Setting

Pakistan is a South Asian nation with population of 185.1 million; based upon World Bank criteria it is a low income nation and it ranks is 147 out of 188 countries and territories according to the human development index (HDI), a summary measure of healthy life expectancy, standard of living and access to knowledge. Life expectancy from birth (66.2 years) is lower than for the South Asian region (68.4 years) and world (71.5 years) [28]. Health expenditure in Pakistan is estimated by the World Bank [29] as 2.8% of gross domestic product (GDP), and 0.4% of health expenditure is devoted to mental health [30].

Lahore is the second largest city in Pakistan, with an estimated population of 7.6 million. This survey was administered at six hospitals in Lahore: Mayo hospital, Jinnah hospital, combined military hospital and services hospital, doctors hospital, Lahore general hospital, between June 2015 and September 2015.

Measures

A single self-report questionnaire was used consisting of the R- DAQ scale [25]; questions concerning participant demographics and medical specialism education and training; and items relating to the possible causes of depression based on a format previously used in a study of university students views about mental illness [20]. Findings concerning the relationships between casual attributions for depression and attitudes measured by the R-DAQ are reported in a further paper [27].

The R-DAQ is a 22-item scale derived from the 20-item DAQ [26]. The original DAQ has been used in studies conducted in Europe [31], Australia [32], Japan [33] and elsewhere [34], with GPs and other health professionals including nurses [35–37] and pharmacists [38]. Despite its extensive use, the DAQ has been identified as having psychometric weaknesses, as well problems in relation to its complex wording and its development with a specific professional group based only in Britain which potentially limited is suitability for use with the wider professional workforce involved in depression identification, support and management. The revised version was developed following a pooled analysis of study findings [31] and a Delphi consensus study involving experts from UK, the USA, Australia, Belgium, Finland, Estonia, and Italy [25]. The 22-item R-DAQ incorporates 9 items from the original 20-item DAQ scale, together with additional items derived and adapted from other mental illness attitude measures including the defeat depression campaign Mori poll questionnaire [39], European alliance against depression (EAAD) instruments [40], the survey derived from the community attitudes toward the mentally Ill (CAMI) measure [41] used within the research surveys of Great Britain (RSGB) Omnibus [42], as well as items proposed and reviewed by the Delphi panel. The scale items are attitude statements with response options noting level of agreement between ‘strongly agree’ and ‘strongly disagree’, scored with a five-point Likert scale.

The R-DAQ initial study involved a sample of 1193 health professionals but these were largely nurses and almost entirely from the UK. In this development study psychometric adequacy was evident: the total scale internal consistency was 0.84, and there was clear construct validity, easy readability, and minimal floor and ceiling effects. Three sub-scales were evident in the R-DAQ development sample concerning respondents’ attitudes about ‘professional confidence in depression care’, ‘therapeutic optimism about depression’ and ‘generalist perspective about depression occurrence, recognition and management’.

Procedure

Ethical approval for the study was sought and provided by the ethical review committee of CMH Lahore medical college, Lahore, Pakistan. Convenience sampling was employed and all available medical practitioners, excluding any who had formal diploma or fellowship training in psychiatry or were practicing as psychiatrists, were approached by four medical students who visited the sites on several occasions over a 3-month period. These researchers provided the participants with study information, a consent form and the survey questionnaire, and all potential participants were ensured anonymity and informed that the reporting of survey responses would be not enable identification of any individual.

The R-DAQ did not require translation as English is the second official language of Pakistan and widely spoken, especially among professionals, and is used as the medium of instruction in all the medical schools. The questionnaire was pilot tested with six Pakistani medical students to ensure that the phrasing, terminology, layout, and time taken to complete the survey were understandable and appropriate to the target population of Pakistani medical practitioners. Some minor modifications of the demographic items were required, and there were no issues with the readability of the R-DAQ (and examination established the Flesch–Kincaid grade level was 9.4, indicating it to be understandable to a typical 14–15 year-old student).

Analysis

The survey sample size was determined in relation to exploratory factor analysis (EFA) of the R-DAQ: calculation was based on the pattern of communalities in the variables being relatively wide, the expected ratio of variables to factors being between 4 and 5, and an expected three-factor solution. These considerations indicated that a sample of between 200 and 350 was necessary [43]. We anticipated, on the basis of response rates in similar studies conducted within a similar population [20, 21], that a total of 700 medical practitioners would need to be contacted to provide our required sample.

The study data were analysed with the FACTOR (10.03.01) programme [44], using unweighted least squares (ULS) with promin rotation (an oblique method). The freely available FACTOR programme enables the factor analysis to be computed from a polychoric dispersion matrix, which is of particular value in regard to the common situation of analysing ordinal data such as derived from Likert-type scales. EFA based on a dispersion matrix using Pearson correlations (as in the commonly used analysis packages and techniques such as SPSS) can lead to underestimation of the strength of the relationships between variables, with reduced factor loadings compared to EFA based upon on the polychoric correlations matrix. Additionally the FACTOR programme provides alternative reliability estimates to the conventional Cronbach’s alpha [45] which, despite its being the most frequently reported measure of internal consistency, has been subject to sustained critical commentary—not simply because it underestimates the reliability of a test and overestimates the first factor saturation, but because it reveals only the average degree of “interrelatedness” between scale items, which has a limited relationship to scale internal consistency [46].

Some additional analyses were conducted with SPSS v. 22 [47] using principle axis factoring (PAF) with oblique rotation using the direct Oblimin method, to obtain scree plots for the data, compare EFA findings derived from different procedures and matrices, and known groups scale validation in which differences in R-DAQ sub-scale mean scores were analysed between participant sub-groups (formed on the basis of their involvement in psychiatry education and training), using independent samples t tests.

Prior to EFA, the distributions of the R-DAQ items were examined for skewness and kurtosis. Multicollinearity was assessed by examining the tolerance values (which should be close to 1.0) for each item within a linear regression model and the variance inflation factor (VIF) values, which should be less than 10.0, the recommended VIF value to further examine the data for multicollinearity.

The Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy (which may vary between 0 and 1, with values closer to 1 are better, and a value of 0.6 the usual minimum) and Bartlett’s (1954) test of Sphericity (a significant finding enables rejection of the null hypothesis that the correlation matrix is an identity matrix) were used to examine the data for suitability for EFA. The anti-image matrix of correlations was examined to ensure that all elements on the diagonal of the matrix were greater than 0.5, indicating the sample was adequate [48]. Additionally, the suitability of the correlational matrix for factor analysis was checked by examining the number of off-diagonal elements in the anti-image covariance matrix greater than 0.09: the count of off-diagonal elements in the anti-image covariance should be less than 30% [48].

Catell’s scree plot (using SPSS) and parallel analysis [49], a widely recommended Monte Carlo simulation technique (using FACTOR with minimum rank factor analysis and polychoric correlations), were used to determine the number of factors to retain. The adequacy of factor solutions was assessed on the basis the percentage of variance explained, the theoretical coherence of factors, and the simplicity of the factor loadings. Bentler’s [50] simplicity index and the loading simplicity index [51] were used to assess the level of factor simplicity attained in the rotated solution. We tested the goodness of fit of the explanatory model using goodness of fit index (GFI), which ranges between 0 and with values in excess of 0.9 considered an indication of adequate model fit [52], and the root mean square of residuals (RMSR), which was assessed taking into account Kelley’s criterion [53], and also by applying the proposition that if the RMSR is of the order of four divided by the square root of sample size, then a test of significance would not reject the hypothesized model [54].

The reliability and internal consistency of the derived factor solution was examined using a formula based on the standard error of factor scores solutions [55] to provide an estimate reflecting the proportion of variance the items’ factor score accounted for by the underlying common latent variable that drives these item scores, as well as by Cronbach’s coefficient alpha [45], Guttman’s Lambda 6 coefficient (particularly recommended when inter-item correlations are low in relation to squared multiple correlations), and by McDonald’s Omega coefficient (widely considered a more robust estimate of reliability than alpha, based upon square of the correlation between the scale score and the latent variable common to all the indicators in the infinite universe of indicators of which the scale indicators are a subset) [56]. Convergent validity was assessed by performing item-scale correlations corrected for overlaps using Pearson’s product moment correlation coefficient: corrected item-total correlations were considered acceptable if ≥0.2 [57].

Respondent characteristics: demographics, training and practice

R-DAQ analysis

Dimensionality

Suitability for EFA was confirmed by the Kaiser–Meyer–Olkin measure of sampling adequacy which was 0.691, and Bartlett’s test of sphericity was statistically significant χ² (231) = 1936.6; P < 0.001. The individual measures of sampling adequacy for each of the 22 items were examined in the anti-image of the correlation matrix, and item values ranged between 0.56 and 0.82 and mostly (18/22) in excess of 0.60; the number of off-diagonal elements in the anti-image covariance matrix greater than 0.09 was checked, and the count of off-diagonal elements in the anti-image covariance was 19% (i.e. less than 30%) [48]. Multicollinearity assessed by examination of the tolerance values for each item within a linear regression model showed values between 0.64 and 0.89, and the VIF values of all the variables were less than 1.60.

Initial EFA showed eight components had eigen values above 1, explaining 58.8% of the variance; but the scree plot (Fig. 1) indicated three components. Parallel analysis (PA) based on minimum rank factor analysis (MRFA) [59] of the 22 item scale likewise indicated three dimensions based on the random explained common variance. The overall percentage of common variance explained, estimated with MRFA for the three factor model, was 49.8%.

Oblique rotation methods are generally recommended for EFA with social science data [60], and Promin rotation (Lorenzo-Seva 1999), the option advised by the developers of FACTOR, was used. Items that fitted poorly with the emergent factor structure in terms of their contribution to the theoretical meaning, or because of weak factor loadings or cross loadings between factors, were successively removed and the analysis repeated. This process led to the removal of seven items: it was apparent for this sample that five items, which in the R-DAQ development process and analysis were indicative of pessimistic or deterministic conceptualisations of depression, fitted poorly for the Pakistani clinicians. Item 9 (Becoming depressed is a natural part of being old) strongly cross-loaded with the generalist perspective factor items; whereas items 12 (Becoming depressed is a way that people with poor stamina deal with life difficulties), 13 (Once a person has made up their mind about taking their own life no one can stop them), 6 (Depression treatments medicalise unhappiness) and 5 (One of the main causes of depression is a lack of self-discipline and will-power) exhibited weak and inconsistent loadings.

There was also weak loading for item 19 (It is rewarding to spend time looking after depressed patients), whilst item 8 (I am more comfortable working with physical illness than with mental illnesses like depression) loaded with the generalist factor rather than (when reversed) fitting with professional confidence in depression management.

The final model for this population was comprised of three factors and contained 15 items (5 items in each of the dimensions). The three factors explained 46.3% of the 15-item scale variance, and these three factors were the same as identified in the R-DAQ development sample: professional confidence in depression care, therapeutic optimism/pessimism about its course and treatment, and a generalist perspective about the occurrence, recognition, and management of depression (Table 2).

R-DAQ item	Factor
	Generalist perspective	Professional confidence	Therapeutic optimism
22: Anyone can suffer from depression	0.641*	−0.216	−0.003
2: Depression is a disease like any other (e.g. asthma, diabetes)	0.632*	0.026	−0.162
10: All health professionals should have skills in recognising and managing depression	0.620*	0.164	0.006
14: People with depression have care needs similar to other medical conditions like diabetes, COPD or arthritis	0.593*	0.034	−0.021
16: Recognising and managing depression is often an important part of managing other health problems	0.555*	0.143	0.214
15: My profession is well trained to assist patients with depression	0.018	0.631*	0.082
17: I feel confident in assessing suicide risk in patients presenting with depression	−0.070	0.613*	−0.035
11: My profession is well placed to assist patients with depression	0.063	0.590*	0.045
7: I feel confident in assessing depression in patients	0.104	0.553*	−0.071
1: I feel comfortable in dealing with depressed patients’ needs	0.033	0.417*	−0.012
4: Antidepressant therapy tends to be unsuccessful with people who are depressed (reversed)	−0.102	0.059	0.614*
3: Psychological therapy tends to be unsuccessful with people who are depressed (reversed)	−0.089	0.119	0.545*
20: Becoming depressed is a natural part of adolescence (reversed)	−0.039	−0.102	0.440*
18: Depression reflects a response which is not amenable to change (reversed)	0.095	−0.106	0.392*
21: There is little to be offered to depressed patients who do not respond to initial treatments (reversed)	0.290	−0.139	0.336*

* Factor loading ≥ 0.32

The EFA procedure was also conducted using the principal axis factoring with direct Oblimin rotation, providing an identical structure (the structure matrix is shown in Table 3).

	Factor
	Generalist perspective	Professional confidence	Therapeutic optimism
10: All health professionals should have skills in recognising and managing depression	0.621*	0.256	0.055
22: Anyone can suffer from depression	0.594*	−0.083	0.001
2: Depression is a disease like any other (e.g. asthma, diabetes)	0.569*	0.082	−0.068
16: Recognising and managing depression is often an important part of managing other health problems	0.548*	0.262	0.233
14: People with depression have care needs similar to other medical conditions like diabetes, COPD or arthritis.	0.498*	0.137	0.035
15: My profession is well trained to assist patients with depression	0.111	0.620*	0.133
17: I feel confident in assessing suicide risk in patients presenting with depression	0.010	0.581*	−0.003
7: I feel confident in assessing depression in patients	0.169	0.547*	−0.016
11: My profession is well placed to assist patients with depression	0.113	0.545*	0.004
1: I feel comfortable in dealing with depressed patients’ needs	0.072	0.404*	0.032
4: Antidepressant therapy tends to be unsuccessful with people who are depressed (reversed)	−0.034	0.096	0.520*
3: Psychological therapy tends to be unsuccessful with people who are depressed (reversed)	−0.010	0.148	0.475*
20: Becoming depressed is a natural part of adolescence (reversed)	−0.014	−0.046	0.403*
18: Depression reflects a response which is not amenable to change (reversed)	0.092	−0.036	0.357*
21: There is little to be offered to depressed patients who do not respond to initial treatments (reversed)	0.277	−0.025	0.323*

* Factor loading ≥ 0.32

Main findings

This study provides details of the first use of the R-DAQ scale among Pakistani medical practitioners. Psychometric analysis indicates the same three-factor structure as apparent in the predominantly UK-based scale development sample. Item scores for each of the 22 R-DAQ items indicated a less positive perspective about depression and less endorsement of a generalist view about its occurrence and management. The difference in item scores between the Pakistan and UK samples was largest for the pessimistic and deterministic statements that (when reverse scored) comprised the therapeutic optimism sub-scale. These differences were most marked for items 5 (One of the main causes of depression is a lack of self-discipline and will-power), 9 (Becoming depressed is a natural part of being old), and item 12 (Becoming depressed is a way that people with poor stamina deal with life difficulties). Among this sample of Pakistani doctors, these negative views that depression was related to personal weaknesses and a natural part of growing old were not only widely held, but also appeared to link in the factor analysis with the other scale factors. This necessitated removal of seven items from the scale to provide the most parsimonious solution for its use with this population of clinicians, involving a theoretically sound factor structure and adequate measures of internal consistency and reliability.

The internal consistency for the total scale measured with Cronbach α was lower (0.64, or 0.69 using polychoric matrix and based on 15 items) than obtained within the R-DAQ development sample (0.84). In the current sample, of the three sub-scales, the five item generalist perspective sub-scale provided highest reliability (0.77) and internal consistency (0.71) values in contrast to the UK sample where the Cronbach’s α value was 0.57 (total sample) or 0.62 (GPs and adult nurses).

Strengths and weaknesses

This study was based on a convenience sample of doctors based in six hospitals in Lahore rather than a probability sample representative of the medical staff in Pakistan; this limits generalisations based on these findings, though the selection procedure which sought all of the available medical staff in the selected hospitals, together with the high response rate (86%) provide some indication of the representativeness of the findings. The study was located in the capital of the Punjab and 70% of Pakistani medical students are enrolled in medical colleges in this region, which may add to the generalisability of findings to the wider population. The excess of female doctors (52.4%) in the sample is reflective of the proportion of women (58.7%) registered with the Pakistan medical and dental council [61].

In this study, we did not conduct test-re-test reliability analysis; and our study design did not enable validation of the R-DAQ scale using criterion measures such as (concurrent validity) the extent of correlation with other measures of similar constructs (such as other mental health attitude or stigma scales), or (predictive validity) observations of relevant clinical behaviours (such as extent and quality of depression care delivery, prescribing treatments, or providing referrals). However, in this study, the sample was sufficient to enable a validation procedure using the known groups method, which showed scale variations associated with participant characteristics in accordance with the hypothesised differences.