Community-Based Management of Acute Malnutrition (CMAM)

In a nutshell

Community-based management of acute malnutrition (CMAM) involves identifying and treating cases of acute malnutrition. This report focuses on donation opportunities to fund non-governmental organizations (NGOs) to support government-run CMAM programs.

We expect the cost-effectiveness of these grants to vary significantly by location and by program. Using the example of one NGO's program in Niger, we estimate that while treatment costs are high (~$70 per child), CMAM programs target children at a very high risk of death (~6% annual mortality rate) and substantially reduce the risk of death (~45%). In some cases, NGOs leverage their work to support additional health programs. This can lead to high cost-effectiveness in some locations.

We’re highly uncertain about our estimate of the impact of CMAM programs, compared to those of our top charities. That’s because:

  • We rely on historical data and indirect estimation methods, rather than direct evidence, to estimate mortality rates of malnourished children and the effect that the program has on mortality.
  • We're uncertain about the number of additional children who will be treated for malnutrition as a result of NGO support.

We're also unsure of how much of the unmet need for malnutrition treatment is above our cost-effectiveness bar for recommending grants.

We've recommended six grants to support NGOs working on CMAM programs. We may direct additional funding to organizations that implement CMAM programs in the future.

Published: March 2024
(Previous versions of this page: July 2021 version, May 2018 version, November 2016 version)

Table of Contents

Summary

Basics

Globally, 47 million children are estimated to have experienced moderate acute malnutrition (MAM) or severe acute malnutrition (SAM) in 2019. This is believed to raise their risk of developmental delays and death.

Community-based management of acute malnutrition (CMAM) identifies and treats cases of uncomplicated malnutrition primarily on an outpatient basis. Cases are often identified by caregivers, who screen children by measuring their mid-upper-arm circumference (MUAC). Cases may also be identified when health care workers measure childrens' weight and height, allowing for calculation of a weight-for-height Z score (WHZ), a measure of nutritional status. Treatment includes specific therapeutic foods and standard medications, including a short course of antibiotics for children with SAM. (More)

The NGOs that we have supported work in government-run health facilities. With GiveWell-directed funding, the NGOs provide training and support for government clinic staff, improve health facilities, provide medical supplies, and conduct community outreach, among other activities.(More) Some programs also support additional pediatric care, such as the provision of routine childhood vaccinations. (More)

How cost-effective is it?

We think community-based management of acute malnutrition can be highly cost-effective, though we expect the cost-effectiveness to vary significantly by program and by location.

Intuitively, we think this program can be highly cost-effective because:

  • Children with MAM and SAM are at greatly elevated risk of dying compared to non-malnourished children. In the locations we support, we estimate that children with untreated MAM are roughly 2.5 times more likely to die than children without malnutrition over a one-year period following measurement of their weight and height. We estimate that children with untreated SAM are roughly 7 times more likely to die than children without malnutrition. Our estimates are based on evidence from historical studies of mortality in children with untreated malnutrition. (More)

  • Children who are malnourished are significantly less likely to die if they receive treatment. We estimate that MAM treatment reduces the relative risk of mortality over the next year by about 40% and SAM treatment reduces it by about 70%. Our model incorporates inputs from a literature review of the impact of CMAM programs on weight-for-height Z scores, a measure of malnutrition status. (More)
  • NGOs' support for CMAM programs in government-run health facilities leads to more children being treated for malnutrition and higher quality care. We largely think this because there is an intuitively plausible story for how the activities they conduct lead to higher numbers of children receiving treatment and better quality care. For example, NGOs can help caregivers to check whether their child is malnourished by equipping them with a tool for measuring children's mid-upper-arm circumference and training them how to use it; it seems plausible to us that this drives additional cases of malnutrition being identified and treated.

    In the case of a program we supported the NGO ALIMA to conduct in Niger, we estimate that about 200,000 children will receive treatment supported by ALIMA's program. We expect that about 35% of these children would have otherwise gone untreated and that the remaining 65% receive higher quality treatment than they otherwise would have. This is based on ALIMA's estimates. (More)

  • Some NGOs also provide support for other pediatric care, such as vaccination and malaria treatment. Some of the NGOs we fund support government health facilities with a broader set of treatment and prevention programs. In the case of a program we supported ALIMA to conduct in Niger, we estimate that this work makes up approximately 30% of total benefits. (More)

  • Although malnutrition treatment is more expensive than many other interventions we support (for ALIMA's program in Niger, ~$70 per child treated per year, compared to ~$6 to $9 to provide a full year of seasonal malaria chemoprevention to one child), CMAM also reduces the risk of death more for each child treated (for ALIMA's program in Niger, by about 45%, compared to ~10% for SMC), which leads us to estimate high cost-effectiveness. (More)

Our cost-effectiveness analysis quantifies this intuition. Here is a sketch, using projections for a three-year program run by one NGO we've supported, ALIMA, in Dakoro and Mirriah departments, Niger, as an example. We report cost-effectiveness estimates as multiples of unconditional cash transfers (GiveWell’s benchmark for comparing different programs). This sketch represents our estimate as of April 2023.

What we are estimating Best guess (rounded) Confidence intervals (25th - 75th percentile) Implied cost-effectiveness
Grant size to charity $7,866,165
Total program cost (includes contributions from government and other philanthropic funders) $13,622,611
Cost per malnourished child reached (more) $67
Number of malnourished children reached 202,106 150,000 - 250,000 8x - 13x
Percent of children who would have received malnutrition treatment in the absence of the charity's program (more) 65% 80% - 40% 7x - 17x
Annual mortality rate from all causes among children 6-59 months with untreated malnutrition (more) 5.9% 3% - 9% 6x - 17x
Reduction in all-cause mortality from receiving NGO-supported malnutrition treatment, instead of no treatment (more) 45% 25% - 70% 7x - 16x
Increased reduction in all-cause mortality from receiving NGO-supported malnutrition treatment, instead of standard treatment (more) 2% 0% - 15% 10x - 16x
Total number of deaths averted among malnourished children 2,047
Initial cost-effectiveness estimate (malnutrition-related mortality benefits only)
Moral weight for each death averted 119
Initial cost-effectiveness estimate (malnutrition-related mortality benefits only) 9x
Summary of primary benefits (% of modeled benefits)
Reduced mortality among malnourished children (more) 66%
Income increases in later life (more) 1% 0% - 2% 11x
Vaccines provided to children under age two (more) 26% 15% - 35% 9x - 12x
Reduced mortality among children receiving malaria treatment (more) 6% 1% - 10% 10x - 11x
Additional adjustments
Adjustment for additional program benefits (e.g. pediatric care) and downsides (e.g. wastage of therapeutic food) (more) +13% -5% - +25% 9x - 12x
Adjustment for diverting other actors' spending into malnutrition treatment ("leverage") and away from malnutrition treatment ("funging") (more) -31% -45% - -25% 9x - 12x
Overall cost-effectiveness (multiples of cash transfers) 11x

This is one example, and we expect wide variations in cost-effectiveness among malnutrition treatment programs, primarily driven by:

  • Underlying population mortality rates (the higher the underlying mortality rate, the more cost-effective malnutrition treatment is likely to be);
  • Cost per child treated (the lower the cost per child treated, the more cost-effective malnutrition treatment is likely to be); and
  • Whether an NGO leverages its malnutrition work to support additional health programs (layering additional benefits on top of the same activities can drive up the cost-effectiveness of the program).

How could we be wrong?

Our analysis contains a number of open questions and uncertain assumptions. The most important of these are:

  • The mortality rates of untreated children with malnutrition. We do not have any direct evidence of the mortality rates of untreated children with malnutrition, nor the effect that malnutrition treatment has on mortality.

    We instead use data from several observational studies conducted in the 1980s and 1990s to estimate the mortality rates of children with untreated malnutrition and the impact of CMAM on mortality (see next bullet). Specifically, we use the observational data to estimate the ratio between the mortality rates of malnourished and non-malnourished children, and we then use that ratio (together with mortality rates and malnutrition prevalence) to estimate mortality rates of children with untreated malnutrition today. (More)

    In order to sense-check our estimates, we developed a ceiling analysis to estimate the maximum plausible mortality rates for malnourished children. This is based on overall mortality rates among 6-59 month-old children, the prevalence of untreated malnutrition, and an estimate of the maximum plausible percentage of deaths in this group that occur among malnourished children. In some contexts, the mortality rates resulting from our ceiling analysis are substantially lower than those we estimate in our main cost-effectiveness analysis. We're unsure what explains this discrepancy. (More)

    For ALIMA's malnutrition treatment program in Dakoro and Mirriah, Niger, our best guess of the 25th/75th percentile range for mortality rates of untreated children with malnutrition is 3% to 9%, which implies an overall cost-effectiveness of 6x to 17x.

  • The treatment effect of CMAM. To estimate the effect of malnutrition treatment, we use the observational data mentioned in the bullet above to estimate mortality rates associated with average WHZ before and after CMAM treatment, and we then calculate the ratio between the two. We use that ratio to estimate mortality rates in treated and untreated children today.

    We have a number of uncertainties about this method. Importantly, it assumes that the effect of malnutrition treatment on mortality is captured by the correlation between WHZ and mortality. This might not be true for a number of reasons. For example, the mortality risk associated with malnutrition could be partially a result of confounding by socioeconomic conditions. While we adjust our estimate for this concern, we are unsure about the size of the adjustment we should make. (More)

    For ALIMA's malnutrition treatment program in Dakoro and Mirriah, Niger, our 25th/75th percentile range for the effect of CMAM treatment on mortality is a 25% to 70% reduction in mortality, which implies a cost-effectiveness of 7x to 16x.

  • We're highly uncertain about the number of additional children who receive malnutrition treatment as a result of NGO support (more). To calculate this, we estimate the number of children that would receive malnutrition treatment with NGO support and without NGO support.
    • Our estimates of the number of children reached with NGO support are based on historical caseloads, where available, and population and incidence estimates where that information is not available. We then adjust these estimates to account for potential future trends. For example, if a program treats MAM cases, we make adjustments for our assumption that SAM cases will decrease as treatment of MAM cases increases because children with MAM will progress to SAM less frequently.
    • Our estimates of the number of children that would be reached without NGO support are based on NGOs' best guesses. We're funding some surveys to estimate malnutrition coverage before and after NGO support, and we expect this data will help us refine this input in the future.

    Our estimates of the number of additional children who will receive malnutrition treatment as a result of our funding are highly uncertain because:

    • Historical caseload is often not a good proxy for future caseload, since malnutrition incidence can change significantly year-to year.
    • Estimates of malnutrition incidence are highly uncertain.
    • We lack data on the number of children that would be reached without NGO support.

    For ALIMA's malnutrition treatment program in Dakoro and Mirriah, Niger, our best guess of the 25th/75th percentile range for the number of additional children who receive malnutrition treatment with ALIMA support is 150,000 to 250,000, which implies a cost-effectiveness of 8x to 13x. We expect to improve our caseload estimates as we get information from the NGOs we support about how many children they are treating.

  • We're also unsure of how many total funding opportunities for malnutrition treatment are above our funding bar. Our understanding is that there is likely to be substantial room for more funding for malnutrition treatment due to a large remaining global need. We're unsure what share of this would be above our funding bar.

What is the problem?

Acute malnutrition refers to excessive thinness for one’s height1 and/or the presence of nutritional edema (swelling caused by excess fluid retention in tissues)2 . It is associated with an increased risk of illness and death.3 Moderate acute malnutrition (MAM) is characterized by low weight-for-height or mid-upper arm circumference.4 Severe acute malnutrition (SAM) is characterized by very low weight-for-height, very low mid-upper arm circumference, and/or the presence of nutritional edema.5 If the underlying causes are left untreated, the condition of a child with MAM may become more severe and cross the threshold into SAM.6

Globally, estimates suggest that about 47 million children under five experienced acute malnutrition in 2019, with 14.3 million of those experiencing SAM.7 Estimates of the impact of malnutrition on mortality draw on historical observational studies, and for this reason among others, are very uncertain (see our discussion here). In addition, the World Health Organization and UNICEF argue that malnutrition can cause cognitive developmental delays that could impair functioning in adulthood among children who survive.8

What is the program?

Until recently, SAM was treated in a hospital setting.9 This approach is expensive and offers limited coverage in regions where hospitals are scarce. Community-based management of acute malnutrition (CMAM) is a strategy for identifying and treating uncomplicated cases of malnutrition without requiring hospitalization, reducing cost and increasing coverage.10

Standard treatment protocols include treatment of both MAM and SAM but recommend different treatment strategies for each (e.g., the protocols for Chad and Mali). Uncomplicated cases of SAM are treated with ready-to-use therapeutic food (RUTF), an all-in-one food product that is designed to provide malnourished children with the nutrients they need to recover,11 along with a course of antibiotics.12 The MAM component of CMAM protocols relies on the use of ready-to-use supplementary food (RUSF) or enriched flours.13

There are no globally accepted guidelines for treating MAM, and in areas of scarce resources, moderately malnourished children may not receive treatment at all.14 This raises the concern that without treatment, a portion of children with MAM will progress to SAM.15 Adding to the complexity for healthcare practitioners who encounter both types of malnutrition, UNICEF supports treatment of SAM with provision of RUTF, whereas the World Food Programme supports treatment of MAM with a different product, ready-to-use supplementary food (RUSF).16

Recently developed “combined” protocols for CMAM treat children with both SAM and MAM using the same food products.17 These combined protocols aim to expand coverage by reducing the complexity of implementation for healthcare workers, as well as reducing the dosage of supplemental foods given to each child, which reduces treatment costs per child.18

Recent and planned trials are testing slightly different combined protocols,19 but broadly these programs involve:

  • Identifying SAM and MAM in the community, often by using community health worker networks or by training caregivers (usually mothers) to screen children using colored plastic strips to measure mid-upper-arm circumference (MUAC), a marker of nutritional status.20
  • Assessing children who meet the criteria for SAM or MAM and referring children who are too sick for CMAM to hospital inpatient care.21
  • Providing a standardized set of medical treatments for SAM children, including antibiotics22 to reduce infections, speed up nutritional recovery, and prevent mortality.23
  • Providing ready-to-use therapeutic food (RUTF), a nonperishable, calorie- and micronutrient-dense food designed for treating malnutrition, until children meet criteria for discharge.24 Combined protocols provide RUTFs to both MAM and SAM cases. They generally provide children with SAM a lower dosage of RUTF than standard protocols in order to expand coverage while still providing enough nutrition for children to recover.25

There is substantial overlap in the evidence for standard and combined protocol CMAM. We have modeled both variants; our published cost-effectiveness analysis focuses on combined protocol CMAM.

Does the program have strong evidence of effectiveness?

Our primary outcome of interest is the impact of CMAM on all-cause mortality in children 6-59 months old with malnutrition, relative to no treatment. We have not found direct estimates of this outcome, since it is widely considered unethical to study children with malnutrition without providing treatment.26

We use historical observational data on the mortality rate of children with untreated malnutrition, relative to children without malnutrition, to estimate the mortality rate of children with untreated malnutrition and the impact of CMAM on mortality.27 Additional inputs into our cost-effectiveness calculations include current local all-cause mortality rates and the prevalence of malnutrition, as well as several adjustments to account for key limitations of these estimates.28

This estimation method has major limitations but suggests that MAM treatment reduces all-cause mortality by about 40% over the following year, and SAM treatment reduces all-cause mortality by about 70%.29 Paired with the program’s highly plausible mechanism of action (see below), we believe CMAM is very likely to avert child mortality, but we are uncertain about the size of the effect.

Typically, at the start of CMAM, children are also screened and treated for malaria, and provided standard preventative vaccines,30 which we believe contribute to averting deaths.

Mortality reduction

Biological plausibility

Malnutrition treatment has a highly plausible mechanism of action. Low body energy stores and nutritional deficiencies increase the risk of death from infectious diseases.31 RUTF addresses deficiencies of energy and essential nutrients, while antibiotics treat infections and may also work through less well-understood mechanisms.32 We also believe the standard care that is typically provided at initiation of CMAM, such as screening and treatment for malaria and administration of preventative vaccines,33 is likely to be beneficial. Overall, we have a strong prior that CMAM will avert deaths among malnourished children to some extent.

Evidence from historical studies of mortality in children with untreated malnutrition

For modeling purposes, a key variable determining the cost-effectiveness of CMAM is the mortality rate of children with untreated malnutrition. In other words, what happens to children with malnutrition in the absence of the intervention? Several studies have estimated the impact of untreated child malnutrition on mortality risk using data from observational cohorts.34 These have typically expressed the mortality risk of children with malnutrition as a ratio with the mortality risk of non-malnourished children (relative risk, hazard ratio, or odds ratio),35 and this is the method we use to generate a key input into our estimate of absolute mortality rates of children with untreated malnutrition.

The most recent study to estimate excess mortality of children with untreated malnutrition is Olofin et al. 2013. This study pools observational data from ten historical cohorts of children under five years old in low-income settings to estimate the excess risk of mortality associated with untreated MAM and SAM.36 It reports that relative to children with weight-for-height z-scores (WHZ) in the well-nourished range (WHZ greater than or equal to -1), the hazard ratio for death was 3.4 in children with MAM (WHZ between -2 and -3), and 11.6 in children with SAM (WHZ less than -3).37 These numbers represent multipliers on the risk of dying at a specific point in time.38 Our previous malnutrition models were based on figures from this paper.39

Given the importance of Olofin et al. 2013 to our cost-effectiveness model and our limited understanding of its methods, GiveWell asked senior advisor David Roodman to replicate it. David was able to access five of the data sets used in Olofin et al. 2013, and one additional data set that was not used in the study.40

This effort was able to approximately replicate the main findings of Olofin et al. 2013,41 but it also identified a major limitation of the Olofin estimates that make them inappropriate for estimating malnutrition mortality rates in our cost-effectiveness analysis. The limitation is that the hazard ratios generated by the Olofin method depend heavily on the amount of time between measurements in the original cohorts.42 This causes two related problems:

  • The hazard ratios reported in Olofin et al. 2013 represent the ratio of mortality rates at a specific, but unspecified, point in time after measuring WHZ.43 Since mortality ratios in this context depend heavily on how long one waits to measure mortality after measuring WHZ,44 the hazard ratios reported in Olofin et al. 2013 are challenging to interpret for our purposes. In addition, for our modeling purposes, ratios representing cumulative mortality over a one-year period are more helpful than ratios representing mortality at a specific point in time.45
  • How often children were measured for WHZ varies between cohorts, and this has a large impact on mortality ratios that was not controlled for in Olofin et al. 2013.46 Therefore, the mortality ratios reported in the paper are in part an artifact of how often the original investigators took measurements, rather than wholly reflecting underlying mortality rates.

To address this limitation and others, David used a different estimation method called mixture inverse Gaussian (MIG) modeling,47 following Aalen 1994. This method yields cumulative one-year mortality ratios that are easier to understand and use in our cost-effectiveness modeling.48 For example, a one-year mortality ratio of 5 for children with SAM vs. non-malnourished children would mean that over a one-year period after measuring WHZ, a child with SAM at baseline who does not receive treatment has a risk of dying that is five times that of a child who does not have malnutrition at baseline.

In addition, generating estimates in-house allows us to use relatively recent, country-specific WHZ inputs for MAM, SAM, and non-malnourished children. For Chad, Ethiopia, Mali, Niger, and Nigeria, the MIG model yields mortality ratios that are smaller than the hazard ratios reported by Olofin et al. 2013: roughly 2.5 for untreated MAM and roughly 7 for untreated SAM, compared to the 3.4 for untreated MAM and 11.6 for untreated SAM reported by Olofin et al. 2013.49 Although the underlying data are more limited,50 results are similar when classifying children according to mid-upper-arm circumference (MUAC).51 This implies that children with MAM and SAM are at greatly elevated risk of dying compared to non-malnourished children, but the risk is not as high as we previously estimated.

Although this work has reduced our uncertainty about the mortality rates of untreated malnourished children, we continue to have substantial uncertainty about the output of this method, for the following reasons:
  • Confidence intervals around mortality ratio estimates are wide.52
  • The historical data sets that underlie this analysis are from the 1980s and 1990s,53 when the landscape of mortality, infectious diseases, and medical services was different from today. It is not clear that mortality ratios for MAM and SAM are the same today as they were several decades ago, when overall under-5 mortality rates were much higher.54
  • It is not entirely clear that we should be modeling mortality ratios from historical data, rather than absolute mortality rates for SAM and MAM.55 We believe mortality ratios are likely to be more stable over time than absolute mortality rates, given a changing landscape of infectious diseases and population-wide mortality rates, but we are not aware of evidence one way or the other.
  • The communities represented in the data sets had varying degrees of malnutrition treatment available.56 The intensity of treatment availability ranged from negligible to modest, compared with modern malnutrition treatment programs, but we did identify a possible relationship between treatment availability and mortality ratio.57 This does not necessarily imply that the MIG mortality ratios are substantially inaccurate, as we do not have specific reason to believe that treatment intensity in these historical settings was substantially different from background treatment intensity in children not enrolled in formal malnutrition treatment programs today. Nevertheless, we apply a small adjustment for this in our model.58

Despite these major uncertainties, we are not aware of a better way to estimate excess mortality experienced by children with MAM and SAM.

The MIG method has undergone extensive review both internally at GiveWell and externally by experts in statistics and malnutrition modeling. Its results are somewhat higher than a much simpler probit model that provides a sense-check on the findings, but with widely overlapping confidence intervals.59 We spoke with over 17 external experts, who were mostly, though not entirely, supportive of the modeling approach.60 Our statistics consultant Megan Higgs raised concerns about relying too heavily on the model’s results based on concerns about generalizing from historic data, assumptions that were difficult to check and justify, and a general lack of transparency to most stakeholders due to the overall complexity of the approach used to generate cost-effectiveness analysis inputs from historic data.61 One expert in malnutrition modeling, Mark Myatt, told us he prefers drawing absolute mortality rates from historical data rather than mortality ratios.62 Experts from the International Rescue Committee (IRC) had concerns about the initial mortality rates we generated using the mortality ratio from the MIG model. Specifically, they flagged the rates as being lower than they thought plausible. This concern was mostly addressed when we updated subsequent estimates, which were higher as a result of using higher estimates of population mortality rates.

How we estimate absolute mortality rates of children with malnutrition

To estimate absolute mortality rates of children with untreated malnutrition in our locations of interest, we use the mortality ratios from the MIG model, the local prevalence of untreated SAM and MAM, and local population-wide under-5 mortality rates.63 Using our model of ALIMA's programs in Northern Nigeria and Southern Niger as an example, this method yields annual mortality rates for untreated malnutrition of 3% to 4% for MAM and 8% to 10% for SAM, after adjustments.64 For ALIMA's program in Dakoro and Mirriah, Niger, we estimate a mortality rate of 5.9% for all children with untreated malnutrition.

There is a constrained relationship between total under-5 mortality rates, malnutrition mortality rates, and malnutrition incidence that is not reflected in our cost-effectiveness model. In other words, due to uncertainty in inputs or modeling choices, our model could return malnutrition mortality rates that are unrealistically high, given the amount of malnutrition in a population and its overall mortality rate.

To address this, we adjust our estimate on the basis of a ceiling analysis we developed to account for this constrained relationship.65 In some contexts, this analysis implies our estimate of mortality rates based on the MIG model is indeed unrealistically high. In those cases, we make an adjustment to account for this.66 In these instances, this is the most impactful adjustment we make to our estimates.67 We discuss the ceiling analysis below.

Sense-check of malnutrition mortality rates using a ceiling analysis

In order to sense-check our estimates of mortality rates among untreated malnourished children, we developed a ceiling analysis estimating the maximum plausible mortality rates for untreated malnourished children. This estimate is based on:

  • Mortality rates among children 6-59 months old in target areas. We calculate this on the basis of under-five mortality rates, and our estimate of what percentage of those deaths occur in the 6-59 month age bracket.68
  • Prevalence of untreated malnourished children. We estimate this on the basis of point-in-time estimates of the prevalence of malnutrition, and malnutrition treatment coverage estimates.
  • Maximum plausible percentage of deaths that occur among malnourished children. We estimate a maximum of 90% of the deaths that occur in children 6-59 months annually occur among malnourished children, based on a subjective guess. We then estimate the maximum plausible percentage of deaths that occur among children who are malnourished at each point in time, to match the prevalence estimates. To do so, we divide 90% by the “incidence correction factor,” a factor used to calculate malnutrition incidence, which corresponds to the inverse of the average duration of a malnutrition episode.69

The model can be found here. We use as an example the work planned by ALIMA (Alliance for International Medical Action) to support the implementation of the combined protocol for CMAM in Dakoro and Mirriah departments, Niger and Kaita local government area, Nigeria.

In some contexts, the mortality rates resulting from this analysis are substantially lower than what we estimate using the method in our cost-effectiveness analysis (CEA).70 Because our main estimates fail this sense-check, we use the ceiling analysis to generate an adjustment that limits our CEA estimates to a range that’s consistent with the ceiling analysis.71

The result of the ceiling analysis differs from our main estimates in a number of ways:

  • It is an upper bound, rather than a best guess.
  • It estimates mortality rates of untreated malnourished children on the basis of the ratio between (1) the probability of death among those who will be malnourished at any point in the year and (2) the probability of death among those who will never be malnourished; conversely the mortality ratios based on historical evidence capture the ratio between (a) the probability of death among those malnourished at that point in time and (b) the probability of death among those not malnourished at that point in time.
  • It employs mortality rates for 6-59-month-old children as a whole; conversely, our CEA employs mortality rates for children with an age distribution weighted to approximate CMAM admissions. We expect the latter to be higher than the former, but we would guess this is offset by malnutrition prevalence rates being lower in the age bracket used in the ceiling analysis than the age bracket used in the main CEA model.72

We do not believe these differences fully explain the discrepancy in the estimates. This is because the first two differences would imply a discrepancy in the opposite direction (the mortality rates in the ceiling analysis should be higher than mortality rates estimated on the basis of historical data);73 the third difference does explain some, but only a limited amount, of the discrepancy.

We remain unsure of what explains the discrepancy between the estimates in our CEA and those in the ceiling analysis. Our best guess is that this could be explained by:

  • The mortality ratio estimates based on historical data are too high. This could be explained by some of the limitations outlined above.
  • Our estimate of the incidence correction factor is too high. This would imply we are overestimating the number of children who become malnourished during a year, and thus underestimating the risk of death per child.74

We have major uncertainties about the ceiling analysis. The maximum plausible mortality rates it yields are very sensitive to the incidence correction factor, prevalence of untreated malnutrition, and coverage estimates, all of which are uncertain. In addition, the analysis assumes the maximum plausible percentage of deaths in the cohort that occur in untreated malnourished children during a year is 90%.75 This is intended to be a plausibility constraint rather than a best guess. It is possible this is too high to be plausible.

As a result, we view the ceiling analysis as a method for generating plausibility caps rather than best guesses for malnutrition mortality rates. Nevertheless, in cases where mortality rates are constrained by the ceiling analysis, it is determining the values we use in our CEA, ultimately influencing our cost-effectiveness estimates.

It is possible we should also use the ceiling analysis to adjust the treatment effect. A possible explanation for the discrepancy between the ceiling analysis and the main model is that the ratio between mortality rates of untreated malnourished children and non-malnourished children estimated by the MIG model is too high. Our estimate of the treatment effect is based on the ratio between mortality rates of untreated malnourished children and treated malnourished children generated by the MIG model.76 It is therefore possible that, to adequately account for the ceiling analysis results, we should adjust our estimate of the treatment effect (as well as our estimate of mortality rates). We decided not to do that because the ceiling analysis is based on a number of uncertain inputs (see above), though we remain uncertain about this.

How we estimate the effect of malnutrition treatment on mortality

To estimate the impact of CMAM on mortality in our CEA, we use the MIG model to generate a mortality ratio that represents the comparison of average WHZ before vs. after CMAM treatment.77 WHZ inputs come from a literature review of the impact of CMAM programs on WHZ, adjusted for differences in recovery rates between government-only and nongovernmental organization (NGO)-supported malnutrition programs.78 This method estimates that MAM treatment reduces the relative risk of mortality over the next year by about 40% and SAM treatment reduces it by about 70%.79

Since this method of estimating a CMAM treatment effect is based on the MIG model described above, it shares all of its limitations.

In addition, this method makes a number of simplifying assumptions when using the MIG model to estimate treatment effect. We believe these are the main limitations of the method:

  • It assumes that the effect of malnutrition treatment on mortality is captured by the observational correlation between WHZ and mortality. We believe this assumption is highly uncertain, and likely incorrect.80
  • The communities represented in the data sets the MIG model is based on had varying degrees of malnutrition treatment available.81 The intensity of treatment availability ranged from negligible to modest, compared with modern malnutrition treatment programs, but we did identify a possible relationship between treatment availability and mortality ratio.82 Ignoring the impact of malnutrition treatment in these historical cohorts could cause our method to underestimate the treatment effect.
  • The method we use to estimate the treatment effect assumes that children are classified as SAM, MAM, or non-malnourished based on their weight-for-height z-score (WHZ) at time 0 and followed up for one year. Each child remains in the original category for the purpose of assigning mortality, regardless of how WHZ changes over the course of the year. In reality, children could be "reclassified" more than once per year if WHZ is measured more often than once per year.83 This would allow a better treatment effect on mortality because children would be treated more accurately based on need. This means our model might underestimate the treatment effect.

We roughly correct for these limitations in our cost-effectiveness analysis,84 but since we have little basis for estimating their impact, we have a large amount of remaining uncertainty about our treatment effect estimate. We recognize that this estimation method has major limitations, but given the absence of direct estimates of the impact of malnutrition treatment on mortality, we are not aware of a better method.

Increasing and improving treatment

We think that NGOs' support for CMAM programs in government-run health facilities leads to more children being treated for malnutrition and higher quality care. We largely think this because there is an intuitively plausible story for how the activities they conduct lead to higher numbers of children receiving treatment and better quality care. For example, NGOs can help caregivers to check whether their child is malnourished by equipping them with a tool for measuring children's mid-upper-arm circumference and training them how to use it; it seems plausible to us that this drives additional cases of malnutrition being identified and treated.85

In the case of a program we supported the NGO ALIMA to conduct in Niger, we estimate that about 200,000 children will receive treatment supported by ALIMA's program. We expect that about 35% of these children would have otherwise gone untreated and that the remaining 65% receive higher quality treatment than they otherwise would have. This is based on ALIMA's estimates.

Vaccination

Malnutrition treatment programs sometimes give admitted children vaccines the children have not yet received.86 We believe vaccination is effective for reducing the risk of the diseases it targets, and associated illness and death.87 We estimate the benefits of vaccination provided by malnutrition treatment programs by adapting part of our New Incentives cost-effectiveness analysis.88 This estimate implies that fully vaccinating 29 children at recommended ages is equivalent to saving the life of one child under five.89 In the case of ALIMA's program in Niger, we estimate that the benefits of vaccinations make up 26% of the program's total benefits.

Malaria testing and treatment

Some CMAM programs test children admitted for malnutrition treatment for malaria and offer malaria treatment as necessary. ALIMA has told GiveWell that all children admitted to its programs in Southern Niger and Northern Nigeria are tested for malaria and infections are treated.90 Since malaria mortality rates are high in these areas, we estimate that this is a significant benefit of the ALIMA program.91 However, we are uncertain about the size of the benefit because we do not have a direct estimate of the number of deaths this activity averts.

Effects on adult income

We are not aware of any studies that have estimated the impact of receiving CMAM in childhood on income in adulthood. However, there is a large body of research that is broadly consistent with the idea that early childhood growth restriction inhibits physical and cognitive development, which in turn negatively affects outcomes in adulthood (GiveWell reviews this literature here). There is also evidence from natural experiments and randomized controlled trials that health shocks in childhood, such as malaria and parasitic worm infections, reduce adult income.92

A systematic review of randomized controlled trials reports small but statistically significant effects of nutrition supplementation on measures of physical growth, and cognitive, language, motor, and social-emotional development in children in low- and middle-income countries.93 Despite the lack of direct evidence, we believe CMAM probably increases income in adulthood by improving early childhood development, so we include it as a benefit in our cost-effectiveness analysis.94

Potential adverse effects

We consider antibiotic resistance to be the most serious potential adverse effect of CMAM programs. Short-term side effects resulting from antibiotics or RUTF appear rare and relatively mild (see below). Long-term side effects are potentially more serious, but we believe they are greatly outweighed by the benefits of reduced mortality.

  • Antibiotic resistance. In Africa, high levels of resistance to amoxicillin have been observed in the bacteria that cause pneumonia, a common cause of childhood death, making resistance to amoxicillin an urgent public health priority.95 However, in a study of children with SAM who were administered oral amoxicillin and then followed for two years, sustained antibiotic resistance was not observed.96 We believe based on this evidence that the benefits of treatment with amoxicillin outweigh the risks in this context. We have included a rough adjustment to account for this in our cost-effectiveness model.97
  • Short-term side effects. A systematic review using data from three studies comparing RUTF to other foods found no difference in risk of adverse events (including cough, diarrhea and fever).98 Three out of 1,847 children treated with antibiotics (924 treated with amoxicillin + 923 treated with cefdinir) in a Malawi study of children with SAM reported adverse drug reactions including rash, thrush, and bloody diarrhea.99 In general, common side effects of amoxicillin, a WHO-recommended antibiotic for treatment of uncomplicated SAM,100 are diarrhea, rash, vomiting, and nausea.101
  • Long-term side effects. Antibiotic exposure in childhood has been linked in observational studies to increased risk of asthma and allergies102 and overweight and obesity.103 In addition, the long-term effects of use of RUTF are not well-studied, and some speculate that RUTF may increase risk of obesity and associated diseases.104 We believe these risks are outweighed by potentially substantial benefits but have included an adjustment in our cost-effectiveness analyses to attempt to account for them.105

How cost-effective is it?

Our cost-effectiveness analyses suggest that the cost-effectiveness of CMAM varies greatly by location and program. Some locations/programs are in the range of cost-effectiveness of programs we would consider directing funding to, while others are not.

Note that our cost-effectiveness analyses are simplified models that do not take into account a number of factors. There are limitations to this kind of cost-effectiveness analysis, and we believe that cost-effectiveness estimates such as these should not be taken literally due to the significant uncertainty around them. We provide these estimates (a) for comparative purposes and (b) because working on them helps us ensure that we are thinking through as many of the relevant issues as possible.

The opportunities we have identified focus on supporting government CMAM programs to improve coverage and quality of care. A sketch of the key components of the cost-effectiveness model is below. Throughout, we use as an example the work planned by ALIMA (Alliance for International Medical Action) to support the implementation of the combined protocol in Dakoro and Mirriah departments, Niger.
  • Cost. In 2022, we estimated an annual program cost of $2,622,055, or $39 for each child treated for malnutrition.106 This includes the cost of providing training and mentoring to government clinic staff, providing medical supplies to fill in gaps in procurement, refurbishing facilities, providing transportation for medical staff and equipment, community outreach, surveys, and providing additional human resources and incentives for existing staff.107 This excludes costs shouldered by government facilities, which we estimate to be around $345,000, and the cost of RUTF, which we estimate to be about $885,000 and financed by UNICEF.108 These additional costs are accounted for in our final cost-effectiveness figures.109 Including these additional costs, we estimate that ALIMA's program in Dakoro and Mirriah costs $67 per malnourished child reached.110
  • Number of children treated. Annually, we estimate that the program will treat about 18,000 children with MAM and 5,500 children with SAM who would not have been treated without the program.111 In addition, we estimate that the program will annually treat about 28,000 children with MAM and 15,000 children with SAM who would have been treated for malnutrition less effectively without the program.112
  • Mortality rate of untreated MAM and SAM. Our estimate is that 4% of those with MAM and 10% of those with SAM would die without treatment in the year beginning with a malnutrition episode.113
  • Treatment effect. We estimate treatment would lower the probability of death by about 42% for children with MAM and 69% for children with SAM.114 In the case of ALIMA's program in Dakoro and Mirriah, Niger, we estimate the reduction in all-cause mortality to be 45%.115
  • Additional benefits. For ALIMA's program in Dakoro and Mirriah, we estimate that 26% of total benefits are due to vaccination, 6% are due to malaria testing and treatment, and 1% are due to improvements in early-life health leading to higher adult income.116
  • Other. Downside adjustments, excluded effects, and leverage/funging adjustments reduce the program's benefits by 22%.117 The excluded effects include a rough adjustment for non-CMAM work ALIMA plans to conduct to support pediatric care in the targeted areas, but we have not modeled this benefit explicitly.118

Overall, we estimate that combined protocol CMAM in Dakoro and Mirriah departments, Niger, is within the range of cost-effectiveness of programs we would consider recommending funding.119 Although the cost per treated child is high relative to our top charities, the intervention reduces the risk of death to a greater degree per child treated.120

However, we have several major uncertainties about our cost-effectiveness estimate:

  • Mortality rate of untreated children with malnutrition. Our estimate relies on extrapolation from historical cohorts from the 1980s and 1990s and is uncertain for several reasons, discussed here.
  • Treatment effect of CMAM. Since direct estimates are not available, we use an indirect estimation method that is very uncertain. Its limitations are discussed here.
  • Caseload.We rely on historical data and projections to estimate MAM and SAM caseloads, and these are uncertain, particularly the latter.121 In our CEA, we also estimate the number of children that would receive malnutrition treatment in the absence of NGO support. These estimates are based on NGOs' best guesses, and are uncertain.122 For ALIMA's program in Niger, we estimate that 65% of children reached would have received government treatment without the program.
  • Vaccination benefits. The evidence underlying our estimate of vaccination benefits is weak, and in addition we rely on uncertain assumptions to estimate the benefits of vaccinating children closer to recommended ages.123

We expect the cost-effectiveness of CMAM programs to vary significantly by location and by program. We expect the drivers for this variation to be: mortality rates among children 6-59 months old in target areas, the cost per child treated, and whether the NGO leverages its malnutrition work to support additional health programs.

Does the program have room for more funding?

Our understanding is that there is likely to be substantial room for more funding for malnutrition treatment due to a large remaining global need.124 We are currently unsure what share of this would be above our funding bar.

Key questions for further investigation

If we pursue additional funding opportunities in this space, we would like to gather additional evidence on the following:

  • How much do pediatric emergency treatment and other non-CMAM program components contribute to the cost-effectiveness of funding opportunities?
  • We would like to refine our caseload estimates by gathering additional caseload data from current and future CMAM programs.
  • Our complementary ceiling analysis suggests our CEA might be overestimating mortality rates.125 To what extent can we reduce our uncertainty about key variables in the CEA by doing additional work on the ceiling analysis?
  • If high-quality long-term follow-up data become available on the mortality rate of children in the year following CMAM, we may be able to use them to refine our estimates of mortality averted by CMAM.
  • What are the longer-term outcomes for children who recover from malnutrition? Are there any permanent impacts of a temporary period of malnutrition on later-life outcomes?
  • What is the relationship between food insecurity, mortality rates, and malnutrition rates?126
  • Does cost-effectiveness vary between settings where children are chronically malnourished as opposed to malnourished due to a temporary crisis?127
  • What fraction of opportunities to fund malnutrition treatment would meet our cost-effectiveness bar?

Our process

  • We performed an in-depth literature review of studies reporting mortality from malnutrition with and without treatment. We performed a medium-depth literature review to find studies on individual components of CMAM and published estimates of costs of CMAM programs. We performed a medium-depth literature review to identify estimates of the impact of malnutrition treatment on WHZ.
  • We completed an initial model of malnutrition treatment, and redesigned it as part of an update on this topic.
  • We commissioned David Roodman to complete a reanalysis of Olofin 2013, the study we were using to estimate mortality rates of malnourished children.
  • We reviewed David's work internally and asked Megan Higgs, an external statistician we work with, to review David's work.
  • We had phone calls with more than 17 experts, including three experts recommended by the International Rescue Committee (IRC), to discuss David's model, as well as our broader approach to modeling the cost-effectiveness of malnutrition treatment programs. During some of these calls, we also gut-checked the plausibility of our untreated SAM mortality rate.
  • We had eight calls with the IRC and three calls with ALIMA to discuss their feedback on the model.
  • We completed three rounds of written feedback with the IRC on the model.
  • We conducted three internal peer-reviews of our CEA model.
  • We constructed a ceiling analysis to sense-check our CEA estimates.

Sources

Document Source
Aalen 1994 Source
Alé et al. 2016 Source
ALIMA, 2023-2025 Budget: Niger Source
ALIMA, conversation with GiveWell, July 26, 2022 (unpublished)
Bailey et al. 2020 Source
Baird et al. 2016 Source
Bazzano et al. 2017 Source
Black et al. 2008 Source
Bleakley 2010 Source
Bourke, Berkley, and Prendergast 2016 Source
Caulfield et al. 2004 Source
Collins et al. 2006 Source
Cutler et al. 2010 Source
Das et al. 2020 Source
Daures et al. 2020 Source
David Roodman, On the association between anthropometry and mortality in children, 2022 Source
Emergency Nutrition Network, "Simplified approaches to the treatment of wasting," July 2020 Source
FDA, "AMOXIL (amoxicillin) capsules, tablets or powder for oral suspension," 2015 Source
Fishman et al. 2004 Source
Frison, Checchi, and Kerac 2015 Source
Gavi and Scaling Up Nutrition, Equity From Birth: An integrated approach to immunisation and nutrition policy brief Source
GiveWell, "Alliance for International Medical Action (ALIMA) – Treatment of Malnutrition in Niger" Source
GiveWell, 2022 cost-effectiveness analysis – version 5 Source
GiveWell, Ceiling analysis of mortality rates for untreated malnourished children Source
GiveWell, CMAM combined & standard protocols with family MUAC Source
GiveWell, Impact of CMAM on WHZ Source
GiveWell, List of experts consulted for malnutrition CEA Source
GiveWell, Malnutrition treatment CEA (combined protocol ALIMA) Source
GiveWell, New Incentives (Conditional Cash Transfers to Increase Infant Vaccination) Source
GiveWell, The impact of early life growth on adult economic status Source
GiveWell's non-verbatim summary of a conversation with ALIMA, March 5th, 2021 Source
GiveWell's non-verbatim summary of a conversation with ALIMA, November 13, 2020 Source
HealthDirect, "Fluid retention" Source
Higgins, Li, and Deeks, Cochrane Handbook for Systematic Reviews of Interventions, Chapter 6: Choosing effect measures and computing estimates of effect, 2022 Source
Higgs, Summary of concerns with modeling of malnutrition mortality effect, April 2022 (unpublished)
Horton et al. 2010 Source
IHME, GBD Compare tool, under-5 all cause mortality in Niger Source
Imdad et al. 2022 Source
Isanaka et al. 2021 Source
James et al. 2016 Source
Jones et al. 2014 Source
Keenan et al. 2018 Source
Kevin Phelan, Nutrition Advisor, ALIMA, email to GiveWell, August 9, 2022 (unpublished)
Langdon et al. 2018 Source
Mark Myatt, Conversation with GiveWell, April 26, 2022 (unpublished)
Olofin et al. 2013 Source
Pelletier et al. 1994 Source
Phelan 2019 Source
Prado et al. 2019 Source
Republic of Chad, Ministry of Public Health, Social Action, and National Solidarity, National protocol for management of acute malnutrition, 2014 Source
Republic of Mali, Ministry of Health, Protocol for integrated management of acute malnutrition, 2011 Source
Rogers et al. 2015 Source
Schaible and Kaufmann 2007 Source
Shao et al. 2017 Source
Tadesse et al. 2017 Source
The GiveWell Blog, "Why I mostly believe in Worms," 2016 Source
Trehan et al. 2013 Source
U.S. National Library of Medicine, ClinicalTrials.gov, "Optimizing Acute Malnutrition Management in Children Aged 6 to 59 Months in Niger (OptIMA Niger)," 2021 Source
UNICEF, WHO, World Bank, "Joint child malnutrition estimates — levels and trends," 2020 Source
WHO and UNICEF, WHO child growth standards and the identification of severe acute malnutrition in infants and children, 2009 Source
WHO, "Guideline: updates on the management of severe acute malnutrition in infants and children," 2013 Source
WHO, "Malnutrition" Source (archive)
WHO, "Supplementary foods for the management of moderate acute malnutrition in children aged 6–59 months," 2019 Source (archive)
WHO, "WHO child growth standards and the identification of severe acute malnutrition in infants and children," 2009 Source
World Health Organization, the World Food Programme, the United Nations System Standing Committee on Nutrition and the United Nations Children’s Fund, Community-Based management of severe acute malnutrition, 2007 Source
Yamamoto-Hanada et al. 2017 Source
  • 1

    "Wasting refers to a child who is too thin for his or her height. Wasting is the result of recent rapid weight loss or the failure to gain weight." UNICEF, WHO, World Bank, "Joint child malnutrition estimates — levels and trends," 2020, p. 2.

  • 2

    “Acute malnutrition is a major public health issue in low-income countries. It includes both wasting and edematous malnutrition, but the terms wasting and acute malnutrition are often used interchangeably." Frison, Checchi, and Kerac 2015, Abstract.
    “Fluid retention is also called oedema or water retention. It occurs when parts of the body swell due to a build-up of trapped fluid. The fluid gets trapped and makes the area swollen or puffy.” HealthDirect, "Fluid retention".

  • 3
    • "Restricted growth as a result of inadequate nutrition and infections is an important cause of morbidity and mortality in infants and children worldwide. . . . Several prospective studies have shown associations of undernutrition with increased risk of various disease outcomes, and reduced survival, in children." Olofin et al. 2013, Introduction.
    • "All degrees of underweight, stunting and wasting were associated with significantly higher mortality. The strength of association increased monotonically as Z scores decreased. Pooled mortality HR was 1.52 (95% Confidence Interval 1.28, 1.81) for mild underweight; 2.63 (2.20, 3.14) for moderate underweight; and 9.40 (8.02, 11.03) for severe underweight. Wasting was a stronger determinant of mortality than stunting or underweight." Olofin et al. 2013, Abstract.

  • 4

    "In children aged 6–59 months, moderate acute malnutrition is defined as moderate wasting (i.e. weight-for-height between –3 and –2 Z-scores of the WHO Child Growth Standards median) and/or mid-upper-arm circumference (MUAC) greater or equal to 115 mm and less than 125 mm." WHO, "Supplementary foods for the management of moderate acute malnutrition in children aged 6–59 months," 2019.

  • 5

    "Severe acute malnutrition (SAM) is defined as a weight-for-height measurement of 70% or less below the median, or three SD [standard deviations] or more below the mean National Centre for Health Statistics reference values, the presence of bilateral pitting oedema of nutritional origin, or a mid-upper-arm circumference of less than 110 mm in children age 1–5 years." Collins et al. 2006, Summary.
    In 2009, WHO recommended increasing the MUAC cut-off point for defining SAM from 110 mm to 115 mm. "WHO standards for mid-upper arm circumference (MUAC)-for-age show that in a well nourished population there are very few children aged 6–60 months with a MUAC less than 115 mm. Children with a MUAC less than 115 mm have a highly elevated risk of death compared to those who are above. Thus it is recommended to increase the cut-off point from 110 to 115 mm to define SAM with MUAC." WHO and UNICEF, WHO child growth standards and the identification of severe acute malnutrition in infants and children, 2009, p. 2.

  • 6

    One study in Ethiopia tracked children identified as MAM whose home districts were ineligible for food supplementation programs and found that only slightly over half of the children recovered without experiencing an episode of SAM during the 28 week tracking period:
    “We prospectively surveyed 884 children aged 6–59 months living with MAM in a rural area of Ethiopia not eligible for a supplementary feeding programme. Weekly home visits were made for seven months (28 weeks), covering the end of peak malnutrition through to the post-harvest period (the most food secure window), collecting anthropometric, socio-demographic and food security data. . . . Only 54.2% of the children recovered with no episode of SAM by the end of the study.” James et al. 2016, Abstract.

  • 7

    See UNICEF, WHO, World Bank, "Joint child malnutrition estimates — levels and trends," 2020, “Number (Millions) Affected Tables,” p. 13, row "Global."

  • 8

    “Stunting is the devastating result of poor nutrition in-utero and early childhood. Children suffering from stunting may never attain their full possible height and their brains may never develop to their full cognitive potential. Globally, 144.0 million children under 5 suffer from stunting. These children begin their lives at a marked disadvantage: they face learning difficulties in school, earn less as adults, and face barriers to participation in their communities.” UNICEF, WHO, World Bank, "Joint child malnutrition estimates — levels and trends," 2020, p. 2.

  • 9

    “Severe acute malnutrition remains a major killer of children under five years of age. Until recently, treatment has been restricted to facility-based approaches, greatly limiting its coverage and impact.” World Health Organization, the World Food Programme, the United Nations System Standing Committee on Nutrition and the United Nations Children’s Fund, Community-Based management of severe acute malnutrition, 2007, p. 2.

  • 10

    “Severe acute malnutrition remains a major killer of children under five years of age. Until recently, treatment has been restricted to facility-based approaches, greatly limiting its coverage and impact. New evidence suggests, however, that large numbers of children with severe acute malnutrition can be treated in their communities without being admitted to a health facility or a therapeutic feeding centre. The community-based approach involves timely detection of severe acute malnutrition in the community and provision of treatment for those without medical complications with ready-to-use therapeutic foods or other nutrient-dense foods at home. If properly combined with a facility-based approach for those malnourished children with medical complications and implemented on a large scale, community-based management of severe acute malnutrition could prevent the deaths of hundreds of thousands of children.” World Health Organization, the World Food Programme, the United Nations System Standing Committee on Nutrition and the United Nations Children’s Fund, Community-Based management of severe acute malnutrition, 2007, p. 2.

  • 11

    "Ready-to-use Therapeutic Food (RUTF) has revolutionized the treatment of severe malnutrition – providing foods that are safe to use at home and ensure rapid weight gain in severely malnourished children. The advantage of RUTF is that it is a ready-to-use paste which does not need to be mixed with water, thereby avoiding the risk of bacterial proliferation in case of accidental contamination. The product, which is based on peanut butter mixed with dried skimmed milk and vitamins and minerals, can be consumed directly by the child and provides sufficient nutrient intake for complete recovery." WHO, "Malnutrition"

  • 12
    • "Children with uncomplicated severe acute malnutrition, not requiring to be admitted and who are managed as outpatients, should be given a course of oral antibiotic such as amoxicillin." WHO, "Guideline: updates on the management of severe acute malnutrition in infants and children," 2013, p. 29.
    • Table 1, p. 37 of Chad's national protocol for management of acute malnutrition displays the differences between SAM and MAM treatment:
      • The "Produits [Products]" row for the "Traitement MAS [SAM treatment]" column states, "Différent produits Aliment Thérapeutique Prêt à l’Emploi (ATPE) – F75 – F100," which translates in English to "Different Ready-to-Use Therapeutic Food (RUTF) products - F75 - F100." The "Antibiothérapie systématique [Systematic antibiotic therapy]" row for the SAM column states "Oui" or "Yes" in English.
      • The "Produits [Products]" row for the "Traitement MAM [MAM treatment]" column states, "Différents produits : Farines fortifiées (ex.CSB), Aliment de Supplément Prêt àl’Emploi (ASPE)," which translates in English to "Different products: Fortified flour (e.g. CSB), Ready-to-use supplementary food (RUSF)." The "Antibiothérapie systématique [Systematic antibiotic therapy]" row for the MAM column states "Non," or "No" in English. Republic of Chad, Ministry of Public Health, Social Action, and National Solidarity, National protocol for management of acute malnutrition, 2014.
    • Mali's national CMAM protocol indicates that RUTF and antibiotic therapy should be used to treat SAM:
      • "Les patients sévèrement malnutris ont des besoins très spécifiques en terme de nutriments qui sont différents de ceux des patients normaux… L’Aliment Thérapeutique Prêt à l’Emploi (ATPE) est un composant essentiel des URENAS, permettant le traitement à domicile." p. 200.
        • English translation from original French: "Severely malnourished patients have very specific nutritional needs that differ from those of normal patients… Ready-to-Use Therapeutic Food (RUTF) is an essential component of URENAS, allowing treatment at home."
      • "Les antibiotiques doivent être donnés aux patients souffrant de MAS systématiquement, même si le patient ne présente pas de signes cliniques d’infections généralisées." p. 72.

    • 13
      • Table 1, p. 37 of Chad's national protocol for management of acute malnutrition displays the differences between SAM and MAM treatment:
        • The "Produits [Products]" row for the "Traitement MAS [SAM treatment]" column states, "Différent produits Aliment Thérapeutique Prêt à l’Emploi (ATPE) – F75 – F100," which translates in English to "Different Ready-to-Use Therapeutic Food (RUTF) products - F75 - F100." The "Antibiothérapie systématique [Systematic antibiotic therapy]" row for the SAM column states "Oui" or "Yes" in English.
        • The "Produits [Products]" row for the "Traitement MAM [MAM treatment]" column states, "Différents produits : Farines fortifiées (ex.CSB), Aliment de Supplément Prêt àl’Emploi (ASPE)," which translates in English to "Different products: Fortified flour (e.g. CSB), Ready-to-use supplementary food (RUSF)." The "Antibiothérapie systématique [Systematic antibiotic therapy]" row for the MAM column states "Non," or "No" in English. Republic of Chad, Ministry of Public Health, Social Action, and National Solidarity, National protocol for management of acute malnutrition, 2014.
      • Section IV, p. 118 of Mali's national CMAM protocol, which discusses management of MAM (Section IV is titled "PRISE EN CHARGE DE LA MALNUTRITION AIGUE MODEREE," which translates to "Medical Management of Moderate Acute Malnutrition" in English, see p. 117), states that the types of treatment provided for MAM include enriched flour and RUSF:
        • "Les aliments de supplémentation utilisés par les URENAM sont à base de:
          • Farines industrielles améliorées en complexes minéralo-vitaminiques répondant aux normes internationales (Supercerealplus, Supercereal2),
          • Farines locales enrichies (Exemple: Misola)53.
          • Aliments Supplémentaires prêts à l’emploi(ASPE): Pâte à base de lipides (Exemple, «Supplementary Plumpy» ou PlumpySup)."
        • English translation from original French: "The supplementation foods used by URENAM are based on:

      • 14
        • “SAM and MAM are managed in separate programs, using different food products and protocols. There is currently no globally accepted guidance for the treatment of MAM, and MAM is not always routinely treated. International mandate adds an additional layer of complexity: UNICEF supports the treatment of SAM and provides ready-to-use therapeutic food (RUTF) for use in outpatient therapeutic programs (OTPs); the World Food Programme supports the treatment of MAM and provides ready-to-use supplementary food (RUSF) or fortified blended flours for use in supplementary feeding programs (SFPs). In humanitarian settings, providing treatment for both SAM and MAM adds to the logistical and financial burden of health systems. When resources are scarce, and in the many settings where prevalence is not high enough to reach emergency thresholds, treatment of SAM is often prioritized, and children with MAM may not be eligible to receive care unless they deteriorate.” Bailey et al. 2020, p. 4.
        • "In addition to the tangle of agencies and case definitions, SAM and MAM programmes are chronically underfunded with only 25% of SAM cases treated globally in 2016 and 16% of MAM cases reached by the World Food Program in 2017." Daures et al. 2020, p. 757.

      • 15

        Reliable data about children who do not receive treatment appears scarce. However, one study in Ethiopia tracked children identified as having MAM whose home districts were ineligible for food supplementation programs and found that during the 28-week tracking period, “only 54.2% of the children recovered with no episode of SAM by the end of the study.” James et al. 2016, p. 2.

      • 16

        “SAM and MAM are managed in separate programs, using different food products and protocols. There is currently no globally accepted guidance for the treatment of MAM, and MAM is not always routinely treated. International mandate adds an additional layer of complexity: UNICEF supports the treatment of SAM and provides ready-to-use therapeutic food (RUTF) for use in outpatient therapeutic programs (OTPs); the World Food Programme supports the treatment of MAM and provides ready-to-use supplementary food (RUSF) or fortified blended flours for use in supplementary feeding programs (SFPs). In humanitarian settings, providing treatment for both SAM and MAM adds to the logistical and financial burden of health systems. When resources are scarce, and in the many settings where prevalence is not high enough to reach emergency thresholds, treatment of SAM is often prioritized, and children with MAM may not be eligible to receive care unless they deteriorate.” Bailey et al. 2020, p. 4.

      • 17

        We are aware of two variants of the combined protocol: ComPAS and OptiMA:

        • "The Combined Protocol for Acute Malnutrition Study (ComPAS) assessed the effectiveness of a simplified, unified SAM/MAM protocol for children aged 6–59 months. . . . Combined protocol clinics treated children using 2 sachets of ready-to-use therapeutic food (RUTF) per day for those with mid-upper arm circumference (MUAC) < 11.5 cm and/or edema, and 1 sachet of RUTF per day for those with MUAC 11.5 to <12.5 cm." Bailey et al. 2020, Abstract.
        • "Although acute malnutrition is a continuum condition, it is arbitrarily divided into moderate (MAM) and severe (SAM) categories defined by mid upper arm circumference (MUAC) or weight-for-height Z-score (WHZ). . . . We piloted a new MUAC-based and oedema approach for treating acute malnutrition in Burkina Faso with a single-arm proof-of-concept trial called Optimising treatment for acute MAlnutrition (OptiMA). . . . only one product was used for treatment (RUTF) at a gradually reduced dose based on a child’s weight and MUAC status." Daures et al. 2020, pp. 756-757.

      • 18
        • "As part of the solution, practitioners and experts have recognised the need to simplify approaches to wasting treatment and have identified key research priorities, such as “reviewing appropriate entry and discharge criteria for treatment of acute malnutrition” and “investigating the safety, effectiveness and cost-effectiveness of reduced dosage ready to-use therapeutic food dosages” (No Wasted Lives, 2018). The aim is to achieve greater coverage and improved efficiency of services (including cost-effectiveness) for children at high risk of illness and death, while maintaining quality of care." Emergency Nutrition Network, "Simplified approaches to the treatment of wasting," July 2020, p. 1.
        • "ALIMA’s Optimizing treatment for acute MAlnutrition (OptiMA) is one such strategy, proposing three main changes to current protocols:
          • Earlier detection by training mothers and caregivers how to use mid-upper arm circumference (MUAC) bands to screen children regularly for malnutrition in the home (i.e., family MUAC.)
          • Simplification and easier management by using only one anthropometric measure (MUAC <125 mm (and/or oedema)) for admissions and one product (RUTF) for treatment.
          • More intelligent use of the costliest input (RUTF) by gradually reducing the dosage based on a child’s MUAC status and weight to increase the number of children with access to treatment at no extra or similar cost." Phelan 2019, p. 40.

      • 19

        The Combined Protocol for Acute Malnutrition Study (ComPAS) (Bailey et al. 2020) tested a protocol using MUAC (along with presence of edema) as the single criterion to diagnose acute malnutrition, determine dosage of RUTF, and diagnose recovery in South Sudan and Kenya. The Optimizing treatment for acute MAlnutrition (OptiMA) protocol has been trialed in Burkina Faso (Phelan 2019) and DRC (Phelan 2019), with another trial planned for Niger (U.S. National Library of Medicine, ClinicalTrials.gov, "Optimizing Acute Malnutrition Management in Children Aged 6 to 59 Months in Niger (OptIMA Niger)," 2021). The OptiMA protocol uses MUAC (and presence of edema) to diagnose and discharge patients and uses both weight and MUAC to determine dosage of RUTF.

        • "A cluster-randomized non-inferiority trial compared a combined protocol against standard care in Kenya and South Sudan. Randomization was stratified by country. Combined protocol clinics treated children using 2 sachets of ready-to-use therapeutic food (RUTF) per day for those with mid-upper arm circumference (MUAC) < 11.5 cm and/or edema, and 1 sachet of RUTF per day for those with MUAC 11.5 to <12.5 cm." Bailey et al. 2020, Abstract.
        • "ALIMA’s Optimizing treatment for acute MAlnutrition (OptiMA) is one such strategy, proposing three main changes to current protocols:
          • Earlier detection by training mothers and caregivers how to use mid-upper arm circumference (MUAC) bands to screen children regularly for malnutrition in the home (i.e., family MUAC.)
          • Simplification and easier management by using only one anthropometric measure (MUAC <125 mm (and/or oedema)) for admissions and one product (RUTF) for treatment.
          • More intelligent use of the costliest input (RUTF) by gradually reducing the dosage based on a child’s MUAC status and weight to increase the number of children with access to treatment at no extra or similar cost." Phelan 2019, p. 40.
        • "The OptiMA-DRC trial, an individual RCT that will begin in 2019, will determine how well this strategy works in a region with a high prevalence of oedematous malnutrition." Phelan 2019, p. 41.
        • "This community-based non-inferiority trial will compare two strategies for the treatment of AM to the Niger protocol for SAM and MAM. The Optimizing treatment for acute MAlnutrition (OptiMA) strategy uses MUAC < 125 mm or nutritional oedema as admission criteria and optimizes RUTF by adapting doses to the degree of malnutrition. RUTF dose for MUAC < 115 mm or oedema is 170 kcal/kg/d and progressively reduces to 75 kcal/kg/d as MUAC increases. The Combined Protocol for Acute Malnutrition Study (ComPAS) uses the same eligibility criteria like OptiMA, but simplifies more the RUTF ration by providing 1000 kcal/d for children with oedema or MUAC < 115 mm and 500 kg/d for children with MUAC 115-124 mm. Children are considered recovered if they have 2 consecutive weekly MUAC measures ≥ 125 mm." U.S. National Library of Medicine, ClinicalTrials.gov, "Optimizing Acute Malnutrition Management in Children Aged 6 to 59 Months in Niger (OptIMA Niger)," 2021, Study Description.

      • 20
        • "Community health workers or volunteers can easily identify the children affected by severe acute malnutrition using simple coloured plastic strips that are designed to measure mid-upper arm circumference (MUAC). In children aged 6-59 months, a MUAC less than 110 mm indicates severe acute malnutrition, which requires urgent treatment. Community health workers can also be trained to recognize nutritional oedema of the feet, another sign of this condition." World Health Organization, the World Food Programme, the United Nations System Standing Committee on Nutrition and the United Nations Children’s Fund, Community-Based management of severe acute malnutrition, 2007, pp. 2-3.
        • Alé et al. 2016 provides evidence that mothers and other caregivers can be trained to accurately measure MUAC and identify malnourished children: "A total of 12,893 mothers and caretakers were trained in the Mothers Zone and 36 CHWs in the CHWs Zone, and point coverage was similar in both zones at the end of the study (35.14 % Mothers Zone vs 32.35 % CHWs Zone, p = 0.9484). In the Mothers Zone, there was a higher rate of MUAC agreement (75.4 % vs 40.1 %, p <0.0001) and earlier detection of cases, with median MUAC at admission for those enrolled by MUAC <115 mm estimated to be 1.6 mm higher using a smoothed bootstrap procedure. Children in the Mothers Zone were much less likely to require inpatient care, both at admission and during treatment, with the most pronounced difference at admission for those enrolled by MUAC < 115 mm (risk ratio = 0.09 [95 % CI 0.03; 0.25], p < 0.0001). Training mothers required higher up-front costs, but overall costs for the year were much lower ($8,600 USD vs $21,980 USD.)" Alé et al. 2016, Abstract.

      • 21

        “Children who are identified as having severe acute malnutrition should first be assessed with a full clinical examination to confirm whether they have medical complications and whether they have an appetite. Children who have appetite (pass the appetite test) and are clinically well and alert should be treated as outpatients. Children who have medical complications, severe oedema (+++), or poor appetite (fail the appetite test), or present with one or more Integrated Management of Childhood Illness (IMCI) danger signs should be treated as inpatients (strong recommendation, low quality evidence).” WHO, "WHO child growth standards and the identification of severe acute malnutrition in infants and children," 2009, p. 3.
        Our understanding is that the assessment procedures described above for standard CMAM are followed in the combined protocols as well.

      • 22

        “Community-based treatment of severe acute malnutrition (SAM) allows the majority of malnourished children to be treated at an outpatient clinic on a weekly or fortnightly basis with only the most severe cases being admitted to inpatient centres for short periods. The outpatient clinic monitors the child’s response to the treatment and provides antimicrobial, antihelminthic, and antimalarial drugs; vitamin A supplementation; and measles vaccination (if required) before sending them home with sufficient ready to use therapeutic food (RUTF) to last until the following visit, allowing recovery to take place in the community.” Rogers et al. 2015, p. 2.

      • 23

        It is hypothesized that reducing infection in children with SAM can speed nutritional recovery and prevent mortality via prevention of colonizing microorganisms, reduction of the inflammatory response, direct reduction of inflammation, reduction in intestinal disease, and alteration of the gut microbiome. This mechanistic understanding is partially based on findings from broad-spectrum antibiotic use in children with HIV (see Jones et al. 2014).

        • “The increased childhood mortality associated with undernutrition is almost entirely due to the elevated risk of death from common infectious diseases such as pneumonia, diarrhea, and bacterial sepsis. . . . A vicious cycle between malnutrition and infection has been long recognized. Episodes of infection potentiate undernutrition via anorexia, reduced nutrient absorption, nutrient losses (such as vitamin A and proteins in diarrhea), diversion of nutrients to inflammatory responses, and tissue repair. Diarrhea is associated with malabsorption and marked losses of protein, vitamin A, zinc, and other micronutrients. All infections are associated with net protein loss with diversion of amino acids to acute phase and immune response proteins. Activation of inflammatory cascades also causes reduced appetite and loss of lean tissue and fat. Thus, episodes of infection, especially diarrhea, result in both linear and ponderal growth-faltering.” Jones et al. 2014, p. S65.
        • “The mechanisms by which these trials not only reduced mortality, but improved growth, are not clear. They may include treatment of active (but covert) infection; prevention of colonizing microorganisms causing disease; reduction of inflammatory responses, resulting in less nutrient diversion and less cytokine-mediated impairment of growth through hormonal control; nonantibiotic effects, including direct anti-inflammatory effects; reduction in enteropathy; and alterations in gut microbiome.” Jones et al. 2014 (on trials of antibiotics in children with HIV), p. S67. Also see Figure 3 on p. S68.

      • 24

        "Ready-to-use Therapeutic Food (RUTF) has revolutionized the treatment of severe malnutrition – providing foods that are safe to use at home and ensure rapid weight gain in severely malnourished children. The advantage of RUTF is that it is a ready-to-use paste which does not need to be mixed with water, thereby avoiding the risk of bacterial proliferation in case of accidental contamination. The product, which is based on peanut butter mixed with dried skimmed milk and vitamins and minerals, can be consumed directly by the child and provides sufficient nutrient intake for complete recovery. It can be stored for three to four months without refrigeration, even at tropical temperatures." WHO, "Malnutrition"

      • 25
        • The Optimizing treatment for acute MAlnutrition (OptiMA) strategy uses MUAC < 125 mm or nutritional oedema as admission criteria and optimizes RUTF by adapting doses to the degree of malnutrition. RUTF dose for MUAC < 115 mm or oedema is 170 kcal/kg/d and progressively reduces to 75 kcal/kg/d as MUAC increases. The Combined Protocol for Acute Malnutrition Study (ComPAS) uses the same eligibility criteria like OptiMA, but simplifies more the RUTF ration by providing 1000 kcal/d for children with oedema or MUAC < 115 mm and 500 kg/d for children with MUAC 115-124 mm. Children are considered recovered if they have 2 consecutive weekly MUAC measures ≥ 125 mm." U.S. National Library of Medicine, ClinicalTrials.gov, "Optimizing Acute Malnutrition Management in Children Aged 6 to 59 Months in Niger (OptIMA Niger)," 2021, Study Description.
        • ComPAS: “The Combined Protocol for Acute Malnutrition Study (ComPAS) unified the treatment of uncomplicated SAM and MAM for children 6–59 months into one protocol, with simplified diagnostic criteria and a single therapeutic food product. . . . Several recent studies of children with SAM in outpatient settings indicate that recovery with a reduced dosage is similar to recovery with standard treatment. . . . The rationale for a reduced dosage is to facilitate increased coverage—and in turn increased public health impact—of treatment in a resource-constrained environment. The optimal dosage achieves the right balance between meeting individual energy needs and extending treatment to more children. This study contributes to the evidence on the impact of different dosage regimes. This study aimed to test the hypothesis that the combined protocol would be non-inferior to the standard protocol in terms of recovery, and improve cost-effectiveness." Bailey et al. 2020, pp. 4-5.
        • OptiMA: “More intelligent use of the costliest input (RUTF) by gradually reducing the dosage based on a child’s MUAC status and weight to increase the number of children with access to treatment at no extra or similar cost.” Phelan 2019, p. 40.

      • 26

        “The assessment of the risk of death associated with different degrees of wasting can be carried out only by community based longitudinal studies with a follow up of untreated malnourished children. This can be analysed only from a limited number of existing studies. For ethical reasons, these observational studies cannot be repeated, as an effective community-based treatment of severe acute malnutrition is now possible.” WHO, "WHO child growth standards and the identification of severe acute malnutrition in infants and children," 2009, p. 4, footnote 1.

      • 27

        This work was conducted by GiveWell senior advisor David Roodman and is described in the following report: David Roodman, On the association between anthropometry and mortality in children, 2022.

      • 28

        See the additional inputs into our calculations here.

      • 29

        Percent mortality reduction can be calculated by taking the inverse of the mortality ratios in table 10 of David’s report, and subtracting them from 1.
        Using Niger as an example, the mortality ratio for NGO-supported malnutrition treatment vs. no treatment is 1.71 for MAM and 3.23 for SAM.
        Mortality reduction from MAM treatment: 1 - (1 / 1.71) = 0.42 (0.58 relative risk of mortality with MAM treatment)
        Mortality reduction from SAM treatment: 1 - (1 / 3.23) = 0.69 (0.31 relative risk of mortality with SAM)
        David Roodman, On the association between anthropometry and mortality in children, 2022, table 10, p. 45.

      • 30
        • ALIMA, conversation with GiveWell, July 26, 2022 (unpublished).
        • “ALIMA treats a large number of malaria cases. Children who seek treatment for malnutrition are automatically given a malaria rapid test. Half of the children admitted for malnutrition in ALIMA’s program in Kamwesha, DRC had a positive test. Some of those children were asymptomatic, and parents would not have known to seek treatment without the screening.” GiveWell's non-verbatim summary of a conversation with ALIMA, March 5, 2021, p. 6.

      • 31
        • “In response to infection, the immune system first executes innate and then subsequently acquired host defense functions of high diversity. Both processes involve activation and propagation of immune cells and synthesis of an array of molecules requiring DNA replication, RNA expression, and protein synthesis and secretion, and therefore consume additional anabolic energy. Mediators of inflammation further increase the catabolic response. . . . Consequently, the nutritive status of the host critically determines the outcome of infection. Apart from deficiencies in single nutrients, such as vitamins, fatty acids, amino acids, iron, and trace elements, undernourishment based on [protein energy malnutrition] PEM greatly increases susceptibility to major human infectious diseases in low-income countries, particularly in children.” Schaible and Kaufmann 2007, p. 0806.
        • “Undernourished children principally die of common infections, implying that mortality is related to underlying immunodeficiency, even in mild forms of undernutrition. . . . The precise nature of immunodeficiency in undernutrition therefore remains uncertain; however, the consensus from the available evidence is that both innate and adaptive immunity are impaired by malnutrition. Defects in innate immune function include impaired epithelial barrier function of the skin and gut, reduced granulocyte microbicidal activity, fewer circulating dendritic cells, and reduced complement proteins, but preserved leukocyte numbers and acute phase response. Defects in adaptive immune function include reduced levels of soluble IgA in saliva and tears, lymphoid organ atrophy, reduced delayed-type hypersensitivity responses, fewer circulating B cells, a shift from Th1-associated to Th2-associated cytokines, and lymphocyte hyporesponsiveness to phytohemagglutinin, but preserved lymphocyte and immunoglobulin levels in peripheral blood. Despite this, most malnourished children seem to respond adequately to vaccination, although the timing, quality, and longevity of vaccine-specific responses may be impaired.” Bourke, Berkley, and Prendergast 2016, pg. 386-87.
        • Caulfield et al. 2004 measured the correlation between nutritional status (weight-for-age z-score) in children and death from any cause, diarrhea, pneumonia, malaria, and measles. All-cause mortality and mortality from all four infectious diseases were correlated with nutritional status. “The RR of mortality because of low weight-for-age was elevated for each cause of death and for all-cause mortality. Overall, 52.5% of all deaths in young children were attributable to undernutrition, varying from 44.8% for deaths because of measles to 60.7% for deaths because of diarrhea.” Caulfield et al. 2004, Abstract.
        • The most recent Cochrane meta-analysis of RCTs of vitamin A supplementation reports that vitamin A supplementation in areas with a high prevalence of vitamin A deficiency reduces all-cause mortality in children. “A meta-analysis for all-cause mortality included 19 trials (1,202,382 children). At longest follow-up, there was a 12% observed reduction in the risk of all-cause mortality for VAS compared with control using a fixed-effect model (risk ratio (RR) 0.88, 95% confidence interval (CI) 0.83 to 0.93; high-certainty evidence).” Imdad et al. 2022, Abstract.

      • 32

      • 33
        • ALIMA, conversation with GiveWell, July 26, 2022 (unpublished).
        • “ALIMA treats a large number of malaria cases. Children who seek treatment for malnutrition are automatically given a malaria rapid test. Half of the children admitted for malnutrition in ALIMA’s program in Kamwesha, DRC had a positive test. Some of those children were asymptomatic, and parents would not have known to seek treatment without the screening.” GiveWell's non-verbatim summary of a conversation with ALIMA, March 5, 2021, p. 6.

      • 34

        See Pelletier et al. 1994, Fishman et al. 2004, Black et al. 2008, and Olofin et al. 2013.

      • 35
        • “The mean and SE of relative risk for severe malnutrition is 8.4 +/- 2.1, for moderate malnutrition it is 4.6 +/- 0.9, and for mild malnutrition it is 2.5 +/- 0.3.” Pelletier et al. 1994, abstract.
        • Table 2.6 of Fishman et al. 2004 presents the relative risk of all-cause mortality for different weight-for-age categories. Fishman et al. 2004, Table 2.6, p. 64.
        • Table 2 of Black et al. 2008 presents odds ratios for all-cause mortality for different weight-for-age categories. Black et al. 2008, Table 2, p. 247.
        • “Pooled mortality HR was 1.52 (95% Confidence Interval 1.28, 1.81) for mild underweight; 2.63 (2.20, 3.14) for moderate underweight; and 9.40 (8.02, 11.03) for severe underweight. Wasting was a stronger determinant of mortality than stunting or underweight. Mortality HR for severe wasting was 11.63 (9.84, 13.76) compared with 5.48 (4.62, 6.50) for severe stunting.” Olofin et al. 2013, abstract.

      • 36
        • “Pooled analysis involving children 1 week to 59 months old in 10 prospective studies in Africa, Asia and South America. Utilizing most recent measurements, we calculated weight-for-age, height/length-for-age and weight-for-height/length Z scores, applying 2006 WHO Standards and the 1977 NCHS/WHO Reference. We estimated all-cause and cause-specific mortality hazard ratios (HR) using proportional hazards models comparing children with mild (−2≤ Z <−1), moderate (−3≤ Z <−2), or severe (Z <−3) anthropometric deficits with the reference category (Z ≥−1).” Olofin et al. 2013, Abstract.
        • "[W]e collated and analyzed data from 10 large prospective studies in low- and middle-income countries." Olofin et al. 2013, p. 2.

      • 37

        See Olofin et al. 2013, p. 5, table 3, section "WHO 2006," column "HR (95% CI)."
        The specific figures depend on which analysis we select from the paper (tables 3 and 4), but the estimates are all fairly similar. Olofin et al. 2013.

      • 38

        “Hazard is similar in notion to risk, but is subtly different in that it measures instantaneous risk and may change continuously (for example, one’s hazard of death changes as one crosses a busy road). A hazard ratio describes how many times more (or less) likely a participant is to suffer the event at a particular point in time if they receive the experimental rather than the comparator intervention.” Higgins, Li, and Deeks, Cochrane Handbook for Systematic Reviews of Interventions, Chapter 6: Choosing effect measures and computing estimates of effect, 2022.

      • 39

        See the "Counterfactual: mortality rates from untreated malnutrition" sections on the "CEA - Combined Protocol CMAM" and "CEA - Standard CMAM" sheets of our previous model here.

      • 40

        “The present review works with five of the Olofin data sets, as well as one from the Democratic Republic of Congo (Van den Broeck, Eeckels, and Vuylsteke 1993; Schwinger et al. 2019).” David Roodman, On the association between anthropometry and mortality in children, 2022, p. 1.

      • 41
        • Olofin et al. 2013 and David Roodman’s work both generated mortality hazard ratios for WHZ categories using cox regressions for Nepal, Philippines, Guinea-Bissau, and Senegal (both analyses included additional countries as well but the overlap in countries was limited to the four listed). The estimates are “recognizably similar” but not identical. See Olofin et al. 2013, Supporting Information, Table S2(C), and David Roodman, On the association between anthropometry and mortality in children, 2022, p. 20, figure 10.
        • “I include the Olofin controls and, as reported in the top and bottom halves of the figure, run regressions on the both the multispell and single-spell data. The former constitute my closest replication of the WHZ regressions in Olofin et al. (2013, Table S2(C)). Original and replication are recognizably similar: for example, the highest-to-lowest hazard ratio for the Philippines is 39.0 there and 33.3 here—in both cases the largest of all.” David Roodman, On the association between anthropometry and mortality in children, 2022, p. 18.

      • 42

        “The present review works with five of the Olofin data sets, as well as one from the Democratic Republic of Congo (Van den Broeck, Eeckels, and Vuylsteke 1993; Schwinger et al. 2019). The main finding is that aging of measurement is likely a major and unrecognized source of disagreement across data sets in the size of the relationship between anthropometry and death. When standard methods are applied to the data set with the most frequent follow-up—Adair et al. (1993), from the Philippines, with bimonthly measurements of children—the methods are assessing the risks associated with anthropometry taken 0–2 months before, or 1 month ago on average. When they are applied to the data from Senegal (Garenne et al. 2000), most of which was gathered at 6 months intervals, the methods are assessing a different set of risks, those associated with anthropometry taken on average at least 3 months ago. Just as a Covid test result quickly loses relevance, the hazard ratios associated with indicators of wasting decline with time since measurement.” David Roodman, On the association between anthropometry and mortality in children, 2022, pp. 1-2.

      • 43

        “Hazard is similar in notion to risk, but is subtly different in that it measures instantaneous risk and may change continuously (for example, one’s hazard of death changes as one crosses a busy road). A hazard ratio describes how many times more (or less) likely a participant is to suffer the event at a particular point in time if they receive the experimental rather than the comparator intervention.” Higgins, Li, and Deeks, Cochrane Handbook for Systematic Reviews of Interventions, Chapter 6: Choosing effect measures and computing estimates of effect, 2022, 2022.

      • 44

        “Again, where children are followed up on more quickly, hazard ratios appear higher. All of these results suggest that frequency of follow-up is a major determinant of apparent cross-country differences in hazard ratios. It appears that much of the cross-country variation within published meta-analyses is an unrecognized artifact of aging of measurement.” David Roodman, On the association between anthropometry and mortality in children, 2022, p. 22.
        See figures 11-13 in David’s report (pp. 23-24), which illustrate the dependence of hazard ratios on measurement frequency.

      • 45

        There are two reasons for this. First, most of the input data we use are annualized and our CEAs often use one-year increments to calculate cost-effectiveness, so it is convenient to have an annualized figure. Second, we want to estimate the risk of death over the entire one-year period we are modeling, not a specific time within that year.

      • 46

        Figures 11-13 (pp. 23-24) of David’s report illustrate that when the average follow-up period is modified within data sets, it has a large impact on the resulting mortality hazard ratio.
        “All of these results suggest that frequency of follow-up is a major determinant of apparent cross-country differences in hazard ratios. It appears that much of the cross-country variation within published meta-analyses is an unrecognized artifact of aging of measurement.” David Roodman, On the association between anthropometry and mortality in children, 2022, p. 22.

      • 47

        “The other approach retains all the timing information and marshals a more sophisticated estimation model that, to the extent that its mathematical assumptions hold, is robust to changes in follow-up frequency. This is the stochastic process survival model of Aalen (1994). I call it the MIG model because it generates a mixture inverse Gaussian distribution for time of failure. The model generates unobserved heterogeneity by endowing subjects with 'health trajectories' with random slopes within each episode. Hazards and hazard ratios fluctuate in time since last measurement, as subjects with downward slopes experience a burst of mortality early on and the survivors then escape most risk of death.” David Roodman, On the association between anthropometry and mortality in children, 2022, p. 2.

      • 48

        "For the sake of consistent interpretation, and with an eye toward assessing the impact of CMAM programs, I focus on ratios in cumulative mortality in the year following measurement that are predicted by fitted models." David Roodman, On the association between anthropometry and mortality in children, 2022, p. 2.

      • 49

        For the mortality ratios calculated by the MIG model, see David Roodman, On the association between anthropometry and mortality in children, 2022, p. 45, table 9.
        For the Olofin et al. hazard ratios, see Olofin et al. 2013, p. 5, table 3, section "WHO 2006," column "HR (95% CI)."

      • 50

        Olofin et al. 2013 uses 10 data sets and David uses six.

      • 51

        “The meta-analytic bottom lines for the one-year mortality ratios are 5.61 for WHZ = –3.5 vs. –0.5 and 1.81 for WHZ = –2.75 vs. –1.75. The next two figures do the same for MUAC, and produce similar overall averages despite the changes in metric and sample: 4.56 for MUAC = 11cm vs. 14cm and 1.73 for MUAC = 11.75 vs. 12.75cm.” David Roodman, On the association between anthropometry and mortality in children, 2022, p. 37.

      • 52

        See the confidence intervals in Figure 27, p. 41, of David Roodman, On the association between anthropometry and mortality in children, 2022.
        Also see the standard error values (in parentheses) for the mortality ratios in Table 9. David Roodman, On the association between anthropometry and mortality in children, 2022, p. 45, table 9.

      • 53

        “GiveWell requested data from the corresponding authors of Olofin et al. (2013) as well as of the underlying studies. We obtained data for five of the ten: Adair et al. (1993) from the Philippines, Garenne et al. (2000) from Senegal, Katz et al. (1989) from Indonesia, Mølbak et al. (1992) from Guinea-Bissau, and West et al. (1991) from Nepal. We also obtained the data for one study not in Olofin et al., that of Van den Broeck, Eeckels, and Vuylsteke (1993), which took place in what is now the Democratic Republic of Congo. That data set is in the public archive for Schwinger et al. (2019).” David Roodman, On the association between anthropometry and mortality in children, 2022, pp. 3-4.

      • 54

        For example, in Niger annual under-5 child mortality rates are estimated to have fallen from approximately 7.3% in 1990 to approximately 2.4% in 2019. IHME, GBD Compare tool, under-5 all cause mortality in Niger.

      • 55

        An absolute mortality rate would be, for example, an observation that 10% of children with untreated SAM die in the year after developing SAM. If, in this same context, 2% of age-matched children without malnutrition die in a year, the mortality ratio between the two categories of children would be 5.

      • 56

        This understanding is based on unpublished communications with the authors of Olofin et al. 2013 and the underlying studies.

      • 57

        We conducted a rough, unpublished analysis of the impact of possible treatment on SAM mortality ratios, based on communications with the authors of Olofin et al. 2013 and the underlying studies.

      • 58

        See the adjustment detailed in the cell note here.

      • 59

        “To check the MIG results I return to the simpler approach that emerged in the analytical narrative: a probit model fit to single-spell data sets. As in the MIG modeling, I enter anthropometric variables quadratically, control for age and gender, and include the Olofin controls where available. Since timing information for death is not needed, the Indonesia data are now incorporated.
        Parameter estimates and model-based one-year risk ratios appear in Table 6 and Table 7. The forest plots for WHZ are in Figure 27 and Figure 28 and those for MUAC in Figure 29 and Figure 30. All follow the formats of corresponding MIG displays. The meta-analytic bottom lines change remarkably little, coming modestly lower. The mortality ratio for the narrower comparison is 1.58 for WHZ and 1.56 for MUAC. Given all the differences in sample, data structure, and method, this concordance is reassuring.” David Roodman, On the association between anthropometry and mortality in children, 2022, p. 39.
        For the overlap in confidence intervals, see Figures 23 and 24 (MIG model) and Figures 27 and 28 (probit model) for WHZ, and Figures 25 and 26 (MIG model) and Figures 29 and 30 (probit model) for MUAC. David Roodman, On the association between anthropometry and mortality in children, 2022, pp 38-42.

      • 60

        See a list of the experts we spoke with here.

      • 61

        Megan Higgs, Summary of concerns with modeling of malnutrition mortality effect, April 2022 (unpublished)

      • 62

        Mark Myatt, Conversation with GiveWell, April 26, 2022 (unpublished)

      • 63

        More precisely, we:

        • Adjust under-5 all-cause mortality rates for the age distribution of CMAM admissions, using program admissions data and Demographic and Health Surveys (DHS) microdata on the age distribution of mortality.
        • Estimate mortality ratios for malnourished children (including both SAM and MAM), by calculating the average of SAM and MAM mortality ratios, weighted by local prevalence of untreated SAM and MAM.
        • Calculate mortality rates among non-malnourished and treated malnourished children, on the basis of the two inputs above. Note that we assume mortality rates among non-malnourished and treated malnourished children are the same, for simplicity.
        • Calculate mortality rates among untreated malnourished children, by multiplying the above by SAM and MAM mortality ratios from the MIG model.

        See calculations here.

      • 64

        See our raw mortality estimates here and our plausibility adjustment here (described in more detail below). The adjusted figures of 3% to 4% and 8% to 10% result from multiplying the raw mortality estimates by the plausibility adjustment from our ceiling analysis.

      • 65

        This analysis was first suggested by our consultant Megan Higgs as a way to obtain realistic bounds on the number of deaths averted as part of a larger strategy of triangulating toward a “best guess" using multiple methods.

      • 66

        The ceiling analysis suggests our estimate of mortality rates is unrealistically high in most locations we have modeled. Thus we apply a plausibility adjustment (see here) based on the ceiling analysis.

      • 67

        See the plausibility adjustment we make here and the calculation of the internal and external validity adjustment spelled out in the cell note here.

      • 68

        We estimate local under-5 mortality rates on the basis of recent population surveys and IHME estimates, adjusted for local food insecurity projections. We then adjust those mortality rates on the basis of Demographic and Health Surveys (DHS) data to estimate mortality rates among 6-59 months old children. For reference, increasing both mortality rates and prevalence by the same factor leaves the adjustment unchanged.

      • 69

        “The derivation of the incidence correction factor is based on the simple mathematical relationship between incidence, prevalence and average duration of an episode of illness. Assuming prevalence is low and incidence and duration of the illness do not vary over time, incidence can be approximated as prevalence multiplied by the inverse of the average duration of an episode of illness (Equations 2-3). The inverse of the average duration of an episode (Kt) is known as the incidence correction factor, as it is shown here to ‘correct’ or be multiplied by an estimate of prevalence to arrive at an approximation of incidence.” Isanaka et al. 2021, p. 2.

      • 70

        This is the case for all locations we use as an example above. However, further unpublished work indicates there are locations where the mortality rates resulting from the ceiling analysis are higher than what we estimate based on the mortality ratios from the MIG model.

      • 71

        See our internal and external validity adjustment here.

      • 72

        More precisely: (a) the ceiling analysis is highly sensitive to malnutrition prevalence rates (mortality rate per child decreases as prevalence increases); (b) we expect mortality rates among children with an age distribution weighted to approximate CMAM admissions to be higher than all children in the 6-59 month-old bracket because children admitted to CMAM are on average younger (see here), and (in the relevant locations) mortality rates are higher among younger children (see here); (c) we lack information on prevalence rates for children with the age distribution of CMAM admissions; however, we expect prevalence for those ages to be higher than they are for the broader 6-59 month-old bracket, since we would guess this is what explains children admitted to CMAM being on average younger than the general population in the 6-59 month-old bracket.

      • 73

        The reason why we believe the second difference would imply the ceiling analysis should be higher than the mortality rates estimated on the basis of historical data is that we would expect the ratio between (1) the probability of death among those who will be malnourished at any point in the year and (2) the probability of death among those who will never be malnourished to be higher than the ratio between (a) the probability of death among those malnourished at that point in time and (b) the probability of death among those not malnourished at that point in time.
        That’s because we expect a malnutrition episode to last less than a year, so that children who are not malnourished at one point in time might become malnourished later in the year. When we group children on the basis of whether they have ever been malnourished during a one year period, we are separating children who suffered from one or more episodes from children who suffered no episodes at all. When we group children on the basis of whether they are malnourished at one point in time, we are comparing children who are ill at that point with children who are not ill at that point, but might become ill later on. This means that the ratio looking at a whole year (comparison between (1) and (2)) is likely to be higher than the ratio looking at the snapshot (comparison between (a) and (b)).

      • 74

        The incidence correction factor is the only input that we use in the ceiling analysis and not in our estimate of mortality rates based on historical evidence. However, we use the incidence correction factor as an input in our cost-effectiveness model, to estimate the number of cases treated, when we lack historical data. See caseload calculations using the incidence correction factor here.

      • 75

        See here in our ceiling analysis.

      • 76

        See the "Treatment effect" sheet of our CEA here.

      • 77

        See the "Treatment effect" sheet of our CEA here.

      • 78

        See our modeling of the impact of malnutrition treatment programs on WHZ here.

      • 79

        Percent mortality reduction can be calculated by taking the inverse of the mortality ratios in table 10 of David’s report, and subtracting them from 1.
        Using Niger as an example, the mortality ratio for NGO-supported malnutrition treatment vs. no treatment is 1.71 for MAM and 3.23 for SAM.
        Mortality reduction from MAM treatment: 1 - (1 / 1.71) = 0.42 (0.58 relative risk of mortality with MAM treatment)
        Mortality reduction from SAM treatment: 1 - (1 / 3.23) = 0.69 (0.31 relative risk of mortality with SAM)
        David Roodman, On the association between anthropometry and mortality in children, 2022, table 10, p. 45.

      • 80

        There are at least three specific reasons to believe this:

        1. CMAM treatment programs include micronutrient supplementation, which could impact mortality risk independently of WHZ. For example, vitamin A is a component of RUTF (see the "Nutritional composition" table on page 6 of this WHO report), and vitamin A supplementation reduces mortality in children with low vitamin A status by reducing susceptibility to certain infectious diseases. This could cause the MIG model to underestimate the treatment effect. “At longest follow-up, there was a 12% observed reduction in the risk of all-cause mortality for VAS [vitamin A supplementation] compared with control using a fixed-effect model (risk ratio (RR) 0.88, 95% confidence interval (CI) 0.83 to 0.93; high-certainty evidence). Nine trials reported mortality due to diarrhoea and showed a 12% overall reduction for VAS [vitamin A supplementation] (RR 0.88, 95% CI 0.79 to 0.98; 1,098,538 children; high-certainty evidence).” Imdad et al. 2022, abstract.
        2. CMAM treatment programs include additional medical care such as antibiotic treatment, which could impact mortality risk independently of WHZ. This could cause our current model to underestimate the treatment effect. “Few studies have directly targeted immune pathways in malnourished children; however, standard protocols for SAM treatment include antibiotics, which can reduce mortality and improve nutritional recovery. The mechanisms through which antibiotics improve outcomes in malnutrition are unclear, but may include treating clinical and subclinical infections, reducing chronic inflammation or ameliorating enteropathy through changes in the microbiota; antibiotic effects on immune function have not been evaluated.” Bourke, Berkley, and Prendergast 2016, p. 396. “The overall annual mortality rate was 14.6 deaths per 1000 person-years in communities that received azithromycin (9.1 in Malawi, 22.5 in Niger, and 5.4 in Tanzania) and 16.5 deaths per 1000 person-years in communities that received placebo (9.6 in Malawi, 27.5 in Niger, and 5.5 in Tanzania). Mortality was 13.5% lower overall (95% confidence interval [CI], 6.7 to 19.8) in communities that received azithromycin than in communities that received placebo (P<0.001)[.]” Keenan et al. 2018, abstract.
        3. Elevated mortality risk associated with malnutrition in the data sets underlying the MIG model may not be entirely causally attributable to malnutrition per se. It could be partially a result of confounding by socioeconomic conditions or other factors. This would cause our current model to overestimate the treatment effect.

      • 81

        This understanding is based on unpublished communications with the authors of Olofin et al. 2013 and the underlying studies.

      • 82

        We conducted a rough, unpublished analysis of the impact of possible treatment on SAM mortality ratios, based on communications with the authors of Olofin et al. 2013 and the underlying studies.

      • 83

        "Sticking with the simple approach here, I reorganize the data so as to represent each child with a single spell starting with the first measurement after reaching six months of age and running one year. Children are only retained in the data if they are followed for at least that long, or die before then. Subsequent measurements within these 12 months are ignored, as is any information pertaining to events happening after." David Roodman, On the association between anthropometry and mortality in children, 2022, p. 11.

      • 84

        We roughly adjust for socioeconomic confounding, malnutrition treatment in historic cohorts, and frequency of malnutrition classification in our internal validity adjustment for mortality effects here.
        We roughly adjust for additional health benefits from malnutrition treatment programs, such as those resulting from provision of antibiotics and micronutrient supplementation, in our excluded effects adjustments here.

      • 85

        "In addition to directly treating malnourished children, ALIMA engages in some or all of the following activities, depending on the context:

        • Conducting training and coaching for Ministry of Health staff
        • Hiring and training new staff
        • Training caregivers in identifying malnutrition by measuring children’s mid-upper arm circumference (MUAC)
        • Determining the level of acute malnutrition, especially among displaced populations
        • Supporting the referral system
        • Supporting screening at the community level by community health workers
        • Improving water and sanitation as well as infection prevention and control
        • Raising awareness about malnutrition and feeding practices among caregivers"

        GiveWell's non-verbatim summary of a conversation with ALIMA, March 5th, 2021

      • 86
        • For example, ALIMA’s programs in Niger and Nigeria include routine vaccinations.
        • “A wealth of evidence stresses that immunisation and nutrition interventions complement each other, making the case for stronger integration. They also indicate that integrated efforts to reach vulnerable children and missed communities with these essential services will be key to breaking the vicious cycle of malnutrition and preventable diseases, an imperative condition for countries to 'ensure healthy lives and promote wellbeing for all at all ages' in line with SDG 3.” Gavi and Scaling Up Nutrition, Equity From Birth: An integrated approach to immunisation and nutrition policy brief, October 2021, p. 5.

      • 87

        “We estimate that the vaccines incentivized (either directly or indirectly) by New Incentives reduce recipients' likelihood of contracting the diseases targeted by 76%.
        This estimate is primarily based on meta-analyses of the effects of vaccines incentivized by New Incentives. A weighted average of vaccines' efficacy against targeted diseases and those diseases' contribution to child mortality in Nigeria yields an overall efficacy of 67%.” GiveWell, New Incentives (Conditional Cash Transfers to Increase Infant Vaccination)

      • 88

        See our calculations of the vaccination benefits from ALIMA's malnutrition program here. Our New Incentives CEA is part of our main CEA here.

      • 89

        The units of value generated from fully vaccinating one child are calculated from the New Incentives CEA, by dividing total units of value generated after adjusting for leverage and funging (4,578) by the increase in the number of children who are treated as a result of the intervention (1,123). This yields ~4 units of value per fully vaccinated child equivalent. These units of value come mostly, but not entirely, from preventing deaths. See calculations here.
        In this context, "vaccination" is defined as having received all recommended vaccines, or partial vaccinations of more than one child that add up to a full vaccination equivalent. This is detailed in our New Incentives charity page here.
        In our main CEA, we assume that the value of averting the death of an individual under five from vaccine-preventable diseases is 116 (i.e., 116 times the value of doubling consumption for one person for one year). See here. 116/4 = 29.

      • 90

        Admitted children are tested for malaria at least once using rapid diagnostic tests, and the program plans for 1.25 tests per admitted child, according to ALIMA. Children who test positive receive malaria treatment. Kevin Phelan, Nutrition Advisor, ALIMA, email to GiveWell, August 9, 2022 (unpublished)

      • 91

        For malaria mortality rates in Southern Niger and Northern Nigeria, see this sheet. In ALIMA’s programs in Dakoro and Mirriah, Niger, we estimate that malaria testing and treatment accounts for 6% of CMAM program benefits.

      • 92
        • “This study uses the malaria-eradication campaigns in the United States (circa 1920), and in Brazil, Colombia and Mexico (circa 1955) to measure how much childhood exposure to malaria depresses labor productivity. The campaigns began because of advances in health technology, which mitigates concerns about reverse causality. Malarious areas saw large drops in the disease thereafter. Relative to non-malarious areas, cohorts born after eradication had higher income as adults than the preceding generation. These cross-cohort changes coincided with childhood exposure to the campaigns rather than to pre-existing trends.” Bleakley 2010, abstract.
        • “We examine the effects of exposure to malaria in early childhood on educational attainment and economic status in adulthood by exploiting geographic variation in malaria prevalence in India prior to a nationwide eradication program in the 1950s. We find that the program led to modest increases in household per capita consumption for prime age men, and the effects for men are larger than those for women in most specifications.” Cutler et al. 2010, abstract.
        • “This study estimates long-run impacts of a child health investment, exploiting community-wide experimental variation in school-based deworming. The program increased labor supply among men and education among women, with accompanying shifts in labor market specialization. Ten years after deworming treatment, men who were eligible as boys stay enrolled for more years of primary school, work 17% more hours each week, spend more time in non-agricultural self-employment, are more likely to hold manufacturing jobs, and miss one fewer meal per week. Women who were in treatment schools as girls are approximately one quarter more likely to have attended secondary school, halving the gender gap. They reallocate time from traditional agriculture into cash crops and non-agricultural self-employment.” Baird et al. 2016, abstract.
        • GiveWell senior advisor David Roodman discusses the randomized controlled trial reporting that deworming increases adult income (Baird et al. 2016) here.

      • 93

        “In studies that provided nutritional supplements, we observed positive significant pooled effect sizes on all five outcomes of length-for-age z-score or height-for-age z-score (effect size 0.05, 95% CI 0.01–0.09; p=0.01; n=50), cognitive or mental (0.06, 0.03-0.10; p<0.01; n=38), language (0.08, 0.03–0.13; p=0.01; n=21), motor (0.08, 0.04–0.12; p<0.01; n=41), and social-emotional (0.07, 0.02–0.12; p=0.01; n=20) scores.” Prado et al. 2019, p. e1398.
        "Additionally, we stratified nutrition supplementation studies by setting, timing, and type. Small, but significant, positive pooled effects on both LAZ or HAZ and development scores were found in nutrition studies that were done in low-income and middle-income countries, but not in those done in high-income countries[.]" Prado et al. 2019, p. e1399. “Of the 75 studies included, 61 were done in low-income or middle-income countries and 14 in high-income countries.” Prado et al. 2019, pg. e1404.

      • 94

        See the "Development effects" section of our CEA here.

      • 95

        “This review, using a public health focused approach, aimed to understand and describe the current status of AMR [antimicrobial resistance] in Africa in relation to common causes of infections and drugs recommended in WHO treatment guidelines. . . . Overall resistance to commonly used drugs, like amoxicillin (MR 72.9%, IQR9.1%–87.3%) and trimethoprim/sulfamethoxazole (MR 75.0%, IQR 49.5%–92.3%) was high. . . . The high levels of resistance to amoxicillin and penicillin in S. pneumoniae and H. influenzae are also concerning given that pneumonia is a leading cause of death in children.” Tadesse et al. 2017, pp. 1, 3-4, 7.

      • 96

        “Stool samples were collected from children with SAM between 6 and 60 weeks of age who received either one week of amoxicillin or placebo (n=164). The children were followed for 12 weeks with longitudinal sampling, and a subset were followed out to 2 years. . . . The use of amoxicillin to treat uncomplicated SAM has therapeutic benefits visible by anthropometry and by content of the gut microbiota. The main concern with the use of prophylactic antibiotics for this purpose is the effect on antibiotic resistance gene enrichment in the children’s microbiota. This concern was not supported here. The benefit/cost ratio for the use of prophylactic antibiotics for individuals in this cohort is positive when weighing effects on anthropometry, microbiome, and antibiotic resistance.” Langdon et al. 2018, Abstract.

      • 97

        See the drug resistance adjustment here.

      • 98

        See Das et al. 2020, Supplementary Materials, Figure 15: Forest plot for the impact of RUTF compared to other foods on Adverse Events.

      • 99

        “A total of 924 children were randomly assigned to the amoxicillin group, 923 to the cefdinir group, and 920 to the placebo group. . . . No cases of severe allergy or anaphylaxis were identified. A total of three adverse events that were presumed to be drug reactions were reported: a generalized papular rash in a child who received amoxicillin, thrush in a child who received cefdinir, and bloody diarrhea that resolved spontaneously while treatment continued in a child who received cefdinir.” Trehan et al. 2013, pp. 428, 430.

      • 100

        “Children with uncomplicated severe acute malnutrition, not requiring to be admitted and who are managed as outpatients, should be given a course of oral antibiotic such as amoxicillin.” WHO, "Guideline: updates on the management of severe acute malnutrition in infants and children," 2013, p. 29.

      • 101

        “The most common adverse reactions (> 1%) observed in clinical trials of AMOXIL capsules, tablets or oral suspension were diarrhea, rash, vomiting, and nausea.” FDA, "AMOXIL (amoxicillin) capsules, tablets or powder for oral suspension," 2015, p. 1.

      • 102

        “Antibiotic exposure in children within the first 2 years of life was associated with current asthma (adjusted odds ratio [aOR] 1.72, 95% confidence interval [CI] 1.10–2.70), current atopic dermatitis (aOR 1.40, 95% CI 1.01–1.94), and current allergic rhinitis (aOR 1.65, 95% CI 1. 05–2.58) at 5 years of age. Analysis of the associations by type of antibiotics showed that cephem was associated with current asthma (aOR 1.97, 95% CI 1.23–3.16) and current rhinitis (aOR 1.82, 95% CI 1.12–2.93), and macrolide was associated with current atopic dermatitis (aOR 1.58, 95% CI 1.07–2.33).” Yamamoto-Hanada et al. 2017, Abstract.

      • 103

        “Antibiotic exposure in early life significantly increased risk of childhood overweight [relative risk (RR) = 1.23, 95% confidence interval (CI) 1.13–1.35, P < 0.001] and childhood obesity (RR = 1.21, 95% CI 1.13–1.30, P < 0.001). Antibiotic exposure in early life also significantly increased the z-score of childhood body mass index (mean difference: 0.07, 95% CI 0.05–0.09, P < 0.00001). Importantly, there was an obvious dose–response relationship between antibiotic exposure in early life and childhood adiposity, with a 7% increment in the risk of overweight (RR = 1.07, 95% CI 1.01–1.15, P = 0.03) and a 6% increment in the risk of obesity (RR = 1.06, 95% CI 1.02–1.09, P < 0.001) for each additional course of antibiotic exposure.” Shao et al. 2017, Abstract.

      • 104

        “The same countries that are receiving the most RUTF are those experiencing rapid changes in the nutrition landscape and increasing levels of obesity and chronic disease. The programs chosen to combat malnutrition must account for potential negative long-term effects across the lifespan that are relevant in the context of the nutrition transition and the double burden of over and undernutrition. Growth restriction in utero, rapid weight gain and undernutrition in childhood have shown some association with increased risk of obesity, diabetes, metabolic syndrome and associated chronic diseases in adulthood. Is it responsible to be introducing high-fat, high-sugar ‘nutrition’ supplements to these populations?” Bazzano et al. 2017, p. 7.

      • 105

        See that adjustment here.

      • 106

        See our figures and calculations here.

      • 107

        See ALIMA’s full budget for Niger here.

      • 108

        See our figures and calculations here.

      • 109

        The costs shouldered by other actors are accounted for via our leverage and funging adjustment here.

      • 110

        See here in our "Simple CEA" tab for this total cost per child reached.

      • 111

        See our figures and calculations here.

      • 112

        See our figures and calculations here.

      • 113

        See our adjusted mortality estimates here. The figures of 4% and 10% result from multiplying our raw mortality estimates by the ceiling analysis plausibility adjustment.

      • 114

        The relative risks are a bit lower for NGO treatment relative to government treatment. See our estimates for government malnutrition treatment here and NGO treatment here. The treatment effect expressed as percent reduction in mortality is calculated as 1 minus the relative risk of mortality. For MAM treatment: 1 - 0.58 = 0.42. For SAM treatment: 1 - 0.31 = 0.69.

      • 115

        This estimate incorporates our adjustment for internal and external validity. See here in our cost-effectiveness analysis.

      • 116

        See the figures here.

      • 117

        Our unmodeled program benefits and downsides yield a positive adjustment of 13%, and our leverage/funging adjustment is -31%. Combining these figures gives us a total adjustment of -22%. (1- (1.13 x (1-.31)) = .22). See figures and calculations for these unmodeled adjustments here.

      • 118

        See the adjustment we make for pediatric care support here.

      • 119

        See our cost-effectiveness estimate here.

      • 120
        • The cost per child treated for malnutrition in Dakoro and Mirriah is $67, whereas the cost per child treated with a full year of seasonal malaria chemoprevention (SMC) is around $6 to $9 in countries where we support SMC programs.
        • We estimate that malnutrition treatment reduces the annual risk of death in children with MAM or SAM by about 45% from a baseline mortality rate of 6%. In contrast, we estimate that seasonal malaria chemoprevention reduces annual all-cause mortality for children under five by around 10% (based on the ~75% effect of SMC on malaria-related mortality, the ~65% of malaria cases that occur during the rainy season, the ~13% of under-five deaths that occur due to malaria in sub-Saharan Africa, and the .75 indirect deaths we estimate are averted by SMC programs for every malaria death averted; 0.75 x 0.65 x 0.13 x 1.75 = ~10%). See our SMC cost-effectiveness analysis here for more details.

      • 121

        We asked implementing organizations to estimate caseloads based on historical data, then made several adjustments to account for future predictions.
        For example, in Niger, based on caseloads provided by ALIMA, we assume that the total malnutrition caseload corresponds to 25% of the total population between 6 and 59 months old. We assume that SAM cases constitute 25% to 37% of the total. For programs that treat MAM cases, we assume that SAM cases will go down as treatment of MAM cases increases as a result of the Combined Protocol, because children with MAM will not progress to SAM as often. We calculate the total number of MAM cases by subtracting SAM cases from total malnutrition caseload. Our estimates of caseload in Dakoro and Mirriah, Niger are lower than ALIMA’s by 23% (Our estimate is 77% of ALIMA's estimate).
        For Nigeria, we estimate SAM caseload in the first year of the program on the basis of the 2022 caseload, adjusted for ALIMA’s best guess of the increase in coverage. We assume that the number of SAM cases decreases year-to-year, due to the increased coverage of MAM cases. We estimate MAM cases on the basis of population, prevalence, coverage and the incidence correction factor (ICF). We then adjust MAM caseload to account for capacity constraints, based on ALIMA’s claim that treating all MAM cases would overwhelm the facilities it works in. Our estimate for Nigeria is 8% higher than ALIMA’s because we rely on historical data to estimate SAM, while ALIMA relies on a formula based on population and prevalence (Our estimate of malnutrition caseload in Nigeria is 108.3% of ALIMA's estimate; 108.3% - 100% = 8.3%).
        We assume the number of children treated is equal to 95% of cases treated, based on estimates provided by another NGO indicating Nigeria is at low risk of relapse (we assume the same is the case for neighboring Niger).

        See our calculations here.

      • 122

        See the "Caseload" sheet of our CEA here for our calculations of how many malnutrition cases would be treated without NGO support. We are currently funding surveys to estimate malnutrition coverage before and after NGO support, which we expect will help us refine this input in the future.

      • 123

        ALIMA has provided evidence that in an area of Niger where it implements malnutrition treatment programs, cumulative vaccination rates are about five percentage points higher than in a neighboring area where ALIMA does not work. See data and calculations here. This is a relatively weak form of evidence so we are uncertain about its accuracy. However, we think the size of the effect is plausible given the activities of the ALIMA program. We estimate that vaccination rates in Niger are about 82% (see here). A five percentage point increase in vaccination rate implies that the ALIMA program would reduce the number of unvaccinated children by about one quarter (1 - 0.82 = 0.18 ; 0.05 / 0.18 = 0.28).
        ALIMA has also provided unpublished evidence that in areas of Niger where it implements malnutrition treatment programs, the relative rate of vaccination in children under two is about twice as high as in neighboring areas where ALIMA does not work. This offers additional support to the idea that the intervention increases cumulative vaccination rates and implies that infants are also likely to be vaccinated closer to recommended ages in ALIMA areas, averting some of the deadly diseases that disproportionately impact infants when vaccination is delayed. The evidence we have does not allow us to quantify how much earlier this program causes children to be vaccinated on average, so we rely on a subjective guess of four months. See our calculations here.

      • 124
        • Niger Need: “In Niger, 350,000 to 400,000 children are treated for SAM every year; ALIMA estimates that ideal coverage would require doubling this figure.” GiveWell's non-verbatim summary of a conversation with ALIMA, November 13, 2020, pp. 1-2.
        • Global Need: A 2010 World Bank report estimated $2.6 billion in additional funding is needed to treat SAM, without considering costs of treating MAM (Horton et al. 2010, p. 31). This report used the optimistic assumption that SAM would fall by half because of successful prevention measures and estimated that treating 80% of SAM children would cost $2.6 billion more than already allocated. “Community-based management of severe acute malnutrition: This is an expensive intervention, which costs US$200 per child treated. The prevalence of severe acute malnutrition is 4.8 percent across the 36 countries in the 6–59 months age group (implying an incidence of 9.6 percent, using the incidence: prevalence ratio of 2:1). We assume that if all the other interventions are funded, that prevalence of severe acute malnutrition will fall to 50 percent of present levels (Isanaka et al. 2009, reporting on the effect of an intensive complementary feeding program). Unlike other interventions (where we aim for 100 percent coverage) we aim for 80 percent coverage, since there are no existing programs at scale achieving higher coverage, and we cost the intervention accordingly. Current coverage is approximately one million children (Stephane Doyon, Médecins Sans Frontières, personal communication). Additional annual cost = US$2.6 billion.” Horton et al. 2010, p. 31.

      • 125

        In order to sense-check our estimates of mortality rates among untreated malnourished children, we use a ceiling analysis. This model estimates maximum mortality rates for untreated malnourished children, given our best guess for the number of untreated malnourished children and overall mortality rates in the area, and an upper-bound estimate of the percentage of deaths in the population that occur among malnourished children.

      • 126

        We currently use a measure of food insecurity published by the Famine Early Warning Systems Network (FEWS NET) to adjust our estimates of all-cause under-5 mortality and malnutrition prevalence in children under five for current conditions (see here). This is because the latter two measures are usually reflective of conditions several years prior. However, this adjustment relies on crude estimates of the relationship between these variables.

      • 127

        Most of the studies we have reviewed come from contexts where children are chronically malnourished, but we have not investigated how this is likely to affect the cost-effectiveness of the program. Our prior is that, in crisis settings, children are more likely to be able to weather malnutrition or infection in the absence of the program, so the program's impact will be smaller relative to the counterfactual. However, we also believe children would require fewer rounds of treatment on average to avoid death from malnutrition (as the period of malnutrition is shorter in a temporary crisis), so the program would be lower cost too.

Based on our level of uncertainty about the best guesses calculated in our cost-effectiveness analysis, GiveWell staff gave their subjective 25th - 75th percentile confidence interval for each parameter. This column is an aggregation of these intervals. The implied cost-effectiveness column shows, for each parameter, what the program's overall cost-effectiveness would be at the 25th and 75th percent level of confidence, holding all other parameters constant.
We use multiples of direct cash transfers as a benchmark for comparing the cost-effectiveness of different programs.
(202,106 x (1-65%) x 5.9% x 45%) + (202,106 x 65% x 5.9% x 2%)
(Multiples of the value of direct cash transfers)
(2,047 x 119 / $7.9m) / 0.00335)
(9x / 66%) x (100% + 13%) x (100% - 31%)
($13.6m / $67)