Overweight Is More Prevalent Than Stunting and Is Associated with Socioeconomic Status, Maternal Obesity, and a Snacking Dietary Pattern in School Children from Bogotá, Colombia^,

Abstract

The objectives of this study were to estimate the prevalence of overweight in school-aged children from Bogotá, Colombia and to examine its associations with sociodemographic characteristics, dietary patterns, and indicators of physical activity. We measured height and weight in 3075 children between 5 and 12 y of age who attended public primary schools in 2006 and we obtained information on maternal sociodemographic and anthropometric characteristics. The survey was representative of children from low and middle socioeconomic backgrounds. The prevalences of child overweight (including obesity) and obesity according to the International Obesity Task Force criteria were 11.1 and 1.8%, respectively. The prevalence of stunting was 9.8%. In multivariate analysis, child overweight was positively associated with indicators of higher socioeconomic status (SES), including low maternal parity and ownership of household assets. The prevalence of overweight was 3.6 times greater in children whose mothers were obese compared with children whose mothers had an adequate BMI (adjusted prevalence ratio = 3.61; 95% CI = 2.64, 4.93). Child overweight was positively associated with adherence to a “snacking” dietary pattern (P-trend = 0.06) and to frequent intake of hamburgers or hot dogs (adjusted prevalence ratio for at least once per week vs. never = 1.93; 95% CI = 1.03, 3.62), independent of total energy intake and other potential confounders. Time spent viewing television or playing outside the household were not significantly related to the prevalence of child overweight. In conclusion, child overweight in Bogotá is more common than stunting and is associated with higher SES, maternal obesity, and a snacking dietary pattern.

Introduction

Overweight and obesity in school-aged children are of considerable public health importance. Not only is childhood obesity associated with an increased risk of high blood pressure, diabetes, respiratory disease, and orthopedic disorders during childhood, but it can also have adverse effects on psychosocial development and academic performance (1). Furthermore, obese school-aged children are between 3.9 and 6.5 times more likely to become obese adults compared with nonobese school-aged children (2). While these issues have been well recognized in most developed countries, less information is available from developing countries.

Several Latin American countries have undergone considerable economic growth in recent years. Whereas this growth coincided with declines in rates of infant mortality, malnutrition, and infectious diseases, rates of overweight and obesity are on the rise. In fact, the rate of overweight among preschool children in several Latin American countries has now surpassed the rate of stunting (3). A variety of lifestyle factors are contributing to this nutrition transition. Traditional diets that were rich in whole grains, fiber, fruits, and vegetables are being replaced with more Westernized diets that have higher contents of saturated fat and refined carbohydrates (4). Compounding the negative changes in diet quality is the problem of reduced physical activity. For example, 9- to 16-y-old children living in Mexico City reported watching more than 4 h/d of television (TV)⁷ and engaging in moderate or vigorous physical activity for <2 h/d (5).

Little is known about the rates of child overweight and obesity in Colombia, a country undergoing the nutrition transition. The demographic, socioeconomic, and dietary factors that are associated with child overweight and obesity in this country are also uncertain. Ascertaining these associations is an important step to inform public health policy and to design appropriately targeted interventions that will effectively address the problem of childhood overweight. Our primary aim was to determine the prevalence of child overweight and its associations with demographic characteristics, socioeconomic status (SES), and indicators of physical activity in Bogotá, Colombia. Our secondary aim was to examine dietary correlates of overweight in a subsample of the children.

Methods

Study population.

This study was conducted as part of a health and nutrition project involving primary school children in Bogotá, Colombia. Details of the study design have been previously published (6). To summarize, in February 2006, we enrolled 3202 children 5–12 y of age at public schools in Bogotá with the use of a cluster sampling strategy. The sampling units were the classrooms in all public schools. The study population is representative of low- and middle-income families from Bogotá, considering that the public school system enrolled 57% of all primary school children in the city by the end of 2005 and 89% of them were from low- and middle-income families (7).

At the start of the study, a self-administered questionnaire was distributed to parents to collect information on sociodemographic characteristics, including age, parity, education level, and household SES; the frequency and duration of the children's periods spent viewing TV or playing videogames; and the time spent playing outdoors. The questionnaire also asked each mother to report her weight and height. We received completed questionnaires from 2446 (81%) households (i.e. the families of 2637 children after accounting for siblings). During the following weeks, trained research assistants visited the schools and collected anthropometric measurements from the children using standardized techniques (8). Height was measured to the nearest 1 mm using wall-mounted portable Seca 202 stadiometers and weight was measured to the nearest 0.1 kg using Tanita HS301 solar-powered electronic scales.

Between May and June, trained dietitians administered a 38-item FFQ to a group of 1027 mothers who attended meetings regularly scheduled by the school, to obtain information on the children's usual dietary intake. Children with dietary information were slightly younger and more likely to live in lower SES neighborhoods than children without dietary information. There were no other differences between these groups. We described reference portion sizes in natural units or standard measures for commonly consumed foods in this population and inquired about the frequency of consumption during the month prior to the interview. Additional details on the FFQ have been published (6). Anthropometric measurements were also taken for a subsample of 671 of the mothers at the time of the FFQ interview.

Data analyses.

For this analysis, we excluded data from 127 children who were missing height or weight measurements; thus, the final sample size was 3075 children.

Child overweight or obesity was defined according to BMI cut-off points for sex and age corresponding to ≥25 in adults, following the International Obesity Task Force (IOTF) recommendations (9). In addition, height-for-age Z-scores were calculated according to the WHO/National Center for Health Statistics reference (10) and stunting was defined as height-for-age Z < −2. Maternal BMI was calculated as kg/m² from measured height and weight in 26.1% of the mothers and from reported values otherwise. Maternal nutritional status was then classified according to BMI categories as underweight (<18.5), adequate (18.5–24.9), overweight (25.0–29.9), or obese (≥30.0) (11).

We used the binary outcome ‘child is overweight or obese’ for all analyses. Predictors of interest included child's sex, age, and birthplace, whether the child was stunted, average weekly number of hours spent watching TV or playing video games, and average weekly number of hours spent playing outside; mother's age, BMI, education level, and parity; number of household assets from a list that included refrigerator, bicycle, blender, TV, stereo, and washing machine, and the socioeconomic stratum of the household according to the city's classification of neighborhoods' public services fees. Continuous predictors including child's age, time spent viewing TV, playing video games, and playing outside the home, mother's age, education, parity, and BMI, and number of household assets were categorized into ordinal groups defined a priori (Table 1). In univariate analyses, we estimated the prevalence of child overweight by categories of each predictor and tested the association using the chi-square test for dichotomous predictors and the Cochrane-Armitage test for ordinal predictors. We obtained adjusted prevalence ratios and 95% CI by fitting multivariate binomial regression models with the correlates that were significant in the unadjusted analyses at the P < 0.10 level or that seemed relevant from a mechanistic point of view. We retained variables that remained significant at the P < 0.05 level as well as child's age and sex in the final multivariate models. We used an exchangeable correlation matrix in all models to account for within-household correlations among siblings. The effect of clustering due to the sampling strategy was negligible and not considered in the models for parsimony.

TABLE 1.

Sociodemographic correlates of overweight in school children from Bogota, Colombia

		Child is overweight or obese2		Adjusted prevalence ratio (95% CI)4
Characteristic	n1	%	P3		P5
Child characteristics
Sex			0.44		0.52
Female	1585	10.7		1.00
Male	1490	11.5		1.07 (0.88, 1.30)
Age, y			0.38		0.45
5–6	550	10.2		1.00
7–8	963	12.4		1.23 (0.91, 1.66)
9–10	1259	11.5		1.13 (0.85, 1.52)
11–12	277	6.9		0.72 (0.44, 1.17)
Born in Bogota			0.01		0.07
No	298	7.1		1.00
Yes	2231	11.8		1.49 (0.97, 2.30)
Stunted			0.0002		0.01
No	2773	11.8		1.00
Yes	302	4.6		0.51 (0.31, 0.85)
TV viewing/video game playing, h/wk			0.37		0.25
≤10.0	756	11.1		1.00
10.1–20.0	608	10.0		0.87 (0.64, 1.18)
20.1–30.0	469	13.2		1.16 (0.86, 1.57)
>30.0	352	11.9		1.08 (0.76, 1.53)
Playing outside, h/wk			0.62		0.32
≤2.00	650	11.1		1.00
2.1–5.0	445	12.1		1.06 (0.76, 1.47)
5.1–8.0	330	10.6		0.90 (0.62, 1.31)
>8.0	638	10.5		0.92 (0.68, 1.25)
Maternal characteristics
Age, y			0.55		0.13
20–29	609	11.5		1.00
30–34	677	9.6		0.88 (0.65, 1.21)
35–39	586	12.0		1.12 (0.81, 1.55)
≥40	627	11.8		1.18 (0.85, 1.64)
Education			0.006		0.15
Incomplete primary (1–4 y)	200	7.0		1.00
Complete primary (5 y)	492	8.7		1.21 (0.68, 2.16)
Incomplete secondary (6–10 y)	642	11.4		1.42 (0.82, 2.44)
Complete secondary (11 y)	997	13.1		1.50 (0.88, 2.55)
University (≥12 y)	171	10.5		1.20 (0.60, 2.40)
Parity			<0.0001		<0.0001
1	301	16.9		1.00
2	902	14.0		0.79 (0.59, 1.06)
3	735	9.1		0.51 (0.37, 0.71)
4	313	7.4		0.44 (0.28, 0.69)
≥5	232	4.3		0.25 (0.13, 0.49)
BMI, kg/m2			<0.0001		<0.0001
<18.5	78	3.9		0.50 (0.16, 1.57)
18.5–24.9	1473	7.9		1.00
25.0–29.9	584	16.6		2.22 (1.73, 2.85)
≥30.0	158	26.6		3.61 (2.64, 4.93)
SES
Number of home assets6			<0.0001		0.008
0–1 item	217	6.0		1.00
2 items	314	8.3		1.24 (0.66, 2.34)
3 items	409	10.3		1.34 (0.73, 2.43)
4 items	498	10.0		1.32 (0.73, 2.37)
5 items	553	12.3		1.50 (0.85, 2.65)
6 items	540	15.0		1.83 (1.04, 3.21)
Household socioeconomic stratum7			0.002		0.10
1 (lowest)	229	7.4		1.00
2	934	9.9		1.09 (0.67, 1.77)
3	1311	12.5		1.26 (0.79, 2.03)
4 (highest)	62	17.7		1.53 (0.78, 3.01)

¹Totals may be <3075 due to missing values.

²According to the IOTF classification (9).

³For variables representing ordinal categories, P is from the Cochrane-Armitage test. For binary variables, P corresponds to the chi-square test.

⁴Prevalence ratios and 95% CI are from binomial regression models with ‘child is overweight or obese’ as the outcome and predictors that included indicator variables for the following covariates: sex, age, stunting, maternal BMI, parity, and number of household assets. Estimates for TV viewing and physical activity were adjusted for each other. The indicator method was used for variables with missing values.

⁵Test for trend when a variable representing ordinal categories of the predictor was introduced into the multivariate model as continuous. The adjusted P for dichotomous predictors corresponds to the Wald test.

⁶From a list that included refrigerator, bicycle, blender, TV, stereo, and washing machine.

⁷According to the city's classification of neighborhoods' public services fees.

To assess the relation between diet and child overweight, we conducted principal component analysis to identify dietary patterns, using as input the 38 items in the FFQ, in servings per day. The factors obtained were rotated by an orthogonal transformation to achieve a simpler structure that assists interpretability. We considered eigenvalues > 1, the Scree test, and the general interpretability of the factors to determine the number of factors to retain. The standardized frequencies of intake for each food group were multiplied by the factor score coefficients and the sum of these products was the score for each derived factor. We identified 4 dietary patterns: cheaper protein (e.g. hamburgers/hotdogs, freshwater fish, cow tripe/liver/spleen, cassava, chicken giblet), snacking (e.g. candy, ice cream, packed fried snacks, soda, sugar-sweetened fruit-flavored drinks), traditional/starch (e.g. rice, potato, plantain), and animal protein (e.g. milk, yogurt, red meat, cheese, poultry). The factor scores produced in the principal component analysis were categorized into quartiles to represent adherence to each pattern. We also examined the prevalence of overweight in relation to the frequency of intake (in servings per day, week, or month) of commonly eaten snacks and fast foods in this population, including soda, packed fried snacks (potato, plantain, or cassava chips, fried pork skin, etc.), and hamburgers or hotdogs using the information collected with the FFQ. Each dietary pattern or food was used as an independent predictor in binomial regression models with child overweight or obesity as the outcome. These models were adjusted for total energy intake estimated from the FFQ, the child's age and sex, and significant sociodemographic correlates of overweight in the population. All analyses were conducted using SAS software version 9.1 (SAS Institute).

The parents or primary caregivers of all children gave written informed consent prior to enrollment. The study protocol was approved by the Ethics Committee of the National University of Colombia Medical School. The Human Subjects Committee at the Harvard School of Public Health approved the use of data from the study.

Results

Children were 8.9 ± 1.8 y old (mean ± SD) and 48.5% of them were male. Eleven percent of the children were overweight or obese (n = 341), 1.8% were obese (n = 54), and 9.8% were stunted (n = 302).

Sociodemographic correlates of child overweight.

Ownership of household assets was positively associated with child overweight (adjusted P-trend = 0.008), whereas maternal parity was negatively related to the prevalence of child overweight; the adjusted prevalence of overweight was 75% lower in children with ≥4 siblings compared with children with no siblings (P < 0.001) (Table 1). The adjusted prevalence of overweight was 49% lower in stunted than in nonstunted children (P = 0.01). This association did not change after further adjustment for household income or dietary patterns. Overweight was >3.5 times more prevalent in children whose mothers were obese compared with those whose mothers had an adequate BMI, after adjustment (P < 0.001). Although city of birth, maternal education, and neighborhood SES were significantly related to overweight in univariate analyses, these variables were no longer significant predictors after adjustment. Child's age or sex were not significantly associated with overweight, nor was the amount of time spent watching TV/playing video games or playing outside the home.

Dietary correlates of child overweight.

Adherence to the snacking pattern was positively associated with child overweight after controlling for age, sex, total energy intake, maternal BMI, and number of household assets; the prevalence of overweight was twice as high in the most adherent quartile as in the least adherent quartile (P = 0.05) (Table 2). There appeared to be some threshold in the association, as children in the lowest quartile of adherence to the snacking pattern had approximately one-half the prevalence of overweight compared with those in the combined 2nd, 3rd, and 4th quartiles (adjusted prevalence ratio = 0.53; 95% CI = 0.30, 0.91; P = 0.02). This apparent threshold was not accounted for by specific foods in the snacking pattern, when they were analyzed separately in relation to overweight. In the analysis of individual snacks and fast food intake, the prevalence of overweight was nearly twice as high among children who consumed a hamburger or hotdog at least once per week compared with those who never consumed either food (adjusted prevalence ratio = 1.93; 95% CI = 1.03, 3.62; P-trend = 0.03), after adjusting for age and sex, SES, maternal BMI, and total energy intake. The prevalences of overweight according to the frequency of soda intake were 8.1% for never, 10.5% for ≤1–3 times/mo, 12.7% for 1–6 times/wk, and 11.3% for once or more per day (adjusted P-trend = 0.29). The prevalences of overweight according to fried snacks intake (including potato, plantain, and pork skin chips) for the same frequency of intake categories were 7.6, 8.9, 13.3, and 10.0% (adjusted P-trend = 0.06). In supplemental analyses, we examined whether adherence to the snacking pattern could mediate the positive association observed between ownership of household assets (an indicator of SES) and child overweight, because children from families with higher purchasing power could have pocket money available to buy these foodstuffs outside their homes. When we included the snacking pattern with household assets into the same multivariate model, the association between home assets and overweight was only slightly attenuated, which does not support that snacking was necessarily in the causal pathway between SES and overweight.

TABLE 2.

Dietary correlates of overweight in children 5–12 y of age in Bogotá, Colombia¹

		Child is overweight or obese2		Unadjusted prevalence ratio (95% CI)4		Adjusted prevalence ratio Model 1 (95% CI)6		Adjusted prevalence ratio Model 2 (95% CI)6
Adherence to dietary patterns	n	% (n)	P3		P5		P5		P5
Snacking pattern			0.17		0.13		0.03		0.06
Q1 (lowest adherence)	234	6.0 (14)		1.00		1.00		1.00
Q2	240	13.3 (32)		2.23 (1.22, 4.07)		2.10 (1.15, 3.84)		1.98 (1.09, 3.59)
Q3	241	13.7 (33)		2.29 (1.26, 4.16)		2.11 (1.15, 3.86)		1.80 (0.99, 3.26)
Q4 (highest adherence)	239	10.0 (24)		1.68 (0.89, 3.17)		2.09 (1.03, 4.23)		1.95 (0.99, 3.84)
Cheaper protein			0.01		0.007		0.03		0.04
Q1 (lowest adherence)	240	12.5 (30)		1.00		1.00		1.00
Q2	239	15.1 (36)		1.20 (0.77, 1.89)		1.22 (0.77, 1.94)		1.21 (0.78, 1.89)
Q3	241	8.7 (21)		0.70 (0.41, 1.18)		0.66 (0.37, 1.16)		0.71 (0.42, 1.22)
Q4 (highest adherence)	234	6.8 (16)		0.55 (0.31, 0.98)		0.64 (0.35, 1.19)		0.65 (0.35, 1.19)
Traditional/starch			0.08		0.07		0.22		0.66
Q1 (lowest adherence)	238	11.3 (27)		1.00		1.00		1.00
Q2	238	14.3 (34)		1.26 (0.79, 2.02)		1.18 (0.73, 1.90)		1.35 (0.84, 2.17)
Q3	241	10.0 (24)		0.88 (0.52, 1.47)		0.88 (0.53, 1.46)		0.96 (0.58, 1.59)
Q4 (highest adherence)	237	7.6 (18)		0.67 (0.38, 1.18)		0.71 (0.37, 1.39)		0.94 (0.49, 1.81)
Animal protein			0.77		0.76		0.31		0.65
Q1 (lowest adherence)	236	10.6 (25)		1.00		1.00		1.00
Q2	240	8.8 (21)		0.83 (0.48, 1.43)		0.79 (0.45, 1.39)		0.82 (0.47, 1.44)
Q3	240	14.2 (34)		1.34 (0.82, 2.17)		1.37 (0.83, 2.27)		1.33 (0.82, 2.16)
Q4 (highest adherence)	238	9.7 (23)		0.91 (0.53, 1.56)		1.09 (0.60, 1.97)		0.92 (0.51, 1.65)

¹In a subsample of 954 children with dietary information.

²According to the IOTF classification (9).

³Cochrane-Armitage test.

⁴Unadjusted prevalence ratios and 95% CI are from binomial regression models with ‘child is overweight or obese’ as the outcome.

⁵Test for trend.

⁶Adjusted prevalence ratios and 95% CI are from binomial regression models with ‘child is overweight or obese’ as the outcome. Covariates in Model 1 included child's age, sex, and total energy intake. Covariates in model 2 (n = 893) included child's age, sex, total energy intake, maternal BMI, and number of household assets. The indicator method was used for missing values in the number of household assets and maternal BMI variables.

There was an inverse trend between adherence to the cheaper protein pattern and child overweight (P-trend = 0.04). The traditional/starch and animal protein patterns were not significantly associated with the prevalence of child overweight.

Discussion

This study examined the prevalence of child overweight in a representative sample of low- and middle-income school children from Bogotá, Colombia, in relation to their SES, dietary patterns, and indicators of physical activity. The prevalence of child overweight (including obesity) was 11.1%. This figure is larger than the 5.4% reported for children 5–9 y of age living in urban areas by the Colombian National Nutrition Survey of 2005 (12). However, the figures cannot be compared directly, because the National Survey did not report data for Bogotá, also included children from higher SES than those in our sample, and used weight-for-height Z-scores > 2 to define overweight instead of the IOTF criteria that employs BMI-for-age centile curves. Our prevalence estimate is somewhat lower than rates in other Latin American settings, including Florianópolis, Brazil, where 22% of 7- to 10-y-old children were overweight or obese in 2005 (13); Chile, where 18.5% of school children were obese by 2005 (14); and Mexico, where the national rate of overweight among school children 5–11 y of age was 26% in 2006 (15).

In this low- and middle-income group, we found that child overweight or obesity was positively associated with SES, as indicated by ownership of household assets, neighborhood of residence, and low maternal parity. Hernandez et al. (5) similarly found that the odds of child obesity were higher in middle-income vs. low-income areas in Mexico City by 1999. Martorell et al. (16) also reported that child overweight was more prevalent in families of higher SES compared with poorer families in their review of studies published between 1982 and 1996 describing obesity trends in Latin American children. The consistency of our findings with these reports and the greater prevalence of overweight than stunting support the notion that Colombia is experiencing the nutrition transition with regards to child overweight along with several of its Latin American neighbors.

Our finding of a reduced prevalence of overweight among stunted children is consistent with a recent analysis of 79 population-based surveys from 24 countries in Latin America and the Caribbean (17) among preschool-aged children. By contrast, Popkin et al. (18) reported an increased risk of overweight among stunted children aged 3–9 y in Brazil, China, Russia, and South Africa. The income-adjusted risk ratios ranged from 1.7 in Brazil to 7.8 in Russia. It is possible that the association between stunting and overweight changes as countries go through different stages of the nutrition transition.

Maternal BMI was a particularly strong predictor of child overweight. The prevalence of child overweight was >3.5 times greater in children with mothers who were obese compared with children whose mothers had an adequate BMI. This positive association with maternal BMI had been previously observed in populations from developed countries (19–25), in people of Mexican origin living in the US (26), and in rural Mexico (27). This association likely reflects the rapid changes in socioeconomic conditions and lifestyle patterns that affect the shared family environment and that have already resulted in increased rates of overweight and obesity among Latin American adults in recent years. Parity, which may reflect both maternal and socioeconomic characteristics, was also strongly and inversely related to child overweight. Reilly et al. (24) did not observe this association in a cohort of children in the United Kingdom; however, the difference in findings could be due to the fact that parity may be a stronger indicator of SES in our population than in Britain.

The prevalence of overweight was positively associated with adherence to a ‘snacking’ pattern. Field et al. (28) did not find a significant association between the consumption of snack foods and BMI in 9- to 14-y-old children in the US, nor did Phillips et al. (29) in a longitudinal analysis of nonobese adolescent girls. The difference in findings across studies could be due to different definitions of snack foods; whereas these authors reported a list of certain foods they considered as snack foods in their FFQ, we used principal components analyses to define a snacking pattern. We also found that frequent intake of hamburgers or hotdogs was related to child overweight. These foods, which are typically eaten away from home, are becoming increasingly popular in Colombia. Taveras et al. (30) found that higher consumption of fried food away from home was associated with increasing BMI among older children and adolescents. The results highlight the possibility that replacement of traditional diets with fast foods eaten away from home may be an important contributor to childhood obesity in this setting.

We did not find that soda consumption was significantly related to the prevalence of child overweight in Bogotá, consistent with observational studies conducted mostly in developed countries (31–34). A recent randomized trial in Chile showed that replacement of sugar-sweetened beverages with milk for 16 wk among children 8–10 y of age did not affect BMI or fat mass but resulted in an apparent increase in lean body mass (35); the intervention also increased linear growth in boys. Whereas such specific interventions have promising potential to curb the growing obesity epidemic in Latin America, it may be necessary to address additional dietary factors that are being identified in the etiologic pathways leading to overweight and obesity.

Unexpectedly, the amount of time spent playing outside the household or watching TV and playing video games was not associated with the prevalence of overweight. Several previous studies found that child or adolescent overweight was positively associated with TV viewing and negatively associated with indicators of physical activity (5,36,37). Lack of precision in the ascertainment of TV viewing and physical activity in our study could have affected our ability to observe an underlying association. A review of physical activity assessment methods in children concluded that self or parental reports of child physical activity have low to moderate validity (38). Longitudinal studies with more precise measures of these exposures would be needed to ascertain these relations.

Our study has some limitations. Because it was cross-sectional in nature, it is not possible to draw conclusions about causal relations between sociodemographic characteristics or dietary patterns and child overweight or obesity. Also, the FFQ was administered to the children's mothers and this may have resulted in misclassification of the children's dietary intake, especially of foods eaten away from home. This could have limited our ability to detect associations between dietary factors and overweight. Finally, we cannot generalize the results to all children of school age in Bogotá, including those of higher SES.

In conclusion, we found that child overweight or obesity was more common than stunting and was positively associated with indicators of higher SES, maternal BMI, and a snacking dietary pattern. These findings suggest that school children in urban Colombia are experiencing the adverse effects of the nutrition transition. Longitudinal studies are warranted to examine the associations between sociodemographic and lifestyle patterns and changes in nutritional status over time in this population.