Estimating effects on maternal and child undernutrition and child mortality
In order to estimate the overall indirect effects of COVID-19 on maternal and child undernutrition and child mortality and explore possible mitigation strategies we used a five-step process across three assumption scenarios (pessimistic, moderate, optimistic) (Table 1).
In the first step, we estimated changes in gross national income (GNI) per capita and poverty (% population<$1.90 per day) due to COVID-19 disruptions across 2020, 2021 and 2022 in 118 LMICs, using three different economic recovery scenarios. Second, we translated these different projections for GNI per capita into country-specific changes in the prevalence of different levels of child wasting and maternal low BMI status using a recent statistical analysis that linked growth shocks to wasting and low BMI risks. Third, we inputted the estimates of poverty and wasting, along with assumptions about levels of disruption of health and nutrition services, into the Lives Saved Tool (LiST) to estimate changes in under-five mortality, stunting and maternal anemia. Finally, we used the Optima-Nutrition model to explore possible strategies to mitigate the predicted increases in undernutrition and child mortality. The methods and inputs for each of these steps are detailed below.
For the first step we used MIRAGRODEP, a global computable general equilibrium model linked to country-specific household survey data, to predict the effects of COVID-19 disruptions on GNI per capita, household incomes and $1.90/day poverty rates between 2020-2022.38 MIRAGRODEP inputs include international trade and financial disruptions, as well as projected country estimates of COVID-19 cases and associated prevention measures (e.g. lockdowns). The economic channels of disruption included changes in labor force participation and labor productivity due to lockdown policies, increased morbidity and mortality, reduced efficiency of domestic and international transportation of goods, reduced consumer demand for in-person services (e.g. restaurants), and international shocks to trade and financial flows (e.g. reduced demand for oil). For 2020, we modelled a single GNI scenario, as there is less uncertainty on the extent of economic disruption. (MIRAGRODEP results for 2020 were compared to the most recent economic forecasts from a range of international agencies.) However, we opted to model three different scenarios for economic recovery from the pandemic in 2021 and 2022. The optimistic scenario assumed a fast V-shaped economic recovery with economic activity accelerating quickly from 2021 onwards. The moderate scenario assumed a second infection wave into 2021 results in a “stop-start” W-shaped recovery. The pessimistic scenarios assumed a protracted U-shaped recovery with continued economic disruptions in 2021 and most countries not returning to pre-COVID-19 per capita income levels by 2022 (Table 1). For this analysis, the magnitude of macroeconomic disruption for the three scenarios is measured as the percentage difference between GNI per capita with and without COVID-19, while impacts on poverty – which are also used as a proxy for food insecurity (see below) – are measured as the difference in $1.90/day poverty rates with and without COVID-19.
Second, we used elasticities from a recent study39 that linked economic growth (GNI) shocks to child wasting and maternal low BMI risks to generate predicted changes in wasting and low BMI prevalence. Specifically, that study linked 177 Demographic Health Surveys from 52 LMICs to national level GNI growth shocks, and uncovered statistically significant negative relationships between changes in GNI per capita and: 1) changes in the prevalence of any (<-1SD WHZ), moderate/severe (<-2SD WHZ), and severe (<-3SD WHZ) wasting among children 0-59 months old and 2) prevalence of low Body Mass Index (BMI, <18.5 kg/m2) among mothers 15-49 years old. The corresponding elasticities from that study were then applied to country-specific GNI growth projections from MIRAGRODEP, population projections and baseline undernutrition estimates from LiST to estimate increases in the number of children with wasting and number of women with low BMI as a result of the COVID-19 crisis.
LiST includes country-specific population projections and other baseline assumptions from publicly available data sources.40 This analysis included 118 LMIC with sufficient coverage data (Web Annex 4); together these countries include over 95% of both global under-five deaths and stunted children. Mean annual coverage reduction assumptions for five preventive interventions across the three scenarios are consistent with the analyses by Roberton et al18 and in line with disruptions reported elsewhere in 2020. (Table 1) We also included treatment of severe acute malnutrition (SAM) with similar coverage reduction assumptions but did not assume changes in coverage of treatment of moderate acute malnutrition (MAM), except in the optimistic scenario for 2022. For household food insecurity assumptions, which influence the type of complementary feeding intervention applied (public provision of complementary foods in food insecure households versus only nutrition counseling in food secure households), we used the annualized MIRAGRODEP < $1.90/day poverty projections.
As detailed in Table 1 we assumed that service disruptions would continue after initial lockdowns but would become less severe over time and in all three scenarios would recover to half of the 2020 drop in 2021. Possible sources of ongoing disruptions include 1) health workers becoming infected and/or are diverted to care with new virus waves, 2) reduced use of available services by the population due to perceived risk of exposure and 3) reduced financing over time due to economic impacts. For 2022, coverage levels in the pessimistic scenario remained lower than pre-COVID-19 baselines, while they returned to 2019 levels in the moderate scenario. For the optimistic scenario in 2022: (1) coverage for the majority of interventions was increased by 5% relative to 2019; (2) for treatment of MAM in children 6-59 months coverage increased to 10%; (3) multiple micronutrient supplements (MMS) replaced IFA during pregnancy; and (4) balanced protein-energy supplementation for pregnant women with low BMI was added at 10% coverage. We estimated the impacts of these last three interventions added in 2022 on adverse birth outcomes and under 5 child deaths.
Effects on long-term productivity
We used the Human Capital Approach to estimate lost productivity from the nutrition-related impacts of the COVID-19 pandemic. This approach calculates the present value of future productivity based on potential wages and potential time in the workforce.41 This approach is similar to that adopted for estimating the benefits of scaling up nutrition interventions in Shekar et al.42 For this study, we used GNI per capita as a measure of average future wages. Estimates of lost productivity were calculated separately for outcomes of stunting, nutrition-related child mortality, and maternal anemia. Total excess cases of each outcome were based on the results of the LiST modelling previously described.
For each child death, years of potential productivity were calculated as the difference between age 18 and the country-specific life expectancy or age 65, whichever is lower. The total number of children with potential lost productivity due to premature mortality was adjusted by the probability of death between age 5 and 18. Each year of productivity was valued using GNI per capita (current US$) adjusted for annual GNI growth, percent lifetime earnings that can be realized43 and labor share of income. Future lost productivity was discounted at 3% per year.44,45 For total lost productivity in each country, the lost productivity from each death is summed across all years of lost productivity and multiplied by the adjusted number of deaths.
The estimates of lost productivity for additional stunting cases used a similar approach as that described for child deaths. However, since stunting is expected to reduce future productive capacity for each child and not prevent future labor force participation, only this reduced productive capacity would be considered as potential lost productivity. Previous research estimated that decreased cognitive development due to early childhood malnutrition marked by stunting could lead to 21 percent reduction in adult earnings.46 Therefore, the approach described for estimating lost productivity due to premature deaths were applied to the additional number of stunting cases and multiplied by 21 percent to derive an estimate for the productivity lost attributable to these additional stunting cases.
Excess cases of anemia in pregnant women were translated into lost earnings due to decreased labor productivity based on the methods described in Horton and Ross (2003)46. The level of productivity lost due to anemia differs depending on whether employment is in heavy manual labor, light manual labor, or service industry, with attributed productivity reductions of 17%, 5%, and 4% respectively. For each country, the distribution of types of labor engaged in by women was derived from the World Development Indicators (WDI) database. The proportion of women employed in agriculture, industry, and services were categorized as heavy manual labor, light manual labor, and service labor, respectively. Lost productivity was calculated separately for each category of employment as a product of per capita GNI, number of additional cases of maternal anemia in each year, percent of female employment in the specific category, and the associated reduction in productivity in that category due to anemia. These estimates were adjusted for female labor force participation and labor share of income. Total lost productivity for each country was the sum of estimated productivity lost across the three employment categories and the three years (2020-2022) modelled, discounted at 3% per year.
Effects on the financing landscape for nutrition
To project how available donor and domestic resources for nutrition are expected to change across the three disruption scenarios, we developed a modelling approach based on Stuckler et al47. First, using Creditor Reporting System (CRS) data on official development assistance (ODA) from 2002-201848, we ran a panel regression to estimate the elasticity of aid to nutrition-specific and nutrition-sensitive sectors with respect to donor countries’ economic growth. Then, we multiplied that elasticity by the MIRAGRODEP GNI growth projections to estimate the expected levels of aid to those sectors between 2020 and 2030. We conducted an analogous procedure to project the effect of domestic economic growth on domestic government health spending, using National Health Accounts panel data from WHO’s Global Health Observatory49 for the 118 countries used in the Optima Nutrition model.
Estimating additional intervention costs and optimized nutrition budget allocations
The Investment Framework for Nutrition estimated that approximately $7 billion per annum was needed to reach the WHA targets for stunting, wasting, maternal anemia, and breastfeeding by 2025.42 Given the disruptions caused by the COVID-19 pandemic, we anticipate that more resources will be needed to address the increased burden of malnutrition. We estimated the additional resource needs to address the expected increased in stunting, wasting, and maternal anemia as the product of additional burden of each outcome estimated from the LiST modelling and the cost per case of stunting or anemia averted or case of severe wasting treated.
We used the Optima Nutrition model50 to illustrate how 2021-2022 financing could be optimally allocated across six nutrition interventions to maximize the number of “alive and non-stunted” children. Optima Nutrition is a cohort model that tracks children until five years of age, categorised according to their mother’s breastfeeding practices, family economic status, height-for-age (stunting) status and weight-for-height (wasting) status. Children in the model can die from a range of age-specific causes, with the relative risks of dying from each cause related to the child’s breastfeeding, stunting and wasting status. Interventions can improve nutritional outcomes directly or indirectly by reducing risk factors including birth outcomes and diarrhoea incidence. The model includes an optimisation algorithm, which can be used to incrementally shift a fixed amount of funding between interventions until it achieves a budget allocation that maximizes (or minimizes) a given objective. For this analysis the objective of maximizing alive and non-stunted children was selected to identify a priority mix of interventions to both reduce mortality from severe wasting and reduce stunting.
For each of the 118 LMICs, the total expenditure on the six nutrition interventions in 2019 and 2020 was estimated by multiplying coverage estimates from LiST from the moderate scenario (Table 1) by country-specific unit costs (Web Annex). Given evidence for a lagged decline in ODA starting in 2022 (Web Annex), we assumed total available financing for the nutrition interventions in 2021 and 2022 would equal total 2019 expenditure. The model was projected for 2020-2022 inclusive. The pre-optimization epidemiological projections and intervention coverages were aligned with the moderate scenario (Table 1) and were compared to outcomes when the funding was optimized to determine the maximum number of stunting cases and child deaths that could be averted through allocative efficiency.
Data availability statement:
Data that supported the final findings of this study have been deposited here: https://drive.google.com/drive/folders/1AVqEF7gYTGmJENlya3UbUxO6A2fKBQm6?usp=sharing
Code availability statement:
All software (modelling tools) used in the analyses of this study have been referenced before and are available here: (1)MIRAGRODEP: https://www.ifpri.org/publication/miragrodep-model; (2) The Lives Saved Tool (LiST): https://www.livessavedtool.org/resources ; (3) The Optima Nutrition model: https://optimamodel.com/nutrition/ .