This analysis presents treatment coverage for acute malnutrition using representative cluster-based population survey methods in fourteen administrative areas of five high-burden countries. Coverage of treatment for acute malnutrition was universally below global standards9 but variable across and within countries. The point estimates of coverage suggest that fewer than 1 in 5 eligible SAM children were enrolled in treatment in eleven administrative areas. These results were obtained in the areas where treatment was available prior to the survey. In Chad and Somalia, one administrative area had significantly lower SAM coverage than the other areas assessed in the same country.
Given low coverage, we aimed to characterize individual, household, and population level characteristics associated with enrollment in treatment. Little evidence exists on child-level factors that may be correlated with coverage. Our findings suggest no significant differences in SAM or MAM coverage by child age in the populations assessed. We also did not detect sex differences in SAM or MAM coverage, despite evidence to suggest sex differences in prevalence. 28,29 We did identify differences in MUAC across covered and non-covered populations in three countries; however, the direction of the associations was inconsistent. While the data from DRC and Niger suggested that enrolled cases had lower MUAC than non-enrolled cases, in Somalia, our results showed the opposite. As the time to presentation is a key determinant of successful treatment in the CMAM model, this is an interesting finding that should be monitored with routine program data.30 It is possible that in some contexts, more visible signs of malnourishment (i.e,. lower MUAC) may increase the urgency to take the child to be screened or stay enrolled in the program.
At the household level, we found a significant association between individual treatment status and distance to the health facility for SAM treatment coverage in Chad and for MAM treatment coverage in Somalia. Here, the relationship was as expected – significantly higher percentage of covered SAM cases (Chad) or MAM cases (Somalia) lived within the 1 hour traveling distance to the treatment facility compared to non-covered cases. In all other assessed countries, this relationship was non-significant. Despite evidence that improving geographic access to health facilities increases the use of maternal and child health services,31,32 and distance to the facility is a well-established primary barrier to CMAM coverage,33 little routine work is done to map locations to CMAM sites and understand the spatial distribution of coverage.16 Many CMAM programs track treatment availability (the number of facilities offering services) and effectiveness (cure rates), without understanding accessibility (the population who can reasonably use the service).34 Targets ensuring geographic access to care are rarely, if ever, set.
An association between individual treatment status and caregiver involvement in screening was observed in one country (Chad). To improve early detection and referral of acutely malnourished children, the Family MUAC approach trains caregivers to screen children at home on a regular basis, as opposed to waiting for screening by a community-based volunteer or at the facility. In Chad, a significantly higher percentage of covered children had a caregiver who was previously trained in the Family MUAC approach, compared to non-covered children. While the cross-sectional and observational nature of our study prevents us from making causal inferences, this finding is in line with promising but limited peer-reviewed evidence and operational findings on the approach’s effectiveness.35,36
At the population level, SAM treatment coverage was unrelated to SAM prevalence. Despite their effectiveness, CMAM programs are regularly underfunded and not integrated into the national healthcare systems. Where resources are limited, we would assume areas with higher prevalence might be prioritized for services.37 However, this was not observed in our study, or it may be that there is a disconnect in supply through treatment provision and demand detected in enrollment and retention in the program. Overall, the factors that drive wasting may also be associated with lower coverage, which was supported by our study.
Because we used novel methods to assess coverage, we compared our findings to previous estimates derived using conventional methods and secondary data sources. We first compared our findings to estimates derived using administrative data (Supplemental File 2). Our data is cross-sectional, directly measuring the percentage of eligible children enrolled at one point in time, whereas estimates derived from programmatic data reflect enrollment over a given period (i.e., monthly or annually). Administrative estimates of coverage were available in Burkina Faso, Chad, and Niger (Supplemental Tables 2.1.1, 2.1.2, and 2.1.3). In Burkina Faso, administrative estimates from 2020 were somewhat similar to our findings (Supplemental Table 2.1.1). This was the exception, as administrative coverage estimates from 2021 in Burkina Faso and in other countries ranged from much lower to much higher than our findings, sometimes implausibly exceeding 100%. This may be due to outdated population and/or prevalence estimates. For example, in Niger, coverage estimates were clearly implausible (ranging from 188–235%) (Supplemental Table 2.1.3). It is important to note that in this context, prevalence figures for combined SAM (all anthropometric criteria) were not available- only those disaggregated as SAM by MUAC/ edema, and SAM by WHZ. While this likely contributed to an underestimation of expected SAM admissions, the direction and magnitude of the discrepancy varied by country and year.
In Chad, we obtained monthly admission data from 2019 and 2020. Acute malnutrition prevalence is driven by seasonality, climate, and conflict crises, and if enrollment does not increase during the peaks, coverage will decrease when the burden surges.38 Our findings in Chad suggest the population estimates for prevalence or population are not representative of some months of the year, and/or coverage fluctuated by month. Monthly coverage derived from administrative data varied from thirty to sixty-eight percentage points in the same year (Supplemental Table 2.1.2). In 2020, the median monthly coverage was higher than our findings in two districts and lower in one district. In 2019, the median monthly coverage was higher than our findings in all three districts.
As administrative coverage estimates leverage routinely collected data, they also rely on the accuracy and precision of this data.16 We use the general incidence correction factor 1.6, whereas recent work suggests this may be an underestimate in many contexts.39,40,41 Ultimately, our findings suggest that coverage estimates derived using administrative data are unreliable. The direct measurements from our data help assess the accuracy of coverage estimates from administrative data where they are available.
We also compared our findings to SQUEAC and SLEAC, the two most frequently used methods for direct assessment of acute malnutrition coverage. Both are based on data collected through active and adaptive case finding, which is then adjusted by a Bayesian model that relies on numerous assumptions and estimated parameters. SQUEAC produces coverage estimates with an associated 95% confidence interval, whereas SLEAC classifies treatment coverage as low, medium, or high based on contextualized thresholds.
Coverage estimates produced by SQUEAC or SLEAC were only available for Chad and Niger and were quite outdated (from 2013–2016) (Supplemental File 2, Tables 2.2.1 and 2.2.2). These prior estimates from SQUEAC and SLEAC (except for one in Mangalme, Chad, in 2015) were considerably higher than the coverage estimates obtained in our study. While it is possible that coverage has decreased over time in some areas, that seems an unlikely explanation for the differences, given active work in all countries to scale up coverage. 14 Our findings support the theoretical concern that the active and adaptive case finding at the core of the SQUEAC method carries a risk of upward bias.42 Our findings suggest a similar risk with SLEAC, though the broad classification thresholds make a comparison to our point coverage estimates difficult. In one case in Chad, SLEAC coverage classification was higher than our findings, whereas in the other, it was lower (2015). In Niger, the SLEAC coverage classification was higher than our findings in all cases (2014 and 2015).
Lastly, we compared our findings to practitioner expectations (Supplemental File 2, Table 2.3). Expected coverage estimates were estimated by IRC technical staff in each country office, informed by stakeholder consultation, and a review of factors that can boost or inhibit coverage, including the food security situation, partner presence prior to the project, availability of MAM treatment, insecurity/ physical accessibility, population displacement, and available coverage survey data. The specific approach to this exercise varied by office, but in all cases, it was not as extensive as the formulation of a prior per the SQUEAC and SLEAC methodologies.17 In 7 contexts, predicted coverage for SAM was within the 95% CI for measured coverage. In 7 other contexts, predicted coverage for SAM was higher than the 95% CI for measured coverage, indicating practitioners were overly optimistic regarding coverage. Predicted coverage for MAM in Somalia was higher than measured coverage in all 4 contexts. This suggests that practitioner expectations for coverage are often higher than reality.
In addition to presenting findings with respect to coverage, we demonstrated the feasibility of population-based methods themselves in a variety of contexts. In contexts where SAM prevalence was higher, data collection required five to seven days of fieldwork. In lower prevalence contexts14 to 20 days were required.
The key strength of this study is the direct measurement of acute malnutrition treatment coverage using gold-standard population-based methods. This approach avoids the imprecision and bias potentially introduced by pre-assessment assumptions, post-hoc modeling, or correction factors. These methods resulted in direct population representative estimates for administrative areas where coverage estimates were previously unavailable, outdated, and/or produced by methods with critical risks of bias. The exhaustive sampling of children within selected clusters allowed us to examine individual and household-level factors associated with coverage. We present novel findings on the association of prevalence and coverage using the same data collection method, and we also make comparisons to existing coverage estimates.
Our study has several limitations. First, our survey’s case-identification criteria, while aligned with the community-based MUAC and edema screening protocol in all five countries, did not account for children malnourished by weight-for-height z-score without MUAC deficiency and/or edema. Mid-upper arm circumference, weight-for-height z-scores, and edema identify overlapping but not identical populations, which varies by setting.43 All anthropometric criteria aim to identify the children most at risk of death due to undernutrition. MUAC cut-offs have been shown to effectively identify children at risk of death,44 but MUAC identifies a younger and more female treatment population than WHZ.45,46 Based on programmatic data from the implementing areas, 10.4 to 55.1% of all children admitted for treatment presented with low weight-for-height alone, and therefore, would not have been included in the denominator of our coverage assessments. It is likely that the coverage for SAM children malnourished by weight-for-height z-score and not MUAC and/or edema is even lower than identified than identified in our surveys and community referrals using MUAC and edema criterion, as these children are only detected as malnourished through passive screening at the health facility. Socio-demographic associations with coverage should also be interpreted with caution, as they may differ for children who are malnourished by weight-for-height z-score and not MUAC and/or edema.
Second, we measured a few covariates at the individual, household, and population level to explore in relation to coverage. Additional socio-demographic factors which may be associated with coverage, such as socioeconomic status, health indicators, and infant and young child feeding practices, were not included in the data collection tool. At the population level, we did not systematically collect information on programs in the area operated by other partners, to assess the relationships between programmatic interventions and coverage. Contextual information was, at times difficult to generalize across entire administrative areas, especially regarding screening regularity and supply availability, which can vary across facilities.
Finally, we conducted an additional coverage survey in the catchment area of two health facilities in Banadir, Somalia. Due to the difficulty of obtaining a sampling framework representative of the catchment areas in this urban context, we exclude results from this paper.