Perspective: Growing Up or Growing Out: Growth Faltering in Early Childhood and Subsequent Risk of Overweight

All of us believe in a world where every person can grow to their full potential and, in turn, contribute to society in their own way. As nutritionists, our specific concerns are with the food we provide to children and the policies that we advocate to promote good nutrition. Historically, nutritionists working in low- and middle-income countries have been concerned with measuring the extent of, and developing strategies to prevent, growth faltering and other manifestations of undernutrition. In some locations, particularly in the context of armed conflict or weather-related famines, acute wasting is a public health emergency, which requires specific responses from the global public health community; this is not the focus of the present article, which focuses on linear growth and overweight and obesity, and their interrelations, within contexts of children living their usual lives.

In recent years, the worlds of “undernutrition” and “overnutrition” have been intersecting. We have become increasingly aware of the spread of the obesity epidemic from high-income countries to low- and middle-income economies, predominantly in adults but, increasingly, in children. We have also become aware of the changing pattern of regional, ethnic, and socioeconomic disparities in obesity and its consequent morbidities within countries.

Linear growth velocity in childhood and adolescence is a monotonic, nonlinear, function of age. Most representations of growth velocity characterize it as a smooth curve—over a scale of months or years this is appropriate but, on a finer scale, growth is saltatory (1). Length (or height, if standing), which is what is measured at any one point, is compared to a reference measure. With repeated measurements over time, a child's growth can be compared to that of children who start at the same percentile. Aggregating across a sample of children, we can develop a measure of the nutritional status of a population—typically reported as the proportion of children whose height for age, expressed in z scores (HAZs), falls below a critical value, traditionally −2. In a well-nourished population, such as the WHO's Multicentre Growth Reference Study sample (2), 2.3% of the population are, by this definition, stunted. This provides an anchor for comparisons across populations or within populations over time. Recent estimates from UNICEF confirm that stunting is still widespread (3). In North America (and presumably Europe, which is not included in this report) stunting is rare, but in other regions, such as Southern Asia and Central and Southern Africa, up to 38% of children under the age of 5 y are stunted. Taking a longitudinal perspective, stunting prevalence has decreased globally from 32.6% in 2000 to 22.2% in 2017, with substantial decreases occurring in all regions except Oceania.

The basic tool for estimating adiposity is the BMI, calculated as weight divided by the square of height (kg/m²). Unlike height, BMI does not increase monotonically over childhood (2). However, despite this, a child can be compared to a reference, and the aggregation of data across children provides an estimate of the population prevalence of overweight and obesity. Based on the same recent UNICEF estimates for children aged 0–5 y, the prevalence of BMI z scores (BMIZ) >2 ranges from 2.4% in Western Africa and 3.1% in Southern Asia to 13.7% in Southern Africa and, unlike the trends for stunting, the prevalence of overweight is increasing—globally the prevalence has increased from 4.9% to 5.6%, with some regions showing dramatic increases (3).

So, we can conclude that our long-term efforts to improve child growth are being successful, but that we are witnessing an emerging epidemic of child overweight, with major potential consequences for global health. Child overweight is important because child overweight and obesity tend to persist and become adult obesity. This has been shown using prospective cohorts from both high-income and middle-income countries. Cunningham et al. (4) used US data to show that early childhood overweight and obesity are strong predictors of adolescent overweight and obesity. In a study of children born in Soweto, Johannesburg, in early 1990 and followed prospectively, Lundeen et al. (5) showed that girls who were already obese by age 4 y were 42 times as likely to be overweight or obese at age 18 y as those who were not overweight. This was also true in males, even though boys and girls had very different patterns of BMI change over childhood, with overweight becoming far more prevalent in females than in males.

Let us turn now to the relation between these 2 patterns of linear growth and of overweight. There are in the literature periodic reports of “dual burdens” or of “co-occurrence” of various measures of nutritional status. Common definitions of burdens include measures of growth and growth faltering, overweight and obesity, and anemia, but one could use any number of nutritional status measures based on the data available to the researcher. Co-occurrence, often termed “clustering,” can be characterized as occurring at the ecological level, at the household level, and at the individual level.

At the ecological level, one can describe the group prevalence of, say, stunting and of overweight, both in children under 5 y of age, across geographies or time. Thus, the second Lancet series reported that the global prevalence of stunting was falling at the same time that the prevalence of overweight was increasing (6). However, there is little association between the prevalence of stunting and that of overweight among children aged 0–5 y across UNICEF regions (3). If anything, the association is inverse, but it is weak, and there is also little association between the change between 2000 and 2017 in stunting prevalence and that of overweight prevalence (3). Those observations are consistent with data from a nonindigenous area of Guatemala, where cross-sectional height and weight data for children under 5 y of age were collected on 6 occasions between 1968 and 2007 (7). The HAZ curves shifted systematically up over time, as the prevalence of stunting decreased from 73% in 1968 to 23% in 2007, whereas the prevalence of overweight (in that report considered to be BMIZ >1) did not change substantially. It would appear that as the environment improves, children first respond by increasing linear growth, and perhaps only secondarily by increasing adiposity.

An interesting form of co-occurrence at the country level concerns wasting and overweight, which are opposite tails of the same parameter distribution, and which cannot of course occur in the same person. One would expect, therefore, a strong inverse association between the prevalence of each as the mean BMIZ changes. Although countries with a high prevalence of child wasting indeed tend to have a low prevalence of child overweight, and vice versa (8), it is interesting that this relation is not as strong as would be expected, with several countries having large proportions of the population at both ends of the distribution simultaneously. This suggests some interesting opportunities for within-country analysis, perhaps in relation to measures of equity.

The widespread availability of Demographic and Health Survey data has made the within-household analysis of the dual burden of “malnutrition in all its forms” straightforward. For example, a steady stream of publications has documented risk factors for co-occurrence of child stunting and maternal overweight in the same household. The various studies have tended to focus on 1 or at most a few data sets, and the results have been inconsistent.

Why the inconsistency? Consider a contingency table, which presents the distribution of 2 burdens (Figure 1); these might be child stunting and maternal overweight, but the table is applicable to any binary measures of under- or overnutrition that might co-occur at the individual or household level. What most investigators have done is treated the whole table as a single binary variable (in cell A compared with not) and modeled this variable using logistic regression. There are at least 3 flaws to this approach.

FIGURE 1.

Contingency table for an analysis of any 2 nutritional burdens.

First, the use of simple regression approaches to model the determinants of nutritional status or of co-occurrence does not allow for correct specification of the causal model, as has been developed by UNICEF, the Scaling Up Nutrition movement, and many others, to specify immediate, proximal, and underlying risk factors (6)—logistic regression treats all predictors as coequal. Some form of structural modeling, such as path analysis, is required to make appropriate sense of the pattern of relations among the predictors.

Second, there are in fact 2 binary variables, specifically the individual burdens. Whether or not being in cell A can be meaningfully predicted based on other characteristics of the household, over and above the predictor of either burden considered separately, depends on the joint distribution of the 2 variables and their individual determinants. At the most simplistic level, there can be no causal relation if there is no statistical association. Determination as to whether or not there is a statistical association requires, as epidemiologists emphasize, some referent. In a contingency table, the referent is the null hypothesis of no association between the row variable and the column variable, whereby the expected cell count reflects the marginal distributions. For example, if the prevalence of child stunting is 30% and that of maternal overweight is 50%, one would expect 15% of households to have a “dual burden.” Of course there are individuals in cell A. Not having any would be very unusual and certainly of interest. The question is incorrectly posed. The question should be whether or not the number of individuals in cell A is different from what would be expected based on the marginals.

Dieffenbach and Stein (9) addressed this question a few years ago by analyzing all the then-available Demographic and Health Survey data sets that had child and maternal anthropometry. In these data sets, child stunting prevalence ranged from 6.6% to 50%, and maternal overweight prevalence ranged from 1.6% to 70%. The prevalence of mother–child dyads in which the mother was overweight and the child was stunted, i.e., cell A in Figure 1, ranged from 0.5% to 16%. Of the 121 data sets analyzed, only 2 had a prevalence of dual burden greater than would be predicted on the basis of the marginals and even in these 2 data sets the difference between the observed and the expected prevalence was trivial and within the confidence limits of the expected value. Across data sets, the 2 burdens were inversely associated with an R-squared value of 0.47, but only maternal overweight was associated with the prevalence of being “dual burden.” The authors concluded that the combination of child stunting and maternal overweight is not a statistically distinct phenomenon that needs describing over and above understanding the determinants of the marginal distributions, and that the driver of this particular form of “dual burden” household is maternal overweight.

The co-occurrence of stunted children and overweight women within a household is not the only form of dual burden, of course. Varghese and Stein (10) identified 13 within-individual or within-household dual burdens (Table 1), and analyzed their co-occurrence within individuals in India, using data from the Indian Fourth National Family Health Survey, which allows for analyses at district (n = 640) and State (n = 36) levels. The authors found that measures of undernutrition (stunting, wasting, and anemia) tended to co-occur, but that there was not statistical excess of any dual burden that included both undernutrition and overweight, either of the mother or of the child. They concluded that, at present, the dual burden of undernutrition and overnutrition is not a public health problem in India, but that the public health community does still need to be cognizant of its potential emergence as India's nutrition transition continues to develop.

TABLE 1.

Forms of dual burden at individual or at household levels¹

Within individuals	Child	Anemic and underweight
		Anemic and overweight
		Anemic and stunted
		Stunted and overweight
		Stunted and underweight
	Mother	Anemic and overweight
		Anemic and underweight
Within households	Mother anemic	Child anemic
	Mother anemic	Child overweight
	Mother anemic	Child stunted
	Mother anemic	Child underweight
	Mother overweight	Child anemic
	Mother overweight	Child overweight
	Mother overweight	Child stunted
	Mother overweight	Child underweight
	Mother underweight	Child anemic
	Mother underweight	Child overweight
	Mother underweight	Child stunted
	Mother underweight	Child underweight

¹Individuals cannot simultaneously be both overweight and underweight. These 2 burdens can coexist at the household level.

The third concern is also statistical in nature but is relevant to nutrition programming. In the contingency table, cell D (neither burden present) is potentially equally as informative as cell A. Co-occurrence of “well-nourished” dyads is interesting, especially if one wants to find out what works to improve nutrition. Furthermore, one might be very interested in what predicts agreement (either both well-nourished or both malnourished) as compared with “disagreement” (1 burden present), and researchers might be interested in the direction of such disagreement—does it matter whether the child is stunted OR the mother is overweight? Do these dyads have different predictors? Looking at the whole of the 2 × 2 table, using all the information, may provide insights that a binary focus on cell A alone cannot.

There is no doubt that the prevalence of adult overweight is increasing dramatically, and much of this increase is occurring in low- and middle-income regions (11). A question of major implications for nutritionists concerns the longitudinal relation between linear growth and overweight. Adults in low- and middle-income settings are short primarily because they experienced growth faltering in childhood. Does being short predispose to overweight? The late David Barker, who was influential in highlighting the importance of prepregnancy, intrapregnancy, and early childhood nutrition and protection from infection to the health of the developing offspring, observed that it is short, overweight mothers whose offspring have the highest risk of diabetes (12). Therefore, it is critical to examine whether being short [which is a function of early-life nutrition and cannot be substantially corrected, as far as we know, after the first few years of childhood (13)] is causally related to being overweight, recognizing that overweight can occur at any age and can, in some circumstances, be reversed. One can, and should, ask the question whether growth faltering per se increases the risk of overweight and its consequent complications.

In some recent work, Hanson et al. (14) revisited the data from the Birth to Twenty cohort in South Africa, examining whether individuals who were stunted (i.e., had a HAZ < −2) in early childhood were more likely to be overweight in late adolescence than those not stunted in childhood. In this study sample, the prevalence of stunting at age 2 y was close to 20%. Males and females had very different patterns of BMI over childhood and adolescence. However, in both males and females, there were no meaningful differences in BMI at any age between those who were stunted at age 2 y and those who were not stunted.

A second piece of evidence comes from Peru. The Young Lives study has been following a cohort of children in Ethiopia, India, Peru, and Viet Nam since they were recruited at average age 1 y (the range was 6–18 mo). In the Peru sample, ∼25% of children in the sample were stunted at the time of the baseline measurement (15). Overall, mean BMIZ was >0 at all ages and decreased with age. In the 25% of children who were stunted at Round 1, the BMIZ distribution was shifted downwards at every age, compared with the children who were not stunted at Round 1. Children who were stunted were, if anything, less likely to be overweight later in childhood.

So, in at least 2 very different settings, early life growth faltering was not associated with an increased incidence of overweight and obesity. This does not imply that there is not an increasing problem of overweight. There certainly is, but it is not concentrated in those who are stunted.

Observational research can only take us so far in facilitating causal inference. The Institute of Nutrition of Central America and Panama Nutrition Supplementation Longitudinal Study was launched almost 50 y ago as a cluster-randomized community-based trial of improved nutrition, with the short-term outcomes of interest being cognitive development and, to validate the fidelity of the intervention, linear growth (16). Four villages in the eastern highlands of Guatemala were randomly assigned to receive either Atole or Fresco. Atole was a blend of Incaparina, skim milk powder, and some sugar for flavor. Fresco contained sugar as its sole source of calories and was fortified to provide the same micronutrient content as the Atole. The supplement was provided twice daily in a central location in each village. All women and young children were encouraged to attend. Intake was ad libitum. Intakes were recorded for pregnant and lactating women and for children under 7 y of age. The study also provided primary care in the study villages.

The nutrition intervention was successful in reducing the prevalence of stunting. Children born ≥3 y after the study started, and hence who were exposed to Atole for the whole of their first 3 y of life, had half the prevalence of severe stunting (HAZ < −3) as compared with children in the Fresco villages (17). There was no accompanying increase in weight for height. This corresponds to a mean difference in height at age 3 y of ∼2.5 cm.

After the end of the original study, follow-up waves of field work were conducted in 1988, 1997, 2002–2004, and 2004–2006, and a wave has recently completed. Thus, there are prospective data on the cohort members living in Guatemala that span >40 y. As a group, the cohort is overweight and has a high prevalence of cardiometabolic disease (unpublished observations). In the 12 y in between the 2002–2004 wave and the 2015–2016 wave there was a marked increase in the prevalence of a wide range of measures of adiposity and of cardiometabolic disease. Although the men started from a lower baseline, especially for measures of adiposity, the increases in this 12-y period were also dramatic for them.

We have started to examine the extent to which the current state of health might be related to the intervention. Ford et al. (18) used latent class group trajectory approaches to characterize BMI from birth through adulthood. They identified 2 trajectories in women and 3 in men. Of note, these trajectories diverged very early in life and did not subsequently regress. Hence, factors present in early life are important; socioeconomic status at time of birth was 1 such factor. Ford et al. (19) have also analyzed the role of exposure to the original intervention on measures of adiposity and of cardiometabolic disease. The intervention ran from early 1969 through September 1977. Of primary interest was exposure during the “first 1000 days,” from conception to the second birthday. Ford et al. (19) operationalized this period by assuming that all births occurring >9 mo after the launch of the intervention, and >2 y before the intervention ended, were exposed fully to the intervention, and considered all others to be unexposed. They employed a difference-in-difference approach to isolate exposure to Atole during “the first 1000 days” from exposure to Atole in other periods and exposure to Fresco, whether during the first 1000 d or not. In earlier work with this cohort, this “double-difference” predicted substantial increases in schooling, cognitive performance, and income. Drawing from evidence from famine studies in the Netherlands, China, and elsewhere (20), as well as the work of David Barker and many others on the association between early-life nutrition and later cardiometabolic disease, we hypothesized that individuals exposed to improved nutrition in early life would have better cardiometabolic health status in adulthood.

For diabetes, Ford et al. (19) observed that individuals exposed to Atole in the first 1000 d had approximately half the prevalence of diabetes as compared with the rest of the cohort. Interestingly, this estimate was not altered substantially with control for adiposity. On the other hand, for metabolic syndrome as a whole and for hypertension, there was no substantial difference in prevalence between those exposed to Atole in the first 1000 d and those not so exposed. Finally, for a range of measures of adiposity, there was an increase in prevalence among those exposed to Atole in the first 1000 d. We are still working out why we see these diverging pathways, and in particular what it is about exposure to Atole that leads to both increased adiposity and lower diabetes. At a mechanistic level, results from the Dutch Famine (21) suggest that epigenetic changes may mediate this association. The increase in adiposity might also be a consequence of lifestyle choices that result from the increased income that Hoddinott et al. (22) have reported resulted from exposure to improved nutrition.

From these data, one may conclude that nutritionists concerned with the well-being of preschool children have an increasingly complicated role, if efforts to prevent growth faltering and promote cognitive development also predispose to adiposity. The nutrition community has had major success in reducing the global prevalence of stunting, but there is still more to do. We need to continue to provide integrated programs of nutrition, sanitation, and child development to ensure that children can thrive and develop into productive adults. We need to continue to develop locally grounded interventions to ensure that children of all ages have access to nutritious foods. But we need to recognize possible adverse impacts as well. We need to be working with our colleagues who plan and deliver nutrition programs to school-age children to ensure that they are attuned to the changing needs of children in their communities and to the futility of using school-based systems to reverse earlier linear growth failure. Indeed, evidence from Chile has shown that when undernutrition is no longer a public health concern, preschool and school feeding programs may need to be cut back to prevent the development of child overweight, but this intervention is not always effective (23). Widespread availability in and around schools and the consequent consumption of cheap calorie-dense foods cannot help but promote overweight. As physical activity levels continue to decline in children and adults worldwide as a result of sedentarization of leisure and increased mechanization of labor (24), we need to be promoting nutrient-dense foods, not the calorie-dense, highly processed, and shelf-stable food products that are swamping local food markets, to all people, be they children, adolescents, or adults. How we do that, given the higher cost and reduced shelf-life of high-quality food, is the challenge for the rest of the 21st century.

Notes

This article is adapted from the text of the 2018 EV McCollum International Lecture in Nutrition, delivered at the 2018 meeting of the American Society for Nutrition, Boston, MA, 9 June, 2018.

The author reported no funding received for this study.

Author disclosures: ADS, no conflicts of interest.

Perspective articles allow authors to take a position on a topic of current major importance or controversy in the field of nutrition. As such, these articles could include statements based on author opinions or point of view. Opinions expressed in Perspective articles are those of the author and are not attributable to the funder(s) or the sponsor(s) or the publisher, Editor, or Editorial Board of Advances in Nutrition. Individuals with different positions on the topic of a Perspective are invited to submit their comments in the form of a Perspectives article or in a Letter to the Editor.