Context of the study setting
Sodo district is in the Gurage zone, Southern Nations, Nationalities and Peoples’ Region of Ethiopia, about 100 km south of the capital city, Addis Ababa. Sodo is a rural district organized in 58 villages (or ‘kebeles’) with a total population of about 162,000 . It was the setting for an implementation research project (the Programme for Improving Mental health carE; PRIME) which evaluated the impact of integrating care for priority mental disorders into primary healthcare . As part of PRIME, a participatory process was undertaken with stakeholders to develop, implement and evaluate a district level mental health care plan (MHCP) for people with depression, psychosis, epilepsy and alcohol use disorders . Linked to PRIME, the Emerald project (Emerging mental health systems in low-and middle-income countries) aimed to investigate health system strengthening required to support implementation of the plan .
The mental health care plan (MHCP) intervention packages
The MHCP intervention packages have been described previously [21, 22]. In brief, the interventions were based on integration of mental health at three levels of the health system: the district health care organization (health system level), the health facility, and the community . The aim of the community-level MHCP intervention packages was to improve access to care and social inclusion through community awareness-raising and stigma reduction, community case detection, support to continue engagement in care, and community-based rehabilitation. For this purpose, health extension workers, members of the health development army (a network of health volunteers), faith and traditional healers and community leaders were trained. The intervention was delivered through workshops and awareness-raising events. Building on their routine activities, the health extension workers were trained to provide outreach and adherence support to people with mental disorders. Community leaders and key informants were trained in case detection. The community advisory board brought together community leaders, multi-sectoral representation (e.g. police, education, non-governmental organisations), religious healers and caregivers/people with mental illness. The board played an important role in awareness-raising. Information leaflets were disseminated to households in the community to increase awareness.
At the health care facility level, the intervention packages focused primarily on building the capacity of clinical staff to detect and treat mental disorders using evidence-based guidelines. For this task, sensitization of all staff and training of clinicians was conducted for two weeks aimed at case detection, prescription of psychotropic medications, provision of basic psychosocial care, referral and ongoing care using the WHO mhGAP-Intervention Guide (mhGAP-IG) . At the district organisation level, the intervention packages included sensitization, advocacy, resource allocation and monitoring and supervision to ensure ownership and sustainability. There were no direct interventions targeting improvements in economic status or catastrophic OOP health expenditure for households with persons with SMD and depression.
Study design and participants
A community-based, controlled before-after study design was used to assess changes in economic outcomes and catastrophic OOP healthcare payments at the household level over 12 months. Two household samples were recruited, each with its own comparison group: the SMD sub-study and the depression sub-study. In the SMD sub-study, 290 households containing a member with SMD and 289 comparison households without a person with SMD were enrolled at baseline between January and August 2015. A follow-up interview was conducted during January and August 2016 (Figure 1). For the depression sub-study, the baseline interviews took place between March and November 2015 and enrolled 129 households which included a person with depression and 129 households without depression. The follow-up survey was conducted between March and November 2016 (Figure 2).
The recruitment methods and data collection procedures for the two sub-studies have been previously described . Briefly, the sample for the SMD sub-study comprised households of community-ascertained people with possible SMD who attended the local health centre for treatment and were confirmed to have SMD (schizophrenia or other primary psychotic disorder or affective psychotic disorder) by psychiatric nurses using a semi-structured clinical interview (Operational Criteria for Research, OPCRIT) . A census register of all households in the study area, developed by PRIME , was used as a sampling frame to select a comparison group of households. The comparison household was matched to the household of a person with SMD based on respondent characteristics (household head vs. other position in household), age (+/-5 years), gender, gott (residential unit within the village) and household size. If there was more than one match for a case in a gott we used a lottery method for selection.
Sample two (for the depression sub-study) comprised households of people attending the health centre who were identified by primary care staff as either having a probable diagnosis of depression or who screened positive on the Patient Health Questionnaire, nine item version (PHQ-9) and were thought to require treatment . The primary care workers had a clinical guideline (mhGAP) to assist them with their assessment . The PHQ-9 has been validated in Ethiopia in primary care attendees in health centres in a district neighbouring the location of the current study . The culturally validated cut-off to indicate probable depression is a PHQ-9 score of 5 or more. The control sample was drawn among people who attended the health centers on the same day as the person with depression but who did not have a primary care worker diagnosis of depression and who had a PHQ score <5, matched to the participant with depression by gender, age (±5 years) and gott.
Inclusion criteria for households with a person with SMD or depression were: age 18 years and older, household included person identified by the psychiatric nurse or PHC worker as having SMD or depression, planning to stay resident in the district for the subsequent 12 months, and provided informed consent. The comparison households were included based on the same criteria that were used to select the cases but with no family member with a suspected or confirmed mental health problem.
The sample size for the SMD sub-study was powered to detect a difference in household income level based on a South African study  with alpha = 0.05 and a power of 80%. Using the two-sample test of the mean sample size formula and allowing for loss-to follow-up, the required sample size was 300 per group. For the depression sub-study, based on a study from India  which found catastrophic expenditures were 14.6% and 4.9% for households that had members with depression and those that did not, respectively, to detect a risk ratio of 2.97 the resulting sample size was estimated to be 147 per group.
Primary outcome variables: the primary outcomes were change in economic status (income and consumption expenditure) and catastrophic OOP health care payments between enrollment (T1) and 12 months follow-up (T2)
Primary explanatory variables: Mental health status within the household (i.e. including a person living with SMD or depression vs. matched control households without affected persons)
Data collection and instruments
Household socioeconomic data were collected using an adapted and abbreviated version of the World Health Organization SAGE (Study of global AGEing and adult health) survey instrument, previously used in a study on health and ageing in six LMICs . The SAGE instrument contains information on a variety of individual and household socio-economic attributes such as consumption expenditure, income, assets, outstanding debts, household demographics, employment, health conditions and household coping strategies when experiencing financial difficulty.
Disability was assessed using the 36-item fully structured interviewer administered version of the World Health Organization Disability Assessment Schedule second version (WHODAS–II) . Total WHODAS-II polytomous summary score ranged from zero to 100, with higher numbers indicating greater impairment of day-to-day functioning. The Amharic version of this instrument was validated for people with SMD in Ethiopia previously . The Brief Psychiatric Rating Scale- Expanded (BPRS-E) was used to assess symptom severity in people with SMD. The BPRS-E is a 24-item observer-rated symptom scale covering four domains of symptoms of SMD (positive symptoms, negative symptoms, anxiety and depressive symptoms, and manic excitement or disorganization) and gives an overall indication of clinical symptom severity [32,33] and can detect improvement in response to an intervention .
All instruments were translated into Amharic and pilot tested in a district neighbouring the location of the current study area before use. Household Interviews were conducted by trained data collectors and supervisors within 2 to 4 weeks of screening and recruitment by PRIME. The interview was administered to the head of the household. In the absence of the head of the household the most knowledgeable person on household finance was interviewed. Repeat contact attempts were made for up to three visits to household participants who were unavailable. Completed questionnaires were checked for completeness and consistency. Identified incomplete responses and errors were sent back to the field for verification before data entry.
Income is the value of household agriculture and livestock output, wages, rental property, trade, savings and grants, transfers from families, community groups, government and from other sources for a different time periods (either daily, weekly, monthly or annually). Income from different sources were summed and converted to their annual equivalents and adjusted for household size and composition using a standard (equivalence) scale, the modified OECD scale. This information helps to establish household members are equalized or made equivalent by weighting each according to their age and household position. Accordingly, the scale scores 1 to the first adult, 0.5 to the second and each subsequent person aged 14 years and over and 0.3 to each child aged under 14 years .
Household consumption expenditure
Our measure of consumption expenditure was consumption of food produced by the household or purchased in the market-place or given in kind to the household and consumption of non-food items for daily use, consumption of consumer durables, consumption of health care goods, consumption related to transfers out to the community. The survey collected the monetary value of 36 food items consumed in the last seven days and consumption expenditure of 34 non-food items in the past month or year, depending on the item. All consumption expenditures were then converted to their annual equivalents in terms of Ethiopian Birr and in per adult equivalent terms .
Catastrophic OOP health care payments
In the literature, a number of possible thresholds to define catastrophic expenditure have been proposed [6, 36 - 39]. The most widely used is expenditure that amounts to 10 percent or more of total household expenditure, with the rationale that this represents an approximate threshold at which the household is forced to sacrifice other basic needs [40, 41]. Other studies have used the cutoff value of 40 percent or more of non-food expenditure [42, 43]. In this study, catastrophic OOP payments for health care were measured as the percentage of households incurring health payments in excess of 10% of total household consumption expenditure and ≥40% of a household’s capacity to pay (i.e. non-food expenditures) over one year . We broke down the consumption aggregate into food and non-food consumption expenditures to estimate the household's capacity to pay.
Age of the head of the household, sex, education, residential location (rural/urban), severity scores of mental health symptoms, disability scores and duration of treatment engagement were considered as potential confounding factors. Level of engagement with mental health care was measured by the number of contacts made during the 12-month follow-up period. This information was extracted from a follow-up registry.
We fitted two separate regression models to address two questions
1. Hypothesis-driven analysis that the change in economic status in households of people with SMD or depression will be greater than secular trends in the general population.
2. An exploratory analysis to examine baseline factors associated with change in economic status in households of people with SMD or depression (i.e. not including the comparison groups).
For the first question, the primary analysis of change in economic status (income, consumption expenditure) and catastrophic OOP health care expenditures were conducted using summary statistics. Chi squared test (χ2), Wilcoxon rank sum (Mann–Whitney U test), Wilcoxon signed-rank test, proportions and Student’s t statistics were used. However, before further analysis, propensity scores (PS) were estimated for each treated and comparison subject by means of a probit regression model including potential confounders (age, gender, education, residence and household size). The estimated score was then used to match each household of a person with SMD or depression with comparison households (with no affected person) using a kernel matching estimator . An important precursor to ensure the quality of matches is to impose what is known as “the common support condition” . The common support is the overlapping region of the propensity score for the two groups to be compared (supplementary figures 1 and 2). The regression adjusted estimates were used to identify the independent effects of the district MHCP on household income and consumption expenditure for households of persons with SMD or depression versus comparison households.
For the second question we used Ordinal Least Square (OLS) regression estimates to identify factors associated with changes in income or consumption expenditure in households of people with SMD or depression. The regression analyses were adjusted for household demographic, economic and clinical characteristics. Regression analyses were preceded by normality, multicollinearity and omitted variable bias tests . Variance inflation factor tests confirmed that multicollinearity was minimal (all variance inflation factors < 3.0). All data were analyzed using STATA software, version 14.1 (STATA Institute Inc.) . Results were considered statistically significant at p < 0.05.