Tubaramure, a Food-Assisted Integrated Health and Nutrition Program, Reduces Child Wasting in Burundi: A Cluster-Randomized Controlled Intervention Trial

Background: Little is known about the impact of food-assisted maternal and child health programs (FA-MCHN) on child wasting. Objectives: We assessed the impact of Tubaramure, a FA-MCHN program in Burundi, on child (0 to 24 months) wasting and the differential impacts by socio-economic characteristics and age. The program targeted women and their children during the first 1000 days and included 1) food rations, 2) strengthening and promotion of use of health services, and 3) behavior change communication (BCC). Methods: We conducted a 4-arm, cluster-randomized, controlled trial (2010-2012). Clusters were defined as “collines”(communities). Impact was estimated using repeated cross-sectional data (n = ∼2620 children in each round). Treatment arms received household and individual (mother or child in the first 1000 days) food rations (corn-soy blend and micronutrient-fortified vegetable oil) from pregnancy to 24 months (T24 arm), from pregnancy to 18 months (T18), or from birth to 24 months (TNFP). All beneficiaries received the same BCC for the first 1000 days. The control arm received no rations or BCC. Results: Wasting (weight-for-length Z-score <2 SD) increased from baseline to follow-up in the control group (from 6.5% to 8%), but Tubaramure had a significant (P < 0.05) protective effect on wasting [treatment arms combined, -3.3 percentage points (pp); T18, -4.5 pp] and on the weight-for-length z-score (treatment arms combined, +0.15; T24, +0.20; T18, +0.17). The effects were limited to children whose mother and household head had no education, and who lived in the poorest households. The largest effect was found in children 6 to 12 months of age: the group with the highest wasting prevalence. Conclusions: FA-MCHN programs in highly food-insecure regions can protect the most disadvantaged children from wasting. These findings are particularly relevant in the context of the economic crisis due to the coronavirus disease 2019 pandemic, which is expected to dramatically increase child wasting. Tubaramure was implemented in Burundi's eastern provinces of Cankuzo and Ruyigi. Burundi is among the poorest countries in the world (16). It ranks 184 (out of 188) on the Human Development Index, and its population has very low levels of formal education (17). The country's population has endured genocides and a long history of political unrest. About half of Burundi's national budget came from foreign aid at the time our trial was conducted (18). From 2010 to 2012, inflation of consumer prices increased from 6.4% to 18.0% (19). In 2010, before the Tubaramure intervention started, around half of the study households suffered from hunger (11), and 60% of children 6 to 24 months of age in our study population and 28% of their mothers were found to be anemic (13). The prevalence of stunting in children 24 to 42 months of age at baseline was 65% (14). Tubaramure was implemented by a consortium of nongovernmental organizations (Food for the Hungry, International Medical Corps, and Caritas Burundi) led by Catholic Relief Services. The program was funded by the US Agency for International Development's Office of Food for Peace. Women at or after the fourth month of gestation and mothers of children less than 6 months of age were eligible to enroll in the program. Women needed proof from a health center–conducted pregnancy test and the pregnancy needed to show. Tubaramure included 3 core components. The food component consisted of a monthly family and individual ration of corn-soy blend (CSB) and micronutrient-fortified vegetable oil (Supplemental Tables 1 and 2). The objective of the family ration was to increase household food security in terms of both quantity and quality; the objective of the individual ration, targeted at pregnant women and women up to 6 months postpartum and at children starting at 6 months of age, was to improve maternal and child nutrition. The child ration (3 kg of CSB and 300 g of oil per month) provided 458 kcal/d. Tubaramure's second component focused on health and sought to improve the provision of preventive health services and to increase the use of these services by pregnant and lactating women and by children 0 to 23.9 months of age. The program's third component, the BCC strategy, was designed to promote adequate health, hygiene, and nutrition behaviors and practices. The BCC was implemented by program staff, locally hired Tubaramure health promoters, and leader mothers who were selected as teachers by their fellow beneficiary mothers. Tubaramure beneficiary mothers were expected to participate in the BCC sessions and to attend preventive health services, but neither expectation was enforced as a condition for receiving the food rations. Further details on the program and its impacts on a range of outcomes have been described elsewhere (11–14), and are also provided in the Supplemental Text. The program evaluation used a cluster-randomized, controlled design whereby 60 collines (a colline or hill is an official administrative unit in Burundi) meeting the study criteria were randomly assigned to 1 of 4 study arms. Collines were eligible if they fell between the first and 99th percentiles for colline population size and were not served by >1 health center each. A cluster design was used since individual randomization of the Tubaramure program components was not feasible. Prior to randomization, 210 collines meeting the study criteria were grouped into strata based on population size. The number of strata in each province (5 in Cankuzo, 10 in Ruyigi) reflected the relative population size. Each stratum had 13 or 14 collines in Cankuzo and 14 or 15 collines in Ruyigi. Using random numbers with a fixed random number seed (Stata version 11, StataCorp 2009) (20), 4 collines were randomly drawn from each stratum, for a total of 60 collines. At a public lottery event with representatives from both provinces, the 4 collines in each stratum were randomly assigned to the 4 different study arms: 3 treatment arms and 1 control arm. The study included 3 treatment arms to assess the differential effects of varying the timing (starting during pregnancy or at birth) and duration (full 1000 days or shorter) of receiving food rations. The rationale for studying these was that providing food rations for the full 1000 days period is expensive; if a similar impact is obtained when providing rations for a shorter time period, more beneficiaries can be covered with the same resources. Beneficiaries in the T24 arm received the standard program: that is, all program benefits during pregnancy and up to the age of 23.9 months for the child. Beneficiaries in the T18 arm received the same benefits, but food rations ended at the age of 17.9 months for the child. In the TNFP arm (no food during pregnancy), food rations started only at birth and were provided to the mother for the first 6 months and to the child between 6 and 23.9 months of age. Collines assigned to the control arm did not receive any Tubaramure benefits, but had access to the standard health-care services provided by the Ministry of Health. Since health centers were used by various collines, the health service intervention component was not limited to the treatment arms. We conducted 3 repeated cross-sectional surveys: a baseline survey in 2010 (before the program started) and 2 follow-up surveys in 2012 and 2014 (Table 1). Each follow-up survey was conducted to assess a specific set of outcomes. Outcomes such as anemia and wasting were best measured when children were still eligible to receive program benefits: that is, when they were between 0 to 23.9 months of age. Thus, these outcomes were assessed using data from the baseline surveys (conducted before the program started) and the 2012 follow-up survey (when children 0 to 23.9 months of age were eligible to participate in the ongoing program). The full effect on child linear growth—the main outcome of the study—was expected in children who had been exposed to Tubaramure from early pregnancy to when they reached 23.9 months of age. Impacts on child linear growth were therefore assessed among children 24 months and older, using data from the baseline survey and the 2014 follow-up survey (when the program had ended and children were between 24 and 41.9 months of age) (14). The first follow-up thus included households with children 0 to 23.9 months of age, the second follow-up included households with children 24 to 41.9 months of age, and the baseline survey included both types of households. Survey waves and Tubaramure program implementation The 2010 and 2012 data on children 0 to 23.9 months of age were used in the analyses presented in this manuscript. Shortly after the baseline survey was completed, eligible families were invited to enroll in Tubaramure, and beneficiaries started receiving program benefits. The study in children 0 to 23.9 months of age was powered to detect a program effect (difference between treatment and control) on child anemia (1 of the study’s primary outcomes) in children of 11 percentage points (pp) in the T24 and TNFP groups and of 8.25 pp in the T18 group using a Type 1 error (α) of 0.05 (1-sided), a power of 0.90, an intracluster correlation coefficient of 0.006, and 15 clusters per treatment arm (13). The differences in expected effect size thus resulted in different sample sizes across arms. Using the same parameters and a baseline prevalence of wasting of 7%, we calculated that the study’s sample size allowed us to detect a reduction in the prevalence of wasting of 4 pp and a change in the mean of weight-for-length z-score (WLZ) of 0.24 when comparing the T24 or TNFP arm to the control group. For the T18 to control comparison, the detectable differences for wasting and WLZ were 4 pp and 0.21, respectively. Since all statistical models controlled for covariates, the actual minimum detectable differences were smaller. The International Food Policy Research Institute’s Institutional Review Board and the Ministry of Health of Burundi approved the study. Written informed consent for participation in the study was obtained before the start of each interview. This trial was registered at clinicaltrials.gov as {“type”:”clinical-trial”,”attrs”:{“text”:”NCT01072279″,”term_id”:”NCT01072279″}}NCT01072279. At the start of the 2010 and 2012 surveys, a household census was conducted in all 60 research collines to generate a complete list of households with a child aged 0 to 23.9 months. Using a probability proportional to size approach, we calculated the target sample size for each colline. Colline-specific lists of randomly ordered households to be surveyed were then generated. Households were visited in the order listed until the required sample size in each colline was reached. As our objective was to estimate the intent-to-treat effect, inclusion in the survey was solely based on having a child in the appropriate age group and not on actual program participation. If there was more than 1 child in this age group, 1 “index child” was randomly chosen using the alphabetic order of the children’s first names. A total of 2625 and 2612 households with a child aged 0 to 23.9 months were surveyed at baseline and follow-up, respectively (Figure 1). Trial flow chart. Abbreviations: T18, treatment arm from pregnancy to 18 months; T24, treatment arm from pregnancy to 24 months; TNFP, treatment arm from birth to 24 months. The survey team was trained extensively, which included classroom teaching, field exercises, and repeated testing to assess skill acquisition. Each field team was composed of 4 enumerators, 2 anthropometrists, and 1 team controller. The enumerators used a household questionnaire to collect data on a wide range of variables, including socio-demographic characteristics (such as education, literacy, and household asset ownership) and program participation. Nurses were trained [and standardized (21)] to collect anthropometric data. Length and height were collected using Shorr boards (Weight and Measure). Measurements were taken twice, and were taken a third time if the difference between the first 2 measurements exceeded 6 mm. The 2 closest measurements were used in the analyses. Weight data were collected using a Seca 874 digital scale (Seca), which allowed the weight of the child to be taken when the mother held the child. WLZ was calculated using the WHO 2006 growth standard (22). Wasting was defined as WLZ <2SD. In line with the CONSORT 2010 guidelines, no formal comparison of baseline means between the treatment and intervention arm was conducted (23). The impact of Tubaramure was estimated using a double-difference colline–fixed effect model, which estimates changes over time in the treatment group relative to the control group. This model was used: Here, Tj is time (baseline or follow-up), Si is the assigned study arm (T24, T18, TNFP, or control), and C is a vector representing the colline-level fixed effects. The coefficient β3 represents the estimated treatment effect. Colline-level fixed effects were used to control for unobserved time-invariant colline characteristics at baseline and follow-up. To reduce residual noise and thus maximize power, covariates (Xi) were added to the model. These included maternal and child age, child sex, maternal height, whether the primary caregiver was the biological mother (and the interaction between this variable and maternal height), the education levels of the mother and the head of household, dependency ratio, and household assets. In line with statistical theory, we used 1-sided tests given the a priori hypothesis that the program would lead to improvements in nutritional status (24). In previous analyses, we found that the impact of the program on linear growth faltering was limited to children growing up in wealthier households, to children with literate mothers, and to children with better-schooled parents. These differences were not due to differences in program enrollment or participation in program activities (14). To assess whether the impact on wasting varied by socio-economic characteristics, we estimated the impact models separately by level of maternal education (none vs. some), maternal literacy (illiterate vs. literate), education of the head of household (none vs. some), and household asset ownership (below or above median number of assets, a proxy for socio-economic status). A similar approach was used to evaluate whether the impact differed by child age: separate models were estimated for children 0 to 5.9 months, 6 to 11.9 months, 12 to 17.9 months, and 18 to 23.9 months of age. All treatment arms were pooled in the subgroup analyses. The SEs of all estimated parameters were adjusted for colline-level clustering by using a clustered (Huber-White) sandwich estimator. A P value of 0.05 was considered significant. Analyses were conducted using Stata 16 (StataCorp, version 16) (20). All analyses pertain to the individual level. No clusters were dropped from the analyses. Fewer than 3% of individual observations were excluded from the analysis because of missing or incomplete data (Figure 1).