Food consumption pattern and obesity in preschool children in Feira de Santana , Bahia , Brazil

Objective To evaluate the association between food consumption patterns and obesity in preschool children in Feira de Santana, Bahia, Brazil. Methods Cross-sectional, population-based nested within a live-birth cohort study of 813 children, which has started in 2004 in Feira de Santana, Bahia. The anthropometric status among children under four years of age was evaluated based on their body mass index; obesity/severe obesity was defined as a Z-score >+2. The Food Frequency Questionnaire was used to identify dietary patterns using principal components analysis. The association between obesity and food consumption patterns was assessed using Pearson’s Chi-squared test and logistic regression, adopting p<0.05 to denote statistical significance. Results Obesity was detected in 12.7% of the children investigated. Four food consumption patterns were identified: dietary pattern 1 (milk and other dairy products, vegetables and tubers, cereals, legumes, fruits, and fish); dietary


I N T R O D U C T I O N
Obesity is considered a global epidemic. The risk of an obese child becoming an obese adult increases from 25% when this condition occurs during the first six years of life to 75% when it occurs during adolescence [1]. Genetic and environmental determinants are involved in obesity genesis. Eating habits is a major environmental factor contributing to obsesity [2,3].
Nutritional status assessments of individuals and populations often include anthropometric measurements (direct method), food consumption questionnaires (indirect method), and demographic, socioeconomic, and cultural factors [4]. However, recently, frequency of food consumption has been used to identify dietary patterns using Principal Component Analysis (PCA) as an alternative to studies that include the isolated assessment of food/nutrient consumption [5,6].
According to the World Health Organization (WHO), the identification of dietary patterns of a given population, i.e., the variety of foods consumed by this population [7], is considered a more reliable method than nutrient consumption [8] since it represents the profile of food and nutrient intake, based on their usual intake [9].
The determination of the anthropometric profile and the food consumption pattern of a population is an important public health strategy that can contribute to therapeutic and educational actions, such as changing eating habits and increasing physical activities. The aim is to reduce the risk of morbidity due to nutritionrelated diseases. This study aimed to evaluate the association between food consumption patterns and anthropometric status of preschool children in Feira de Santana (BA), Brazil.

M E T H O D S
This is a cross-sectional, populationbased study nested within a live-birth cohort study carried out between April 2004 and March 2005 in the city of Feira de Santana (BA), whose objective was to evaluate the effects of breastfeeding, complementary feeding, and weaning on eating habits, growth, and health of children under six years of age.
The cohort consisted of a representative sample of live births (Feira de Santana residents) that were being treated in local hospitals. Live births were added to the cohort sample over the period of 12 months; two new hospitals were included every two months using the lottery method, except for two maternity wards that were included because they provided care to a greater number of women. In the present study, a total of 813 children younger than 4 years of age, who were followed-up in the cohort study were evaluated. More details on the cohort can be obtained in other publications [10,11].
Data were collected during home visits using direct interviews conducted by health professionals working in pairs and properly trained university students. The interview questions were concerned with maternal and child characteristics, and they were clear and objective and offered multiple-choice answers.
Food consumption was assessed using a Food Frequency Questionnaire (FFQ) concerning food consumption during the child's fourth year of life. It was administered to the children's mothers or guardians. The FFQ was composed of a list of 133 foods and options for reporting the frequency of intake including the following frequency categories: never/rarely; 1-3 times per month; 1 time per week; 2-4 times per week; 1 time per day; and 2 or more times per day. Food frequency data were transformed into daily servings in order to use only one unit of time. Thus, average intake frequency was calculated for each food by dividing the weekly consumption by seven or monthly consumption by thirty.
Children's body weight and height measurements were obtained using a digital scale with a maximum capacity of 150kg and an accuracy of 100g and a portable, folding stadiometer capable of measurements up to 2.16 meters and capable accuracy of 0.1cm. All measurements were made in triplicate, and the reference value used was the arithmetic mean of the triplicate results.
Obesity was classified based on the anthropometric status evaluated by the direct method, that is, by calculating the BMI-for-age, according to WHO [12] reference curves. The Anthro [13] software was used, adopting the Z-score cutoff point >+2, which includes obesity and severe obesity.
The children characteristics evaluated were: gender, birth weight (<2.500g, ≥2.500g), interruption of exclusive breastfeeding at 4 months, weaning at 2 years of age, bottle feeding, sleep feeding, and consumption of foods in liquid form. The maternal variables were: gestational age (preterm, full term), maternal age at childbirth (<20 years, ≥20 years), maternal skin color (white, black/brown), maternal level of education (incomplete elementary and middle school, high school/higher education), work outside the home, income (≤1 minimum wage, over the minimum wage), and maternal overweight/obesity. The foods included in the FFQ were grouped using factor analysis. Thus, Food frequency data were transformed into a daily servings of each food and grouped into 17 food groups, based on nutritional characteristics and correlations obtained in the preliminary factor analysis. The following food groups were formed: group 1: milk and other dairy products (whole milk, skimmed milk, fermented milk, natural yogurt, petit suisse cheese, queijo prato (a Brazilian soft yellow cheese), mozzarella, queijo minas (a type of cheese produced in the Brazilian state of Minas Gerais), requeijão (a loose, ricotta-like cheese), porridge, smoothies (milk + fruit + chocolate milk powder mix; milk + chocolate milk powder mix; and milk + fresh fruits); group 2: ketchup/mayonnaise (mayonnaise, potato salad with mayonnaise, ketchup); group 3: encased meats (sausage and frankfurter sausage); group 4: fast foods (French fries, acarajé (a dish made from peeled beans formed into a ball and then deep-fried in palm oil), cheeseburger, sandwiches, coxinha de galinha (popular Brazilian food made with shredded chicken meat, covered in dough, battered and fried), cheese bread (traditional baked cheese roll), sfiha (meat pie [open-face or not], pizza, hot dogs); group 5: red meats -cooked meat, liver, fried steak, cooked lamb, fried lamb chops, pork, maniçoba (typical Brazilian dish made with leaves of the cassava plant and salted pork, dried meat, and smoked ingredients, such as bacon and sausage), sarapatel (a Portuguese dish made with offal and meats such as pork, lamb, or even beef), mocofato (traditional rich Brazilian dish made with beef [cow's feet], sausages, and bacon), rabada (a typical dish made with oxtail), buchada (Brazilian dish made with goat offal an bacon), and feijoada (a stew of beans with pork meat); group 6: vegetables and tubers -mashed potatoes, sweet potatoes, cooked yams, yams, lettuce, Chinese cabbage, pumpkin, watercress, cauliflower, beetroot, carrot, spinach, cucumber, tomato, vegetable soup, unstuffed or stuffed beiju (a typical dish made with tapioca), and flour; group 7: cereals -rice, noodles, instant noodles, pasta and pastries, cookies and crackers, French bread, breakfast cereal, couscous, corn, plain cake, popcorn, canjica (Southern Brazil) and/ or mungunzá (Northern Brazil) -sweet dishes or porridge made with white de-germed whole maize kernels (canjica), cooked with milk, sugar and cinnamon until tender; group 8: legumes (peanuts, beans, green beans, soybeans, bean soup), abará (typical dish of Bahia state based on a paste made from mashed black-eyed peas wrapped into pancakes inside banana leaves and steamed); group 9: fruits -pineapple, avocado, banana prata (burro or chunky banana), banana da terra (plaintain or cooking banana), guava, orange, apple, papaya, melon, mango, strawberry, sugar apple, grape, sugarsweetened fruit juice; Group 10: eggs (fried egg and quail egg); group 11: chicken (cooked chicken); group 12: fish (fried fish, shellfish, crab, and other foods used to prepare fish, especially coconut milk); group 13: packaged snacks (potato chips); group 14: soft drinks/ artificial fruit juices (traditional soft drink, diet soft drink, artificial juices, artificial beverages); group 15: sweets (chocolate, ice cream, freeze pop, chocolate milk powder mix, candies, fruit preserves, desserts, jello, pudding, sugar); group 16: oils and fats (margarine, butter, olive oil, palm oil, vegetable oil); group 17: coffee/tea. Some foods were excluded from the analysis because their reported consumption among the participants was less than 5% (non-fat yogurt, pea soup, unsweetened juice, sweetener, natural or veggie sandwich, alcoholic beverages).
The daily servings from each food were grouped into a single value for each child by the sum of the servings of the foods from each food group. Therefore, it was possible to obtain a continuous variable that was standardized, according to Z-score of the standard normal curve. It is the input variable in the principal component analysis used to identify the dietary patterns.
The dietary patterns were identified by exploratory factor analysis using PCA. Principal Component Analysis is a multivariate statistical analysis, which allows the combination of variables based on the degree (strength) of the relationship between them. Thus, the variables grouped in each factor are strongly correlated with each other [14,15].
Initially, the suitability of the sample for the application of for PCA was verified using the Kaiser-Meyer-Olkin test and the Bartlett's sphericity. Orthogonal rotation (Varimax) was used to determine the dietary patterns; this method provides nearly uncorrelated, distinct factors, improving the interpretability of the factor loadings.
The multivariate statistical approach of the exploratory factor analysis allowed the grouping of the food items in the FFQ according to the degree (strength) of correlation between them. The Kaiser's eigenvalue-greater-thanone rule and the Cattel's scree test were used to determine the number of factors to retain in the factor analysis. The foods or food groups that contributed to the characterization of each dietary pattern had factor loading ≥0.30, considering the highest saturation factor, significance level of 0.05, and power of 80%, as recommended by Hair Jr. et al. [15]. The dietary patterns obtained based on the PCA were categorized into tertiles of the individual consumption scores of these dietary patterns and were denominated as follows: 1st tertile (low adhesion to the dietary patterns), 2nd tertile (moderate adhesion to the dietary patterns), and 3rd tertile (high adhesion to the dietary patterns).
To evaluate the association between obesity and dietary pattern, four logistic regression models were constructed because each dietary patterns identified was considered a main independent variable.
Statistical analysis was carried out using the Software Statistical Package for Social Science (SPSS, Inc., Chicago, Illinois, United States) 9.0 version [16] for Windows and validated using the EpiData software (Atlanta, Georgia, United States) program. The R programming language [17] was also used in the data analysis. Initially, the frequencies of the child and maternal characteristics and the frequency of children's anthropometric status were calculated. In the bivariate analysis, the prevalence, Prevalence Ratio (PR), and their respective confidence intervals were calculated to determine the factors associated with obesity, considering 95% confidence interval as the threshold for statistical significance, according to the Chi-squared test.
In the bivariate analyses, the value of p≤0.25 was used for the inclusion of variables in the logistic regression model. To select the predictor variables that should remain in the model, variables with a p-≤0.20, obtained using the likelihood ratio test, were considered as candidate variables. A backward selection process and p-value ≤0.05 (statistically significant associations) criterion were used to select the variables in the final model.  [18]. The data of this research were used with the research coordinator's authorization. Olkin value of 0.807 indicated the applicability of the factor analysis; the Bartlett's test of sphericity was significant (p-value ≤0.001), indicating correlation between the variables.

R E S U L T S
The dietary pattern 1 was characterized by a predominance of milk and other dairy products, vegetables and tubers, cereals, legumes, fruits, and fish. In dietary pattern 2, there was a predominance of deep-fried or baked snacks, soft drinks/artificial fruit juices, sweets, oils and fats, and coffee/tea. Dietary pattern 3 was characterized by the predominance of encased meats, fast foods, ketchup/mayonnaise, and eggs. In dietary pattern 4, there was a predominance of chicken and red meats ( Table 2).
Obesity was not associated with any of the tertiles of dietary pattern (four) adherence in the crude (unadjusted) analysis (Table 4).
In the multivariate analysis, there was a statistically significant association between obesity and high adherence to dietary pattern 3 (OR=1.92) after adjustment for potential confounders (Table 5).

D I S C U S S I O N
In the present study, there was a higher prevalence of obesity (12.7%) and lower prevalence of thinness (0.5%) among children younger than four years old. The increasing prevalence of obesity and decreasing prevalence of malnutrition have been reported by other researchers in recent decades in Brazil [19], a fact that characterizes nutrition transition. Changes in eating habits with increased consumption of high-calorie foods may help explain this phenomenon.
The dietary patterns identified in the present study are similar to those reported by D'Innocenzo et al. [5] in the city of Salvador (BA).
The dietary patterns 2 and 3 identified in the present study show a global trend that has been observed over the last three decades, which reflects contemporary living habits, including preference for industrialized foods and increased caloric intake [20,21]. These two patterns were characterized by foods high in fat and sugars. Nobre et al. [9] reported similar findings in a study involving preschoolers in the city of Diamantina (MG). These authors identified dietary patterns denominated "snacks" and "unhealthy". Similarly, D'Innocenzo et al. [6] found a dietary pattern characterized by fried foods, packaged snacks, soft drinks/artificial fruit juices, and another one characterized by encased meats, eggs, and red meats.
In the present study, there was a positive association between obesity and high adherence to dietary pattern 3 (encased meats, fast-food. ketchup/mayonnaise and eggs). Increased energy intake and decreased fiber intake are major factors contributing to obesity [22]. These dietary habits can be modified through intervention measures on healthy eating awareness [22,23].
However, the present study showed some limitations such as the method used to obtain and analyze food consumption. Although the FFQ is the commonly used method to estimate individuals' usual frequency of food consumption, it has some limitations, such as it depends mostly on the memory of the subject being interviewed, presence of interview-related difficulties, and the fact it does not allow precise estimation of portion size of foods consumed. Moreover, the factor analysis carried out to identify the dietary patterns involved making arbitrary decisions regarding the number of factors that were not retained, the choice of orthogonal rotation, the determination of factor loading significance, and the identification and interpretation of the dietary patterns.
On the other hand, it is worth emphasizing the value and importance of the present study, which is a cross-sectional, population-based study nested within a cohort study that was carried out using factor analysis to identify patterns of food consumption, a more reliable method than the consumption of nutrients, in order to evaluate the eating habits of a specific group.
The internal consistency of the identified dietary patterns reinforce the pre-established scientific knowledge that the consumption of processed and high-calorie foods, such as encased meats, fast foods, ketchup, and eggs is a predictor of obesity in childhood [24,25].
Aiming to encourage the adoption of healthy eating habits among people, at the individual and population levels, and to prevent childhood overweight and obesity, public policies and food education programs are necessary [12], especially in health units and public and private schools to provide educators, parents, and caregivers with information about the nutritional quality of foods and the increased risk of obesity due to the consumption of some industrialized and high-calorie foods [7].
It is also necessary to warn the population that the prevention of obesity in adulthood starts at as early as the day a child is born with breastfeeding and healthy eating habits in the first years of life so that they can make informed decisions due to globalization and changes in social structures that contribute the adoption of harmful eating habits with increased consumption of industrialized and energy-dense foods [5].