Time trends (1992-2005) in undernutrition and obesity among children under fi ve years of age in Alagoas State, Brazil

The aim of this study was to describe time trends in stunting and obesity in children under five years of age in Alagoas State, Brazil. Two surveys were conducted with representative samples, the first in 1992 (n = 1,228) and the second in 2005 (n = 1,384). Stunting was defined as heightfor-age < -2 standard deviations and obesity as weight-for-height > 2 standard deviations. Prevalence of stunting decreased from 22.5% to 11.4% (PR = 0.50; 95%CI: 0.42; 0.60), while obesity increased from 6.7% to 9.3% (PR = 1.36; 95%CI: 1.04; 1.77). During the same period there was a decrease (p ≤ 0.05) in the proportions of the following variables: rural residence, households without running water, households with more than four members, mothers with more than two children, low birth weight, and mother’s lack of access to prenatal care. Adjustment for these variables significantly reduced the magnitude of associations, and the 95% confidence included 1.0, suggesting that changes in the prevalence of stunting and obesity were mediated by these characteristics. During the period, there was a striking reduction in the prevalence of stunting and an increase in the frequency of obesity. Protein-Energy Malnutrition; Nutrition Surveys; Preschool Child ARTIGO ARTICLE

Time trends (1992-2005) in undernutrition and obesity among children under fi ve years of age in Alagoas State, Brazil Tendência temporal (1992-2005) da desnutrição e da obesidade em menores de cinco anos no Estado de Alagoas, Brasil Tendencia temporal (1992-2005) de la desnutrición y de la obesidad en menores de cinco años en el estado de Alagoas, Brasil

Abstract
The aim of this study was to describe time trends in stunting and obesity in children under five years of age in Alagoas State, Brazil.Two surveys were conducted with representative samples, the first in 1992 (n = 1,228) and the second in 2005 (n = 1,384).Stunting was defined as heightfor-age < -2 standard deviations and obesity as weight-for-height > 2 standard deviations.Prevalence of stunting decreased from 22.5% to 11.4% (PR = 0.50; 95%CI: 0.42; 0.60), while obesity increased from 6.7% to 9.3% (PR = 1.36; 95%CI: 1.04; 1.77).During the same period there was a decrease (p ≤ 0.05) in the proportions of the following variables: rural residence, households without running water, households with more than four members, mothers with more than two children, low birth weight, and mother's lack of access to prenatal care.Adjustment for these variables significantly reduced the magnitude of associations, and the 95% confidence included 1.0, suggesting that changes in the prevalence of stunting and obesity were mediated by these characteristics.During the period, there was a striking reduction in the prevalence of stunting and an increase in the frequency of obesity.

Introduction
Beginning with the National Survey on Family Expenses (ENDEF) in 1974, periodic surveys have been conducted in Brazil, allowing identification of the prevalence, regional distribution, and time trends in the most relevant nutritional problems in the country as a whole and in the major geographic regions.The results of these studies indicate that the Brazilian population is undergoing a nutritional transition, characterized by a reduction in the prevalence of undernutrition and a slow increase in obesity 1 3 .
These surveys were planned to obtain indicators for the Brazil as a whole and its major geographic regions, but do not allow disaggregating the data according to States of the country 4 .Considering that historically the prevalence of undernutrition has been higher in regions with lower economic development (stunting is two to four times more common in Northeast as compared to Southeast Brazil, according to the national surveys from 2006-2007 and 1974-1975, respectively) 2 , it is valid to assume that such disparities also exist at the State level, given the important socioeconomic differences among the latter.
The State of Alagoas is located in the Northeast Region of Brazil.Compared to other States, it is known for having the country's most precarious social indicators 5 .In this context, due to the epidemiological, demographic, and nutritional transition, with possible changes in the population's pattern of physical activity and eating habits 6 , the possibility exists of an increase in obesity prevalence without proper control of undernutrition, thus establishing the coexistence of both conditions at the population level and characterizing the so-called double burden of nutritional problems and their respective harm to health 7 .
In 1992, the United Nations Children's Fund (UNICEF) conducted a household survey that evaluated a representative sample of children less than five years of age in Alagoas 8 .A new survey was conducted in 2005, this time by researchers from the Alagoas Federal University (UFAL) 9 .This work aimed to analyze the time trends in stunting and excess weight in children under five years of age in the State of Alagoas from 1992 to 2005.

Sampling plan • 1992 survey
In order to obtain a representative sample, the sampling process used a multiple (three-stage) design.The first stage selected the municipalities (counties) to be included in the study (n = 20); the second selected the census tracts in each municipality (n = 8); and the third stage selected a starting point from which 15 consecutive households were visited.Thus, 2,400 households were visited (20 municipalities x 8 tracts x 15 households).Based on data from the Brazilian Institute of Geography and Statistics (IBGE), which estimated 0.57 children under five years of age per household, by the end of the study 1,370 children in this age bracket would be identified.This sample would allow obtaining estimates of the most common health problems and the health services' coverage, with acceptable margins of error 10 : 95% significance (1-α) and 80% study power (1-β).
The municipalities were selected with probability proportional to size, i.e., municipalities with larger populations had higher odds of being included in the sample when compared to smaller ones.For example, the State capital Maceió was selected six times, which meant visiting 48 census tracts there, and thus 720 households.For each municipality selected, a list of all the census tracts was obtained from the IBGE.This list was then used to systematically select the tracts to be included in the study.After selecting the census tract, its corners were numbered, always clockwise, and one corner was randomly selected as the starting point for consecutively visiting 15 households.
The survey identified 1,351 children less than five years of age in the households visited.Of this total, it was possible to obtain information on 1,228, resulting in a loss rate of 9.1%.

• 2005 survey
Calculation of the sample size considered a prevalence rate of 9.5% for stunting 11 , a 1.5% margin of sampling error, 95% confidence interval (95%CI), and population estimated at 308,000 children.The sample size, calculated in StatCalc from Epi Info (Centers for Disease Control and Prevention, Atlanta, USA), was 1,461 children.The sampling process was multiple (three-stage).In the first stage, 20 municipalities (counties) were selected, with probability proportional to the population (systematic sampling).In the second stage, eight Cad.Saúde Pública, Rio de Janeiro, 29(4):793-800, abr, 2013 census tracts were selected in each municipality.In the third, a starting point was selected in each tract, from which nine consecutive residences were visited that had at least one child in the target age bracket.A total of 1,461 children less than five years of age were identified in the households visited.As a result of sampling losses (n = 75; 5.1%) due to refusals or the child's absence from the household, the final sample consisted of 1,386 children.

Weighting for adjustment of the urbanversus-rural relationship in the samples
Since the rural census tracts tended to be smaller than the urban, the study tended to evaluate a smaller proportion of children living in the rural area as compared to estimates from the census.Thus, the study calculated weighting factors based on the proportion of inhabitants in rural versus urban areas according to data from IBGE and the respective proportions obtained in the surveys.Weighting factors were obtained by simple rule of three.In 1992, 625 children were studied in the urban area and 603 in the rural area (n = 1,228).For correction as a function of the values estimated by IBGE, these numbers were multiplied by 1.1370 and 0.8590, respectively.In 2005, 990 children were studied in the urban area and 394 in the rural area (n = 1,384).For adjustment, these values were multiplied by 0.8755 and 1.3128, respectively.

Data collection
The data from the two surveys were collected through home visits conducted in July and August 1992 and from November 2005 to February 2006.Pretested structured forms were used 10 , which were applied to the child's mother or guardian and included demographic, socioeconomic, environmental, nutritional, and health variables.
Anthropometric data (weight and stature) were measured by properly trained and supervised interviewers.In 1992, body weight was measured with a Salter type portable scale (CMS PBW-235; CMS Weighing Equipment, London, England), accurate to 100g, while in 2005 a portable electronic scale was used, with a capacity of 180kg and sensitivity of 100g (Marte PP180®; Marte Balanças Ltda., São Paulo, Brazil).Children older than 24 months had their stature measured in the standing position with a vertical stadiometer, while in children under two years the length was measured with the child supine, using a pediatric (horizontal) stadiometer.The equipment in both surveys included a non-flexible measuring tape, accurate to 0.1cm.

Data analysis
The information contained in the databanks was processed with Anthro, version 2 (World Health Organization; http://www.who.int/childgrowth/software/en/) in order to generate anthropometric indices using the variables sex, age, weight, and stature, based on World Health Organization guidelines (OMS-2006) 12 .These indices were used to produce indicators for the following conditions: underweight (weight-for-age < -2z); wasting (weight-for-height < -2z); stunting (height-for-age < -2z); and obesity (weightfor-height > 2z).The results were expressed as prevalence rates (%) according to age brackets and year of the survey.The comparison between mean z-scores and the observed height-for-age and weight-for-height indices, respectively, in 1992 and 2005 used the Student t test for independent samples.
According to the study's objectives, the dependent variables were stunting and obesity.Socioeconomic, demographic, and health factors were described according to the prevalence rates of their respective categories.For each of these, distribution of the prevalence rates for the target outcome was presented (stunting and obesity).The F statistic was used to compare these proportions in the two surveys.
The observed increases (or decreases) in the prevalence rates in the second survey were described as percentages, using the following equation: [(prevalence in 2005 -prevalence in 1992)/ prevalence in 1992] x 100.
Analysis of trends in the target indicators used Poisson regression with robust adjustment of variance to estimate prevalence ratios (PR).Increases were defined as PR > 1, and the values obtained in 1992 were used as the reference.Socioeconomic, demographic, and health variables that presented statistically significant variation between the two surveys (p < 0.05) were used for adjustment in calculating the PR for stunting and obesity in 2005.
Importantly, the adjustment did not aim to control for confounders, but to identify factors associated with the observed changes (that is, possible determinants).
Data analysis was performed with Stata 12.0 (Stata Corp., College Station, USA), using the svy command, since the data were from complex samples.

Results
A total of 1,228 children were studied in the 1992 survey (619 boys and 609 girls) and 1,384 in 2005 (677 boys and 707 girls).Table 1 shows the distribution of nutritional disorders as defined by the anthropometric extremes.In the two surveys, the prevalence rates were considered low for weightfor-height and weight-for-age deficits (1.3% and 5.6% in 1992 and 0.9% and 2.7% in 2005, respectively).Stunting was more concentrated in the age brackets greater than 12 months in the 1992 survey, but not in the study in 2005.Although obesity prevalence was higher in the second survey, it was not more predominant in any of the specific age brackets in either of the surveys.
Table 2 shows the distribution of socioeconomic, demographic, and health characteristics in the samples and the respective prevalence rates and measures of association with stunting and obesity.From 1992 to 2005 there was a significant reduction (p ≤ 0.05) in the proportion of: residents in the rural area, families without access to running water, households with more than four residents, mothers with more than two children, children with low birth weight (< 2,500g), and mothers without access to prenatal care.This same period also witnessed an increase in the proportion of adolescent mothers.
The following variables showed statistically significant associations with increased obesity: living in the urban area, mother not having more than two children, and birth weight ≥ 4,000g.Meanwhile, higher prevalence rates of stunting were associated with living in the rural area, households without running water, mothers with more than two children, birth weight less than 2,500g, mothers without access to prenatal care, and maternal age greater than 40 years.
Table 3 shows the time trends in prevalence of stunting and obesity.Prevalence of stunting showed a 49.3% decrease, while obesity increased by 38.8%.However, as shown in Table 2, for both situations, adjustment of the prevalence ratios for the predictors associated with these outcomes, respectively (p ≤ 0.05), reduced the magnitude of associations and the confidence intervals included 1.0, showing that the observed changes in prevalence of stunting and obesity in the second survey were heavily influenced by alterations in the behavior of these predictors.
Changes in prevalence rates of the dependent variables during the period were reflected in the z-score values for the anthropometric indices used to comprise the respective indicators: the mean z-score for the weight-for-height index increased from 0.45 ± 1.05 in 1992 to 0.63 ± 1.12 in 2005 (p < 0.001), while the mean z-score for the height-for-age index changed from -0.92 ± 1.56 to -0.49 ± 1.29 during the same period (p < 0.0001).

Discussion
In Latin America, especially in Brazil, there has been a rapid demographic, epidemiological, and nutritional transition.As a consequence, the population has experienced a sharp change in the distribution of individual nutritional status, characterized by a decrease in the frequency of undernutrition and an increase in obesity 13,14 .This observation is corroborated by the current study's findings, whereby from 1992 and 2005 the prevalence of stunting in children less than five years of age decreased by 49.3%, while the prevalence of obesity increased by 38.8%.
Notwithstanding the socioeconomic and structural differences within and between Brazil's major geographic regions 15 , this process has occurred nationwide, and the results of various population-based surveys conducted on comparable dates corroborate the current study's findings 16,17,18,19,20 .
In Pernambuco State, from 1991 16 to 2006 17 , there was a 61% reduction in the prevalence of stunting, from 24.6% to 9.6%.Meanwhile, in the second survey obesity prevalence reached 8.1% 18 , close to the level observed for stunting, thus displaying opposite trends in undernutrition and obesity, characteristic of the third stage of the nutritional transition, with the appearance of excess weight on a population scale 21 .
In a municipality in South Brazil (Pelotas, Rio Grande do Sul State), Barros et al. 19 compared the trends in stunting and obesity in one-year-old children, based on three birth cohorts launched in 1982, 1993, and 2004.They observed a 27.7% decline in the prevalence of stunting (from 8.3% to 6%).Meanwhile, the number of obese children increased by 26.2% (from 6.5% to 8.2%).
Two nationwide studies were conducted on dates close to the surveys in Alagoas, namely the National Demographic and Health Surveys (PNDS) of 1996 and 2006 20 .During this period undernutrition decreased from 13.4% to 6.7%.However, contrary to the other studies cited above, the prevalence of obesity remained stable between the two surveys (7.2% and 7.3%, respectively).
Concerning the upward trend in prevalence of overweight as reported here, Uauy & Kain 22 call attention to the fact that programs for the prevention of undernutrition require periodic evaluation, including a clear definition of their beneficiaries due to the possible effects on obesity, to the extent that such programs can lead to a positive energy balance in individuals not subject to hunger and undernutrition.Thus, given the current epidemiological reality, characterized by the coexistence of undernutrition and obesity, it is no longer possible to establish generalized programs for the prevention and control of undernutrition.
The nutritional transition is attributed mainly to socioeconomic and demographic changes and those related to public services infrastructure 21,23 .Important changes that have occurred in Alagoas can explain the findings presented here.According to Carvalho 24 , Alagoas presented the second lowest human development index of all the Brazilian States.Nevertheless, during the 1990s it was also the State that achieved the greatest advances in education (31%) and health (16%), with a positive variation on the order of 20% in comparison to other States.Improved management of investments in the social area, especially in primary education and public health programs, were the main reasons for this growth.
The current study showed favorable trends in a series of variables associated with children's nutritional status, which contributed to the decrease in the prevalence of undernutrition, but which favored an increase in obesity prevalence.Monteiro et al. 23 analyzed the causes of the decline in stunting in Brazilian children from 1996 to 2007 and identified the following variables: increase in maternal schooling, growth in families' purchasing power, expansion of healthcare, and improved sanitation.Menezes et al. 16

analyzed
Cad. Saúde Pública, Rio de Janeiro, 29(4):793-800, abr, 2013 the time trend in stunting in children under five in the State of Pernambuco based on surveys conducted in 1991, 1997, and 2006 and found a 64.6% reduction from 1991 to 2006 (from 24.6% to 8.7%).All the determinants analyzed improved during the period analyzed; those that remained after adjustment were socioeconomic variables (per capita family income, maternal schooling, number of household members, and access to consumer goods), besides birth weight and maternal height.
Although the State of Alagoas still presents the country's most precarious social indicators, the investments made in public services infrastructure, especially in health and education, in addition to the large amount of Federal funds invested in the State through conditional income transfer programs, appear to at least partially explain the striking reduction in the prevalence of undernutrition in preschool-age children.Meanwhile, the prevalence of overweight has increased gradually, so that the two conditions now appear with virtually identical magnitude, thus emphasizing not only the importance of maintaining actions to control undernutrition, but also the need to expand measures to prevent overweight, since both represent health problems with serious repercussions in the short and long term.This is a key issue for public policymakers, indicating that access to food should be accompanied by an educational process aimed at adequately meeting nutritional needs, in addition to health promotion.
In conclusion, a nutritional transition was observed during the period studied here, consistent with the transition in Brazil as a whole, characterized by a sharp reduction in the prevalence of undernutrition and an increase in obesity.

Desnutrición Proteico-Calórica; Encuestas Nutricionales; Preescolar
Contributors H. S. Ferreira and M. L. Assunção wrote the initial version.H. S. Ferreira, B. L. Horta, and J. A. Cesar elaborated the analytical plan.All the authors participated in writing and approving the final version of the article.

Table 1
Prevalence (%) of nutritional disorders defi ned by anthropometric extremes in children less than fi ve years of age in Alagoas State, Brazil, by age bracket, according to two surveys (1992 and 2005).

Table 2
Prevalence, prevalence ratios, and 95% confi dence intervals (95%CI) for obesity and stunting, according to categories of socioeconomic, demographic, and health variables in children under fi ve in Alagoas State, Brazil.

Table 3
Time trends in stunting and obesity in children under fi ve years of age in two surveys in Alagoas State, Brazil.(1992-2005).