Article Text
PDF
Abstract
Background Extreme weather events, or natural disasters, present a large and increasing threat to human health, infrastructure and food security, including in sub-Saharan Africa (SSA), where the burden of undernutrition is high. However, research about associations between natural disasters and undernutrition in early childhood is limited.
Methods We combined anthropometric data of children aged 0–59 months from 51 Demographic and Health Surveys datasets collected from 2010 to 2019 in 30 countries in SSA with information on natural disaster events (flood, drought, other) from the Emergency Events Database database to determine disaster exposure. The analytic sample included 320 479 children. We used generalised estimating equations to predict stunting, wasting and anaemia by disaster exposure and selected covariates.
Results Almost 20% (19.7%) of children under five were exposed to a natural disaster in the preceding year. In adjusted analysis, children exposed to at least one disaster in the preceding year had a relative risk (RR) of wasting 1.17 times higher than unexposed children (95% CI 1.12, 1.22). Adjusted models examining exposure to drought or flood consistently estimated higher risks of wasting post-disaster (drought RR 1.36, 95% CI 1.26, 1.47; flood RR 1.07, 95% CI 1.02, 1.12). RRs increased when using a 3-month exposure period. However, exposure to natural disaster was not consistently associated with significant differences in RR of stunting or anaemia.
Conclusion Natural disasters are prevalent in SSA. Given the high risk of wasting associated with disaster exposure, policymakers should prioritise interventions to address wasting in post-disaster settings.
- CHILD HEALTH
- NUTRITION
- ANAEMIA
- CLIMATE CHANGE
Data availability statement
Data are available upon reasonable request. Both datasets used in this analysis (Demographic and Health Surveys; EM-DAT international disaster database) are publicly and freely available. Note to editorial team: both DHS and EM-DAT are freely available on their own websites, but are not in specific data repositories. We are unable to report the datasets in the BMJ format. Thus, on recommendation from the editorial assistant, we have unticked 'public, open access repository'. We cited the data sources as the publishers request on their websites, so they can be easily located by readers.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Climate change and variability are associated with various health risks for children, such as increased infectious disease exposure, poorer respiratory health, increased prevalence of allergies, and undernutrition, particularly in lower- and middle-income country settings, where the risks of morbidity and mortality among children under 5 years of age are much greater relative to high-income countries.
Research in single countries or of single types of extreme weather events shows that extreme weather events like droughts, floods and storms—which are increasing in frequency due to climate change—are associated with higher risks of child malnutrition, including stunting, wasting and underweight.
WHAT THIS STUDY ADDS
This study is one of the first to examine population effects of extreme weather events on undernutrition cross-nationally in a region with increasing climate change and high rates of undernutrition, and one of the first to estimate the effects of extreme weather events on anaemia in early childhood. Results indicate that nearly one-fifth of children in sub-Saharan Africa were exposed to an extreme weather event in the preceding year.
Natural disasters, particularly drought, are consequential for young children’s health and development by elevating their risk for acute malnutrition and in some places, anaemia; however, effects vary across regions.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Undernutrition remains a critical determinant of under-five mortality in SSA, which has the highest global rates of under-five mortality and high rates of undernutrition and anaemia; the region faces potential setbacks in progress reducing wasting prevalence due to increased extreme weather events caused by climate change.
To ensure continued progress in reducing under-five morbidity and mortality, countries and multinational organisations must design and implement effective post-disaster responses to address acute malnutrition and anaemia.
Introduction
Undernutrition and anaemia are persistent global problems, including in sub-Saharan Africa (SSA).1 Among children under 5 years of age in SSA in 2020, 32.3% and 5.9% were stunted and wasted, respectively.2 Iron-deficiency anaemia is highly prevalent in the region, affecting an estimated 64.1% of children 6–59 months.3 Stunting, an indicator of chronic malnutrition, and iron-deficiency anaemia, an indicator of overall nutritional and health status, have significant long-term consequences for children, including impaired cognitive development, poorer educational attainment, greater risk of chronic disease and lower socioeconomic status in adulthood.4 5 Wasting, or acute malnutrition, elevates a child’s risk of mortality with potential for long-term cognitive and developmental damage if untreated.4 6 7
A growing literature finds climate change affects undernutrition among young children. Broadly, studies identify consistent effects across settings of changes in temperature, rainfall and weather patterns on rates of child stunting and wasting.8–12 Environmental factors, including land vegetation, rainfall and temperature, are affected by climate change and associated with prevalence of anaemia among young children across SSA.13 In addition to changes in average temperature and rainfall, climate change is expected to cause an increasing frequency of extreme weather events or natural disasters.14 Natural disasters—including droughts, floods, extreme temperature events, earthquakes, landslides, storms and wildfires—may influence children’s nutritional status directly and/or indirectly. Reduced food security and crop and livestock damage in the aftermath of a natural disaster may increase children’s risk of wasting.15 Persistent food insecurity after a disaster may contribute to a higher risk of stunting. While the aetiology of iron-deficiency anaemia is more complex,16 it may also be more common where children experience micronutrient deficiencies due to persistent food insecurity or insufficient dietary diversity post-disaster. Protracted effects on families’ income, socioeconomic status and/or access to health services or markets may further exacerbate the risks and effects of food insecurity in the aftermath of a natural disaster.
Prior studies estimating the effects of natural disasters—rather than climate variability—on child undernutrition have primarily analysed single disaster types or single countries. For example, in the Horn of Africa, drought exposure was associated with greater risk of wasting.17 Studies in India and Guatemala found associations between exposure to multiple types of natural disasters and greater risk of stunting.18 19 These findings suggest that increased exposure to extreme weather events will increase the prevalence of both acute and chronic undernutrition. However, it is unclear whether these relationships are consistent across SSA, and how different disaster types differentially affect child nutritional status in the region. In addition to acute and chronic undernutrition, studying the relationships between natural disasters and anaemia is important given the high prevalence of anaemia in early childhood in SSA and its large potential for long-term effects over the life course.
To comprehensively assess relationships between natural disaster exposure and child nutritional status at a population level in a region with high vulnerability to climate change, we combine data from the Demographic and Health Surveys (DHS) in SSA from 2010 to 2019 with data from the Emergency Events Database (EM-DAT). The objectives are to quantify the effects of natural disaster exposure on three key indicators of undernutrition and underlying health status in early childhood, stunting, wasting and anaemia; and to identify the most consequential effects of natural disasters on child nutrition to inform global and regional policy and intervention efforts.
Methods
Study population
The DHS are a series of nationally representative repeated cross-sectional household surveys that follow a two-stage probability sampling design with consistent measures over time and across countries.20–22 DHS capture population and health indicators among women of reproductive age (15–49) and their households via computer-assisted interviews administered by trained interviewers. We included publicly available DHS waves with GPS cluster locations in the SSA region where all fieldwork was conducted between 1 January 2010 and 31 December 2019. The final dataset contains 51 survey waves from 30 countries. This 10-year period is intended to capture sufficient exposure to natural disasters prior to the COVID-19 pandemic. Children under 5 years of age at the time of survey with valid height and weight measures and non-missing measures of key covariates were eligible for inclusion in analyses of stunting and wasting; those with valid haemoglobin measures were eligible for inclusion in analyses of anaemia.
The EM-DAT aims to include information on all disasters from 1900 to the present.23 We included all natural disaster weather events reported within 1 year of a DHS wave in the database: droughts, earthquakes, floods, landslides, storms and wildfires. EM-DAT bases their disaster definitions and classification on the Integrated Research on Disaster Risk Peril Classification and Hazard Glossary (online supplemental table S1).24
Supplemental material
Exposure, outcomes and covariates
We used two exposure periods, 3 and 12 months, to determine whether children experienced a natural disaster. These periods are used elsewhere in the literature.17 19 25 While a 3-month period captures short-term effects, the 12-month period may be more appropriate for capturing stunting and/or iron-deficiency anaemia that take longer to develop. DHS cluster and disaster locations were matched to administrative level two units (county, district or equivalent subnational administrative area) provided by the Global Administrative Areas Database using ArcGIS software.26 Children were considered exposed if a disaster occurred in their administrative level two unit in the 3 or 12 months prior to their DHS interview date. Although DHS offsets the exact location of clusters to protect respondent privacy, cluster locations are not moved across level two boundaries, minimising misclassification bias due to cluster location jittering.
Our outcomes of interest are stunting, wasting and anaemia in children under 5 years of age. We calculated height-for-age (HAZ) and weight-for-height (WHZ) z-scores using the zscore06 module in Stata.27 Children with biologically implausible z-scores (less than −6 or greater than 6 for HAZ and less than −5 or greater than 5 for WHZ) were excluded. Children with HAZ below −2 SD were considered stunted, and children with WHZ below −2 SD were considered wasted. Children with an altitude-adjusted haemoglobin count less than 10 g/dL were considered moderately or severely anaemic.28 We excluded children less than 6 months of age from anaemia analyses because haemoglobin counts are higher after birth and may skew prevalence estimates.
The inclusion of child, parent and household covariates in our models was determined a priori. For fully adjusted models of stunting and anaemia, we controlled for child age in months (0–5, 6–11, 12–23, 24–35, 36–47, 48–59), child sex (male vs female), number of household members, low birth weight (<2500 g), birth order, maternal educational attainment, paternal educational attainment, household wealth quintile and the child’s type of residence: urban or rural. Birth order was categorised as first, second or third, fourth through sixth and seventh or greater. Maternal educational attainment was categorised as no education, primary, secondary and higher. Paternal educational attainment was categorised as no education, primary, secondary, don’t know and missing. Household wealth quintiles were estimated within each survey wave based on principal components analysis of a set of durable goods, livestock and physical housing characteristics. For fully adjusted models of wasting, we added a measure of whether the child suffered from acute illness in the past 2 weeks (diarrhoea, fever or cough with fast breathing), based on maternal report.
Statistical analysis
Descriptive analyses used DHS-derived sample weights. We used generalised estimating equations (GEE) to estimate the population average effect of natural disaster exposure on stunting and wasting. GEE, an extension of a generalised linear model, accounts for within-group correlation between observations and relies on relatively few assumptions regarding the correlation structure relative to other multilevel modelling strategies; it produces unbiased estimates of regression coefficients even when the correlation structure is misspecified.29 Additionally, the use of GEE aligned with our goal of estimating population average effects of disaster exposure on our outcomes of interest. GEE is particularly well suited to handle pooled DHS datasets, which have a high number of clusters with small numbers of observations.30 These models specified a Poisson distribution and log link, with robust SEs and independent correlation structures, to estimate relative risks (RR) and 95% CIs.31 For each outcome, model 1 was unadjusted. Model 2 was adjusted for survey year, country and all covariates listed earlier. We estimated average marginal effects (AMEs) based on the model 2 results using a 12-month exposure to compare effect estimates across models and samples. All analyses were completed using Stata V.17.1.
Results
The final sample included 320 479 children under 5 years of age with non-missing, biologically plausible height and weight data to estimate HAZ and WHZ. A subsample of 196 869 children between ages 6 months and 5 years with haemoglobin measures was included in analyses of anaemia. Information on children excluded from analysis and the sample by country and survey year is available in online supplemental table S2. Table 1 contains information on the individual and household characteristics of the children in the analytic sample.
Characteristics of children under 5 years of age in included DHS waves (N=320 479)
Table 2 summarises the exposures of the children in our sample. In total, children in our sample were exposed to six types of disasters and 90 unique disaster events. Floods and droughts accounted for a large proportion of the exposures. In the 3 months and 12 months prior to their DHS interview date, 9.0% and 19.7% of children in our sample were exposed to at least one disaster, respectively. Exposure varied across and within countries (online supplemental figure S1).
Number of events and children under five exposed by disaster type
Tables 3 and 4 describe the prevalence of stunting, wasting and anaemia in our sample. The overall prevalence of stunting in our sample was 32.1% (95% CI 31.9%, 32.4%), and the overall prevalence of wasting was 8.3% (95% CI 8.1%, 8.4%). The overall prevalence of anaemia was 40.7% (95% CI 40.3%, 41.0%). For stunting, wasting and anaemia, children exposed to at least one disaster in the 3 months prior to their DHS interview had higher prevalence of the outcome compared with unexposed children, respectively. Children exposed to at least one disaster in the 12 months prior to their DHS interview had a higher prevalence of stunting and wasting compared with unexposed children, respectively.
Stunting, wasting and anaemia prevalence among children under 5 years of age by timing of exposure to any disaster
Relative risks of stunting, wasting and anaemia by disaster type and exposure period
Table 4 presents the RRs for stunting, wasting and anaemia estimated by our GEE models. In fully adjusted models, we found no evidence for an association between exposure to any natural disaster and stunting, regardless of the exposure period (3-month RR=1.015, 95% CI 0.987, 1.043; 12-month RR=0.981, 95% CI 0.960, 1.002). While RRs are attenuated from the unadjusted model, we found that children exposed to a natural disaster have greater risk of wasting than children who did not experience a natural disaster, using both a 3-month (RR=1.265, 95% CI 1.190, 1.343) and 12-month exposure period (RR=1.169, 95% CI 1.118, 1.223). Examining exposure to drought specifically, we found a protective effect of drought exposure on stunting, while children exposed to drought had a greater risk of wasting. For example, in fully adjusted models using a 3-month exposure period, the risk of stunting was 0.888 times as high (95% CI 0.851, 0.926), while the risk of wasting was 1.565 times higher (95% CI 1.434, 1.709) among children exposed to drought relative to children who did not experience drought. We found evidence of an association between exposure to flood and stunting risk in the fully adjusted model using a 3-month exposure period (RR=1.092, 95% CI 1.056, 1.128); however, this relationship is attenuated when using a 12-month exposure period (RR=0.987, 95% CI 0.965, 1.011). We found more consistent evidence of a relationship between flood exposure and wasting in fully adjusted models, regardless of how the exposure period is defined (3-month RR=1.150, 95% CI 1.066, 1.241; 12-month RR=1.072, 95% CI 1.022, 1.124). Across models, RRs were generally higher using a 3-month exposure period compared with a 12-month exposure period. Using a 3-month exposure period, we found no relationship between exposure to any disaster, drought, or flood, and anaemia. When using a 12-month exposure period, we found a slight protective effect of exposure to any disaster and flood specifically on anaemia (RR=0.961, 95% CI 0.939, 0.983; RR=0.965, 95% CI 0.942, 0.989, respectively).
Figure 1 shows the AMEs obtained from the fully adjusted models for all outcomes using a 12-month exposure period for all disasters, disaggregated by region.
Forest plot.
We found the strongest effects of natural disasters in East Africa, where exposure to a natural disaster in the preceding 12 months is associated with higher risk of wasting and anaemia. The effects of natural disaster exposure on wasting are consistently positive across subregions with the exception of Central Africa, where it was not possible to estimate region-specific effects on wasting given low prevalence of disaster exposure and wasting. However, the relationships between disaster exposure and stunting and anaemia vary in direction and magnitude across regions.
Results are robust to alternate modelling specifications, including mixed-effect regression models with random effects for PSU. Models testing a 6-month exposure period produced effect estimates that ranged between those obtained using the 3- and 12-month exposure periods (online supplemental tables S3–S5). Supplementary analyses restricting exposure to after birth are broadly consistent with the results presented in online supplemental table S6.
Discussion
This study combined household survey and geo-located natural disaster data to investigate relationships between exposure to disasters and undernutrition and anaemia among children under 5 years of age in SSA over a 10-year period, 2010–2019. The results revealed several key findings. We found consistent evidence of an association between natural disaster exposure and increased risk of wasting, particularly between drought exposure and wasting. However, we found differential effects by disaster type. While flood exposure had the greatest effect on stunting risk, drought exposure had the greatest effect on wasting risk, and a counterintuitive protective effect on stunting risk; moreover, the effects varied across sub-regions. Surprisingly, we did not find evidence of a direct relationship between disaster exposure and anaemia in pooled models, though regional models showed that disaster exposure was associated with increased risk of anaemia in East Africa and decreased risk of anaemia in South Africa. With a population of 185 million children under five, our finding that almost one-fifth of children in SSA were exposed to a natural disaster in the preceding year translates to an estimated 36.4 million children—a number highly likely to increase given population growth and an expected increase in natural disasters in the region due to climate change.32 A 56.5% relative increase in the risk of wasting after exposure to drought could have devastating effects on recent gains in reducing undernutrition and under-five mortality.33 34 Taken together, these findings underscore the need for tailored, institutionalised responses to prevent wasting after exposure to floods, droughts and other natural disasters, and where applicable, prevention of stunting and anaemia.
In contrast to studies estimating relationships between natural disaster exposure and undernutrition in Latin America and South Asia,19 25 35 we found consistent effects of natural disaster exposure on wasting but not stunting. This suggests that in SSA, acute malnutrition may be the most important health outcome for young children meriting intervention after a natural disaster. Our study uses data from a more recent period; it is possible that climate change has modified the relationship between natural disaster exposure and child outcomes over time. It may also be that topographical, climatological and/or agricultural characteristics of Africa vary enough from Latin America or South Asia to produce different relationships between disasters and nutrition. Our results are consistent with studies in East Africa, which find drought is associated with increased prevalence of wasting.17 Our findings are also consistent with a study from Uganda that found self-reported drought exposure in the preceding year is associated with a greater risk of wasting but not stunting.12
There are several potential mechanisms that could explain why we found consistent relationships between natural disasters and wasting but not stunting or anaemia, particularly as this contrasts to studies finding effects of deviations from historical rainfall trends on child stunting.36 37 The difference in protracted vs acute exposure may reflect different biological processes, resulting in different estimated effects on undernutrition. It may also be that wasting, which can be caused by insufficient caloric intake, is directly affected by disaster-related food insecurity, while stunting is less responsive to disaster-related disruptions to food systems and other services.12 Coupled with humanitarian response, existing food systems are a key determinant of the degree to which a disaster will affect child undernutrition. In a study of rainfall deviation and stunting across SSA, Cooper and colleagues report that diverse local agriculture systems, stable crop production and greater food imports mitigate the effects of low rainfall on stunting.15 Data from Ethiopia and Kenya further underscore the importance of rainfall in relation to agricultural seasons as a determinant of undernutrition11 37; thus, the timing of disasters relative to agricultural seasons may result in differential effects on stunting and wasting. Finally, regional variation in disaster-undernutrition relationships may be due to underlying variation in the distribution of undernutrition indicators across and within regions relative to where disasters occurred.
This study is subject to several limitations. First, it is possible that some children have misclassified exposure. Major disasters may have affected the timing of DHS data collection. Our estimates may be biased towards the null if our sample includes children who experienced persistent disaster-related wasting, but the DHS interview was delayed to more than 1 year after the disaster. EM-DAT primarily reports the geolocation of disasters as the lower-level administration level. We assumed equal exposure for all clusters within an administrative unit reported in EM-DAT, however, the degree of exposure may have varied within administrative units. We excluded three disaster events because they lacked geolocations and three because we could not identify the geolocation provided in EM-DAT. Second, while EM-DAT provides comprehensive measures of disaster events globally, it may omit some disaster events due to reporting limitations. It is possible that the omission of certain natural disaster events introduces bias into our estimates. Third, the limited number of other disaster types (eg, earthquake, landslide) prevented us from directly assessing their effect on nutritional outcomes. Fourth, to protect participants’ confidentiality, DHS randomly displaces cluster locations within a 5- to 10-kilometre range. However, this potential problem likely had a limited impact on our results because DHS has avoided moving clusters across administrative level two units whenever possible since 2009, prior to the study period. We anticipate this greatly minimises exposure misclassification due to cluster jiggering.
The effects of natural disasters may be particularly acute in SSA, where there is a high underlying burden of undernutrition and multiple regions especially vulnerable to the effects of climate change, and a large population of children under five at risk. Our study finds consistent associations between exposure to natural disaster and increased risk of acute malnutrition. As climate change progresses, estimates of the population-level impacts of natural disasters’ effects on child nutrition and growth will be key to informing appropriate nutrition policies and disaster response efforts to support children and their families.
Data availability statement
Data are available upon reasonable request. Both datasets used in this analysis (Demographic and Health Surveys; EM-DAT international disaster database) are publicly and freely available. Note to editorial team: both DHS and EM-DAT are freely available on their own websites, but are not in specific data repositories. We are unable to report the datasets in the BMJ format. Thus, on recommendation from the editorial assistant, we have unticked 'public, open access repository'. We cited the data sources as the publishers request on their websites, so they can be easily located by readers.
Ethics statements
Patient consent for publication
Ethics approval
This study involves human participants but the University of Michigan IRB exempted this study because it only involves de-identified, publicly available data. Participants gave informed consent to participate in the study before taking part.
Acknowledgments
The authors appreciate guidance from Josh Errickson of the University of Michigan Consulting for Statistics, Computing, and Analytics Research. Any remaining errors are the responsibility of the authors. We also appreciate research assistance from Meera Herle and Sasha Rozenshteyn.
References
Footnotes
X @kavao
Contributors ET and KP conceptualised the study. KP curated data. KP and ET accessed and verified the underlying data. All authors contributed to the methodology. KP created the software and conducted formal analysis with support from ET and MKM. All authors interpreted results. ET and ALW provided supervision to KP. ET provided resources and validated results. KP wrote the original draft. ET, MKM and ALW reviewed and edited the manuscript. All authors reviewed and approved the final version for publication. ET is the guarantor and accepts full responsibility for the work.
Funding This research was supported in part by an NICHD centre grant to the Population Studies Center at the University of Michigan (P2CHD041028). KP was also supported by the University of Michigan Department of Epidemiology Internship Funding (grant number N/A). No funders had any role in the study design; data collection, analysis or interpretation; or the writing or decision to submit for publication.
Map disclaimer The inclusion of any map (including the depiction of any boundaries within it), or of any geographic or locational reference, does not imply the expression of any opinion whatsoever on the part of BMJ concerning the legal status of any country, territory, jurisdiction or area or of its authorities. Any such expression remains solely that of the relevant source and is not endorsed by BMJ. Maps are provided without any warranty of any kind, either express or implied.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.