The prevalence of overweight among children in most industrialized countries is of concern(Reference Janssen, Katzmarzyk, Boyce, Vereecken, Roberts, Currie and Pickett1). In Canada, the proportion of overweight children 2–17 years of age grew from 15 % to 26 % between 1979 and 2004(Reference Shields2) and in a 2002 survey, as many as 52 % of First Nations children aged 2–11 years living on reserve developed excess body weight(3). Changes in food consumption patterns associated with increased average body weight are little understood(Reference Shields2, Reference Taylor, Evers and McKenna4–Reference Newby7).
A study of the relationship between snacking and childhood obesity seemed worth undertaking since nearly all children in North America eat snacks(Reference Nicklas, Baranowski, Cullen and Berenson5, Reference Jahns, Siega-Riz and Popkin8, Reference Adair and Popkin9). Studies of the relationship between snack food intake and overweight in children and adolescents produce contradictory results(Reference Francis, Lee and Birch10–Reference Stockman, Schenkel, Brown and Duncan17).
Since 1994, the Kahnawake Schools Diabetes Prevention Project (KSDPP) based on the promotion of healthy eating and physical activity has been in place in the schools and community of Kahnawake, a Kanien’kehaka (Mohawk) community of 7200 inhabitants near Montréal, Québec, Canada(Reference Macaulay, Paradis and Potvin18). The project also includes a research component along with the intervention. A recent analysis of 24 h recalls indicated that overweight in the community’s school-aged children was associated with overall increased consumption of French fries and potato chips (crisps)(Reference Receveur, Morou, Gray-Donald and Macaulay19). The purpose of the present study was to understand more specifically how the quality, quantity and frequency of snack food consumption differs in different BMI categories, to improve the approach of the ongoing community programme.
Methods
As part of the KSDPP research, a total of 449 school-aged children (grade 4–6) provided a 24 h recall and their height and weight in 1994, 1998 and 2002, in three independent cross-sectional surveys. The 24 h recalls were all collected in the month of October by trained dietitians using the same methods and were coded using the Canadian Nutrient Files (1997), a data set largely based on the US Department of Agriculture database, Handbook No. 8, with adjustments and additions made to reflect Canadian fortification and regulatory standards as well as Canadian-only foods (www.dietsoftware.com/Canada.html). This 1997 database (rather than the most recent 2007 version) was selected since it is close to the mid-point period of data collection (1994–2002) and may therefore better reflect the composition of foods actually available at those times. Snacks were defined by the interviewer based on time of day and whether the food was consumed between two consecutive meals.
Anthropometric measurements were obtained for children wearing light indoor clothing without shoes. Height was measured to the nearest 0·1 cm using a portable wooden measuring board (I.J. Shorr, Olney, MD, USA) and weight to the nearest 0·1 kg using the same numeric scales (model 770 Alpha; SECA Co., Columbia, MD, USA) which were periodically calibrated with standard weights. BMI (kg/m2) was categorized as between the 5th and the 85th percentile (normal weight, n 266), between the 85th and the 95th percentile (at risk of overweight, n 75) or above the 95th percentile (overweight, n 103) based on the National Center for Health Statistics (2000) sex- and age-specific reference population(20). Five children with BMI below the 5th percentile for their age were excluded, leaving a total of 444 children (151 in 1994, 144 in 1998 and 149 in 2002) for the analysis. The variables of age and gender were also considered in the study.
Food categories
In all, 778 snacks were consumed by 444 children between the ages of 8 and 13 years including 306 different foods in eighteen categories described in Table 1. The food items consumed as snacks were grouped into categories that correspond to those already defined in the literature(Reference Phillips, Bandini, Naumova, Cyr, Colclough, Dietz and Must14, Reference Forslund, Torgerson, Sjöström and Lindroos21–Reference Nicklas, Demory-Luce, Yang, Baranowski, Zakeri and Berenson23). Some categories had to be modified since the database was based on 24 h recalls and not on a list of predefined food items on an FFQ(Reference Phillips, Bandini, Naumova, Cyr, Colclough, Dietz and Must14). When possible, food items were grouped to obtain a minimum frequency of five per BMI category as validity criterion for χ 2 tests.
Table 1 Food group composition
*On eleven occasions, children drank non-sweetened beverages such as mineral water or diet sodas which we excluded from the analyses along with water.
Statistical analysis
ANOVA was used to describe the children’s behaviour at mealtime and snack time according to gender, age and year of study, and to compare the children’s BMI categories according to total energy intake and energy density of solid foods. Energy density of beverages was not calculated, since it is difficult to estimate how much water children drink (especially because of the availability of water fountains in schools). In the present study, milk was considered a solid food because of its nutritional value, because it is often consumed with cereal at breakfast and because it is an integral component of many recipes. When P values were significant between BMI groups, multiple comparison tests were conducted using the Tukey procedure to identify group differences and minimize the risk of type 1 errors associated with multiple comparisons. The χ 2 test was used to compare snacking frequency according to BMI category, gender, age group and year of study for the number of snacks per day (0 to 3 maximum) as well as the time of day (morning, afternoon and evening).
For each food category of snacks, frequency of consumption and average daily amounts consumed were compared by BMI category using the χ 2 test and ANOVA, respectively. When ANOVA results were significant, the Tukey procedure was used to identify differences of P < 0·05 between groups of children. Finally, food groups with frequency of consumption trends of P < 0·15 and reported in Table 4 were introduced into two full logistic regression models adjusted for gender, age group and year of study. One model compared normal-weight children with those at risk of overweight, and the other compared normal-weight children with overweight children. Statistical significance was set at P < 0·05 for the interpretation of results.
The SAS for Windows statistical software package version 8·2 (SAS Institute Inc., Cary, NC, USA) was used to analyse results.
Results
General snacking patterns
As previously described(Reference Receveur, Morou, Gray-Donald and Macaulay19), there was no significant difference in distribution of BMI categories by gender or year of data collection. Table 2 indicates results of energy intake and energy density of solid food for meals and snacks by gender, age group and year of study for all three BMI categories. When all children are considered without regard to BMI (n 444), results show that boys and girls had different behaviours with girls consuming less energy during meals and more at snack times. As they get older, children increased their energy intake at meals while reducing energy density from snacks, without changing the amount of total energy.
Table 2 Mean energy intake and energy density from meals and snacks by BMI category for gender, age group and year of study: school-aged children (grade 4–6) from a Kanien’kehaka (Mohawk) community, Canada
Normal weight, BMI between 5th and 85th percentile; at risk of overweight, BMI between 85th and 95th percentile; overweight, BMI > 95th percentile.
a,bMean values within a row with unlike superscript letters were significantly different (P < 0·05).
*P values <0·15 are shown.
†Significant difference between genders for energy intake from meals (P < 0·01).
‡Significant difference between genders for energy intake from snacks (P = 0·03).
§Significant difference between age groups for energy intake from meals (P = 0·05).
∥Difference between age groups for nutrient density of snacks (P = 0·11).
¶Difference for energy intake from meals between 1994 and 2002 (P = 0·08).
Overall, there was no significant difference in energy intake and energy density from meals by BMI group. On the other hand, energy intake from snacks tended to be higher for the children at risk of overweight, compared with the other two categories, in nearly all analyses (except for 2002 where there was no difference). A negative association between energy density of snacks and BMI could be observed in boys.
Table 3 describes behaviour associated with the number of snacks and snacking occasions by BMI category, gender, age group and year of study. Only forty children (9 %) out of 444 had not had a snack in the 24 h recall. Neither the number of snacks nor the number of meals per day (data not shown) was associated with BMI. Children at risk of overweight consumed evening snacks in greater numbers (P = 0·01). Girls consumed a greater number of snacks overall and also at different times of the day than boys, except in the afternoon. Age did not appear to influence the number and time of snacking occasions. Considering the number of snacks and snacking occasions by year of study, the results show that there might be a shift since 1994 towards consuming more morning snacks v. later in the day.
Table 3 Percentage consuming snacks by BMI category, gender, age group and survey year: school-aged children (grade 4–6) from a Kanien’kehaka (Mohawk) community, Canada (n 444)
Normal weight, BMI between 5th and 85th percentile; at risk of overweight, BMI between 85th and 95th percentile; overweight, BMI > 95th percentile.
*Rounded to whole percentage.
Snacking patterns according to food groups
Of the eighteen food groups listed in Table 1, there were no tendencies (P < 0·15) towards differences in intake between BMI categories save for a few exceptions reported in Tables 4 and 5. Food groups with a higher energy density were consumed significantly more frequently by children at risk of overweight (chips and fast food), as were larger average portions of cereal bars and breakfast cereals. Normal-weight children more frequently selected crackers, popcorn and pretzels, which have lower energy density than chips. Only the mean portion size of milk showed a positive association with BMI.
Table 4 Food groups consumed at snacking occasions showing differences in frequency by BMI categoryFootnote *: school-aged children (grade 4–6) from a Kanien’kehaka (Mohawk) community, Canada (n 444)
Normal weight, BMI between 5th and 85th percentile; at risk of overweight, BMI between 85th and 95th percentile; overweight, BMI > 95th percentile.
* Only food groups from Table 1 showing differences in frequency between BMI categories at P < 0·15 are reported.
† Rounded at 0·1 %.
Table 5 Food groups consumed at snacking occasions showing differences in mean portion size between BMI categoriesFootnote *: school-aged children (grade 4–6) from a Kanien’kehaka (Mohawk) community, Canada (n 444)
Normal weight, BMI between 5th and 85th percentile; at risk of overweight, BMI between 85th and 95th percentile; overweight, BMI > 95th percentile.
a,b Mean values within a row with unlike superscript letters were significantly different (P < 0·05).
* Only food groups from Table 1 showing differences in mean portion size between BMI categories at P < 0·15 are reported.
Finally, when all of the variables from Table 4 were included in a logistic regression model together with age, gender and year of study, the only significant effect left was for desserts, with overweight children consuming them less frequently than normal-weight children (P = 0·04).
Discussion
Snacking is a common eating habit for the majority of schoolchildren in Kahnawake, contributing approximately 20 % of their total energy intake, a finding similar to other studies carried out in North America(Reference Nicklas, Baranowski, Cullen and Berenson5, Reference Jahns, Siega-Riz and Popkin8, Reference Adair and Popkin9, Reference Stockman, Schenkel, Brown and Duncan17). Like Nicklas et al.(Reference Nicklas, Morales, Linares, Yang, Baranowski, DeMoor and Berenson11) but contrary to another study(Reference Jahns, Siega-Riz and Popkin8), no increase was observed in energy intake from or energy density of snacks since 1994. This could be explained by the fact that the increase in the energy contribution from snacks was observed, at least for the USA, in the early 1990s, before our study, suggesting that this trend was therefore over(Reference Newby7, Reference Jahns, Siega-Riz and Popkin8). Rather, we observed a shift in the timing of snacks, with an increase in morning snacks and a decrease in afternoon and evening snacks. We also observed that girls consumed more snacks than boys, in agreement with Forslund et al.(Reference Forslund, Torgerson, Sjöström and Lindroos21) and Ovaskainen et al.(Reference Ovaskainen, Reinivuo, Tapanainen, Hannila, Korhonen and Pakkala24) but contrary to Field et al.(Reference Field, Austin, Gillman, Rosner, Rockett and Colditz12) and Jahns et al.(Reference Jahns, Siega-Riz and Popkin8), suggesting that those observations may be highly context-dependent and require more data.
Our results also showed that, as they get older, children increase their energy intake at meals while reducing energy density from snacks, without changing the amount of total energy, number of snacks and time of snacking. The few differences we detected may be due to the narrow age range of our participants and could be expected to become larger with adolescence(Reference Moreno, Kersting, de Henauw, Gonzalez-Gross, Sichert-Hellert, Matthys, Mesana and Ross25).
The main objective of the current study was to better understand the snacking patterns of children across different body weight categories in order to contribute to the ongoing community programme in Kahnawake. Contrary to the results of several other studies(Reference Nicklas, Morales, Linares, Yang, Baranowski, DeMoor and Berenson11–Reference Phillips, Bandini, Naumova, Cyr, Colclough, Dietz and Must14), our study clearly shows that children at risk of overweight consumed a greater number of snacks than the other two BMI categories without differences in total energy intake, energy from fats, energy density or diet diversity, as it was recently reported(Reference Receveur, Morou, Gray-Donald and Macaulay19).
As for types of snacks, normal-weight children consumed more crackers, popcorn and pretzels than heavier children. Children in the at risk of overweight group stood out however by consuming more frequently evening snacks, high-energy-density foods like potato chips and larger portions of breakfast cereals and cereal bars. We had already reported higher consumption of potato chips among children at risk of overweight, but we can now further identify it as a snacking issue(Reference Receveur, Morou, Gray-Donald and Macaulay19).
Results for milk consumption showed that overweight children do not consume milk more frequently, but in larger servings at snack time. Also, no difference was found in consumption of some food groups according to BMI category, such as sugar-sweetened beverages, while this association is increasingly noted in longitudinal studies(Reference Phillips, Bandini, Naumova, Cyr, Colclough, Dietz and Must14, Reference Striegel-Moore, Thompson, Affenito, Franko, Obarzanek, Barton, Schreiber, Daniels, Schmidt and Crawford26, Reference Ludwig, Peterson and Gortmaker27). In a previous study(Reference Adams, Receveur, Mundt, Paradis and Macaulay28) of the same children from 1994 and 1998 only, our group had shown that lean children consumed more milk overall (snacks + meals), in keeping with other studies(Reference Barba, Troiano, Russo, Venezia and Siani29, Reference Weaver and Boushey30). Further ANOVA were then performed (results not shown) in the milk and sugar-sweetened beverage categories: an important decline in consumption of both was observed between 1994 and 2002, especially with meals. Adjusted for age, gender and BMI, total daily average consumption has gone from 343 g to 224 g for milk, a downward trend confirmed elsewhere(Reference Nicklas, Demory-Luce, Yang, Baranowski, Zakeri and Berenson23, Reference Striegel-Moore, Thompson, Affenito, Franko, Obarzanek, Barton, Schreiber, Daniels, Schmidt and Crawford26, Reference Nielsen and Popkin31), and from 546 g to 332 g for sugar-sweetened beverages, a result contrary to the secular trend observed(Reference Nicklas, Demory-Luce, Yang, Baranowski, Zakeri and Berenson23, Reference Striegel-Moore, Thompson, Affenito, Franko, Obarzanek, Barton, Schreiber, Daniels, Schmidt and Crawford26, Reference Nielsen and Popkin31) and which could be attributed, at least in part, to the diabetes prevention programme in place in schools since 1994.
Finally, in our multiple regression models, we found an inverse association between overweight and dessert consumption. Similar results have been reported in other studies. For example, in a systematic review including 137 000 children aged 10–16 years in thirty-four countries, Janssen et al.(Reference Janssen, Katzmarzyk, Boyce, Vereecken, Roberts, Currie and Pickett1) reported a negative relationship between the intake of candies/chocolates and BMI in thirty-one of the thirty-four countries, even when controlled for being on a weight-loss diet. A nutritional survey of Québec children found a negative association between overall candy, chip and fruit drink consumption and overweight(Reference Lavallée32). A low reporting of some food groups such as sweets and sugary foods found in those studies may also reflect recognition of their socially negative connotations, as observed by Klesges et al.(Reference Klesges, Baranowski, Beech, Cullen, Murray, Rochon and Pratt33) in young Afro-American girls aged 8–10 years. This further reduces chances of finding associations between the consumption of specific snack foods and the risk of overweight.
The lack of further differences in snacking behaviour among overweight children could be due to greater bias of under-reporting among overweight than among at risk of overweight children, as observed in adults(Reference Huang, Howarth, Lin, Roberts and McCrory13, Reference Bandini, Schoeller, Cyr and Dietz34, Reference Livingstone and Robson35). We calculated Goldberg factors(Reference Goldberg, Black, Jebb, Cole, Murgatroyd, Coward and Prentice36) across BMI categories. Indeed, these were 1·7 for normal-weight, 1·5 for at risk of overweight and 1·2 for overweight children, suggesting that under-reporting was present among heavier children but not to the extent of invalidating the data. Contrary to Moreno et al.(Reference Moreno, Kersting, de Henauw, Gonzalez-Gross, Sichert-Hellert, Matthys, Mesana and Ross25), we observed no differences in under-reporting according to gender.
Apart from reporting bias, there may be other reasons for the relative lack of association between food consumption and body weight category. Perhaps we should not expect a dose–response relationship between at risk of overweight and overweight children because these two conditions could be associated with different behaviours. For example, overweight children might decrease their physical activity level as their body weight increases while at risk of overweight children might keep the same level of physical activity as their normal-weight peers. In the same way, overweight children may restrict their food intake as they are gaining weight while at risk of overweight children may not. In fact, other studies showed no differences in energy intake according to BMI or percentage body fat(Reference Field, Austin, Gillman, Rosner, Rockett and Colditz12, Reference Phillips, Bandini, Naumova, Cyr, Colclough, Dietz and Must14) and two studies showed an inverse association between energy from snacks and body weight in children and adolescents(Reference Lioret, Touvier, Lafay, Volatier and Maire16, Reference Stockman, Schenkel, Brown and Duncan17).
Our results must be considered with caution because of other limitations of our study. First, since it is a cross-sectional study, causal direction cannot be established. Second, food records included only one 24 h recall in which a misclassification bias may exist between meals and snacks, especially since children do not always eat three meals a day and may skip meals and consume several snacks(Reference Livingstone and Robson35). Third, a relatively small sample size for some snack categories reduces the possibility of detecting differences with sufficient statistical power. Fourth, certain variables associated with risk of overweight in children linked to social environment, lifestyle and hereditary factors were not included in the scope of the study but may explain differences in results. For example, physical or sedentary activities, time spent in front of a computer screen, parents’ weight, mother’s education level, concern about weight, dieting or pubertal development were not accounted for in analysis. Finally, the results of our study apply only to Kahnawake children and cannot be generalized for other populations.
One of the factors making comparisons with other studies difficult is the very definition of what constitutes a snack or snacking occasion(Reference Newby7, Reference Jahns, Siega-Riz and Popkin8, Reference Francis, Lee and Birch10, Reference Nicklas, Morales, Linares, Yang, Baranowski, DeMoor and Berenson11, Reference Phillips, Bandini, Naumova, Cyr, Colclough, Dietz and Must14, Reference Nicklas, Yang, Baranowski, Zakeri and Berenson15, Reference Forslund, Torgerson, Sjöström and Lindroos21, Reference Nicklas, Demory-Luce, Yang, Baranowski, Zakeri and Berenson23). Also, there is no consensus at this time on a standardized method of classifying foods by group in relation to their nutrient intake(Reference Drewnowski22). This makes comparisons difficult, despite our attempt to classify foods into groups according to examples set by recent studies(Reference Phillips, Bandini, Naumova, Cyr, Colclough, Dietz and Must14, Reference Forslund, Torgerson, Sjöström and Lindroos21–Reference Nicklas, Demory-Luce, Yang, Baranowski, Zakeri and Berenson23). Gregori and Maffeis’ call for a clearer definition of a snack is highly relevant(Reference Gregori and Maffeis37) as a food classification system is needed to make food consumption data comparable(Reference Moreno, Kersting, de Henauw, Gonzalez-Gross, Sichert-Hellert, Matthys, Mesana and Ross25).
The present study adds to previous studies by trying to establish an association between specific eating patterns and overweight. In our study, it was possible to detect differences in snack food intake in normal-weight and at risk of overweight children. In addition to an added focus on factors linked to under-reporting, this suggests that future studies of dietary differences associated with body weight would benefit from comparing children at risk of overweight, and not only overweight children, with normal-weight children.
From a practical standpoint, school interventions related to snacks are appropriate since children may have more control over their snack choices than they have at meals, which are generally overseen by parents. For this reason, teaching children to change their snack and/or sugar-sweetened beverage choices can be relevant(Reference Jimenez, Receveur, Trifonopoulos, Kuhnlein, Paradis and Macaulay38). This is also true for food policies that make healthier snack choices available and discourage less healthy choices(Reference Macaulay, Paradis and Potvin18, Reference Paradis, Lévesque, Macaulay, Cargo, McComber, Kirby, Receveur, Kishchuk and Potvin39). In this regard, promoting the consumption of water, milk and other nutritional beverages, instead of sugar-sweetened beverages, and fruits and vegetables, instead of energy-dense foods, is relevant. A particular attention to snacking patterns is also warranted since snacks can contribute important nutrients on a daily basis and snacking patterns highlighted in the present study are dynamic across time, gender and body weight categories.
Acknowledgements
Sources of funding: G.M. was supported through a studentship of the Canadian Institutes of Health Research (CIHR MFE-78173). KSDPP funding came from National Health Research and Development Program, Health Canada (6605-4188-ND and 6605-4187-ND) and the Canadian Institutes of Health Research (CIHR H43274). Conflict of interest declaration: The authors have no conflict of interest. Authorship responsibilities: G.M. wrote the manuscript and assisted in the statistical analyses. O.R. conducted the statistical analyses and revised the manuscript. A.C.M. contributed to the study design, funding and revised the manuscript. Acknowledgements: The authors thank the schools, teachers, parents and children for their ongoing support and for agreeing to participate in this project. They also thank the KSDPP Community Advisory Board for ongoing guidance and support, and especially Chantal Saad-Haddad and Treena Delormier for reviewing the paper and ensuring community relevance.
