Determinants of fruit and vegetable intake in adolescents using quantile regression

RESULTADOS: Dos adolescentes entrevistados, 6,4% consumiram a recomendação mínima de 400 g/dia de FLV e 22% não consumiram nenhum tipo de FLV. Nos modelos de regressão quantílica, ajustados pelo consumo energético, faixa etária e sexo, a renda domiciliar per capita e a escolaridade do chefe de família associaram-se positivamente ao consumo de FLV, enquanto o hábito de fumar associou-se negativamente. Renda associou-se signifi cativamente aos menores percentis de ingestão (p20 ao p55); tabagismo aos percentis intermediários (p45 ao p75) e escolaridade do chefe de família aos percentis fi nais de consumo de FLV (p70 ao p95).


INTRODUCTION
Fruits and vegetables are widely known because of their relevant role in human health, a and their intake seems to have a protective effect on the risk of obesity, 2 diabetes mellitus type 2, 2 cardiovascular diseases 12 and certain types of cancer. 28,29Estimates of the World Health Organization (WHO) points to approximately 2.7 million deaths worldwide in 2000 that could have been prevented with adequate fruit and vegetable intake, which would reduce global burden of disease by 1.8%. 18 a non-communicable chronic disease prevention strategy, the WHO and the Brazilian Ministry of Health recommend a minimum intake of 400 g per day of fruits and vegetables (equivalent to fi ve portions). 29Ministério da Saúde.Secretaria de Atenção à Saúde.Coordenação Geral da Política de Alimentação e Nutrição.Guia alimentar da população brasileira: promovendo a alimentação sudável.Brasília, DF; 2005.(Série A. Normas e Manuais Técnicos).[Cited 2011 Jan 25].Available from: http://dtr2001.saude.gov.br/editora/produtos/livros/pdf/05_1109_M.pdf The 2002-2003 Pesquisa de Orçamentos Familiares (Household Budget Survey) recommends that the caloric contribution of fruits and vegetables in diet be 2.3%. 7In an analysis of food availability trend in Brazilian households, Levy-Costa et al (2005)  17 observed that this contribution has remained low in the last 30 years.
In Brazil, estimates of fruit and vegetable intake among adolescents are scarce and, in their majority, originate from studies conducted with non-representative samples.In addition, the food frequency questionnaire, the method of investigation generally used, does not provide accurate intake measurements. 5,21e present study aimed to analyze fruit and vegetable intake among adolescents and to identify associated factors.

METHODS
A population-based cross-sectional study was conducted in 2003, resulting from the multicenter research project entitled Inquérito de Saúde de São Paulo (ISA-Capital -City of São Paulo Health Survey).The ISA-Capital was developed to collect data on the health and diet of adolescents, adults and elderly individuals of the city of São Paulo, Souhteastern Brazil.The present study used data on 812 adolescents aged between 12 and 19 years.Sample was stratifi ed into clusters, in two stages: census tract and household.More details on the sampling process can be found in Castro et al. 6 Qualifi ed interviewers conducted a household survey with a structured questionnaire to investigate demographic, anthropometric, socioeconomic, family and lifestyle characteristics.
Food intake was measured with the 24-hour dietary recall, a method applied according to the multiplepass method, structured in fi ve steps: 1) quick list, where participants list all the foods and beverages consumed uninterruptedly; 2) forgotten list, where participants are asked about commonly forgotten foods consumed, such as candies, coffees and sodas; 3) time and location of food and beverage intake; 4) detailing cycle, where the way of preparation and amounts consumed are described; and 5) fi nal review, which verifi es whether a certain food consumed during the day was not previously recorded. 20This method helps individuals to recall foods and beverages consumed on the day before the interview and to record them in a detailed way, thus reducing dietary measurement errors. 24Adolescents were instructed to record the foods consumed according to household measures and to mention the way of preparation, seasonings and commercial brands.
Information about food intake was reviewed to identify possible errors of completion and to convert household measures of foods and beverages into grams and milliliters, respectively. 23The Nutrition Data System for Research (NDS) software was used (2007 version, NCC, University of Minnesota, Minneapolis, USA) to obtain nutritional values.
Daily intake (in grams) of fruits and vegetables was calculated to create the fruit and vegetable variable.This variable was comprised of all types of fruits and vegetables, whether they were consumed raw or cooked, peeled or unpeeled, including those used as ingredients in preparations.Researchers chose to quantify foods of the fruit and vegetable group included in preparations such as soups, side dishes and mixed dishes, aiming to measure all and every type of fruit and vegetable consumed, thus avoiding an underestimated intake of this food group.The preparations recorded in the 24-hour recall were quantifi ed, typed and separated into their respective ingredients.Foods from the group of roots and tubers, such as potato, cassava and yam, were excluded because they do not belong to the fruit and vegetable group, according to the Guia Alimentar da População Brasileira (Brazilian Population Food Guide).a Fruit and vegetable intake was considered adequate when equal to or higher than 400 g/day, according to the Ministry of Health recommendations.a Classifi cation of the nutritional status was based on the cut-off points of Body Mass Index (BMI) according to sex and age proposed by the WHO. 22Anthropometric measurements of self-reported weight and height were used to calculate BMI.Overweight or obese adolescents were grouped and classifi ed as weight excess.
Adolescents who reported alcohol consumption, regardless of the type of beverage, frequency and quantity, were categorized as alcohol drinkers.Current or previous smoking habit was considered to defi ne adolescents as smokers.
Per capita household income was estimated by adding the monthly incomes of all individuals of a household, divided by the total number of residents.A total of two categories of per capita household income were defi ned (up to one minimum wage; more than one minimum wage), according to the value of a minimum wage at the time of this study (R$ 240.00 or US$ 75.00).
Head of household's level of education was categorized according to the number of years of school completed (up to eight years; nine years or more).
Fruit and vegetable intake was stratifi ed by sex and described as percentiles.Quantile regression models were constructed to identify the association between fruit and vegetable intake and the variables analyzed.These models enable the effect of independent variables on different points of the distribution of the dependent variable to be observed and they are also appropriate when this variable is asymmetric and heteroscedastic, compared to the remaining variables in the model. 4uit and vegetable intake (grams/day) was the dependent variable.Dichotomous variables (head of household's level of education, per capita household income, smoking habit, drinking and age group) and continuous variables (energy (kcal/day) and BMI-for-age Z-score) were tested as independent variables.
Modeling was performed using the stepwise-forward procedure and the associated variables whose 95% confi dence interval did not include zero remained in the model.Regression coeffi cients were simultaneously estimated at every fi ve percentiles in the interval between the 20 th and 95 th percentiles of the same model to analyze the effect of independent variables throughout the entire distribution.This interval characterizes the distribution points in which fruit and vegetable intake is different from zero.Standard error for each of the estimated coeffi cients was calculated from the 400 replications, using the Bootstrap replication technique.Associations observed in a certain percentile may not be signifi cant in another, because regression coeffi cients were estimated in several points of distribution.The selection of models that achieved the best adjustments was based on the following two criteria: 1) association of explanatory variables in the highest number of percentiles; and 2) lowest confi dence intervals of the coeffi cients of the main associated variables.

RESULTS
The prevalence of overweight was 19%; smoking habit, 10%; and alcohol drinking, 39%.Among approximately 57% of adolescents, heads of households had eight years of education and 52% belonged to families with incomes equal to or lower than R$ 240.00 (up to one minimum wage) (Table 1).About 20% of adolescents did not consume fruits and vegetables on the day reported and 6.5% had an adequate intake (≥ 400 g/day).Intake was also adequate between the 90 th and 95 th percentiles of distribution, in the majority of variables investigated in the analysis stratifi ed by sex (Table 1).The proportion of individuals  with an intake higher than this recommendation was lower than 10% in all strata analyzed, except for female smokers, whose 90 th percentile was 403 g.
The model that best met the criteria had "per capita household income", "head of household's level of education" and "smoking habit" as explanatory variables, adjusted for caloric intake, age group and sex.Figures 1 and 2 show the changes in regression coeffi cients throughout the distribution, with regards to head of household's level of education and income.These coeffi cients show the difference of fruit and vegetable intake in grams among the categories of independent variables analyzed.In these graphs, the gray area represents the 95% confi dence interval of coeffi cients estimated in each percentile.Income was positively associated with fruit and vegetable intake in the 20 th to 55 th percentiles.
Coeffi cients remained almost constant between these percentiles, indicating an intake of approximately 20 g less in the group with the lowest income.Head of household's higher level of education had a signifi cant infl uence after the 70 th percentile and this effect varied with the increase in percentiles.The difference in fruit and vegetable intake between the categories with the highest and lowest level of education was 63 g in the 70 th percentile and 77 g in the 90 th percentile, regardless of the remaining variables.
After the 45 th percentile, the effect of smoking habit on fruit and vegetable intake was signifi cant and it gradually increased until the 75 th percentile (Figure 3).The group of smokers showed an intake of 20 g and 65 g less than that of the group of non-smokers in the 45 th and 75 th percentiles, respectively.

DISCUSSION
In the present study, 6.5% of adolescents met the recommended fruit and vegetable intake (≥ 400 g/ day).Income, head of household's level of education and smoking habit had an effect on the intake of this food group.
Although there are differences in dietary assessment methods, the results of the present study corroborate those of Brazilian studies on adolescents.Crosssectional studies using food frequency questionnaires found low intake of fruits and vegetables in Brazilian adolescents. 1,25In a study conducted by the Instituto Nacional do Câncer (INCA), approximately half (45%) of adolescents aged between 15 and 24 years, living in the city of São Paulo, SP, consumed fruits or vegetables less than fi ve times per week.b Among 8 th grade students of municipal public schools of the city of Rio de Janeiro, Southeastern, raw and cooked vegetables were consumed at least fi ve times per week by 20% and 16.5% of the adolescents studied, respectively. 5cent North American studies have reported a low fruit and vegetable intake among adolescents. 11,15,16owever, differences in the categorization of food groups, with the inclusion of potatoes and other tubers in the fruit and vegetable group, restricted the comparison of results.Data from the 2003-2004 National Health and Nutrition Examination Survey (NHANES III), for example, showed that potatoes represented approximately 41% of the total intake of vegetables. 15antile regression models, used to investigate the association between fruit and vegetable intake and explanatory variables, are extensively applied in econometrics, although being rarely used in health sciences, especially in nutritional epidemiology.The study conducted by Variyam et al 27 on factors associated with nutrient intake using such models is one of the exceptions.In the present study, the choice for this particular model is justifi ed by the asymmetry to the right of the curve of fruit and vegetable intake and to the heteroscedasticity of the dependent variable, in relation to the explanatory variables.In these cases, models based on the method of minimum squares create residues that can have a non-uniform variance, resulting in inferences originated from the biased coeffi cients.When there is no homoscedasticity, certain percentiles may be more infl uenced by explanatory variables than others.Finally, quantile regression models are additive, i.e. regression coeffi cients are interpreted as the change in a certain percentile of the dependent variable, caused by the change of an independent variable unit.
Multiple analyses show differences in the infl uence of income and head of household's level of education on fruit and vegetable intake among adolescents, similarly to what was observed in studies that used linear regression models. 26Income was signifi cantly associated in the lower intake percentiles (p20 to p55), whereas level of education was only associated in the highest percentiles of distribution (p70 to p95).

Percentiles
Income can be a determinant among adolescents who consume low amounts of fruits and vegetables (75 g or less, i.e. lower than one daily portion).Among those with a higher intake (between 140 g and 460 g), head of household's level of education became a determinant.
The role of variables in fruit and vegetable intake has been investigated in different studies. 7,26Income infl uences access to and availability of fi nancial resources to purchase more expensive foods, such as fruits and vegetables. 25In the present study, where 52% of adolescents had a per capita household income equal to or lower than R$ 240.00 (US$ 75.00), low intake of fruits and vegetables can be attributed to the lower purchasing power of families, which would restrict the amount of fruits and vegetables available in the household.
Level of education, in its turn, would infl uence the foods selected, because it provides an individual with the ability to assimilate messages from nutritional education programs and understand the importance of eating as a way to promote health. 26In the present study, the higher level of education of a head of household could infl uence adolescents to consume more fruits and vegetables, as a result of this head of household's choice for healthier foods, once they are main responsible for family decisions.
Smoking habit has an effect on fruit and vegetable intake between the 45 th and 75 th percentiles of intake, regardless of the infl uence of income and head of household's level of education.These fi ndings suggest that the effects of smoking habit on fruit and vegetable intake occur on both low (50 g) and moderate levels of intake (165 g).In a population study conducted in Canada, Birkett 3 observed that smokers consumed a significantly lower amount of fruits and natural fruit juices than non-smokers.In addition, those who consumed a higher number of cigarettes per day were four to six times more likely to eat less fruits than what is recommended.Although less expressive, similar results of fruit and vegetable intake were found.Smokers, however, have increased requirements of antioxidants present in the fruit and vegetable group, because they have lower plasmatic levels of these substances, caused by the excessive amount of free radicals found in cigarette smoke. 14uit and vegetable intake for each individual was obtained from one single 24-hour recall.This procedure fails to capture daily variations, thus increasing distribution variance. 8An intake lower than the median can be underestimated, whereas that which is higher can be overestimated.Considering the fact that the recommendation of 400g/day of fruit and vegetable intake was met in the higher percentiles of distribution, it is possible that the proportion of adolescents with an intake lower than this value is underestimated.Another effect of the increase in variance is the attenuation of measures of association between intake and explanatory variables.This implies a possible underestimation of the coefficients found.It is assumed that the associations observed would continue to exist if more than one 24-hour recall was used one per individual, although with coeffi cients of higher magnitude.Despite the fact that only one dietary measurement reduces the accuracy of percentiles of intake and attenuates regression coeffi cients, its use does not invalidate the conclusions of this study.
Fruit and vegetable intake is a socially desirable behavior, which can result in over reporting of these foods. 13,19However, this type of bias does not seem to be a relevant source of error in the sample studied, once a low percentage of adolescents met the recommendations (6.5%) and a higher percentage did not report fruit and vegetable intake (20%).
Another possible limitation refers to the self-reported weight and height, used to calculate the BMI.;10 Fruit and vegetable intake among adolescents of the city of São Paulo is signifi cantly lower than the current recommendations and it is infl uenced by income, head of household's level of education and smoking habit.Intersectoral actions, aimed to encourage fruit and vegetable intake and the adoption of healthy lifestyles, can improve this situation.The implementation of public policies to reduce the prices of fruits and vegetables, the development of mechanisms that promote an increase in household income and the adoption of multiple nutritional education actions in schools are measures that help to increase fruit and vegetable intake among Brazilian adolescents.
Stata statistical software (Statistics/Data Analysis, version 10.1, Texas, USA) was used in the analysis.The present study was approved by the Research Ethics Committee of the Faculdade de Saúde Pública da Universidade de São Paulo School of Public Health (Process 381, on 2/13/2001).

Figure 1 .
Figure 1.Quantile regression coeffi cients by per capita household income and fruits and vegetables intake in adolescents.Municipality of São Paulo, Southeastern Brazil, 2003.Notes: Coeffi cients adjusted for smoking habit, head of household's level of education, caloric intake and BMI-per-age Z-score.The gray area corresponds to the 95% confi dence interval.Per capita household income: dichotomous variable, reference category: "more than one minimum wage".

Figure 2 .
Figure 2. Quantile regression coeffi cients by head of household's level of education and fruits and vegetables intake in adolescents.Minicipality of São Paulo, Southeastern Brazil, 2003.Notes: Coeffi cients adjusted for smoking habit, caloric intake and BMI-per-age Z-score.The gray area corresponds to the 95% confi dence interval.Head of household's level of education: dichotomous variable, reference category was the group with the highest level of education.

Figure 3 .
Figure 3. Quantile regression coeffi cients by smoking habit and fruits and vegetables intake in adolescents.Municipality of São Paulo, Southeastern Brazil, 2003.Notes: Coeffi cients adjusted for head of household's level of education, caloric intake and BMI-per-age Z-score.The gray area corresponds to the 95% confi dence interval.Smoking habit: dichotomous variable, reference category was the group of non-smokers.

Table 1 .
Percentages of fruit and vegetable intake (in grams) per adolescents, according to sociodemographic characteristics, lifestyle and nutritional status, stratifi ed by sex.Municipality of São Paulo, Southeastern Brazil, 2003.