Study design, sampling and population
A population-based cross-sectional survey was conducted from October to November 2016 in the Yangon Region; the most developed and densely populated area of Myanmar. The population of the Yangon Region is 7.4 million, where 5.2 million reside in urban areas and 2.2 million reside in rural areas [24]. It accounts for 14.3% of the entire Myanmar population [24]. The Yangon Region is composed of four districts, of which only the North and South include both, urban and rural populations, and the East and West are urban only. Therefore, the North and South districts were purposively selected for this study.
We applied a multistage study design. The sampling frame for each district was based on the 2014 census [24], with villages (in rural areas) and wards (in urban areas) used as accounting areas. A “ward” is a segment of an urban township. In the first step of sampling, we selected North and South districts. The second step was to select urban wards and rural villages from these two districts. There are 125 wards and 235 villages in the Northern district and 110 wards and 375 villages in the Southern district. Eight wards and eight villages were randomly chosen from each of the two districts, i.e. a total of 16 wards and 16 villages. In the third step, households were randomly selected from each ward and village. The number of households selected in the two districts was based on the population size of the districts and by the urban–rural distribution, using proportional probability sampling. The population distribution is 64.8 and 35.2% in the North and South district, respectively [24]. We randomly selected one woman and one man from every other household based on a list of households with at least one person aged 18–49 years old. We obtained information from the District Health Department in Yangon Region about the number of wards and villages in the selected districts. The number of men and women aged 18–49 years old in the selected wards and villages were obtained with the help of the local authorities, midwives or lady health visitors. In each household, one person was asked to list all of the family members within the age range of 18–49. From this list, one person was randomly invited to participate in the study using the sealed envelope method [25]. In total, 2400 invitees were identified, and 2391 (99.6%) participated in the present study, 1547 from the Northern district (889 from urban and 658 from rural) and 844 from the Southern district (254 from urban and 590 from rural) (Fig. 1).
Both men and women aged 18–49 years, independent of ethnicity, were invited to the study. The age group of young adults up to middle age was chosen, as the present study is a part of a larger project targeting women in reproductive age. We excluded people who were not ordinary residents such as military personnel, Buddhist monks and nuns and other institutionalized persons, and those who were physically or mentally too ill to participate. The sample size was calculated based on the expected prevalence of exposure to domestic violence among women committed by the spouse (21%), previously reported in the 2016 Demography and Health Survey (DHS, 2015–16) [26]. The prevalence of domestic violence was used in the sample size calculation because it is a main outcome in the larger project, in which the present study is a part of.
Data collection
Data was collected from structured interviews, a modified version of the Myanmar Demography and Health Survey (DHS) [26], which was already translated to Burmese. The Hopkins Symptom Checklist (HSCL-10) [16] was included for recording of mental distress, which was translated from English to Burmese by a psychiatrist and back translated to English by an English professional.
The principal investigator and 12 trained field workers (MDs) carried out the data collection. Both male and female interviewers with experience in population-based surveys were recruited. A two-day course was conducted for the purpose of training in interviewing techniques, to learn about the purpose of the study, sampling methods, interpersonal communication skills, informed consent, and the survey questionnaire.
A pilot survey was conducted on the 1st and 2nd of October 2016 in 54 households of one Ward of the Dagon Seikkan township, a township not included in the survey sample. During this survey, interviews were conducted in order to assess clarity, cultural acceptability and understanding of the questions. Based on these results, a minor amendment was created in the Burmese version of the HSCL-10.
Male fieldworkers interviewed male respondents and female fieldworkers interviewed female respondents. After selecting the eligible person, the interviewer obtained informed consent to undertake the survey at the onset of the interview. In order to ensure privacy, a separate room or area outside of the household was chosen before the interview of the respondent. In the event of the respondent being out of reach at the time of the initial visit, the interviewer made at least two repeat visits on that same day. Due to resource limitations and travel distances, it was decided not to return to the household on another day. In the case of no one being present at the household on the day of the interview, or if the participant did not want to participate, they were regarded as non-responders. If a respondent became distressed during the interview, the interview was stopped, and only continued if the respondent was willing, and able to continue answering the questions. After collecting the data from the interviewer, the supervisor checked the completeness of each questionnaire. When answers were missing (non-refusal), a return visit to the household was made and the questions were asked again. Data entry was completed using Epidata software, version 3.1.
Study variables
Mental distress was assessed using The Hopkins Symptom Check List-10 (HSCL-10), a mental health-screening tool that emphasizes the dimensions of depression and anxiety. This tool is recommended for screening purposes because the instrument has high sensitivity and specificity in identifying ‘non-distressed’ and ‘distressed’ groups in the general population [16]. It was developed from HSCL-25 [27] and consists of 10 items on a four-point scale ranging from 1 (not at all) to 4 (extreme) with a higher mean score indicating increased mental distress. The 10 items included in this shorter version are; feeling panicky, anxious, dizzy, tense, sleepless, sad, worthless, hopeless, fault within self and finding everything a burden. The presence of symptoms during the past week (including the day of the interview) was recorded, of which six items are related to depression and four items are related to anxiety. The reliability and validity of HSCL-10 as a screening measure for mental distress has been demonstrated in community surveys in Norway [16] and Pakistan [28]. In the present study, the internal consistency (Cronbach α: .85) was similar to the Pakistan population-based study (.86) [28]. According to Strand et al. [16] and Sørlie et al. [29], the cut-off criteria of 1.85 is significantly associated with an increased risk of depression and anxiety (mental distress).
Socio-demographic variables collected during the survey included age, gender, location, marital status, family members, educational level, occupational status, income, number of children, migration and household debt. Age (years) of the respondents was categorized into age-groups of 18–29, 30–39 and 40–49. For educational levels, we operationalized the number of years at school into three separate groups: ≤ 5 years (also includes those with no formal schooling); 6–11 years; and more than 11 years of schooling. Age and education (years of schooling) were also used as a continuous variable in multivariable analysis.
With regard to household income, the total monthly household income was divided by the number of residents regardless of age, generating a per capita monthly income. Daily individual income was categorized into three groups according to the World Bank’s cut-off of poverty lines of 1.90 USD/day and 3.10 USD/day [30] i.e. low (≤ 1.90 USD/day), medium (between 1.90–3.10 USD/day) and high (≥ 3.10 USD/day).
Occupational status was divided into dependent, unskilled worker, government staff, non- government staff and small business owners. Unskilled workers included individuals undertaking odd jobs, dependents (including students), unable to work, and housewives. Marital status was categorized into currently married, never married (single), separated, divorced or widowed.
Health status variables: current health status was self-reported based on the question “In general, how would you characterize your current health?” The response options were poor, not very good, good and very good. The responses were operationalized into “poor health” (poor, not very good) and “good health” (good, very good). To determine functional disability, the question “Do you have impairments …” was asked, with possible replies being “no”, “yes, a little”, or “yes, a lot”. With regard to mobility, vision, hearing, the ability to bathe one’s self, dressing and remembering or concentrating, the responses were categorized into “yes” and “no” [31].
Muscle and skeletal pain were measured by the following questions: “Have you in the last 12 months experienced pain several times in: head, neck/shoulders, arms/legs/knees, stomach, and back?” with responses being “yes” or “no”. Based on the answers given, three groups were created: 0 pain sites, 1–2 pain sites and 3 to 5 pain sites [32].
Domestic violence (DV) was recorded if the respondent was exposed to physical, sexual and/or emotional violence committed by the intimate partner (for married / partnered participants only), or if the respondent, after the age of 15 years, was exposed to physical violence committed by anyone else (for all participants), or if the respondent was exposed to childhood sexual violence (for all participants). The domestic violence questions were obtained from the WHO Myanmar Demographic and Health Survey [26].
Data analysis
STATA/IC version 15.0 was used in the data analysis. We used the survey prefix command “svy” for complex survey data. For descriptive analyses, we estimated prevalence with 95% confidence interval (CI), mean with standard deviation (SD) and median values with Inter Quartile Range (IQR). For bivariate analysis, a Pearson’s chi-square test was used for testing differences between proportions. Multiple linear regression was used to estimate the association between years at school (education) and HSCL score (mental distress), with separate analysis for men and women.
We checked the distribution of years at school by mental distress (HSCL score) (Fig. 2). We drew a directed acyclic graph (DAG) (Fig. 3) to illustrate our strategy of analyzing data (multivariable analyses). We identified confounders, mediators and colliders based on prior knowledge about the connection between exposure and outcome. We adjusted for cofounders, however not for mediators since we are interested in estimation of the total effect of the exposure. If the mediating variables are adjusted for in the model, that path will be closed and we no longer see the total effect of the exposure on the outcome. Moreover, that may cause over-adjustment bias in the estimation of the total effect and reduce precision [34]. Based on the DAG (Fig. 3), age, income and marital status were identified as confounders in order to estimate the total effect of education on the HSCL score (mental distress). We detected an interaction between education and age, which we adjusted for in the estimation of the total effect of education on mental distress. We analyzed the marginal effect of the association between years at school and mental distress at low age (18 years) and high age (49 years) in men, using a marginal plot analyses (Fig. 4). The assumptions of a linear model (linear effects and constant error variance) were tested by plotting residuals versus predicted values. We looked for observations with high influence by plotting delta-betas versus observed numbers. All significance tests were two-sided, and p-values < 0.05 were considered to be statistically significant.