The finding of high proportions of Chinese CHD patients screening positive with anxiety and depression using the HADS is consistent with previous studies conducted in China or in western countries [3, 5, 34, 35]
The CFA findings are in many respects, rather surprising. It has been suggested that the bi-dimensional underlying factor structure of the HADS observed in many early psychometric evaluations of this instrument is an artifact of the factor extraction method (exploratory factor analysis) and that the instrument is indeed tri-dimensional, a perspective supportive by more recent studies using CFA . In studies of the HADS in patients with CHD using CFA, a clear advantage of three-factor models over two-factor models is consistently observed [18–20]. In the current study the best performing three-factor and two-factor models were equivalent in terms of model fit indices and offered a good fit to the data. It is interesting however to reflect that the best-performing two-factor model of Moorey et al. , represents a modification of the original model proposed by the instrument developers , suggesting that even in the context of a two-factor model, the traditional HADS scoring system may not be optimal. The best-performing three-factor model of Dunbar et al.  is an important observation because it represents a good performing model based on a cogent theoretical model of anxiety and depression. The observation of a well-fitting theoretical model when applied to clinical data is deemed to be a good test of the model.
Given the observation that tri-dimensional models offered a similar fit to the data as bi-dimensional models, the issue of scoring the instrument as a tri-dimensional instrument is worthy of discussion. It has been previously suggested that the HADS could be scored as a three sub-scale instrument . However, such tri-dimensional scoring approaches that have been proposed are complex and time-consuming for busy practitioners to use routinely since they require factor scores to be regressed to calculate sub-scale scores . A key operational rubric of the HADS is that it is a quickly administered and easily scored measure, therefore implementation of a complicated scoring system would be highly undesirable. Moreover, the extensive use of the HADS in clinical research over the last 20 years has led to the dissemination of several hundred publications reporting the HADS sub-scale means for a broad range of clinical groups. Adopting a tri-dimensional scoring approach would essentially remove this valuable reference data for comparative purposes in new research. Finally, the principle finding of the current study of no clear advantage of tri-dimensional models over bi-dimensional models would suggest that consideration of tri-dimensional scoring approaches is at the very best, highly premature.
The finding of virtually identical fit characteristics of the best performing two and three-factor models also raises the issue of conclusively defining the underlying factor structure of the HADS. The HADS clearly cannot be both bi-dimensional and tri-dimensional within the same data set and further clarification of the structure of the instrument is desirable since this may provide additional evidence not only on the limitations of the HADS, but also the development, enhancement and possible future revision of this widely used measure. The current study was limited by sample size and this may be an important factor in clarifying the relative performance of the competing models tested. It is worthy of note that a number of studies that have utilised factor analysis with large sample sizes have found a clear advantage in model fit of tri-dimensional models over the traditional anxiety/depression bi-dimensional model of the HADS [17, 18, 26]. Large sample sizes are generally desirable in confirmatory factor analysis and the conservative sample size of the current study may have contributed to the absence of differences in model fit between the best-fit two and three-factor models. Further research is necessary to address this particular issue conclusively.
Previous findings of the factor structure of the Chinese version of the HADS in Cantonese-speaking Chinese CHD patients in Hong Kong indicates clear and consistent superiority of three-factor models in fits to data . One possibility that may account for the ambiguous factorial conclusions in the current investigation concerns the issue of translation. Translating English language instruments to Chinese language versions can be problematic in terms of establishing cultural and semantic equivalence [36, 37].
The original validation of the Chinese version of the HADS  identified potential issues of case detection accuracy with the instrument. It is conceivable that problems of case detection accuracy may be artifactual of the original translation process, which may also explain inconsistencies between the underlying factor structure of the instrument between Mandarin-speaking Chinese CHD patients in the current study and those reported in Cantonese-speaking Chinese CHD patients.
It should be acknowledged that the current study had a number of limitations, in particular, the modest sample size and the absence of a comparison to a 'gold standard' such as a structured clinical interview to assess for anxiety and depressive disorder. Further research addressing these limitations is recommended.