Psychometric properties of the Bangla version of the Kessler Psychological Distress Scale (K6)

A self-administered questionnaire survey was conducted among students aged 13–24 years from two high schools and two universities in Dhaka, the capital of Bangladesh. These institutions were purposively chosen based on geographical convenience and connection with the researchers in order to maximize participation. Each of the participating schools and universities approved the survey prior to its administration and nominated their representatives (e.g., lecturers/teachers) to facilitate implementation of the study. For high school students, individual written informed consent was obtained from students and one parent of each student prior to administering the survey. The university students completed the informed consent form prior to completing the survey. Students completed the survey in classroom settings in the presence of one of the research team members and representative lecturers/teachers. The survey was repeated in a week. The survey was available in Bangla, the native language of the study participants. It took approximately 10–15 minutes to complete the survey. Data for this study were collected between August 2017 and April 2018.

Psychological distress of the study participants was assessed using the Bangla K6 scale, which asked participants to rate how often during the past four weeks they felt: (1) nervous; (2) hopeless; (3) restless or fidgety; (4) so depressed that nothing could cheer them up; (5) that everything was an effort; and (6) worthless. A five-point rating scale was used as the response options, indicating 0 (none of the time); 1 (a little of the time); 2 (some of the time); 3 (most of the time); and 4 (all of the time). The Bangla K6 scale, used in the present study, was translated by the New South Wells (NSW) Health, Australia (Transcultural Mental Health Centre 2012). In the translation process, the Australia’s NSW Health made efforts to produce a translated scale that is culturally competent and linguistically appropriate for the target Bangla speaking population. The Bangla K6 was developed following a standardized translation model developed by the Epidemiology and Surveillance Branch of the NSW Department of Health in collaboration with the NSW Multicultural Health Communication Centre. The model included three major steps: (i) pre-translation preparation, (ii) translation, and (iii) verbal back-translation. Translators accredited by the Australian National Association for the Accreditation of Translators and Interpreters (NAATI) conducted the translations. Verbal back-translation was performed by bilingual interviewers and/or interpreters to enhance the quality of the translation, which ensured that the translation was scientifically sound but was neither too formal nor too colloquial. A pilot survey was conducted to detect any obvious language difficulties with the translated version (Williamson et al., 2000). In addition to the Bangla K6, the students in the current study completed the 10-item Center for Epidemiologic Studies Short Depression (CES-D-10) scale twice with a week’s interval between the administrations. Furthermore, students provided information on some selected background characteristics including age and sex.

Cronbach alpha was computed to evaluate the internal consistency of the Bangla K6 scale. Cronbach alpha measures how closely related a set of items are as a group and examines whether the items measure an underlying construct. Higher alpha values indicate greater internal consistency with alpha ≥ 0.70 suggesting acceptable (Bland and Altman, 1997). Principal component analysis (PCA) was used to examine variability in the data and to determine whether the items on the Bangla K6 belong to one or more constructs.

Confirmatory factor analysis (CFA), as used elsewhere (Easton et al., 2017; Bessaha, 2017; Lee et al., 2012), was used to examine how well the Bangla K6 items represent one or more latent constructs. CFA offers the testing of hypothesis that a relationship between observed variables and their underlying latent constructs exists. In particular, CFA was conducted to determine the goodness-of-fit between the original factor structure of the K6 and the sample data collected using the Bangla K6. CFA was also used to examine whether factor structure of the Bangla K6 could be refined to improve the model fit by addition of pathways. The determination of a path addition to a CFA model is based on the examination of modification indices (MI). If a MI between two items is high in relation to other MIs, it suggests that the addition of a path is expected to improve the overall fit of the model. If the addition of a path does not make theoretical or logical sense, then the path should not be included. The following goodness-of-fit indices of CFA, along with their threshold values, were used to assess the degree of fit between the model and the sample: Chi-square test; Tucker Lewis Index (TLI: > 0.90 acceptable, > 0.95 excellent); Comparative Fit Index (CFI: > 0.90 acceptable, > 0.95 excellent); Root Mean Square Error of Approximation (RMSEA: < 0.08 acceptable, < 0.05 excellent); and Standardized Root Mean Square Residual (SRMR: < 0.08 acceptable, < 0.05 good-fit) (Hooper et al., 2008).

Furthermore, differential item functioning (DIF) analysis (Crane et al., 2006) was conducted to examine whether the relationships between psychological distress and the Bangla K6 item responses were influenced by any participant factor (e.g., student group). DIF is a statistical characteristic of an item that shows the extent to which the item might be measuring different abilities for members of separate subgroups. Thus, DIF analyses examine potential bias of the scale items and evaluate whether different population groups perform differently on items of the scale. In presence of DIF (which was not desirable), participants in different groups (school students vs. university students), but with equal underlying levels of psychological distress, would differ in the probabilities of assenting to the Bangla K6 items and their categories. In absence of DIF (which was desirable), the probability of assenting to the Bangla K6 items and their categories would not differ across the student groups.

Prior to computing test-retest reliability coefficients, Bland-Altman (BA) plot was used to assess and display agreement and systematic differences between the measurements at two time points. The BA plot examines heteroscedasticity between the measurements and identifies possible outliers. The 95% confidence interval (CI) of limits of agreement (LOA) was used to examine the magnitude of systematic differences, with a narrower interval indicating a higher stability (Bland and Altman, 1999). The intra-class correlation coefficient (ICC) with two-way mixed effects model was used to estimate the test-retest reliability coefficient of the Bangla K6. In examining test-retest reliability, while the Pearson correlation coefficient can be used to measure of the correlation of measurements, and paired t-test can be used to measure agreement; none of them measures both correlation and agreement. ICC is such an index that measures both correlation and agreement between the measurements, and as such, it has become a widely used measure of test-retest reliability. The ICC values can be interpreted as 0.75–0.90 as good reliability, and > 0.90 as excellent reliability (Koo and Li, 2016).

The predictive performance of the Bangla K6 scale was examined by Receiver Operating Characteristic (ROC) curves (Fassaert et al., 2009) with CES-D-10 as the reference criterion. ROC curve tells how much a fitted model is capable of distinguishing between classes and gives an idea about the benefit of using the test in question. An ROC curve is a plot of true positive rate (sensitivity) against the false positive rate (100-specificity) for different cut-off points of a diagnostic test. Each point on the ROC curve represents a sensitivity/specificity pair corresponding to a particular decision threshold. An ROC curve demonstrates the trade-off between sensitivity and specificity as any increase in sensitivity is accompanied by a decrease in specificity. We used CES-D-10 because it is considered as one of the most common measures of depressive symptoms (Andresen et al., 2013; Radloff, 1977). Participants with a total score of ≥ 10 on CES-D-10 were classified as showing depressive symptoms, and those with scores < 10 were classified as non-depressive (Radloff, 1977). The area under the curve (AUC) was calculated as a measure of the extent to which the Bangla K6 scores predicted depressive symptoms using the dichotomized CES-D-10. The AUC is used as a measure of test accuracy, and it shows how well a test can distinguish between two diagnostic groups (positive/negative). An AUC of 0.50 suggests that the Bangla K6 is no better than chance at predicting depressive symptoms, whereas an AUC of 1.0 would indicate that the Bangla K6 predicts depressive symptoms perfectly. AUCs are typically interpreted as AUC: 0.90–1.00 excellent, 0.80–0.89 good, 0.70–0.79 fair, and < 0.70 poor (Cicchetti, 2001). Furthermore, the relationships between the Bangla K6 and the CES-D-10 scores were examined using the Pearson’s correlation coefficients. CFA was conducted by using the statistical software AMOS 25, and the rest of the analyses were performed by Stata SE 14.

Of the 941 students approached, a total of 718 students completed the survey at time 1 (week 1; response rate = 76.3%) and 712 completed at time 2 (week 2). Half of the participants (51%) were school students, aged 13–17 years, and the rest (49%) were university students, aged 18–24 years. The mean age of the study participants was 18.39 years (SD = 3.52) and 45% were female.

High values of Cronbach’s alpha demonstrated good internal consistency of the Bangla K6 with alpha = 0.87 at time 1 and alpha = 0.88 at time 2. An initial PCA of Bangla K6 revealed that the six items of the scale represented a single factor solution at time 1 with 51% of the total variance explained and the first eigenvalue being 3.04 (second eigenvalue = 0.75). A single factor was also extracted for the Bangla K6 at time 2 with 52% of total variance explained and the first eigenvalue being 3.11 (second eigenvalue = 0.74). The PCA results provided evidence of uni-dimensionality in the Bangla K6 items.

The CFA provided further evidence that the Bangla K6 scale consisted of one factor at both time 1 and 2, after adjusting for relatively high MIs. The goodness-of-fit indices of the Bangla K6 at time 1 were: χ2(7) = 7.92, p = 0.34; CFI = 0.996; TLI = 0.997; RMSEA = 0.014; and SRMR = 0.012. The fit indices at time 2 were: χ2(6) = 9.97, p = 0.13; CFI = 0.997; TLI = 0.992; RMSEA = 0.03; and SRMR = 0.014. Insignificant Chi-square values suggested good fit of the CFA models. Both RMSE and SRMR were below the cut-off point of 0.05, suggesting satisfactory model fit. The values of CFI and TLI (> 0.95) represented an excellent fit. The item loadings ranged from 0.53 (item 6) to 0.68 (item 4) at time 1 (Figure 1), while 0.57 (item 6) to 0.70 (item 4) at time 2 (Figure 2).

Figure 1

Figure 2

Further analyses showed only a minor DIF for item 3 (restless or fidgety) across the student groups at time 1, which was graphically inspected. In case of similar levels of non-specific psychological distress, the university students had a slightly higher probability of affirming item 3 than the school students. Given that the DIF for item 3 was minor at time 1 with no DIF at time 2, the DIF findings were unlikely to influence the current analyses.

The BA plot for test-retest reliability of the Bangla K6 demonstrated small differences on repeated measurements without any evidence of heteroscedasticity (Figure 3). The narrow range of the LOA and a few outliers in the Bangla K6 indicated a high stability and less systematic differences in the scale over time. ICC value for the scale was 0.80 (0.77 – 0.82), which suggested an acceptable reliability.

Figure 3

The Bangla K6 provided good predictions of depressive symptoms, measured by the CES-D-10 scale, with AUC = 0.82 (0.79–0.85) at time 1 and 0.80 (0.76–0.83) at time 2. ROCs for predicting depressive symptoms by the Bangla K6 scores for school and university students are presented in Figure 4 for time 1 and in Figure 5 for time 2. These figures clearly show that the predictive validity of the Bangla K6 scores in predicting depressive symptoms were comparable across student groups at both time points. When investigating the associations, the Pearson’s correlation coefficient between the Bangla K6 and the CES-D-10 scores were considerably high (Pearson’s r = 0.68 at time 1; 0.67 at time 2).

Figure 4

Figure 5