Measuring the psychosocial wellbeing of Rohingya mothers in Cox’s Bazar, Bangladesh: Psychometric properties of an MHPSS assessment battery

Current study

The purpose of the current study was to adapt, translate, pilot, and assess the psychometric properties of a collection of instruments for use with Rohingya refugee mothers enrolled in an intervention study in Cox’s Bazar, Bangladesh. A second aim of this study was to explore demographic characteristics associated with psychosocial health among a sample of Rohingya refugee mothers accessing nutrition treatment services for their children under age two (i.e., women not necessarily seeking MHPSS support).

Setting

This study was carried out in Rohingya refugee camps in Cox’s Bazar, Bangladesh, where the study’s operational partner, ACF, has provided critical humanitarian services since 2008, including efforts to meet the expanded need from the 2017 refugee influx through a number of Integrated Nutrition Centers (INCs). As a core feature of their holistic approach to supporting child wellbeing, ACF also offered maternal psychosocial support at these INCs, including the Baby Friendly Spaces (BFS) program. BFS is a low-intensity psychosocial support intervention aimed at promoting maternal and child wellbeing and improving care practices among mothers of children under age two who are receiving nutritional treatment for acute malnutrition. During the study period, the BFS program completed approximately 50-100 intakes of mother-child dyads per month at each of the 10 included INCs. Resident populations across INC catchment areas were generally comparable, with the exception of length of time in displacement across camps; some camps primarily hosted people who were displaced in the 1990s and others from the more recent 2017 influx. For this study, we prioritized INCs located in camps hosting more recent arrivals. Data were collected from November, 2021 through January, 2022.

Participants

This paper uses baseline data from a clustered randomized controlled trial (RCT) of the BFS program collected among a sample of 600 Rohingya refugee mothers who were newly enrolling in BFS services at 10 participating INCs. RCT eligibility criteria included being a mother, age 18 or older, of a child under two years of age identified as suffering from moderate or severe acute malnutrition without complication by ACF as a part of their routine programming. Mothers who had cognitive impairment or psychosis that would preclude participation in program activities, as well as mothers of children with severe developmental disabilities or severe malnutrition with complications, were excluded. Additionally, while mothers who received individual counseling from a counselor or psychologist within ACF remained eligible, anyone receiving referral care outside of ACF for more severe mental health or protection needs (in accordance with standard ACF protocols) were excluded.

Measure selection and adaptation

Measures were chosen to assess multiple indicators of person-focused mental health and psychosocial wellbeing as outlined in the IASC Common M&E Framework for MHPSS in emergencies.¹² As this study was undertaken as one of five projects involved in the USAID-funded Health Evaluation and Applied Research Development (HEARD) learning collaborative of related MHPSS studies,³¹ key research domains and assessment instruments were aligned for consistent use across projects. Thus, selection prioritized existing, standardized instruments with demonstrated prior utility in trauma-affected populations across multiple cultures and contexts.

Prior to launching the study, all instruments were translated into Bengali language with independent back-translation (English-Bengali) to facilitate local IRB review. Following reconciliation, the Bengali versions were further translated and back-translated to create a written version in the Chittagonian Bengali dialect that is widely spoken in Cox’s Bazar and has a substantial degree of mutual intelligibility with the Rohingya dialect (which has no standard written form). Ten Chittagonian Bengali-speaking interviewers from the Cox’s Bazar district who had conversational fluency in the Rohingya dialect were then trained in the full assessment battery, and in partnership with Rohingya volunteers, collaboratively standardized Rohingya phrasing. The ten data collector pairs (comprised of a Bangladeshi interviewer and Rohingya volunteer; see procedures below) each subsequently piloted the instruments with 1-2 separate Rohingya mothers recruited from the BFS program at the same INCs that would be involved in study recruitment. After noting all feedback, the Bangladeshi interviewers met together with the research team to make collective recommendations on instrument revisions to clarify wording, reduce redundancy, remove problematic items, align response options, and shorten the overall survey.

Psychological distress was measured using two separate scales. To align across the HEARD learning collaborative, we used the Kessler 6-item scale (K-6), a standardized measure of general distress that has been used in diverse settings.³² Responses are based on frequency of experience over a 2-week recall period, with a 5-point Likert type response option of “none”, “a little’, “some”, “most”, and “all” of the time. We supplemented this primary instrument with items from the International Depression Symptom Scale (IDSS), a 29-item measure of depression developed to reflect symptoms that were commonly reported in a global review of qualitative studies on depression and showed reliable cross-cultural performance in quantitative analyses.³³ The IDSS has previously been validated, refined, and tested among four distinct groups of people from Myanmar,^33–36 though not with Rohingya people. That previous work included using item response theory to develop and further test the Myanmar-wide IDSS & AMHR Short Instrument,^35,36 a 19-item abbreviated measure that performed similarly well and included a subset of IDSS items as well as other items assessing anxiety and post-traumatic stress. Although we initially piloted the 29-item IDSS, feedback regarding the length of the assessment battery led us to reduce the IDSS portion from 29 to 12 items to reduce participant burden. This included removing two items that were redundant to the K-6 (feelings of worthlessness and hopelessness) as well as many items that had been removed from the Myanmar-wide short version (e.g., items related to weight, appetite, and sleep, as well as a number of items describing heart-related symptoms).^35,36 One item about concentration and three somatic items from the full IDSS (headache, stomach ache, and other bodily aches) that were not included in the Myanmar-wide short version were retained on an exploratory basis due to the potential relevance in this population. Responses to all versions of the IDSS are typically provided on a 4-point response scale, but for this administration we revised the response scale to match the K-6. Both the IDSS and K-6 were scored by taking the average of all contributing items to generate scale scores ranging from 0-4, with higher scores indicating greater distress. Given conceptual overlap between the K-6 and IDSS as measures of distress, we also explored the psychometric properties of a combined score using items from both scales, again calculated on the 0-4 score range.

Functional impairment was measured using the 12-item version of the WHO Disability Assessment Schedule 2.0 (WHODAS), a widely used assessment of health and disability developed to be applicable across a range of cultures and contexts³⁷ and used previously with Rohingya adults.³⁸ The WHODAS assesses six domains of functioning: cognition, mobility, self-care, getting along/social interactions, life activities, and participation. The WHODAS is typically administered by asking how much difficulty the respondent has had carrying out various activities in the past month, but we adjusted the recall period to two weeks to align with other scales on the assessment battery. Response options included “none”, “mild”, “moderate”, “severe”, and “extreme or cannot do”. Scores were calculated by taking the average of all contributing items for a range of 0-4, with higher scores indicating greater functional impairment.

Subjective wellbeing was measured using items recommended by the Organization for Economic Co-operation and Development (OECD),³⁹ including a single-item rating of global Satisfaction With Life (SWL) that is widely used internationally and the 8-item Personal Wellbeing Index (PWI).⁴⁰ Frequently administered with the PWI, the SWL item asks “Thinking about your own life and personal circumstances, how satisfied are you with your life as a whole these days?” The PWI has one item for each of seven core life domains (e.g., health, safety, and personal relationships) and an optional item on religion, which we included. Previous research with Rohingya people has used similar question stems, i.e., asking “how satisfied” the respondent is with personal relationships and support from friends.¹⁸ Both the SWL item and the PWI used the same 0-10 response scale, accompanied by the image of a ladder with ten rungs that correspond to the ratings with 0 at the bottom (i.e., “no satisfaction at all”) and ten at the top (i.e. “completely satisfied”). After piloting, one item “what you are achieving in life” was removed, as the question was not clear or meaningful to respondents. The PWI was scored by taking the average of all contributing items for a range of 0-10, with higher scores indicating greater subjective wellbeing.

Coping strategies were assessed using items from the Brief COPE,⁴¹ a 28-item instrument that has been translated and validated in numerous languages,^42–44 and which is suggested as a means of verification in the IASC M&E framework.¹⁴ The Brief COPE assesses a broad range of coping strategies, such as denial, humor, active coping, and use of instrumental and emotional support. Responses were provided on a 4-point scale regarding how frequently each strategy has been used, ranging from 0 “I haven’t been doing this at all” to 3 “I’ve been doing this a lot”, with a 2-week recall period. The Brief COPE has previously been used to generate three overarching coping styles of problem-focused, emotion-focused, and avoidant coping. However, studies have shown the factor structure of the Brief COPE to vary across samples; for example, in a previous maternal sample the instrument consisted of two separate sub-scales: active coping and disengaged coping.⁴⁵ For the current study, the items on the original Brief COPE were supplemented with several items depicting additional coping strategies found to be relevant locally through qualitative research with Rohingya adults (e.g., “thinking about a better future for the children”).¹⁸ After piloting, feedback regarding redundancy, item irrelevance, and respondent fatigue led us to remove some items resulting in a 24-item scale that retained 18 items from the Brief COPE and six locally developed coping items. Scores were generated by taking the average of contributing items for a range of 0-3, with higher scores indicating more frequent use of the included coping strategies.

Data collection

As part of their programmatic intake, women enrolling in the BFS program were screened for study eligibility by BFS providers using the inclusion/exclusion criteria listed above. Eligible participants were informed of the study, and those who indicated an interest in study participation were referred to a trained data collector who obtained oral informed consent and administered the assessment interview. We used an oral consent process due to low literacy levels in the target population. As described above, these interviewers were native Chittagonian Bengali speakers who were also proficient in Rohingya, and who were additionally accompanied by a Rohingya volunteer to facilitate comfort and understanding. Engaging Rohingya volunteers has previously been recommended to help overcome language and cultural barriers²⁹; our paired approach was due to both hiring restrictions in the camps as well as low literacy levels among Rohingya volunteers. As such, consent was obtained by the interviewer with the volunteer present as a witness, and assessment interviews were conducted in Rohingya, the participants’ native language. Baseline data collection lasted between 1-2 hours and was completed prior to the participant engaging in BFS program activities. Data were recorded in KoBo Toolbox⁴⁶ using a handheld tablet. As consent was obtained at the same time as the interview, consent was also documented in Kobo Toolbox.

Ethical approval

This study was approved by the Institutional Review Board of the Institute of Health Economics (approval number IHE-IRB/DU/2021/33/Final) at the University of Dhaka.

Analyses

Analyses were conducted in Stata 17.0.⁴⁷ Descriptive analyses included cross-tabulations of demographic variables and histograms of individual item distributions. We then examined each scale’s factor structure using exploratory factor analysis (EFA). An exploratory rather than confirmatory approach was used due the dearth of prior instrument validation research with this population, the substantial revisions made to some of the instruments prior to data collection that could impact factor structure, and prior evidence for different factor structures for some instruments.^45,48 Because item responses were categorical, we first generated and examined polychoric correlation matrices for the items on each instrument. We determined the number of factors to extract based on number of eigenvalues over 1 and Horn’s parallel analysis for principal components. Factor models used iterated principal factor estimation with promax rotation. Items were retained based on high factor loadings (>.4) and low uniqueness. Where there was ambiguity in factor structure, decisions were made based on parsimony and meaningful interpretation. Once refined, scale scores were calculated and summarized to examine sample distributions, and each scale’s internal consistency was assessed using Cronbach’s alpha (α).

We then evaluated construct validity by examining correlations between scales, hypothesizing that different indicators of similar domains (e.g., K-6 and IDSS, both measures of distress) would be highly correlated; indicators of unique but related domains (e.g., IDSS and WHODAS, measuring distress and functional impairment) would be moderately correlated; and, indicators of positive wellbeing (e.g., PWI) would be negatively correlated with distress but positively correlated with each other. Lastly, we examined correlates of psychosocial health by regressing each scale (as separate models) onto a consistent set of demographic characteristics in multiple linear regression models with cluster-robust standard errors to account for clustering. Due to low missingness (e.g., <1%), listwise deletion was used.

Exploratory factor analysis

International Depression Symptom Scale Items. Parallel analysis identified two eigenvalues above 1, although the second was marginal (adjusted: 1.03) and fell below that of the simulated value in Horn’s analysis. To explore potential multidimensionality, we fit a 2-factor EFA model. Factor loadings and uniqueness are provided in Table 2. Most items loaded highly onto a single factor that accounted for 84% of the variance, with only the three exploratory somatic items loading onto the second factor. As these items were excluded from prior short versions of the IDSS in Myanmar, we exclude these items. Additionally, the item assessing recent thoughts of self-harm had high uniqueness (.82) with no substantial loading onto either factor. In prior use of the Myanmar short instrument this item has been retained for clinical purposes but not factored into scale scores; we opted to do the same for the current study. Thus, the final IDSS scale score was comprised of eight items (IDSS-8) with high internal consistency (α = .87).

Kessler 6-item Scale. Parallel analysis of the K-6 identified only one eigenvalue above 1. In a single-factor model, all six items had satisfactory factor loadings (.57-.69) and satisfactory internal consistency (α = .78).

Combined IDSS/K6. Using the items retained from the separate IDSS and K-6 EFAs, parallel analysis of the combined items supported a single factor structure that accounted for nearly 94% of the variance. Factor loadings for contributing items ranged from .55 to .79, and internal consistency for the 14-item combined scale was high (α = .90).

WHO Disability Assessment Schedule 2.0. Parallel analysis identified three adjusted eigenvalues over 1, although the 3^rd was marginal (1.01) and fell below the simulated value. The first two factors together accounted for 95% of variance. As such we fit a 2-factor EFA (Table 3). Six unique items loaded on each factor sufficiently, and internal consistency of the two subscales were .73 and .78, respectively. Factor 1 was comprised of items reflecting aspects of self-care (e.g., washing your whole body) and social interaction (e.g., joining in community activities), and appeared to present less difficulty (mean = .37, SD = .44), whereas Factor 2 was comprised of more task-oriented items (e.g., standing, walking, household activities) for which a higher level of impairment was reported (mean = .78, SD = .63). However, the factors were moderately correlated (r = .49), and internal consistency was improved for the full scale (α = .81); as such, we opted to explore the relative contribution of each factor to the overall scale score and construct validity, but retained the single full-scale score for regression analysis.

Personal Wellbeing Index. Parallel analysis clearly supported a single factor model with only one eigenvalue over 1 and X% of the variance explained. The six standard items all loaded highly, with factor loadings ranging from .61 to .78. The optional item assessing satisfaction with religion did not load at or above .4 and had high uniqueness (.96) and was thus excluded. Cronbach’s alpha with that item excluded was .85.

Brief COPE and Local Coping Items. Parallel analysis for all coping items (inclusive of locally developed items) indicated four adjusted eigenvalues greater than one, although two of these fell at or below those of the randomly generated set. We first fit 4- and 3-factor models, yet each of these models produced one factor with only a single sufficiently loading item. Thus, we fit a more parsimonious 2-factor (shown in Table 4) in which standard COPE items loaded on Factor 1 and locally developed items on Factor 2, with the exception that finding comfort in religion (a standard item) loaded with the locally developed items. A number of items did not load on either factor (e.g., substance use, making jokes). The internal consistency of Factor 1 was satisfactory (α = .84) and all contributing items had positive item-rest correlations. This was unexpected, given that the items reflected a blend of presumably adaptive (e.g., getting comfort, accepting reality) and maladaptive (e.g., criticizing yourself, giving up) strategies which would make a cumulative score not hold utility. Given this, we further examined scale psychometrics using theoretical or previously demonstrated subscales (e.g., emotion-focused, problem-focused, avoidant, adaptive, maladaptive), none of which produced satisfactory scale properties (results not shown). We then examined the internal consistency of only those theoretically adaptive items loading onto Factor 1, and were able to identify a theoretically useful 10-item subscale of adaptive Brief COPE items (i.e., “B-COPE”, α = 81). A similar effort to identify a maladaptive subset was unsuccessful (e.g., 3-items, α = .47). Using Factor 2, we were able to generate a second 4-item subscale of locally derived items (i.e., “L-Cope”) after excluding “solving disputes”, which had low factor loading and high uniqueness. Removal of that item increased the subscale’s internal consistency from α = .74 to α = .76. We then re-ran the parallel analysis and EFA using only those items retained across the two scales, and items continued to clearly organize themselves across the two distinct factors. As the factors were only minimally correlated, (r = .24), we opted to treat these as two separate constructs for analysis.

Construct validity

Descriptions of each of the final scales, including the number of contributing items, internal consistency, possible and observed score ranges, and sample distributions are reported in Table 5. Distress and functional impairment were relatively low, with moderate subjective wellbeing ratings. Comparing score distributions among like constructs, the Myanmar-specific IDSS items had a slightly higher mean relative to the K-6 (1.03 vs. .77). The two subjective wellbeing scores performed relatively similarly, as evidenced by similar means (6.89 vs. 6.67). There was a larger variation between the two measures of coping, with scores indicating more frequent use of strategies included in the local coping scale (mean = 2.3) relative to the Brief COPE items (mean =1.06).

Correlations between scores, as shown in Table 6, aligned with our hypotheses aside from coping. For example, the IDSS-derived scale and K-6 (i.e., measures of distress) were highly correlated with each other (r = .71), and both moderately correlated with total functional impairment (IDSS: r = .65; K-6: r = .58), although these correlations appeared to be driven largely by WHODAS Factor 2 (r range .61 to .70) rather than Factor 1 (r range .31 to .34). PWI scores, reflecting a composite score of satisfaction with various life domains, were likewise positively correlated with a global rating of life satisfaction (r = .68). Both measures of subjective wellbeing were negatively correlated with both measures of distress and functional impairment (r range: -.35 to -.44). However, the two coping measures were not highly correlated with each other (r = .27) and had seemingly divergent relationships with other scales. For example, the Brief COPE items appeared to be more highly – and positively – correlated with functional impairment and distress (r range from.26-.38), with very low – but negative – correlations with subjective wellbeing (r = -.09 and -.17). The local coping scale had lower positive correlations with distress, but also slight positive correlations with subjective wellbeing (r = .09 and .17).

Demographic correlates

Adjusted associations between demographic characteristics and scale scores are reported in Table 7. Controlling for other factors, demographic characteristics most strongly and consistently associated with psychosocial outcomes included household meat consumption, pregnancy status, and marital status. Relative to mothers from households with the least access to meat, mothers from households eating meat on either a monthly or more frequent basis reported significantly lower distress (β range from -.18 to -.30, most p < .05) and significantly higher subjective wellbeing (β range from .75 to 1.85, all p < .05). Mothers who were currently pregnant also reported significantly higher distress (β range from .18 to .22, most p < .05) and lower wellbeing (SWL: β = -.47; PWI: -.32), though not all estimates were not statistically significant at p < .05. Likewise, the small number of unmarried mothers in the sample reported seemingly higher distress (β range from .51 to .53) and functional impairment (β = .36), and lower subjective wellbeing (SWL: β = -1.42; PWI: -1.09), though these associations were not statistically significant (p < .10). Other characteristics were less associated with distress and wellbeing, but meaningfully associated with other outcomes. For example, mothers with higher educational attainment reported lower functional impairment (completed primary: β = -.23, p < .01) and higher use of coping strategies (any primary: β = .31; completed primary: β = .44; both p < .01). Length of time in displacement had a small but positive association both with indicators of distress (all β = .02, p < .05) and coping (β = .01 and.03, p < .05). Other correlates of coping strategies were less consistent and largely not significant.

Limitations

A key limitation of this study is a lack of formative qualitative work that would have helped to better refine instruments and potentially explain difficult findings. Due to our participation in a broader consortium of research projects, we were limited in our choice of measures and faced time constraints that precluded more intensive instrument development work. Additionally, although the instruments were piloted and refined with careful consideration to the interview process and, for the most part, did show good psychometric properties, the complexities of the adaptation and translation process did not fully align with best practices and could have introduced inconsistencies.^66,69 Overall, the field is in great need of more attention to the development of both qualitatively grounded instruments assessing a range of psychosocial outcomes of interest, as well as clinical validation of pathology-oriented measures,^70,71 both needs we were unable to address.

This study also relied on a convenience sample of mothers enrolled in a longer intervention study, which imposed some inclusion restrictions that would not have otherwise been included (e.g., excluding women referred to outside support for severe distress or suicidality). Thus, this study likely failed to capture the full range of distress that would be observed in a representative community-based sample. This could alter the observed factor structures and also makes it an inappropriate sample to attempt to establish clinical cut-points. However, the sample does represent a population that could be targeted with prevention and promotion interventions.

Lastly, we took an exploratory approach to factor analysis given the changes made to the instruments and the limited prior research with this population; future research can build upon this work to take a more confirmatory approach. It is worth noting that given the extent of post-pilot instrument modifications to remove many items from the IDSS and the Brief COPE and the exploratory factor analytic approach, the resulting scales can only be considered as approximations of the originating tools and cannot be presumed to fully represent the constructs those tools were originally designed to measure.

Underlying data

OSF: Measuring the psychosocial wellbeing of Rohingya mothers in Cox’s Bazar, Bangladesh: Psychometric properties of an MHPSS assessment battery, https://doi.org/10.17605/OSF.IO/AMDF2.⁷²

This project contains the following underlying data:

Extended data

OSF: Measuring the psychosocial wellbeing of Rohingya mothers in Cox’s Bazar, Bangladesh: Psychometric properties of an MHPSS assessment battery, https://doi.org/10.17605/OSF.IO/AMDF2.⁷²

This project contains the following extended data:

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).