Ethiopia is the second most populous nation and the fastest growing economy in Africa, with an estimated population of 114.9 million in 2020 and an average economic growth rate of 9.8% between 2009 and 2019. Per capita income of Ethiopians is US$855 making the country one of the world’s low-income countries (27-29). More than three fourth of the population (78.3%) lives in rural areas with scattered population settlement patterns where subsistence farming and animal husbandry are the main sources of livelihood (30). The country is administratively divided into 10 regional states and two city administrations. Regions and cities are divided into woredas (equivalent to districts) which are further divided into Kebeles, which represent the lowest administrative structure serving an average of 5,000 population.
Health service delivery in Ethiopia is predominantly public. The public sector is organized in a three-tier healthcare delivery model that includes primary level care provided by primary healthcare units (composed of health centers and health posts) and primary hospitals, secondary level care provided by general hospitals, and tertiary level care provided by specialized hospitals (31). Healthcare is financed predominantly by external funding (35%), government expenditure (32%), and households’ out-of-pocket payments (31%). The role of health insurance has been insignificant until recently (15).
A comparative cross-sectional study design was conducted to assess the impact of CBHI on health services utilization and financial risk protection among CBHI member households in the regions of Tigray, Amhara, Oromia, and Southern Nations Nationalities and Peoples (SNNP), where CBHI has been implemented for the longest period in the country. Health service utilization and the incidences of catastrophic and impoverishing health expenditures were measured and compared between CBHI member households and non-members from two different settings: CBHI implementing and non-CBHI implementing districts.
Three groups of households were established for the household survey based on the implementation of CBHI at the district level and membership status at the household level. Our primary sampling unit was the enumeration area (EA), a geographically defined cluster of about 250 households, developed and maintained by the Central Statistical Agency of Ethiopia for census and survey purposes. The sampling unit for second stage sampling was the household. Heads of households and their spouses were respondents for the survey.
Study households were identified through a two-stage stratified sampling method. Enumeration areas in the four study regions were first stratified into CBHI implementing and non-CBHI implementing categories based on administrative records of the authority responsible for overseeing insurance schemes. These were further stratified by livelihood categories (urban and rural). Using probability proportional to size method of cluster sampling, we selected 96 EAs (81 rural and 15 urban) from the CBHI implementing stratum and 22 (11 rural and 11 urban) from the non-CBHI implementing stratum. For each selected EA, a complete list of households was constructed to generate a sampling frame of CBHI member and non-CBHI member households. Then, 36 households were randomly selected from each EA using computer-assisted random sampling technique. In CBHI implementing EAs, the sampling was stratified to include 18 member and 18 non-member households. In EAs where the total numbers of members or non-members was less than 18, all available households in that category were taken and the remaining portion of the sample size would be added to the other category so that 36 households are included in total. This resulted in 1586 CBHI members, 1863 non-CBHI members from CBHI-implementing districts, and 789 non-CBHI members from non-CBHI-implementing districts.
Health service utilization was measured in the forms of probability of modern healthcare seeking for the most recent episode of illness during the one-month period preceding the survey and per capita health facility visits during the last one month. Financial risk protection was measured using out-of-pocket health expenditure and the incidences of catastrophic health expenditure and impoverishment due to out-of-pocket health expenditure.
Probability of modern healthcare seeking at time of illness: The proportion of household members with at least one episode of illness during the one-month period preceding the survey who sought health service from a modern healthcare provider (public or private) for their most recent illness.
Per capita health facility visits: The average number of health facility visits that a household member had during the one-month period preceding the survey.
Out-of-pocket health expenditure: Payments made by a household at the point they receive health services.
Catastrophic health expenditure: A household is classified as having catastrophic health expenditure if it’s total OOP health payment equals or exceeds 10% of its total household expenditure.
Impoverishment due to out-of-pocket health spending: A non-poor household is impoverished by health payments when it becomes poor after paying for health services. Impoverishment due to health payments was measured in terms of absolute increases in poverty headcount, poverty gap, and normalized poverty gap after OOP health payments. The 2015/16 national poverty line (7184 birr per capita) (32) was adjusted for general food and non-food inflation based on consumer price indices reported by the Central Statistical Agency for the years 2015/16 to 2019/20. The resulting adjusted poverty line for 2019/20 was 10 053 birr per capita. This poverty line was used to calculate the poverty headcount and the poverty gap and determine the differences in poverty indices before and after health payments.
Data collection tools and data collectors
Data were collected by experienced and trained data collectors using standard questions adapted from related studies in the fields of health service utilization and household consumption surveys. All data collection tools were translated from English to the major local languages spoken in the study regions. Translated versions were back-translated into English for quality assurance purposes. All the tools were pre-tested in similar settings prior to data collection. In areas where languages other than the major local languages are spoken, mainly in SNNP, translators assisted data collectors.
Data were collected through household and market surveys. Household data was collected through face-to-face interviews with heads of households and their spouses and measurement of food items using weight scales. Data on household’s general characteristics including family size and composition, health service utilization patterns, and household food and non-food consumption and expenditure, including OOP health payments were collected through two rounds of household visits. Food consumption data was collected for a recall period of three days in the first round and four days in the second round.
Consumption of own products was valuated using data collected on local market price through local market surveys. Price data from the market survey was used in the consumption aggregate to determine the level of prices for various items in local markets in the study area and allowing for estimation of monetary values for items produced and consumed at home. Open markets, kiosks, groceries, butcheries, pharmacies, supermarkets, and other service establishments where households in the EAs purchase most of their goods and services for household consumption and other purposes were used as sources of data.
Household and market survey data were collected using the electronic Census and Survey Processing System (CSPro) with pre-designed data quality assurance features loaded on Android devices. The program included features such as assignment of data collection tools to data collection sites, listing of households in study EAs, random sampling of households within EAs, collection of household and market data, synchronization of data with supervisors, and online submission of data to a central server. The system was prepared with appropriate sequencing of questions and data validation rules to ensure the collection of valid and consistent data.
Data management and analysis
Data collected using CSPro were synchronized to supervisors’ accounts on a daily basis. Supervisors then submitted quality-checked records to a central server. A central data quality assurance team checked the quality of submitted data on a daily basis. Upon completion of data collection fieldwork, data submitted to the central server was compared with data on individual tablet computers to ensure complete synchronization. The final dataset was cleaned to remove duplicate records and exported to Stata version 16.0 for Windows for analysis. Consumption data collected from households was combined with price data collected through market surveys to create consumption aggregates. Individual-level data collected about members of households were aggregated to household-level variables.
The distribution of outcome variables (health service utilization and health expenditure), exposure variables (CBHI implementation status of woredas and membership of households), and covariates (socio-demographic and other characteristics of households) were summarized using descriptive statistics. The effect of CBHI was estimated by comparing outcomes between CBHI member households and the two categories of comparison groups: 1) non-member households from CBHI-implementing districts and 2) households from non-CBHI-implementing districts. Sampling weights, calculated as the inverse probability of the selection of sampling units, were used while analyzing household data.
We used propensity score matching to estimate the effect of CBHI by accounting for possibilities of bias arising from voluntary enrolment in CBHI. Propensity scores were estimated using the logit model. We used the nearest neighbor matching method, which matched each CBHI member to a comparison household with the closest propensity score. We ran two separate models to estimate the impact of CBHI membership. In model 1 and model 2, we used non-CBHI members from CBHI-implementing woredas non-CBHI implementing woredas, respectively. The average treatment effect on the treated (ATT), the effect of CBHI membership among CBHI members, was calculated as the average difference between matched pairs of households. The distribution of the propensity scores matched satisfactorily between CBHI members and non-members in both models.
Ethics approval and consent to participate
The study was conducted in accordance with applicable ethical standards. The study protocol was reviewed and approved by the Institutional Review Board of the Ethiopian Public Health Institute (Ref No: EPHI 613/624 dated 18 February 2020). Official permissions were obtained from relevant authorities at different levels. Verbal informed consent was obtained from all participants after they were provided adequate information about the study.