Community-Based Health Financing and Child Stunting in Rural Rwanda

Abstract

Objectives. We analyzed the likelihood of rural children (aged 6–24 months) being stunted according to whether they were enrolled in Mutuelles, a community-based health-financing program providing health insurance to rural populations and granting them access to health care, including nutrition services.

Methods. We retrieved health facility data from the District Health System Strengthening Tool and calculated the percentage of rural health centers that provided nutrition-related services required by Mutuelles’ minimum service package. We used data from the 2010 Rwanda Demographic and Health Survey and performed multilevel logistic analysis to control for clustering effects and sociodemographic characteristics. The final sample was 1061 children.

Results. Among 384 rural health centers, more than 90% conducted nutrition-related campaigns and malnutrition screening for children. Regardless of poverty status, the risk of being stunted was significantly lower (odds ratio = 0.60; 95% credible interval = 0.41, 0.83) for Mutuelles enrollees. This finding was robust to various model specifications (adjusted for Mutuelles enrollment, poverty status, other variables) or estimation methods (fixed and random effects).

Conclusions. This study provides evidence of the effectiveness of Mutuelles in improving child nutrition status and supported the hypothesis about the role of Mutuelles in expanding medical and nutritional care coverage for children.

Undernutrition is a leading cause of child morbidity and mortality in low-income countries.¹ Lack of sustainable financial support is one of the major barriers to scaling up nutritional interventions in resource-poor settings.^2,3 Evidence suggests that the most effective nutritional interventions are those with regular financial support via integration into existing local health systems.^2–14 Community-based financing, an approach designed to address financial bottlenecks at both the demand and supply side of medical care, has been proposed to promote financial sustainability of nutrition (and other health-related) programs and equitable access to nutrition care.^2,3 With increasing global efforts invested in child nutrition, evidence of the effectiveness of the community-based financing approach in improving child nutrition status is urgently needed.

Existing studies on community-based health-financing programs, which have mainly focused on their impact toward achieving universal health coverage in developing countries, have found a positive association between the programs and individual-level medical care utilization or household-level financial risk protection in countries such as Burkina Faso, China, Ethiopia, India, Laos, Malawi, Mexico, and Rwanda.^15–23 To our knowledge, little research has been conducted to investigate the role of community-based health financing in improving the nutritional status of children in resource-limited settings. We aimed to fill the knowledge gap through a case study in rural Rwanda.

As an agricultural country in central and east Africa with a per capita gross domestic product of US $595,²⁴ Rwanda has been making remarkable progress in reducing child mortality in the past decade: the mortality among children younger than 5 years in Rwanda has declined from 152 per 1000 live births in 2005 to 76 in 2010.²⁵ Child stunting, however, continues to be a public concern: about 44% of children younger than 5 years were stunted in 2010.²⁵ Stunted children in Rwanda are susceptible to infectious diseases such as diarrhea, malaria, and acute respiratory infections,²⁵ and stunting contributes to approximately 50% of morbidity and mortality among children younger than 5 years in Rwanda.²⁶ Stunting in Rwanda is found to be associated with food shortage and insecurity, repeated illnesses and poor health care, a lack of knowledge on feeding, and inadequate hygiene and sanitation.²⁵

The government of Rwanda has adopted a public health approach to improve child nutritional status. The public health sector in Rwanda provides the majority of health services to its population, especially in rural areas. In recent years, various nutrition interventions, from preventive to curative, were integrated into the local public health systems as a part of the minimum service package (MSP) covered by Mutuelles, a community-based health insurance program providing health insurance to populations in rural and informal economies (Panel A, available as a supplement to the online version of this article at http://www.ajph.org).²⁷ The inclusion of nutrition-related services in the MSP of Mutuelles has sparked our research interest in understanding the relationship between Mutuelles enrollment and child nutritional status. The Mutuelles program has been found to be effective in promoting medical care utilization with financial risk protection among its enrollees.^19,28 A few studies have discussed the possible link between Mutuelles enrollment and health outcomes, but without supporting empirical evidence.^29,30 In this study, taking advantage of the 2010 Rwanda Demographic and Health Survey (RDHS 2010) and the 2010 District Health System Strengthening Tool (DHHST), we determined the availability of nutritional services in the MSP of Mutuelles across rural health centers in Rwanda and conducted the first empirical study at the individual level that quantifies the association between Mutuelles enrollment and the likelihood of being stunted for rural children.

METHODS

Our target population was rural children in Rwanda, who had a much higher prevalence of stunting than urban children in 2010 (47% vs 27%).²⁵ We focused on rural children aged 6 to 24 months for the following reasons. First, it has been suggested that stunting is hard to reverse for children older than 2 years, and the first 2 years of life is the “window of opportunity” for interventions against stunting.^2,31–34 Second, about 93% of rural children younger than 6 months had exclusive breastfeeding, and the prevalence of stunting for this age group was much lower (18%) than for the children aged 6 to 24 months (42%).²⁵

Data and Variables

Data sources.

We used data from the RDHS 2010 to conduct an individual-level analysis. The RDHS is a nationally representative, population-based survey conducted every 5 years to measure indicators of population health and nutrition, with a special emphasis on mothers and on children younger than 5 years. The RDHS also collects information on households’ and mothers’ sociodemographic characteristics, health insurance status, and utilization of health care. It has been widely used to provide national and regional evidence to policymakers in Rwanda.^27,35,36 The RDHS 2010 collected individual information through a 2-stage sampling process. Villages, or primary sampling units, were selected at the first stage. Households in the selected villages were chosen at the second stage.³⁷

We used the DHHST to study the availability of nutritional services included in the MSP across rural health centers in Rwanda. The DHHST is an ongoing Web-based database system built by the Rwandan Ministry of Health in 2009 for monitoring and strategic planning on strengthening health systems.³⁶ The DHHST requires rural health centers to report to the database on an annual basis. The data provide information on (1) medical services (including nutritional care), (2) capacity building (infrastructure and staffing), and (3) revenues and expenditures.

Sample size.

The RDHS 2010 measured the heights of children younger than 5 years from a randomly selected 50% subsample of households. Among the 3474 rural children younger than 5 years with a valid height measure, 1087 were aged 6 to 24 months. To identify the link between children’s Mutuelles membership and their nutritional status, we included only children who were either enrolled in the Mutuelles program or had no insurance. We excluded 23 rural children who reported other types of insurance. The final sample size was 1061 children: 838 enrolled in Mutuelles and 223 uninsured.

In the DHHST data, of 389 rural health centers in Rwanda, 384 reported service provision in 2010.

Outcome variables.

For individual-level analysis, we constructed a dichotomous variable to indicate whether a child in our study population was stunted, defined as height-for-age 2 standard deviations below the median of the international reference population recommended by the World Health Organization in 2006.³⁸

Facility-level analysis identified a list of services that were included in the MSP of the Mutuelles and available in rural health centers. We treated each variable as binary and assigned a value of 0 or 1. The DHHST surveys included 30 questions on nutrition services in 2010: 10 promotional services, 7 preventive services, and 13 curative services. Online Panel B (available as a supplement to the online version of this article at http://www.ajph.org) lists these nutrition services and how they were delivered by community health workers.

Exposure.

For individual-level analysis, we constructed a dichotomous variable to indicate whether a child was enrolled in Mutuelles or uninsured. According to the Mutuelles legislation enacted in 2008,³⁵ enrollees were entitled to access the nutritional care (listed in online Panel B) when the services were available in rural health centers.

Covariates.

Sociodemographic variables included a child’s gender, maternal characteristics (mother’s age, education, and height), and household wealth status. We constructed 2 dummy variables that accounted for a mother’s completion of primary school and age older than 30 years. Previous studies found that a mother’s height was significantly associated with her child’s stunting status.³⁹ Following Özaltin et al.,³⁹ we constructed a categorical variable to indicate the mother’s height (< 150.0 cm, 150.0–154.9 cm, 155.0–159.9 cm, or ≥ 160.0 cm).

The RDHS 2010 had a wealth quintile variable that summarized a household’s assets (e.g., motorcycle), housing construction (e.g., floor), water source, and sanitation. Because about 45% of Rwanda’s rural population lived below the national poverty line (defined as US $0.45 per day per adult) in 2010,²⁴ we regrouped the households into 2 groups: below the poverty line (households in the lowest and next-to-lowest wealth quintile) and above the poverty line (all others). We constructed a dummy variable to indicate whether a child was from a household living below the poverty line. Note that information about a household’s water and sanitation was reflected in its wealth quintile, and we did not construct separate variables to represent these factors. To determine whether Mutuelles enrollment was associated with the nutritional status of children living below the poverty line versus those living above the poverty line, we constructed an interaction variable between Mutuelles enrollment and poverty indicators.

To control for district-level heterogeneity of health systems at the individual-level analysis, we used the reported number of community health workers in the DHSST to construct a variable indicating the number of district-level community health workers per capita in the 27 rural districts. As described in online Panel B, community health workers play a key role in implementing nutritional services in rural areas.

Statistical Analysis

We used a multilevel logistic regression model in individual-level analysis to estimate the association between children’s Mutuelles status and their likelihood of being stunted, controlling for child’s gender, poverty status, maternal factors (age, education, and height), and district- and village-level clustering effects. The multilevel model enabled us to control for clustering effects and correct for standard errors in the higher levels so as to obtain coefficients with more efficiency.⁴⁰ Because of the process by which the RDHS sampled villages at the first stage, children from the same village could be more similar to each other than to children from other villages. As a result of decentralized health care delivery at the district level in Rwanda, children in the same district were more likely to share similarities to each other than to children of other districts.⁴¹ Ignoring these clustering effects could lead to underestimation of standard errors and assumption of statistical significance where it does not exist.⁴² To control for clustering effects, we included random effects at the village and the district level. More details on modeling are presented in the online Supplementary Methods (available as a supplement to the online version of this article at http://www.ajph.org).

We applied a standard 2-step procedure recommended for discrete-outcomes multilevel models and started with a first-order marginal quasi-likelihood approach to get crude estimates, followed by a Bayesian Markov chain Monte Carlo approach in the second stage to improve the approximations.^40,43 Details can be found in the online Supplementary Methods (see also Figures A and B, available as a supplement to the online version of this article at http://www.ajph.org). We report odd ratios and credible intervals. The credible intervals derived from the Markov chain Monte Carlo method indicate that, with 100 000 simulations, the true estimate will lie in the credible interval with a probability of 95% (Table A, available as a supplement to the online version of this article at http://www.ajph.org).^43,44 We used the Stata version 14 command runmlwin (StataCorp LP, College Station, TX) to perform the multilevel statistical analysis.^45–47

Because the inclusion of different covariates and the use of different estimation methods may alter the association between the exposure and the outcome variable, we examined the sensitivity of the results to model specifications. We estimated the main association between Mutuelles status and stunting (with or without covariates) using the logistic model with multilevel random effects (model 1 to model 3 in Table B, available as a supplement to the online version of this article at http://www.ajph.org). We then estimated the association between Mutuelles status and stunting among children in households below versus above the poverty line by adding the interaction variable between Mutuelles and poverty indicators (model 4 in Table B). We also tested the sensitivity of results to estimation methods by conducting an analysis with a fixed-effects model including the 27 district indicators in regression analysis. A fixed-effects model generates less biased estimates, whereas a random-effects model generates more efficient estimates.⁴⁸

RESULTS

In 2010, among the 384 rural health centers, most of the promotional and preventive services of the MSP were available (Table 1). For instance, about 94% of rural health centers conducted a nutrition campaign in their catchment areas, and more than 90% of rural health centers offered child growth monitoring and malnutrition screening and vitamin A supplements for children. Among the curative services available, only a few health centers offered inpatient nutritional rehabilitation (11%), but most of them had outpatient care (providing mebendazole, iron, ready-to-use therapeutic food, and vitamins) for children with malnutrition.

TABLE 1—

Percentage of Rural Health Centers (n = 384) Providing Nutritional Services: Rwanda, 2010

Service	%
Community-based promotional and educational services
Campaign on hygiene	96.8
Campaign on Integrated Management of Childhood Illness	78.0
Campaign on ITN	87.0
Campaign on malaria	98.0
Campaign on nutrition	93.9
Campaign on vaccine	97.9
Campaign on water and sanitation	85.0
Kitchen demonstration	48.0
Garden demonstration	68.0
Nutrition education in school	37.0
Community-based preventive care
Breastfeeding support	75.0
Children’s malnutrition screening	96.0
Feeding program in school	9.8
Growth monitoring	90.0
Nutrition monitoring in school	13.0
Oral rehydration salts for children	70.9
Vitamin A supplements for children	92.9
Curative care
Inpatient nutritional rehabilitation	11.9
Outpatient food support	89.0
Ordinograms for severe acute malnutrition	49.0
Protocol for severe acute malnutrition	56.0
Providing amoxicillin to children with malnutrition	38.0
Providing vaccine to children with malnutrition	83.0
Providing folic acid to children with malnutrition	46.0
Providing HIV care to children with malnutrition	76.0
Providing iron to children with malnutrition	48.0
Providing mebendazole to children with malnutrition	89.0
Providing TB care to children with malnutrition	54.0
Providing vitamin to children with malnutrition	86.2
Ready-to-use therapeutic foods as food support	84.0

Note. ITN = insecticide-treated bed net; TB = tuberculosis.

Summary statistics of the variables used in individual-level regression analysis revealed that of the 1061 children in the sample, approximately 42% were stunted and 79% were enrolled in Mutuelles, 50% were female, 51% lived in poverty, 42% had mothers older than 30 years, and 19% had mothers who completed primary school (Table C, available as a supplement to the online version of this article at http://www.ajph.org). The prevalence of stunting by children’s sociodemographic variables is presented in Table D (available as a supplement to the online version of this article at http://www.ajph.org), which shows that boys had higher prevalence of stunting than girls. Children living in poverty, or with mothers who had not completed primary school, had higher prevalence of stunting than children living above the poverty line or with mothers who had completed primary school. Compared with children with mothers in the lowest height group, children whose mothers were in the other height groups had a significantly lower mean prevalence of being stunted (Table D).

Table 2 gives the estimated odds ratios of Mutuelles enrollment derived from the multilevel random-effects logistic models controlling for covariates and clustering effects. Odds ratios for covariates are presented in Table E (available as a supplement to the online version of this article at http://www.ajph.org).

TABLE 2—

Association Between Children’s Mutuelles Enrollment and Risk of Being Stunted, From Multilevel Random-Effects Logistic Model: Rwanda, 2010

Model Specification	OR (95% CrI)
Model 1	0.56 (0.39, 0.76)
Model 2	0.58 (0.41, 0.79)
Model 3	0.60 (0.41, 0.83)
Model 4
Aged 6–24 mo, below the poverty line	0.61 (0.39, 0.95)
Aged 6–24 mo, above the poverty line	0.58 (0.33, 0.96)

Note. CrI = credible interval; OR =odds ratio. Model 1 shows the main association. Model 2 shows the main association after adjustment for poverty status. Model 3 shows the main association after adjustment for all covariates. Model 4 is the fully adjusted model.

The odds ratio of Mutuelles enrollment was 0.56 (95% credible interval [CrI] = 0.39, 0.76; model 1 in Table 2), which remained significant after adjustment for covariates. For instance, the odds ratio of Mutuelles enrollment was 0.58 after adjustment for poverty status (95% CrI = 0.41, 0.79; model 2 in Table 2). When all covariates were adjusted, the odds ratio of Mutuelles enrollment was 0.60 (95% CrI = 0.41, 0.83; model 3 in Table 2).

The inverse association between Mutuelles enrollment and the risk of being stunted remained unchanged regardless of a child’s poverty status. For children living below the poverty line, the odds ratio of Mutuelles enrollment was 0.61 (95% CrI = 0.39, 0.95); for those above the poverty line, it was 0.58 (95% CrI = 0.33, 0.96; model 4 in Table 2). Estimates for other covariates are presented in Table E.

Controlling for all covariates and clustering effects, we derived the probabilities of rural children being stunted (Figure 1). The probability of being stunted was 39% for Mutuelles enrollees, significantly lower than that for uninsured children (53%; Figure 1). Regardless of their poverty status, there was a significant difference between Mutuelles enrollees and uninsured children. After we controlled for covariates and clustering effects, the probability of being stunted was lowest among Mutuelles enrollees living above the poverty line (34%) and highest among uninsured children living below the poverty line (57%).

FIGURE 1—

Predicted Probability of Being Stunted, by Mutuelles Status, After Controlling for All Covariates and Multilevel Clustering Effects: Rwanda, 2010

Table 2 shows that the inverse association between Mutuelles enrollment and likelihood of being stunted was not sensitive to model specifications. When we applied different estimation methods, the inverse association remained the same; the estimated odds ratios of Mutuelles enrollment derived from the fixed-effects models were between 0.53 and 0.62 and statistically significant (Table F, available as a supplement to the online version of this article at http://www.ajph.org).

DISCUSSION

Using the most updated and comprehensive data and multilevel models, we have presented 2 salient findings in this study. First, most rural health centers in 2010 offered nutrition-related promotional and preventive services covered by the Mutuelles program. Second, an inverse association between Mutuelles enrollment and the likelihood of being stunted was found among rural children aged 6 to 24 months, regardless of their poverty status. The estimated odds of being stunted among the Mutuelles enrollees were significantly reduced, by around 39% to 44%, compared with the uninsured children. This finding was robust when we applied various model specifications or estimation methods.

The inverse association between Mutuelles enrollment and likelihood of being stunted among children aged 6 to 24 months is consistent with previous findings of Mutuelles’ effect on increasing medical care utilization and reducing households’ catastrophic health spending.^19,28 It is also consistent with studies in 54 developing countries in Africa and Southeast Asia on the window of opportunity (from conception to 24 months) for effectively reducing child stunting.^31–34 Our study shows a similar magnitude of reduction in risk of being stunted among Mutuelles enrollees aged 6 to 24 months, regardless of their poverty status, which confirms the importance of providing comprehensive medical and nutritional care to children in this critical age period.

Note that the association between the Mutuelles enrollment and reduced likelihood of being stunted was found in a context in which child and maternal health, especially child nutrition, has become a priority for the government of Rwanda, reflected by the provision of promotional and preventive nutrition care by most rural health centers in our study. In 2008, a law on Mutuelles defined the minimum medical and nutritional services for children younger than 5 years to be provided by rural health centers and made the risk-pooling funds of Mutuelles a regular funding source for rural health centers.³⁵ In the same year, pay-for-performance was scaled up at the national level to improve the quality of care provided by health facilities.²⁹ In 2009, the National Protocol on Management of Malnutrition was enacted in health facilities and the National Emergency Program to Eliminate Malnutrition (NEPEM) was launched to eliminate all forms of undernutrition in Rwanda, especially stunting, with a community-based nutritional intervention approach.⁴⁹ Children aged 6 to 24 months in the RDHS 2010 survey (born in 2008 or later) could enjoy the increased and improved nutrition care provided by rural health centers during their critical window of opportunity. Our analysis using the RDHS 2005, the year before all major health system reforms were implemented, found no association between Mutuelles enrollment and stunting for children aged 6 to 24 months (Table G, available as a supplement to the online version of this article at http://www.ajph.org). This suggests the necessity of strengthening rural health systems for any policy instruments to work.

Even with the aforementioned policy measures, the prevalence of stunting for rural children aged 6 to 24 months remained high in 2010 (42%). We posit the following plausible explanations. First, it takes time to roll out the interventions at the national level. In 2010, important policy instruments such as the NEPEM had just started its second stage of preventing chronic undernutrition. The RDHS 2010 data used in this study could not capture its full-scale effect. Second, Mutuelles could promote the utilization of care among its enrollees with reduced out-of-pocket payments; however, for households living under the poverty line (175 Rwandan Francs [RwF], or US $0.32, per adult per day), paying premiums (1000 RwF) and copayments (200 RwF) for each visit to a health center, as well as 10% of hospitalization costs, could still be a burden. The RDHS 2010 data showed that the Mutuelles enrollment rate was significantly lower in the lowest wealth quintile than in the highest (67% vs 86%). To ensure equitable access and financial contribution to health care for the poor, the Rwandan Ministry of Health launched a new policy in 2011 to subsidize the premiums and copayments (with funds from the government and external donors) for the population in the lowest income quartile.⁵⁰ We expect to see an improvement in nutrition outcomes for the poorest children after 2011. Third, reducing and preventing chronic undernutrition could not be achieved only with the increased nutritional care offered by the health sector. It requires a multisectoral approach, with emphasis on food, health, economic planning, education, agriculture, social support, and water and sanitation at multiple levels, which is a long-term task and may require a greater time commitment.

The study has 3 limitations. First, the outcome variable was an anthropometric indicator that we constructed using children’s height; it was not based on laboratory tests or clinical diagnosis. Rwanda has no nationally representative data on child stunting measured by clinical diagnosis. In developing countries, the anthropometric indicators have been commonly used to measure child nutritional status for trend analysis and international comparisons. Second, with an observational study, we cannot infer causal associations between Mutuelles enrollment and stunting; therefore, generalizability cannot be warranted. Our observational study provides baseline evidence for further studying the effectiveness of a community-based health-financing approach in reducing stunting. Third, although the RDHS 2010 has the most recent data for measuring child nutritional status in Rwanda, our analysis was not able to document the full-scale change in child nutritional status because of the lagged effects of rolling out national-level nutrition policies.

Despite these limitations, the present study has multiple strengths. First, it is the first empirical study to our knowledge that quantifies the association between Mutuelles enrollment and stunting for rural children in Rwanda aged 2 years or younger. Knowledge gained from rural Rwanda is necessary for evidence-based promotion of the approach. Second, taking advantage of the availability of the DHHST, this study is the first to our knowledge to provide detailed information about community-based nutrition services that were integrated into the MSP of Mutuelles and provided by rural health centers in Rwanda. Given that the community-based health-financing approach has been implemented in other sub-Saharan countries,^51–54 this information may be of great interest to policymakers there. Third, the study addressed methodological issues such as clustering effects at the different levels with multilevel models and yielded robust estimates using various model specifications and estimation methods.

This study contributes to knowledge of the role of the community-based financing approach to improving child nutritional status: by integrating a series of nutritional activities into a community-based health insurance program, enrolled children in their first 2 years of life had lower risk of being stunted than children without any insurance. Although the mechanism by which Mutuelles had an effect on child nutrition status is beyond the scope of the study, the observed evidence supports the hypothesis that Mutuelles’ financial support to both the supply and demand sides may lead to the expansion of medical and nutritional care coverage for children, and subsequently improve their nutritional status, especially those living below the poverty line. In 2013, the European Union announced a €30 million program to support Rwanda’s goal of eliminating undernutrition with a special focus on child stunting.⁵⁵ Future studies will focus on identifying the pathway of Mutuelles’ effects on child nutritional status and explore the sensitivity of child nutritional status to different combinations of nutritional services.