The present study compared the psychometric properties and concordance between two anxiety/depression instruments- the HSCL-10 and AD-scale of the PADQ- in a Pakistani rural community sample. Overall, both instruments seem to be adequate screening tools for anxiety/depression and show concordance acceptable to the conventional standards.
All the HSCL-10 inter-item correlations and corrected item-total correlations demonstrated that no single item deviated in any significant way from the overall scale functioning. This was not true, however, for the PADQ, wherein "mental problem" and "mind in peace", were inadequately correlated to other items. In contrast to other items, which describe specific symptoms of anxiety and depression, the "mental problems" and "peace in mind" items are of general nature, and may indicate an underlying construct such as "mental illness". Prejudices against mental illness in the population, conceived of as a state of arrested or incomplete development of the mind which results in impaired intelligence and social functioning, and often aggressive or seriously irresponsible conduct, may explain the low correlation to the other items .
Factor analyses were preformed to examine whether the HSCL-10 and the PADQ should be perceived as one dimensional (distress) or two dimensional (anxiety and depression) measures. One factor solutions accounted for 48% and 39% of the variance in the HSCL-10 and PADQ, respectively. In these solutions, all items on both scales (with the exception of "mind in peace" in the PADQ) seemed to load relatively evenly on the distress-dimension.
The two-factor solution with oblique rotation accounted for 60% and 49% of the variance in the HSCL-10 and PADQ, respectively. Whereas the factor loadings for some items on both scales increased, several factor loadings decreased. This indicates that the HSCL-items "blaming yourself", "difficulty in sleep" and "feeling blue" account for variance in both anxiety and depression, and thus do not seem to differentiate between these two conditions. The same seems to be the case with some of the items from the PADQ: "anxious amongst a lot of people", "lost your self-confidence", "get frightened" and "mind in peace".
The two-factor solution for HSCL-10, besides overlapping of factor loadings for items described above, did show a clustering pattern in factor loadings both for anxiety and depression related items, respectively. The two-factor solution for PADQ, however, showed a less clear pattern. Whereas the two anxiety related items ("anxious among a lot of people" and "get frightened") loaded on both factors, the same was true for one of the depression related items ("lost your self-confidence"). For the rest of the depression related items, there was not clear pattern on which factor they loaded.
From these analyses it could be concluded that both the HSCL-10 and PADQ may be meaningfully interpreted as a one-dimensional measure of distress. However, the HSCL-10 could also be interpreted as a two-dimensional measure because of the emerging pattern of clustered factor loadings for both anxiety and depression. A two- dimensional interpretation of the PADQ was not evident.
The HSCL-10 and PADQ were found to be moderately to highly correlated (0.62, 0.73 after correction for attenuation). Although this level of correlation is acceptable when comparing psychological instruments, a great deal of variance remains unexplained. It is proposed that problems comparing the measurement of psychiatric morbidity across the two screening instruments may have arisen from differences in the constructs they use. It appears that the construct used for anxiety and depression in the HSCL-10 was suitable for general population in Pakistan. This observation is in accordance with other studies describing the HSCL as a valid and effective screening tool for use across clinical , non-clinical and in community settings [29, 39] as well as from a cross- cultural perspective .
In contrast, some items of the PADQ seem to be more valid in the clinical settings, where subjects define themselves as psychiatric patients. This might be due to a difference between psychiatric symptom constellations that exist in a particular community to those experienced in hospitals, which are related to define certain diagnosis . This is further supported by the fact that high clinical validity for an instrument defined and tested in a patient population has often not been achieved in field studies in general populations . Moreover, previous studies support this assertion and report that difficulties arise when one instrument, developed to measure a given construct in one particular group, may not validly assess the same construct in other groups due to conceptual or metric differences [43, 44].
By correspondence analysis (Table 7), 35% concordance was observed on diagnosed cases between PADQ and HSCL-10 at its conventional cut-off score 1.85 . A gradual improvement was observed as we go downwards from the conventional cut-off score of HSCL-10. More than 50% concordance on diagnosed cases was achieved at 1.65 cut-off score. Previously, a study conducted by Sandanger et al. has shown 46% concordance between cases diagnosed by the Composite International Diagnostic Interview (CIDI) and cases identified by HSCL-25 at its conventional cut-off 1.75 . It is however important to keep in mind that reports of the agreement between psychiatric symptom screening and medical diagnosis are seldom better than 50% .
An increased concordance by reducing cut-off score of HSCL-10 could be attributed to the fact that HSCL-10 is a self-inventory but in our study it was used in interview format. The use of HSCL in interview format has been reported in other studies while dealing with non-literate populations . It has been generally reported that social desirability and other factors related to the interview technique can have negative impact on response patterns compared to self-reporting . Sandanger et al. has discussed a possibility of lower score on HSCL-25 in general population when list of symptoms was presented orally by an interviewer, rather than the respondents checking off her/his symptoms on sheet by her/himself , probably due to social desirability bias .
Another explanation of this observation could be related to the cultural aspects linked with the expression of psychological distress. A change in the cut-off score is in accordance with other studies, suggesting a more cautious approach to the cross-cultural use of pre-determined cut-off scores . A change to the optimal cut-off scores has also been observed with other well established instruments in various studies . A recently conducted study in Afghanistan, a neighbour country at the northern border of Pakistan, has raised concerns over administration of self-administered questionnaires by interview format and the role of culture in defining response on mental health symptoms by the respondents .
Our sample was self-motivated and respondents were invited to the field camps with a message to get a free medical consultation and blood test for diabetes. Hence it might be possible that the participants were overrepresented with respect to somatic disorder and poor socio-economic status. In order to examine the effect of this sampling bias on the main findings in this article, regarding the concordance between HSCL-10 and PADQ, we calculated correlations between the sum scores of the two measures separately among people with low, middle and high income and among people with self-reported poor and good somatic health, respectively. These correlations showed a slight tendency in the direction of biased results with regard to both hypothesized sources of bias, however in opposite directions. In other words; when comparing correlation between the two scales in the different income strata, this was largest for those with the highest income (the group underrepresented in this sample). Whereas comparing correlations in the groups of people reporting somatic health problems, the correlation was highest among those with poor health (the group overrepresented in this sample). Thus, we have one effect suggesting that there is a bias decreasing the strength of our results due to sampling bias and one effect suggesting an increase. We therefore hypothesize that these opposite effects approximately equal each other so that the effect of sampling bias on our results taken together is relatively small.
In mental health studies, the term "gold standard" usually attributed to the clinical interviews and they are relatively irrefutable standards those constitute recognised and accepted evidence that a disease exists  and it is customary to use them in validation studies. Also in our study it would have been a better option to include clinical interviews but limited resources and available working conditions in community settings and the purpose of the study contained us to conduct this study without clinical interviews.
Another possible limitation of the present study might be due to overrepresentation of the females in our sample, 63% of the study's sample was female. Yet, we were unable to detect any bias due to this imbalance when conducted exploratory factor analysis stratified by gender (data not shown).