Risk Factors for Childhood Stunting in 137 Developing Countries: A Comparative Risk Assessment Analysis at Global, Regional, and Country Levels

1Child survival has improved substantially over the past fifty years. The annual number of child deaths under age 5 y declined from 17.6 million in 1970 to 6.3 million in 2013, and under-five mortality declined from 143 per 1,000 live births to 44 during the same period. Global progress in improving childhood growth has been less impressive. While the prevalence of stunting (height-for-age z-score less than two standard deviations below the global median, as defined by the 2006 World Health Organization Child Growth Standards) among children under 5 y declined from 47% in 1985 to 30% in 2011 globally, only minor improvements have been achieved in some of the poorest regions of the world, especially South Asia and sub-Saharan Africa. In recognition of the large disparities across the globe in the areas of early life nutrition and development, the World Health Assembly set a target to reduce by 40% the number of stunted children worldwide by 2025.

2To reach this target, information on ways to alleviate stunting in each country is essential. Randomized trials and observational studies have identified a large number of risk factors for poor childhood growth. However, the impact of these risk factors on stunting at the population level (globally, regionally, and at the country level) is not known.

3To address this gap, we conducted a global comparative risk assessment analysis of stunting risk factors. We used country-level data on the prevalence of risk factors from global pooling projects of population health surveys, in combination with effect sizes for each risk factor on stunting from recent meta-analyses of epidemiological studies. This report focuses on 18 risk factors for stunting, while a forthcoming paper uses similar methodology to examine four psychosocial risk factors.

4We estimated the burden of stunting among children 2 y (24–35 mo) of age (i.e., right at the end of the first 1,000 days of life) that is attributable to 18 risk factors in 137 developing countries. The selected countries included all countries designated as developing by the Global Burden of Disease Study, which closely correspond to the countries designated as developing by the United Nations for tracking progress towards the Millennium Development Goals. These risk factors were selected from an extensive list of modifiable (i.e., behavioral or environmental; nongenetic) risk factors for stunting based on (i) the availability of high-quality exposure data (i.e., nationally representative data using standard measurements such as measured weight rather than self-report, and using appropriate statistical methods for pooling and imputing data), (ii) strong evidence for an association with stunting, and (iii) the availability of evidence on the effect size on stunting from recent meta-analyses of epidemiological studies (criteria described in detail in S1 Text; see also and S2 Table). Estimating the burden of stunting attributable to various risks does not in itself establish causality, but because we have included only risk factors for which there is convincing evidence of a causal relationship with stunting, the relationships examined here can be interpreted as our current best estimates of their causal effect. Stunting was defined as height-for-age z-score < −2 based on the World Health Organization Child Growth Standards. We grouped risk factors into five clusters: maternal nutrition and infection, teenage motherhood and short birth intervals, fetal growth restriction (FGR) and preterm birth, child nutrition and infection, and environmental factors (i.e., unimproved water and sanitation and use of biomass fuels) (Table 1). These categories were based on the similarity of risk factors and of their corresponding interventions. We estimated the proportion of stunting that is attributable to each risk factor and cluster of risk factors in each country, as detailed below.

5We derived the prevalence of exposure to each risk factor for the year 2010 (or as close to 2010 as possible) from published literature and from available surveys such as Demographic and Health Surveys (DHS) (Table 1). Estimates of stunting prevalence for children under 5 y for each country for the year 2011 were provided by the Nutrition Impact Model Study, which provides regional and global levels similar to those estimated by WHO. We chose 2010 as the index year for risk factor exposure and 2011 as the index year for stunting exposure to allow a temporal sequence such that risk factors are measured or estimated before stunting. To estimate the prevalence of stunting and number of stunted children at age 2 y (i.e., 24 to 35 mo of age), we calculated the ratio of stunting prevalence among children age 2 y to that among children under 5 y in 104 surveys available from the WHO Global Database on Child Growth and Malnutrition (available from the Nutrition Landscape Information System). For 33 countries without surveys, we used population-weighted sub-regional averages to generate a correction factor (see S2 Text for more detail and S3 Table for the country-specific ratios). Data on cohort size (population of children at age 2 y) were provided by the United Nations Population Division World Population Prospects 2015 Revision. Data on prevalence of teenage motherhood and short birth spacing were available for 64 countries with recent DHS surveys (73 countries did not have a recent DHS survey). For countries without DHS data, we used a sub-regional average if estimates from at least one country in the region were available (or a regional average when no data were available within the sub-region; see S3 Fig for sub-region and region classifications). The analysis of child HIV infection without highly active antiretroviral therapy (HAART) before 2 y of age (untreated HIV infection) was conducted only for 45 countries with available HIV prevalence data. To inform the effect size of each risk factor on stunting, we identified the most recent meta-analyses of epidemiological studies or conducted de novo systematic reviews (S1 Text; S2 Table).

6We calculated the population attributable fraction (PAF) of stunting for each risk factor using methods that have been described elsewhere. The estimated PAF for each risk factor quantifies the independent effect of that specific risk factor on stunting (holding all other risk factors constant); the PAF is estimated using the following relationship: where RR i is the relative risk of stunting comparing category i with the reference (or optimal) category and P i is the prevalence of exposure to the risk factor of interest.

7For most risk factors, epidemiological studies have reported effect sizes on stunting. For maternal malaria, underweight, and anemia, as well as biomass fuel use, meta-analyses of effect sizes were available only on low birth weight (LBW), which is itself associated with stunting. For these risk factors, we calculated the PAF of stunting by multiplying the proportion of LBW attributable to the risk factor (i.e., the PAF) by the proportion of stunting attributable to LBW in each country. For example, if in a particular country 20% of LBW is attributable to maternal underweight and 30% of stunting is attributable to LBW, then it can be easily inferred that 6% of stunting is attributable to maternal underweight. Similarly, the effects of nonexclusive and discontinued breastfeeding were reported only on diarrhea, which itself is associated with stunting. We used these reported effect sizes to estimate the effect on stunting through diarrhea by multiplying the PAF of diarrhea attributable to these risk factors by the PAF of stunting attributable to diarrhea.

8We further estimated the combined effect of several risk factors within each of the five clusters. The number of stunting cases attributable to multiple risk factors in one cluster is less than the sum of the number of stunting cases attributable to each risk factor because one case of stunting can be due to multiple risk factors (i.e., multicausality) and because one risk factor may affect stunting partly through another risk factor in the same cluster (i.e., mediation). To prevent “double counting” due to multicausality, we used a relationship to estimate the combined effects of multiple risk factors based on their individual PAFs. The relationship is captured by the following formula, which assumes that there is no correlation and no effect modification for relative risks: where PAFi is an individual PAF for a risk factor in cluster j, and all individual risk factors (from 1 to R) in cluster j are combined using the formula above to estimate PAFj, which is the fraction of stunting attributable to all the risk factors in cluster j.

9To use this relationship in the child nutrition and infection cluster of risk factors, discontinued and nonexclusive breastfeeding were excluded because their effects on stunting are mediated through diarrhea, and we did not include childhood untreated HIV infection because data on this risk factor were not available for all countries. For this risk factor cluster, we had to make an additional modification because part of the effect of zinc deficiency on stunting is mediated by diarrhea. Therefore, using the above formula for this cluster would lead to overestimation if the effect of diarrhea and zinc deficiency were both included in the combined PAF (PAFj). To correct this, we replaced the overall effect of zinc on stunting with the part of the effect of zinc on stunting that is not mediated through diarrhea. We conducted a literature review but did not find any studies that quantified this relationship and therefore assumed that 50% of the excess relative risk of stunting due to zinc deficiency is mediated by diarrhea. For example, in Nigeria, the overall relative risk of zinc deficiency for stunting was estimated to be 1.04 (based on a 0.06 lower mean height-for-age z-score in zinc-deficient children; see notes below Table 1, and S3 Text, for more details), which corresponds to an excess relative risk of 0.04. Using the 50% proportion mentioned above, we calculated that the excess relative risk of zinc deficiency not mediated through diarrhea was 0.02, and that calculation in turn yields a “direct” relative risk of 1.02. This relative risk does not include the effect of zinc deficiency on diarrhea and can be combined with the effect of diarrhea on stunting to estimate the combined effect of both risk factors on stunting without creating a bias due to overestimation. We conducted a sensitivity analysis using 0% and 100% as the mediated proportion (S4 Table).

10We calculated the attributable prevalence of stunting by multiplying the PAF by the prevalence of stunting in each country. We also calculated the attributable number of stunting cases in children aged 2 y by multiplying the PAF by the number of stunted children at age 2 y. To quantify uncertainty, we used the mean and standard error for each exposure prevalence and effect size, separately, to generate 1,000 simulations for the prevalence of the risk factor and its odds ratio for stunting, then calculated the PAF and attributable prevalence of stunting at age 2 y 1,000 times (once per simulation). The 95% confidence intervals of PAFs and numbers of attributable stunting cases were calculated by using the 2.5th and the 97.5th percentiles of draws. All analyses were conducted using STATA SE version 13.1. This study was conducted solely using secondary and existing datasets and therefore did not require institutional review board review.

11In 2011, we estimated that 44.1 million children aged 2 y in the selected 137 developing countries were stunted, corresponding to 36% of the 2-y-old population. The most important individual risk factor for stunting was being term, small for gestational age (TSGA), with 10.8 million (95% CI 9.1 million–12.6 million) stunting cases attributable at age 2 y in 2011. Unimproved sanitation, with 7.2 million (95% CI 6.3 million–8.2 million) attributable cases of stunting, and diarrhea, with 5.8 million (95% CI 2.4 million–9.2 million) attributable cases, were the second and third most important risk factors for stunting worldwide, respectively (Fig 1). When clusters of risk factors were considered, FGR and preterm birth (preterm, small for gestational age; TSGA; preterm, appropriate for gestational age [PAGA]) were the leading risk factors for stunting prevalence, with 32.5% of stunting prevalence being attributed to these factors (14.4 million cases, 95% CI 12.6 million–16.2 million). This cluster of risk factors was followed by environmental factors (unimproved water, unimproved sanitation, and biomass fuel use), with 21.7% (9.6 million cases, 95% CI 8.4 million–10.8 million), maternal nutrition and infection risk factors, with 14.4% (6.4 million cases, 95% CI 5.3 million–7.5 million), and child nutrition and infection risk factors, with 13.5% (6.0 million cases, 95% CI 2.6 million–9.4 million) of attributable stunting cases. Teenage motherhood and short birth intervals had the fewest attributable stunting cases, with 1.9% (0.86 million cases, 95% CI 0.77 million–0.95 million) (Table 2; Fig 2).

12The risk factor cluster FGR and preterm birth was associated with the largest attributable burden of stunting in all regions. The cluster of environmental factors, most importantly unimproved sanitation, was the second leading cluster of risk factors in South Asia, sub-Saharan Africa, and East Asia and Pacific, whereas in Central Asia, Latin America and Caribbean, and North Africa and Middle East, the second leading cluster was child nutrition and infection (mostly childhood diarrhea). Central Asia had the highest proportion of stunting attributable to child nutrition and infection risk factors across all regions, at 18.9%, and sub-Saharan Africa had the largest proportion of stunting attributable to environmental risk factors across all regions, at 27.0%. Of all stunting cases attributable to each risk factor cluster, South Asia and sub-Saharan Africa had the largest shares across all regions (ranging from 34% to 50%, and 29% to 43%, respectively, across risk factor clusters) due to having both high exposures to the selected risk factors and large numbers of stunted children.

13Within these larger regions, there were important differences across sub-regions in the burden of stunting attributable to risk factors and risk factor clusters. Within the sub-Saharan Africa region, the attributable prevalence of stunting associated with unimproved sanitation in southern Africa was less than half that of central, east, and west Africa. Similarly, diarrhea was associated with almost three times the burden of stunting in Andean and central Latin America as in tropical and southern Latin America. The burden of diarrhea also differed substantially within Asia, with much smaller attributable prevalence of stunting in the East Asia sub-region (1.9 percentage points) than in the Central Asia, South Asia, and Southeast Asia sub-regions (all greater than 3.5 percentage points).

14At the country level, nations with high stunting prevalence such as Niger, Burundi, Yemen, Eritrea, Ethiopia, Afghanistan, Timor-Leste, and Zambia (all with prevalence greater than 50%) had, as expected, a large burden of stunting attributable to all risk factor clusters (Fig 3; S4 Fig displays the PAFs). However, several countries with a relatively lower prevalence of stunting also had a relatively large attributable burden due to high exposure to specific risk factors. For example, Somalia had the largest prevalence of stunting attributable to discontinued breastfeeding and the second largest prevalence attributable to nonexclusive breastfeeding (0.9 and 1.4 percentage points, respectively). Malawi had the top rank for PAGA, at 4.9 percentage points, and Bangladesh had the top rank for teenage motherhood, at 0.9 percentage points (country-level PAFs and numbers of attributable stunting cases are available in S5 Text and on the study website: http://www.healthychilddev.sph.harvard.edu/).

15The ranking of risk factors within each country was generally consistent with global and regional ranking of risk factors (S5 Table). Nevertheless, specific risk factors imposed a larger burden of stunting in particular countries irrespective of the burden of stunting in the country: unimproved sanitation surpassed TSGA as the leading risk factor in China and in many sub-Saharan African countries, and nonexclusive and discontinued breastfeeding had a substantially higher rank in middle-income countries (e.g., Mexico, South Africa, Iran, Turkey, and Argentina) than in low-income countries.

16Our results suggest that a large proportion of childhood stunting in developing countries could be prevented if exposure to a few key risk factors could be eliminated. Globally, TSGA was associated with the largest stunting burden, followed by unimproved sanitation and childhood diarrhea. This pattern highlights the success of current clinical and public health interventions to prevent and manage childhood infections and improve childhood nutrition in many developing countries, but also calls for a new focus on interventions before and during pregnancy to address intergenerational effects of malnutrition among girls and women, as well as interventions to improve the environment in which mothers and families live, with specific attention to improving sanitation.

17A previous analysis of FGR and preterm births in developing countries reported smaller fractions of stunting attributable to TSGA (e.g., a PAF of 16% for TSGA compared with 24% in our analysis). However, the value reported in this previous analysis may be an underestimation, as the regional and global effects were estimated using regional prevalence of stunting as opposed to estimating the attributable stunting in each country, as we did here.

18The large burden of stunting attributable to FGR is perhaps unsurprising given that prenatal restricted growth is logically strongly related to postnatal restricted growth; nevertheless, our findings serve to further emphasize the importance of early intervention during pregnancy. Several recent reviews have identified maternal iron, balanced protein-energy, and multiple micronutrient supplementation as the most effective interventions to alleviate FGR. However, providing these interventions before pregnancy or in its early months is logistically difficult because in many developing countries, the majority of pregnant women start attending antenatal clinics in their second or third trimester.

19Environmental factors (i.e., unimproved water, unimproved sanitation, and biomass fuel use) had the second largest global attributable burden. Particularly, 7.2 million cases of stunting worldwide were attributable to unimproved sanitation. The attributable burden of unimproved sanitation for stunting was larger (though not significantly) than that of childhood diarrhea, as some of the effects of improved sanitation may be through prevention of other childhood infections and improvement of maternal health and nutrition in pregnancy. This further underscores the importance of ongoing water, sanitation, and hygiene (WASH) programs to increase access to, and use of, safe water and sanitation for children and families worldwide.

20We found that one in seven cases of stunting was attributable to child nutrition and infection risk factors. Programs to promote adequate complementary feeding and rotavirus and cholera vaccines may reduce the attributable burden of this group of risks. Childhood HIV infection remains a significant contributor to child mortality but was not a major risk for stunting at the population level. In Swaziland and Lesotho, the countries with the highest prevalence of this risk factor, the prevalence of stunting attributable to untreated HIV infection was estimated to be 0.7 and 0.8 percentage points, respectively. We may have slightly underestimated the fraction of stunting attributable to HIV as we were only able to estimate the effect of untreated child HIV infection, due to lack of data on the growth of HIV-infected children who receive HAART before 2 y of age compared with HIV-uninfected peers in developing countries (S2 Fig). Nevertheless, the contribution of HIV-infected children who receive HAART before 2 y to population-level stunting is likely small, given that young children initiating HAART exhibit rapid linear growth catch-up. Teenage motherhood and short birth intervals had a fairly small effect on stunting at the population level due to both lower prevalence of exposure in many developing countries and smaller effect sizes for stunting compared with other risk factors.

21The global scope of this analysis and the large number of risk factors included result in several limitations. Although we included only risk factors with strong evidence of an association with stunting, causality can never be guaranteed by observational studies (which were the sources of nearly all effect sizes available for the risk factors analyzed here). Although the list of risk factors is extensive, we had to exclude several risk factors because country-level data on exposure or effect size on stunting were not available. Examples are maternal smoking, prenatal alcohol use, and illicit drug use. Similarly, global data on environmental pollutant exposures in childhood, such as lead and arsenic exposure, were not available. In addition, we had to use approximate estimates for prevalence of use of biomass fuels, maternal malaria, maternal short stature, and maternal underweight. We used the prevalence of relying mainly on biomass fuels for cooking as a proxy for all exposure to biomass fuels, underweight among women of reproductive age as a proxy for maternal underweight, height among women aged 18 to 49 y as a proxy for maternal height, and prevalence of P. falciparum infection among children aged 2 to 10 y as a proxy for maternal malaria prevalence. Estimates of uncertainty in the exposures to maternal malaria, maternal short stature, childhood zinc deficiency, untreated HIV infection, LBW, and PAGA were unavailable, so uncertainty in the estimates of the effect of these risk factors is underestimated. Relatedly, we did not incorporate the uncertainty from converting the prevalence of stunting among children under 5 y to the prevalence of stunting among children age 2 y. Another limitation is that we estimated the effect of six risk factors through either LBW (maternal anemia, malaria, and underweight and biomass fuel use) or diarrhea (nonexclusive and discontinued breastfeeding), which may have led to underestimation of the effect of the maternal and childhood nutrition and infection risk factor clusters. As is standard in global risk factor analyses, we assumed that proportional effects are fairly similar across countries. Evidence from analyses of FGR and preterm birth supports this assumption as the odds ratios for stunting across different regions were fairly similar. Multi-country studies are required to evaluate variation of effect sizes for other risk factors, but such studies are currently not available. Finally, the effect sizes for most risk factors were reported as odds ratios, which overestimate the relative risk when exposure is not rare and could lead to overestimating the effect of these risk factors on stunting. Correcting this bias requires information on the incidence of stunting among unexposed children for each risk factor and in each country, and these data are currently not available.

22Our analysis also had several major strengths. We included all major risk factors for stunting after considering an extensive set and limiting our analysis to those with convincing evidence on their effect on stunting and high-quality data on country-level exposure. We reviewed and did not include risk factors with insufficient evidence on their effect on stunting, including child HIV infection with HAART, maternal and childhood iodine deficiency, child hookworm infection, acute lower respiratory infection, and childhood malaria. We also excluded risk factors for which meta-analyses of observational studies and/or randomized trials identified no significant effect on childhood stunting. These were childhood anemia, maternal hookworm infection, and vitamin A deficiency. For the included risk factors, we used the most recent and reliable meta-analyses of effect sizes on stunting. We quantified uncertainty at each step of estimation and reported overall uncertainty in the final results.

23Our results represent a consistent and comparable set of global estimates of the impact of 18 risk factors on stunting. FGR, unimproved sanitation, and diarrhea are the leading risk factors for stunting globally, with larger estimated impacts on stunting in sub-Saharan Africa and South Asia compared with other regions. According to our findings, reducing the burden of stunting requires continuing the current efforts to diagnose and treat maternal and child infections, especially diarrhea, along with a new focus on clinical and public health interventions that focus on improving nutrition and sanitation among mothers and families.