Impact of community-based health insurance on health services utilisation among vulnerable households in Amhara region, Ethiopia

Background: Ethiopia piloted community-based health insurance in 2011, and as of 2019, the programme was operating in 770 districts nationwide, covering approximately 7 million households. Enrolment in participating districts reached 50%, holding promise to achieve the goal of Universal Health Coverage in the country. Despite the government’s efforts to expand community-based health insurance to all districts, evidence is lacking on how enrolment in the programme nudges health seeking behaviour among the most vulnerable rural households. This study aims to examine the effect of community-based health insurance enrolment among the most vulnerable and extremely poor households participating in Ethiopia’s Productive Safety Net Programme on the utilisation of healthcare services in the Amhara region. Methods: Data for this study came from Amhara pilot integrated safety net programme baseline survey in Ethiopia and were collected between December 2018 and February 2019 from 5,398 households. We used propensity score matching method to estimate the impacts of enrolment in community-based health insurance on outpatient, maternal, and child preventive and curative healthcare services utilisation. Results: Results show that membership in community-based health insurance increases the probabilities of visiting health facilities for curative care in the past month by 8.2 percentage points (95% CI 5.3 to 11.1), seeking care from a health professional by 8.4 percentage points (95% CI 5.5 to 11.3), and visiting a health facility to seek any medical assistance for illness and check-ups in the past 12 months by 13.9 percentage points (95% CI 10.5 to 17.4). Insurance also increases the annual household per capita health facility visits by 0.84 (95% CI 0.64 to 1.04). However, we find no significant effects of community-based health insurance membership on utilisation of maternal and child healthcare services. Conclusions: Findings that community-based health insurance increased outpatient services utilisation implies that it could also contribute towards universal health coverage and health equity in rural and informal sectors. The absence of significant effects on maternal and child healthcare services may be due to the free availability of such services for everyone at the public health facilities, regardless of insurance membership. Outpatient services use among insured households is still not universal, and understanding of the barriers to use, including supply-side constraints, will help improve universal health coverage. In 2011, the government of Ethiopia piloted CBHI in 13 rural districts (covering about 1.6 million people) targeting rural households and people working in the informal sector. This was scaled up to 161 districts after three years of piloting [32]. As of 2019, the programme covered 7 million households residing in 770 districts throughout the country (i.e., 75% district coverage nationwide). CBHI is currently operating in all regions and Addis Ababa except Somali, Gambella, and Dire Dawa. In the programme districts, 50% of eligible households are currently enrolled and the programme has an 82% renewal rate [33]. Nevertheless, the national level enrolment is still below the target set by the government: 80% of household enrolment and 80% coverage of districts by 2020 [11]. Enrolment in CBHI is conducted voluntarily. The programme uses the core principles of risk-sharing, a community-based decision-making process, and community support. Enrolment is conducted at the household level and all rural households in the district, excluding those formally employed, can join the programme. The CBHI is a yearly contractual agreement with advance premium payments by the members, and all renewals and new member registrations are conducted for a period of up to three months every year. Currently, the programme has two member types – self-paying and indigent members. The regional and district governments jointly fund the enrolment premiums of indigent households such as the permanent direct support (PDS) clients in the productive safety net programme. For paying members, annual premiums are set based on household sizes. In 2019, the premiums were ETB 240 (USD 8.6) for 1 to 5 member households, ETB 290 (USD 10.4) for 6–7 member households, and ETB 340 (USD 12.2) for households with 8 or more members. The benefit package of CBHI programme includes all outpatient and inpatient services available in health centres, treatment for cancer, dialysis and organ transplant for renal failure, treatment of major trauma, intensive care unit, hip and knee replacement, and major burns [32]. Services sought at primary, general, and referral public hospitals are also covered following appropriate referral procedures [34]. All services must be sought from public healthcare facilities with contractual agreements with the district CBHI office. CBHI does not cover costs related to tooth implantation, eyeglasses for ophthalmic cases, cosmetic procedures [32], aesthetic surgery, infertility treatment, and organ transplants (except renal, heart, and bone marrow) [34]. Ethiopia’s government also enacted its flagship poverty-targeted social protection programme, the rural Productive Safety Net Programme (PSNP), in 2005. About 85% of the programme beneficiaries are required to work on labour-intensive Public Works (PW) for payments while the other 15%, called the Permanent Direct Support (PDS) clients, who lack labour to participate in public works, receive unconditional cash and/ or food transfers [35]. To integrate various social protection programmes, the government endorsed its National Social Protection Policy (NSPP) in 2014 and launched the National Social Protection Strategy (NSPS) in 2016. However, Hirvonen et al. [36] found limited linkages between these large-scale social protection programmes. The Integrated Safety Net Programme (ISNP) is designed to address this gap. This pilot project, with the technical support from the United Nations Children’s Fund (UNICEF) Ethiopia country office (ECO), aimed to reinforce the linkages between the PSNP and CBHI and leverage the impacts of PSNP to reduce poverty and improve the multidimensional well-being of PSNP-participating households. The efforts to integrate the social protection programmes also assume that increasing coverage of CBHI among PSNP-participating households will increase their health services utilisation and improve health outcomes. The ISNP was launched in 2019 [37]. This study used cross-sectional data from the ISNP impact evaluation baseline survey in Amhara region, Ethiopia [38]. The ISNP evaluation is being carried out in 4 rural districts of Amhara region, namely, Libo Kemkem and Dewa Chefa as treatment districts and Ebinat and Artuma Fursi as comparison districts. Households in treatment districts receive additional (‘plus’) interventions on top of PSNP cash transfers including facilitation to CBHI enrolment, nutrition information through behavioural change communication (BCC) sessions, and case management through social workers and community care coalitions, while those in comparison districts do not get these plus components. While the treatment districts were selected purposively based on the availabilities of CBHI in the district, UNICEF ECO nutrition interventions and linkages to other UNICEF interventions, districts’ accessibility and practicality for UNICEF ECO support, comparisons districts were selected based on their similarities with treatment districts in socio-demographic, health service supply, programme organization, culture/ ethnicity, and ecological characteristics. Thus, the treatment and their respective comparison districts are geographically close and similar culturally and economically. The trial is registered on November 5, 2018, in the Pan African Clinical Trial Registry with trial registration ID—PACTR201902876946874. More information about the overall ISNP evaluation and interventions can be found in the online Additional file 1: Appendix 1. However, in the current study, we do not examine programme impacts of the INSP, but rather we use the baseline data to examine the effects of CBHI on health services utilisation. The ISNP evaluation employed a mixed-method study approach. However, this study used the quantitative data generated through household, community, and health facility surveys. Households eligible for the survey include all PSNP-participating rural households in the four districts. The sample size was determined using power calculation based on estimates of baseline means and the expected impacts of indicators. The indicators included individuals’ health services utilisations during the last month, visiting or consultation of a health service provider in the last 4 months, enrolment in CBHI, child nutrition and preventive health indicators, and mothers receiving antenatal care from a skilled provider during the last pregnancy. For each indicator, the sample size was calculated to detect a desired change of delta (δ) with minimum power of 80% under the assumption of simple random sampling and zero non-response rate. Accordingly, a target sample size of 5,400 households was decided, of which 5,398 were interviewed. The household questionnaire was designed to capture a broad range of information both at the individual and household levels such as demographics, educational attainment, health status and utilisation, PSNP participation, asset ownership, food security, and dwelling characteristics. Questionnaire items were drawn from previously implemented questionnaires and validated measures, including from the Transfer Project and other surveys implemented in Ethiopia and Eastern Africa (see Online Additional file 1: Appendix 6 for details about the variables) [39]. Some sections draw directly from other standard surveys such as the Multiple Indicator Cluster Survey (MICS), and instruments were tested in Ethiopia during piloting of the questionnaire at data collection trainings and then adapted as needed. A proxy female respondent from each household (priority was given to the main woman of the household or caregivers of children) was interviewed. Enumerators used electronic tablets installed with programmed survey (Survey Solutions) tools to input data and interviews were administered face-to-face in local languages (Amharic in Libo Kemkem and Ebinat districts and Afan Oromo in Dewa Chefa and Artuma Fursi districts). Baseline data collection was conducted between December 2018 and February 2019. For the community surveys (one per kebele (village) – the lowest administrative level in Ethiopia), community leaders and knowledgeable individuals in each sector were interviewed. Health care workers or facility administrators were interviewed for the health facility surveys on the facility characteristics/ infrastructure, personnel, and supplies. Data were also collected from official logbooks in all government health care facilities in study communities. The CBHI enrolment is the treatment variable. It is defined as holding a currently valid or renewed CBHI card, which is determined at the household level (i.e., once a household enrols all members of that household are automatically enrolled, except for additional fees required for adult children). Households were coded 1 if they were currently enrolled in CBHI and 0 if they were not enrolled. Outcomes of interest included primary preventive health services (child received all vaccinations (BCG, three doses of Polio vaccine, three doses of Pentavalent vaccine, and Measles) and mother received at least four ANC services and PNC visits in the past 12 months for children, children sleeping under long-lasting ITBNs, delivery at a health facility, births attended by skilled professionals, and children given deworming in the past 6 months). Child curative services considered in the study included health facility visits to seek treatment for child illness last month and any health facility visits for children in the last 12 months). We also considered outpatient services by members including any facility visits for curative services for illness in the past one month and if they also sought curative cares from health professionals. Data were collected on members’ facility visits to seek medical assistance for an illness and check-ups from health facilities in the past 12 months, and, if yes, the number of visits to a health facility for illness by all members in the household in the past 12 months. We excluded behaviours related to seeking medications over the counter and alternative care services from our analyses. Since CBHI enrolment is at the household level, we aggregated all outcomes at the household level. Accordingly, for outcomes observed at the individual level (adult members and under-five children), we consider the household as a service user if at least one member utilized the service. Covariate selection for the propensity score matching analysis was guided by the principles that: 1) omission of important variables could seriously increase bias in estimates [40, 41], 2) only those variables that simultaneously influence participation decision and the outcome should be included [42], and 3) selected covariates should not be affected by participation decision, that is variables should either be time-invariant or measured before participation took place [42]. Accordingly, we used previous studies, economic theory, and study context to select covariates. Household head-related factors included sex, age, current marital status, disability, and literacy status. The household-level factors were wealth status, number of household members by age, access to improved water during winter (the dry season in Ethiopia), whether the household worried about food in the last 4 weeks, number of food insecurity months in the last 12 months, having outstanding debt, drought in the last 12 months, total annual income received from PSNP, number of ill household members in the last month, and indices on perceptions and understandings about CBHI generated using Factor Analysis (see Online Additional file 1: Appendix 6 for details). The study also controlled for community and health facility-related characteristics including distance from the village to the nearest health centre (kilometres), distance from the village to the nearest health facility with a doctor (kilometres), whether the nearest health facility admits people covered with CBHI, number of years the village has been in PSNP, and village distance from district capital (kilometres). Estimations also included district fixed effects. The study was approved by the Amhara Public Health Institute (APHI) Research Ethics Review Committee (Reference Number HRTT—03/192/2018). Enumerators received instruction during data collection training about ethical data collection, informed consent, and referral services and procedures. Informed consent was obtained from all survey participants to use their anonymised information. This study does not involve patients. An inception workshop was conducted to select treatment districts using several social and economic indicators. Findings from the baseline data collection were disseminated in a consultative workshop conducted in August 2019 with the Amhara region and district administrators, Amhara Public Health Institute (APHI) experts, UNICEF Ethiopia staff, and stakeholders from district Bureau of Health (BoH), Bureau of Labour and Social Affairs (BoLSA) and Bureau of Women and Children Affairs (BoWCA), and district CBHI and PSNP coordinators. We first describe the characteristics of the target population by applying the sampling weights in the descriptive analyses. Individual-level data were aggregated at the household level. All data processing and analyses were conducted using STATA software version 15.1. To examine the impacts of CBHI enrolment on utilisation of healthcare services, we used propensity score matching (PSM) [43, 44], to account for selection into CBHI based on observable covariates, and then estimate the effect of enrolment on outpatient, maternal, and child healthcare services utilisation. PSM allows us to construct a comparison group that comprises PSNP participating households but did not join the CBHI programme (non-treated) but with the same probability of participating in CBHI as their enrolled counterparts (treated) based on observable and controlled characteristics. The attainment of PSM’s fundamental assumptions (conditional independence assumption (CIA) or unconfoundedness, and common support) are key to reducing bias arising from observed differences between groups. Accordingly, for CIA to be met, the factors associated with CBHI enrolment among PSNP households and those factors affecting outcomes related to CBHI must be observed, i.e., the selection is solely based on observable characteristics. Further, the common support or overlap assumption also requires that households with the same characteristics (X) have a positive probability of being in both arms and have the same probability of participation between 0 and 1, such that (0 < P(T = 1|X) < 1) [42, 45]. We first calculate the average treatment effect (ATE), at the population level constituting differences between the treated and non-treated groups as E[YiT- Yi C] [46]. Next, following Smith and Todd [44] and Caliendo and Kopeinig [42] and given the above assumptions, we estimate the average treatment effect on the treated (ATT) as follows. where ATT is the average treatment effect on the treated for outcome Y (mean difference in outcomes between groups over the common support weighted by propensity scores), and T and C denote CBHI enrolled and non-enrolled households. P(X) is the probability of CBHI enrolment given the set of observable covariates X. Both ATE and ATT are calculated using Stata’s Treatment Effects command. In PSM analyses, we employed a nearest neighbour algorithm with replacement (described in more detail in Additional file 1: Appendix 2). We further performed a sensitivity analysis to examine the robustness of estimates to hidden bias, described in more detail in Additional file 1: Appendix 2.