The study was a cross-sectional survey of community-ascertained people with SMI in an Ethiopian district as part of the Programme for Improving Mental health carE (PRIME) study (16). PRIME was a consortium of mental health researchers, the World Health Organization, Ministry of Health representatives and non-governmental organisations in five low and middle-income countries, including Ethiopia. In PRIME, participatory district level mental health care plans were developed in order to evaluate the impact of task-shared mental health care on disability and symptom severity for people with priority disorders (17). The larger PRIME study includes participants with depression, alcohol use disorder, psychosis, and epilepsy; the current study only includes those with psychosis.
The study took place in Sodo District, Gurage Zone, of the Southern Nations, Nationalities and Peoples’ Region (SNNPR), Ethiopia, between December 2014 and July 2015. The district is located 100 km from the capital city, Addis Ababa. Reflecting Ethiopia as a whole, 90% of the district inhabitants live rurally with the main sources of livelihood being farming and animal husbandry. The official language is Amharic. At the time of the study, the district had a population of approximately 165,000 with no mental health specialists. However, through PRIME, a district level mental health care plan was being implemented with the goal of expanding access to integrated primary mental health care (18).
Fig 1 shows a flow chart of the participant recruitment process. Health Extension Workers (HEWs) and community key-informants first identified probable cases of SMI. The key informant method is a sensitive case detection technique previously used in a neighbouring district. HEWs are women with at least a grade 10 education and one year of health care training in health promotion and illness prevention activities. HEWs live in the community they serve and visit each household in their catchment area every month and therefore have close community ties (18). The use of community-based case ascertainment methods allowed us to recruit a more representative sample with minimal selection bias; this is particularly important in a setting such as Sodo District where access to facility-based health care is low. Probable cases of SMI were referred to local PHC services and were then evaluated by nurses or health officers who had been trained in use of the World Health Organization mental health Gap Action Programme Intervention Guide (mhGAP-IG) to diagnose and treat people with psychosis or bipolar disorder (19). A trained psychiatric nurse confirmed the diagnoses using the semi-structured OPerational CRITeria for research (OPCRIT) interview (20).
People with confirmed SMI and their caregivers were then recruited into the study if they met the following criteria:
- Aged 18 years or older,
- Planning to continue living in the district for the next 12 months,
- Provided informed consent (evaluated by trained psychiatric nurses) or, if lacked capacity to consent, did not refuse and caregiver permission was obtained, and
- Able to understand Amharic.
Those diagnosed with a mental, neurological and substance use (MNS) disorder prior to the study were not excluded.
Sample size and power calculation
The sample size was powered for the primary objectives of the PRIME study: detection of a 20% reduction in symptom severity after 12 months, with 90% power, two-sided significance level of 5%, and an assumed attrition rate of 20% (21). The final analytic sample had 300 participants; this was the baseline psychosis cohort for the PRIME study.
The primary outcome of the current study was discrimination experienced by people with SMI and the primary exposure was residence. Potential confounders were clinical diagnosis, symptom severity, disability, alcohol use, social support, poverty, age, sex, marital status, and education level. All assessments were administered directly to participants. Most participants had caregivers present during interviews. The caregivers were able to contribute to the responses so that the assessor had the most complete information available to them.
Primary outcome: discrimination
Discrimination was assessed by lay interviewers using section 1 of the Discrimination and Stigma Scale (DISC-12), which asks about the frequency of negative experienced discrimination over the past year (22). Responses are based on a four-point Likert scale ranging from “not at all” to “a lot”. The DISC-12 has been validated (22) and adapted in numerous countries, including Nigeria and Kenya (9). Responses for two items, ‘unfair treatment in getting welfare benefits or disability pensions’ and ‘unfair treatment in the level of privacy’, were not collected due to lack of local face validity. After conducting exploratory factor analysis with pairwise polychoric correlation on the study dataset, the following two items were found to have low item-factor loading (<0.3) as well as being endorsed with low frequency (<5%): ‘Unfair treatment when getting help for physical health problems’ and ‘unfair treatment from mental health professionals’. This left 17 of the original 21 items loading onto one factor, indicating construct validity of the scale. The 17 DISC-12 items were, therefore, summed for a total score.
Primary exposure: residence
Residence was self-reported as either urban or rural neighbourhood.
Information on potential confounding variables was obtained from measures administered by (1) psychiatric nurses and (2) lay data collectors.
(1) Psychiatric nurse-administered measures
Clinical diagnosis. Categorised as affective psychosis (bipolar, schizoaffective, major depressive disorder with psychotic features and postpartum psychosis), or primary psychotic disorder (schizophrenia and other non-affective psychotic disorders) based on the OPCRIT.
Symptom severity. Symptom severity was determined by using the Brief Psychiatric Rating Scale - Expanded Version (BPRS-E) translated into Amharic (23). The BPRS-E is a 24-item tool with seven possible responses ranging from “absent” to “extremely severe”, based on self-reported concerns and clinician observation. The 24 items were summed for a total score. The BPRS-E has been used in Ethiopia previously.
(2) Lay data collector-administered measures
Disability. The 36-item World Health Organization Disability Assessment Schedule (WHODAS-2.0) questionnaire measures difficulties performing activities over the last 30 days due to all health problems (24). The complex scoring method was used to determine a total score, ranging from 0 to 100, where 0=no disability and 100=full disability (24). The WHODAS-2.0 has been validated in a neighbouring district (25).
Alcohol use. The Alcohol Use Disorders Identification Test (AUDIT) self-reported version measures alcohol use in the past 3 months (26). Total score was categorized as either no alcohol use problem (<8) or hazardous use (≥8).
Social support. The Oslo 3-item Social Support Scale (OSSS) asks about ease of getting practical help, number of close acquaintances, and level of concern from others (27). Responses are categorized as poor support (3-8), intermediate support (9-11), or strong support (12-14). The OSSS has been previously used in the district.
Poverty. Calculated based on indicators used in the 2011 Ethiopia Demographic and Health Survey (28). Using exploratory factor analysis with maximum likelihood estimation, the following were found to load onto one factor and were summed to form a poverty index: thatched (vs. corrugated iron) roof, non-improved toilet, no separate room for cooking, no electricity, unprotected water source, not possessing a radio or television, or mobile phone. The index was dichotomized at the median (4) to categorise households into higher vs. lower poverty status.
Age, sex, marital status, and education level. These variables were self-reported.
Data management and analysis were done using STATA version 15.1 (29). Categorical variables were summarized by frequency and percent, continuous variables by mean and 95% confidence intervals (CIs). Zero-inflated negative binomial regression models were used to test the relationship between residence and discrimination, including adjustments for the potential confounders listed above. The negative binomial distribution is appropriate for modelling over-dispersed count data. The distribution of discrimination was determined to be zero-inflated upon visual examination.
The potential confounders were determined a priori, as described above, and were included in the adjusted model regardless of statistical significance.
Coefficients are on a log scale; they have been exponentiated and presented as risk ratios for ease of interpretation. The risk ratio represents the increase in total discrimination score for a one-unit increase in explanatory variable.