Assessing the impact of community-based health insurance on health service utilization and out-of-pocket payments in Dangila Wereda, Awi zone, Ethiopia

Community-based health insurance (CBHI) schemes have gained significant attention as a viable approach to improve healthcare access and reduce financial barriers in low- and middle-income countries. CBHI involves voluntary enrollment and the pooling of financial resources at the community level to provide health coverage, particularly targeting the informal sector and rural populations. In Ethiopia, where out-of-pocket (OOP) payments constitute a major portion of healthcare financing, CBHI presents a promising strategy to enhance health service utilization and financial protection¹.

In Africa, the implementation of CBHI schemes has met with varying degrees of success. For instance, in Rwanda, one of the early adopters of CBHI, the scheme has been shown to increase healthcare utilization and reduce financial barriers to accessing care². However, challenges such as low enrollment rates, sustainability issues, and regional disparities in impact persist. Evaluations in Ghana have similarly highlighted both the benefits and limitations of CBHI schemes, emphasizing the need for continuous monitoring and adaptation to local contexts³.

Ethiopia, a low-income country with a predominantly rural population, faces significant healthcare financing challenges. The health sector in Ethiopia is largely funded through OOP payments, which pose a substantial financial burden on households and limit access to essential health services^4,5. To address these challenges, the Ethiopian government has integrated CBHI into its health sector reform strategy, aiming to achieve universal health coverage (UHC)⁶. By 2020, CBHI schemes were operational in numerous woredas (districts) across the country, targeting vulnerable populations, particularly in rural areas⁷. The Ethiopian CBHI initiative aims to provide financial protection, enhance health service utilization, and improve health outcomes for rural and informal sector populations^8,9.

Despite the theoretical benefits of CBHI, empirical evidence on its actual impact is limited and varies across different regions. In Dangila Wereda, the introduction of CBHI is relatively recent, and there is a lack of comprehensive data on its impact on healthcare behaviors and financial outcomes. Key questions remain unanswered: Has the CBHI scheme in Dangila Wereda led to increased utilization of health services? Has it effectively reduced OOP payments for healthcare? Addressing these questions is critical for informing policy decisions and improving the implementation of CBHI schemes^8,10.

CBHI has demonstrated significant positive impacts on healthcare access and financial protection in various contexts. For instance, a study in Senegal indicated that CBHI members were more likely to utilize health services and less likely to experience financial hardship due to health expenses¹¹. However, the success and impact of CBHI schemes can vary significantly based on local implementation strategies, community engagement, and the overall health system context. This highlights the need for localized studies to understand the specific dynamics and outcomes of CBHI in different regions.

Despite numerous studies on Community-Based Health Insurance (CBHI) in Ethiopia, few focused on quantitative analysis, and none thoroughly examined the impact of CBHI using statistical models. This study filled that gap by evaluating factors that influenced enrollment and assessing the scheme’s effect on health service utilization and out-of-pocket payments.

Therefore, this study aimed to identify the factors that determine the demand for enrollment in the CBHI scheme and examined its impact on health service utilization and out-of-pocket payments of households.

Study setting

The study was conducted in Dangila Woreda, Ethiopia, a district in the Awi Zone covering 772.3 square kilometers, making it the fourth largest in the zone by area. Dangila is bordered by Mecha district (West Gojjam) to the east, Jawi district to the west, Fageta Lekoma (Adis Kidam) district to the south, and Achefer district (West Gojjam) to the northeast. The capital, Dangila town, is 38 km from Enjebara, the Awi zonal town, 78 km from Bahir Dar, the capital of the Amhara Region, and 475 km northwest of Addis Ababa, the capital of Ethiopia. The district consists of 29 rural kebeles and 6 urban kebeles.

Dangila Woreda was selected for this study due to its relatively high CBHI enrollment rates and mix of urban and rural communities, providing a diverse sample for assessing program effectiveness. The district has a well-structured CBHI scheme that has been operational for several years, allowing for an in-depth evaluation of its impact on healthcare access and financial protection. Additionally, Dangila Woreda is known for its community engagement in public health programs, making it an appropriate setting for analyzing the factors influencing CBHI participation and its broader implications for healthcare utilization and out-of-pocket expenditures.

Study design

This study employs a cross-sectional design to identify factors determining a demand for enrolling at community-based health insurance schemes and the impacts of community-based health insurance schemes on health service utilization and out-of-pocket payments of households. The cross-sectional approach allows for the collection of data at a single point in time and is suitable for identifying relationships between socio-economic, demographic, environmental, and health-related variables.

Study population

The target population of this research included all households in Dangila Wereda, both participants and non-participants of the Community-Based Health Insurance (CBHI) scheme. The study included households from both urban and rural settings within the zone.

Sampling technique

The main objective of designing the sampling strategy was to collect data that is representative of the population. In this study cluster sampling method would be adopted as an appropriate sampling method. The cluster sampling method is an appropriate sampling design for selecting a representative sample of households. It is a technique that attempts to restrict the possible samples by taking some parts of the population represented in the sample in order to increase the cost efficiency. The clusters are Kebeles in Wereda and the final sampling units are households in selected Kebeles.

Sample size determination

In the planning of a sample survey to investigate the impact of Community Based Health Insurance (CBHI) on health service utilization and out-of-pocket payments in Dangila Wereda, Awi Zone, careful consideration was given to determining an appropriate sample size. Following¹², the sample size was calculated using formula (1), n = \(\:\frac{{{z}_{\frac{\alpha\:}{2}}}^{2}\:\left(pq\right)}{{d}^{2}\:}\), where d: Level of margin or precision error? (d = 5%=0.05), α: Level of significance or probability of making a certain error (α = 0.05), P: Proportion of households enrolled to CBHI (P = 0.455), Z: The inverse of the standard normal cumulative distribution (\(\:{z}_{\frac{\alpha\:}{2}}=1.96\)). The initial computation yielded a sample size of 381.04, rounded up to 419 to account for potential non-response.

Given the adoption of cluster sampling for efficiency, 20% of the kebeles (local administrative units) in the study area were selected. By cluster sampling principles outlined by¹², this led to the selection of 7 kebeles. The sample allocation to each kebele was determined proportionally based on the total number of households, ensuring a representative distribution of the overall sample size across the population. The resulting sample sizes per kebele are as follows: Alefa (49 households), Gudi (61 households), Gisa (80 households), Sehara (56 households), Kansen (67 households), Jibana (64 households), and Ginjama (42 households), summing up to the total sample size of 419 households.

Data collection

This cross-sectional study utilized both primary and secondary data. Secondary data from the Dangila Wereda CBHI Coordinating Office included enrollment records, claims data, and financial reports. Primary data was collected through household surveys in seven kebeles using a structured questionnaire, developed based on a literature review and expert input. The questionnaire was pretested on a small sample, refined for clarity, and assessed for reliability using Cronbach’s alpha, while validity was ensured through expert review and comparison with secondary data. To enhance accuracy, primary and secondary data were triangulated to verify CBHI participation trends, healthcare utilization, and financial burden.

Variables included in the current investigation

This study examines two dependent variables: the demand for enrolling in Community Based Health Insurance (CBHI) and specific outcomes related to participation, including health service utilization and financial protection against out-of-pocket payments. Independent variables encompass a range of factors identified through literature review and past research. Socio-demographic variables such as age, sex, marital status, household headship, family size, educational attainment, and place of residence are considered influential in CBHI enrollment. Economic factors such as income levels, health expenditures, and insurance premiums are also examined. Health-related variables including perceived health status, availability and quality of healthcare facilities, and public health infrastructure are integral to understanding enrollment behaviors. Additional variables such as geographical distance to healthcare services, levels of trust in the CBHI scheme, and community awareness and perception of the insurance program are also analyzed for their impact. This comprehensive approach aims to provide insights into the multifaceted determinants of CBHI participation, informing strategies to enhance healthcare access and financial security for participants.

About the model

This study employs two statistical models to evaluate the impact of community-based health insurance (CBHI) on health service utilization and out-of-pocket payments. The Classical Logistic Regression Model is used to predict enrollment in CBHI based on various predictor variables, accommodating both dichotomous and continuous independent variables. This model is well-suited for analyzing binary outcomes and identifying factors influencing CBHI enrollment. Complementarily, the Propensity Score Matching (PSM) Model addresses potential biases from non-random selection into CBHI by matching insured and non-insured individuals based on their propensity scores, derived from logistic regression on observed covariates. This approach creates a control group statistically similar to the treatment group, enhancing the accuracy and validity of the program’s impact assessment. Together, these models provide a comprehensive analysis of CBHI’s effects on health outcomes and financial burdens.

Binary logistic regression

Logistic Regression is useful to predict an outcome or dependent variable from a set of predictor variables. It is also useful when some or all of the independent variables are dichotomous; others can be continuous¹³. Binary Logistic Regression is used when the dependent variable is dichotomous, meaning it can take the value 1 (success, such as enrolling in CBHI) or 0 (failure, such as not enrolling in CBHI). The Logistic Regression Model is mathematically represented as follows:

Given data on a binary outcome variable \(\:Y\) and a set of explanatory variables \(\:{X}_{1},{X}_{2},\ldots,{X}_{k},\) the model estimates the probability \(\:Pi\) of a positive outcome:

\(\:Pi\:=\:Pr(Mi=1/X=xi),\) i.e.

Where,

• \(\:Pi\) ​is the probability that the i^th household enrolls in CBHI,

• \(\:Mi\) is the observed enrollment status (1 if enrolled, 0 otherwise),

• \(\:\beta\:\) is a vector of unknown coefficients.

The logit transformation of \(\:Pi\) ​provides a linear relationship between the predictor and response variables:

The coefficient \(\:\beta\:\) can be interpreted as the change in the log odds corresponding to a one-unit change in the respective continuous independent variable.

Assumptions of the logistic regression model:

1. 1.

   The dependent variable is dichotomous.

2. 2.

   Error terms are independent.

3. 3.

   The relationship between the independents and the log odds of the dependent is linear.

4. 4.

   Absence of perfect multicollinearity.

5. 5.

   The sample size must be large.

Odds ratios

The odds ratio is the exponential of the parameter estimate\(\:\:\beta\:\), denoted as\(\:\:\:{e}^{\beta\:}\). This represents the predicted change in odds for a one-unit increase in the predictor

Fitting the logistic regression model

The parameters \(\:\beta\:\) are estimated using Maximum Likelihood Estimation (MLE) rather than least squares, as MLE provides estimators with desirable statistical properties for logistic models¹⁴.

Assessment of the fit of the model

Goodness of fit is assessed through various statistical tests, including the Hosmer-Lemeshow Test, Pearson’s Chi-square test, the Wald test, and the Likelihood Ratio Test (LRT)^13,14. The LRT compares the fit of a full model with a reduced model and is defined as.

The Hosmer-Lemeshow Test evaluates the fit by comparing observed and expected frequencies across deciles of predicted probabilities.

Propensity score matching (PSM) model

Propensity Score Matching (PSM) is utilized in this study to assess the impact of Community-Based Health Insurance (CBHI) on health service utilization and out-of-pocket payments. Given that CBHI enrollment is not randomly assigned, biases may stem from factors such as self-selection, economic status, health conditions, and household characteristics^15,16. This method matches insured and non-insured individuals based on their propensity scores, derived from logistic regression on observed covariates, to create comparable groups for evaluating the effects of CBHI more accurately¹⁷. The aim is to mitigate selection biases and provide robust estimates of the program’s impact on health outcomes and financial burdens.

Constructing the counterfactual

To create a credible counterfactual, the treatment group (insured members) is compared to a control group (non-insured members) with similar characteristics. PSM pairs insured members with non-insured members based on their propensity scores, which are estimated using a logit model. The logit model used to estimate the propensity score is given by:

Where:

• \(\:Pi\) is the probability that the i^th household enrolls in CBHI,

• \(\:\beta\:\) are the coefficients,

• \(\:Xi\) are the independent variables affecting CBHI enrollment.

Propensity scores are used to match insured and non-insured households, ensuring common support and comparability. The impact is measured by the mean difference in outcomes between the two groups:

Where \(\:ATT\:\:\)the Average Treatment Effect on the Treated is, \(\:Yi\left(1\right)\) is the outcome for insured households, and \(\:Yi\left(0\right)\) is the outcome for non-insured households^17,18.