an expectation of a presumed two-factor model makes it difficult to reconcile an unexpected three-factor model emerging from the data, therefore it is explainable why these researchers might choose to dismiss a third factor.

The findings from the CFA revealed the best model fit to be provided by Friedman et al.'s [33] three-factor mode (see figure 1). The 'next best' fit to the data is offered by Caci et al.'s [31] three-factor model 2. It was also observed that the remaining three-factor models tested [31, 32] not only offered a good fit to the data but also provided a superior fit to the data compared to the two-factor models evaluated on a number of estimates of model fit. The two-factor models of Zigmond and Snaith [12] and Moorey et al. [43] did however offer an acceptable fit to the data. The uni-dimensional model of Razavi et al. [44] was found to offer a poor fit to the data, a finding consistent with previous research on the HADS across a variety of clinical groups [19, 24, 26, 30, 35]. There remains little doubt from the CFA analysis that the best fit to the data is offered by three-factor models irrespective of the clinical population from which the three-factor model was derived. The findings from the CFA have furnished compelling support of the HADS as a tri-dimensional instrument, consistent with contemporary research with this instrument across diverse clinical presentations [19, 26, 30, 31, 33, 35].

In conclusion, the current study found the German language version of the HADS to have an underlying three-dimensional factor structure following CFA in CHD patients, an observation consistent with UK [26] and Chinese [24] CHD populations. The traditional interpretation of the HADS as a two-factor (anxiety and depression) structure was also found to offer an acceptable fit to data, though inferior to that of the three-factor models. It can be concluded that the HADS may serve as useful screening purpose by being scored as two sub-scales of anxiety and depression. The clinical utilisation of the HADS continues to be invaluable in screening for mental disorders. Our results suggest that the assessment of the efficacy of interventions in evaluation studies by the HADS may be biased by problems in construct validity. Two decades have passed since the HADS was introduced to the clinical screening battery. The findings of this study and those of others, suggests that despite the clinical usefulness in screening the individual results of the HADS could be interpreted more precise in clinical routine. The differentiation of the anxiety scale in "psychomotor agitation" and "psychic anxiety" in the best fitting model may be helpful in the interpretation of individual results of patients. These results may improve our understanding of the process of adaptation in patients with somatic illness. A separate analysis of subscales in clinical trials may reduce bias caused by somatic medical conditions of the patients. Agitation might be more likely biased by the medical status of the patients.

The study was funded by the Federal Ministry of Education and Research, Germany; Regional Pension Insurance Institute, Baden-Wuerttemberg, Germany, LVA 02 804 and is part of the Rehabilitation Research Network South-West (Germany).