![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
Department of Human Nutrition and Department of Epidemiology and Biostatistics, University of Illinois at Chicago, Chicago, IL;
*
Department of Nutrition and Carolina Population Center, University of North Carolina at Chapel Hill, Chapel Hill, NC; and
Institute of Nutrition and Food Hygiene, the Chinese Academy of Preventive Medicine, Beijing, China
4To whom correspondence should be addressed. E-mail: youfwang{at}uic.edu.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
(a statistic measuring agreement beyond chance). About half of those who initially consumed high fat, high carbohydrate, high vegetable and fruit, and high meat diets continued such diets 6 y later. Family income, urban-rural residence, mother’s education and baseline dietary intakes were important predictors of children’s dietary intake patterns. In conclusion, even under conditions of rapid socioeconomic change, children are likely to maintain their dietary intake patterns from childhood into adolescence. Efforts to promote healthy eating behaviors may be more effective if focused on younger children, and parents should be involved in these efforts.
KEY WORDS: • children • adolescents • dietary intake • tracking • China
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
–6
). A few small-scale cohort studies from industrialized countries indicate tracking (i.e., maintenance) of dietary patterns from childhood into adolescence (7
–12
). The types of diet linked with chronic diseases are found to prevail before pubertal maturation, and children’s dietary patterns change only marginally during pubertal development (13
). However, these findings of dietary tracking have not been conclusive due to the limitations of the studies or their research design such as small sample sizes, short follow-up periods and homogeneous study populations.
On the other hand, many researchers believe that considerable changes in an individual’s eating behavior may occur between childhood and adolescence due to the physiologic changes related to growth and maturation as well as the increasing independence and interaction of adolescents with their social environment (14
,15
). Research from industrialized countries, predominantly the United States, shows that adolescents have unique dietary patterns (16
–22
). Little research has been conducted to study specific factors that may influence changes or predict the tracking of dietary patterns from childhood into adolescence. Of equal importance, our understanding about the dietary patterns of children and adolescents in developing countries is very limited.
Of great concern is the increase in the prevalence of child obesity, which has become an important public health challenge in many industrialized countries. Obesity is also increasing at an alarming rate in many developing countries (23
,24
). Childhood and adolescence are critical periods for individuals to lay the foundation for their future good health (25
). Childhood and adolescence are also two critical periods in the development of obesity, and the prevention of obesity in children and adolescents has been suggested as a public health priority (2
,6
). Understanding the dietary patterns of children and adolescents will enhance our efforts to promote healthy eating behavior in these age groups to prevent chronic diseases.
The present study examines the tracking of dietary patterns from childhood into adolescence as well as the predictors of these patterns with a focus on examining macronutrients and food groups linked to obesity and chronic diseases. Longitudinal data from the China Health and Nutrition Surveys (CHNS),5
collected between 1991 and 1997, were examined. China, whose population accounts for about one fifth of the world population (26
), has experienced remarkable socioeconomic changes during the past two decades (27
). Large heterogeneities exist in people’s diet, lifestyles and nutritional status (27
,28
). This provides us with an ideal setting in which to study these questions.
|
SUBJECTS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
The CHNS is a longitudinal study that began in 1989 and covered eight provinces (Guangxi, Guizhou, Henan, Hubei, Hunan, Jiangsu, Liaoning and Shandong provinces). Data were collected in four rounds (in 1989, 1991, 1993 and 1997). School-age children were not included in the 1989 survey but have been surveyed since 1991. The initial survey covered 3780 households. To ensure adequate sample size, new households were added in later surveys to make up for those lost to follow-up (due mainly to migration). Individuals who were lost to follow-up were physically absent when the surveys were conducted. For example, children and adolescents might have been away from home because they were attending school, working or visiting relatives. Data on anthropometry and diet were collected for each family member in each survey. Additional information was collected at household and community levels. More details about the study were published previously (29
,30
).
Subjects
A cohort of 984 children initially aged 6–13 y who had complete dietary data for both the 1991 and 1997 CHNS surveys were included as the core of our study sample. They became 12–19 y of age in 1997. In the 1991 survey, a total of 1946 children aged 6–13 y were surveyed and had complete dietary data for at least two 24-h recalls; 984 of them were resurveyed in 1997 and had complete dietary data. A major reason for the relatively low follow-up rate was because one province (Liaoning) did not participate in the 1997 survey. Overall, the follow-up rate was >60% between 1991 and 1997 for the remaining provinces. Our analysis shows that those who were lost to follow-up in 1997 were not significantly different from those being resurveyed regarding their baseline sociodemographic characteristics and main dietary intake variables (P > 0.05). In addition, we tested potential selection bias using the Heckman selection models (see Statistical Analysis). The University of North Carolina at Chapel Hill School of Public Health and the Chinese Academy of Preventive Medicine reviewed and approved procedures for the protection of human subjects during data collection.
Data collection
Dietary data.
Detailed household food consumption data and individual dietary intake data were collected for three consecutive days in each survey. The sample was randomly allocated from Monday to Sunday and almost equally balanced across the 7 d of the week from each sampling unit. Household food consumption was determined by inventory change from the beginning to the end of each day. Individual dietary intake data (24-h recall) for the same three consecutive days were obtained from all family members. For young children (<10 y old), dietary intake was reported by their mothers. The collection of data on household food consumption and individual dietary intake allowed a check on the quality of each against the other. At the time of data collection, the individual and household dietary data were compared and used to identify major discrepancies. When significant discrepancies were found, the household and the individual in question were revisited and asked about their food consumption to resolve these discrepancies. From the household dietary data, information on added fat (cooking oil represents a significant component of fat intake in the Chinese diet) and other condiments, such as salt, supplemented the individual dietary intake data based on an individual’s meat and vegetable consumption (31
,32
). Great efforts were made in the field to survey all subjects in the same season in each round; the majority of the subjects were interviewed within a short period in the fall in each survey. The 1991 Chinese food consumption table (33
) was used to calculate subjects’ nutrient intakes.
Quality control. The survey team was comprised of 20–23 nutritionists for each of the provinces surveyed in 1991, 1993 and 1997. All interviewers had at least a college degree and they had attended 10-d training sessions for each survey. Most interviewers had the experience of conducting other national health and nutrition surveys. Interviewers were required to follow a carefully developed protocol similar to that used in the U.S. National Health and Nutrition Examination Surveys developed by the National Center for Health Statistics. In addition, inter- and intraobservation reliability and equipment checks were conducted during the training and data collection.
Study variables
Dependent variables—dietary intake. An individual’s diet was characterized by using the individual’s 3-d average macronutrient intakes and food group consumption. Three main food groups [meat, vegetable and fruit (VF) and edible oil] were selected on the basis of the University of North Carolina and the Chinese Institute of Nutrition and Food Hygiene (UNC-INFH) China Food Grouping System, developed by the authors (Wang et al., unpublished data). The UNC-INFH China Food Grouping System separates >1400 foods into five major categories and 39 food groups. Meat, vegetable and fruit were highlighted because they are important sources of protein, minerals and vitamins, which are critical for normal growth and good health. In addition, these nutrients and edible oil influence the energy density of diet and total energy intake, and therefore may affect adiposity.
Relative measures of dietary intakes.
We chose to use relative measures [e.g., percentage of Recommended Dietary Allowances (RDA) and percentage of energy derived from the macronutrient], instead of absolute amounts of nutrient or food intakes as our main measures to study the tracking of children’s dietary intakes over time. This is because biological requirements for nutrients differ by age, gender and maturation during childhood. Normally the absolute amounts of nutrient intakes and food consumption increase with age (14
,15
,34
). Total energy intake was expressed mainly as percentages of the Chinese RDA (33
), whereas dietary fat, carbohydrate and protein intakes were expressed as percentages of the total energy intake (%E). Consumption of food groups was measured as grams of food per 1000 kcal (1 kcal = 4.184 kJ) energy intake (g/kcal). In other words, we examined children’s diet composition (diet structure) instead of the absolute amounts of their dietary intake. There are three main advantages in using these relative measures. First, they were comparable over time because age and gender could be adjusted. Second, the use of children’s macronutrient intake relative to total energy allowed us to study the change and track the children’s diet structure over time. These data can help define some specific types of diet, for example, a "high fat diet," in which the proportion of energy derived from dietary fat is > 30%. Finally, reporting errors are often adjusted when dietary intakes are expressed as a proportion of energy intake (35
,36
). When self-reported dietary data such as 24-h recalls are used to measure people’s diets, it has been of concern that some individuals may omit foods, meals or snacks, for example; as a result, their energy and food consumption might be underestimated (36
–38
).
To study the tracking of energy intake, we calculated each subject’s energy intake relative to his/her body weight (kcal/kg). This allowed a consideration of the large variation in energy requirements due to differences in body size among individuals of the same age. An FAO/WHO/UNU Expert Committee concludes that the most useful index of the basal metabolic rate (BMR) is body weight, and in most (age-sex) groups, height does not help to predict the BMR independently of weight (39
). On the other hand, the use of gender- and age-specific Chinese energy intake RDA could not fully address this problem because it is defined only for broad age groups (33
). Similar analyses were also performed for other dietary intakes, but the results were similar to those based on %E and g/kcal. Therefore they are not presented here.
We also examined an alternative approach to express energy intake as relative to the predicted BMR (kcal/BMR), which also helped us to test the validity of using three 24-h recalls to estimate children’s total energy intake in the study population. The FAO/WHO/UNU Expert Committee recommended that estimates of energy expenditure be expressed as multiples of predicted BMR (predicted BMR = ß · weight + constant) to provide the basis for calculating energy requirements (39
). Later Goldberg and colleagues (37
,38
) revisited the topic and demonstrated that in a large number of studies, people’s actual energy intake had been underestimated based on self-reported data on dietary intake (e.g., 24-h recalls), particularly in wealthy populations. They suggested using the multiples of BMR to test the validity of estimates in energy intake.
We found that on average the multiple factor was 
1.9 for Chinese boys and 1.6 for Chinese girls, which matched the Expert Committee’s suggested figures (39
). This suggests that the subjects’ estimated energy intakes based on the three 24-h recalls seem to be valid. To calculate the predicted BMR, we used the committee’s recommended equations (e.g., for boys aged 10–18: BMR = 17.5 · weight + 651, and for girls aged 10–18: BMR = 12.2 · weight + 746) (39
).
Our results of tracking from the two approaches were similar. For example, estimates of the proportion of tracking were almost identical and = 0.8, which indicates excellent agreement (40
), for children’s dynamic patterns (e.g., classified as "remaining in the top quartile—tracking high," "moved to a higher quartile," "moved to a lower one" and "remaining in the bottom quartile—tracking low"). These were as expected because in fact the two approaches were similar when children’s sex and age were controlled for because we classified tracking patterns using sex-age-specific quartiles. Thus, we chose to use kcal/kg because the calculation is easier and the interpretation is straightforward.
Definition of tracking of dietary intake patterns.
We chose to use gender- and age-specific quartiles (6–9 and 10–13 y at the baseline) for each dietary intake variable to define the tracking of dietary patterns. If an individual remained in the same quartile in both the 1991 and 1997 surveys, this pattern was defined as "tracking." We thus classified six types of diet: "tracking of high fat diet," "tracking of high carbohydrate diet," "tracking of high energy diet," "tracking of high VF diet," "tracking of low VF diet" and "tracking of high meat diet" (see Table 1
). This is of interest particularly due to the linkage between such diets and chronic conditions such as heart disease and obesity.
|
). This allowed us to examine what factors might predict beneficial or adverse changes over time. Main independent variables. Detailed sociodemographic data were collected in the surveys. For this analysis, the children were separated into two age groups: 6–9 y (childhood, called "younger children" in this paper) and 10–13 y (early adolescence, called "older children"). Other measures were urban-rural residence, per capita family income tertiles (which were used to indicate low, medium and high socioeconomic status), maternal education (low, medium and high), father’s occupation, and geographic regions [the seven provinces were separated into three regions, i.e., the Coast (Shandong and Jiangsu), Central(Hunan, Hubei and Henan) and Southwest (Guizhou and Guangxi). Liaoning did not participate in the 1997 survey].
Statistical analysis
Correlation coefficients have been widely used to study associations and tracking (36
,37
). We calculated both Pearson and Spearman’s rank-order correlation coefficients to examine the correlation between individuals’ repeated dietary intake measures in 1991 and 1997. Although the Pearson correlation coefficient, in which dietary intakes are treated as continuous variables, is widely used, Spearman’s rank correlation coefficient is likely a better measure with which to study tracking of dietary intakes because it helps minimize concerns about outliers and data distribution (41
, 42
). To calculate Spearman’s coefficients, we used the relative measures of dietary intakes because they were "comparable" across gender-age groupings. In addition, we performed similar analyses to examine the tracking between 1991 and 1993, and 1993 and 1997 to test whether the association between individuals’ two repeated measures was stronger within a shorter time interval.
The tracking patterns of individuals’ dietary intakes over a 6-y period were examined by using quartiles, i.e., to examine the proportion of subjects who remained in the same quartile over time. -values were calculated to examine the general tracking patterns, testing the agreement between each individual’s relative positions in 1991 and 1997. These were classified on the basis of sex- and age group–specific quartiles in each year. = 0 when the observed agreement equals that expected by chance, and = 1 when the tracking is perfect, i.e., when everyone maintained their quartile positions between 1991 and 1997) (40
,43
). Further, we studied the tracking of the six types of diet by examining the proportions of subjects who maintained the diets over time. When comparing differences between groups, Cochran-Mantel-Haenszel tests were performed for categorical outcome variables to control for potential confounders (43
).
Because there are no well-established cut-off points to define tracking, we chose to use 0.2 as the cut-off point for correlation coefficients. This figure is widely used by statisticians to evaluate the strength of correlations. For , a value > 0.2 was selected to indicate the existence of tracking, 
0.4 for moderate tracking, and 0.8 for excellent tracking (40
). For percentage of tracking, we used the cut-off point of 33.3% because only 25% of subjects will remain in the same quartile between two time points assuming they could move randomly into any of the quartile at time 2 (i.e., proportion of tracking by chance). In general, greater values of correlation coefficient, and proportion of individuals who remained in the same categories suggest stronger tracking. Additionally, we conducted a simulation analysis in an attempt to examine whether some of the tracking patterns we observed were significantly different from the statistically expected ones.
Finally, we conducted multiple ordinary and multinomial logistic regression (M-logit) analyses to study the predictors for tracking of different dietary patterns (43
,44
). We focused on the tracking of the six special types of diet (see Table 1
): first, we studied the predictors of tracking when our outcome variables were coded as dichotomous variables (i.e., tracking: yes/no); then, for dietary fat and vegetable and fruit intakes, we examined the dynamic patterns of children’s dietary intake by conducting multinomial logistic regression analyses (43
,44
). All odd ratios (OR) presented were adjusted for potential confounders.
In most longitudinal studies, there is a concern about selection bias related to systematic lost to follow-up. Our unreported analyses found that although individuals who were lost to follow-up in 1997 were not significantly different than those being resurveyed on main baseline sociodemographic characteristics and dietary intakes, there were significant differences in several other key characteristics. The rate of missing observations was higher in older children and in urban areas. To test the potential selection bias, we conducted linear regression analyses applying the Heckman selection models, which make corrections to the results on the basis of some variables (e.g., at baseline) that strongly affect the chance of observation (missing or not) but not the outcomes under study (44
,45
). Correcting selection bias changed the size of some coefficients but not their statistical significance or direction, and the differences were small (results were not presented). Therefore, we have confidence that the results presented here are not likely to be affected considerably by selection bias. Data management and analysis were performed by using SAS (version 6.12, Cary, NC) and Stata (version 6.0, College Station, TX).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
About half of the 984 subjects were male. Their mean age was 9.4 y, and about one fifth lived in urban areas. Their 3-d average energy intake was 2009 kcal (8406 kJ); the proportions of energy derived from fat and carbohydrate were 22.3 and 66.1%, respectively.
General tracking patterns of dietary intakes from childhood into adolescence.
Pearson correlation coefficients (not presented) and Spearman’s rank-order correlation coefficients showed similar tracking patterns for these children’s dietary intakes over time. Over a 6-y period, there were specific patterns of children’s dietary intakes, particularly in terms of the proportion of energy derived from dietary fat and carbohydrate as well as meat consumption (r = 0.5) (Table 2
). Meanwhile, the tracking of energy, protein, vegetable and fruit, and edible oil was moderate or weak (r 0.2). In general, no notable gender, age or rural/urban differences were found except that girls were less likely to track energy, vegetable and fruit intakes than boys. Moreover, the overall tracking during 1991–1993 and 1993–1997 was stronger than that for 1991–1997.
|
Using gender- and age-specific quartile we examined the agreement between individuals’ relative positions between 1991 and 1997. There were tracking patterns for dietary fat, carbohydrate, meat and edible oil intakes (Table 3
). Close to 40% of these subjects maintained their quartile positions 6 y later. Interestingly, we found a stronger tracking pattern when individuals’ total energy intake was expressed relative to his/her body weight (kcal/kg) than as relative to sex-age-matched Chinese RDA. This indicates that body weight influences energy intake following suggestions that overweight individuals have to consume more food to maintain body weight than normal weight people (23
). This also is consistent with the FAO/WHO/UNU Expert Committee’s conclusion that body weight is the most useful index of the BMR (39
).
|
We found a strong tracking pattern for all six types of diet (Table 4
). Almost half of those who initially consumed high fat, high carbohydrate, high VF, and high meat diets still consumed such diets 6 y later. About one third of those who initially consumed high energy and low VF diets still did so 6 y later. There was no significant gender difference (P > 0.05), but we found a notable urban/rural difference for all diet types except for a high energy diet (P < 0.05). In addition, older children (10–13 y) were less likely to maintain a high energy diet than younger children [27.5%, 95% confidence interval (CI), 23.7%, 31.3% vs. 46.8%, 95% CI, 42.4%, 51.2%] (P < 0.001); the OR and 95% CI were 0.48 (0.32, 0.70). Older girls were least likely to maintain such a diet, with only 23.8% (95% CI, 17.2%, 30.4%) maintaining the high energy diet. This may be due to the fact that the variation in individuals’ energy requirements was larger in the older group than in the younger group in 1997 because many older adolescents might have been involved in different occupations or other activities. There are also other social and behavioral factors that may contribute to such differences. In general, the above results (e.g., correlation, and proportion) suggest the existence of tracking in Chinese children’s dietary intake patterns.
|
–48
). We performed statistical simulation analyses and theoretical calculations to compute the probabilities for individuals to maintain their initial quartile position (48
). Our analysis showed that the proportion of tracking was 38.0% for the top and bottom quartiles, and 26.8% for the 2nd and 3rd quartiles when r = 0.3 (observed correlation between 1991 and 1997 VF intakes); the figures were 48.1 and 29.6%, respectively, when r = 0.5 (observed correlation for meat consumption in 1991 and 1997). Our real data showed that 44.8% (95%CI, 38.6%, 51.0%) of these children maintained a high VF diet (the top quartile) that was significantly higher than predicted (38.0%)(P < 0.05). The observed tracking for a high meat diet was 47.2% (95%CI, 41.0%, 53.4%), which was not significantly greater than expected (48.1%). However, they both were significantly > 25% (tracking by random) and than the predicted tracking for the 2nd and 3rd quartiles. (Note 95% CI was calculated as p ± 1.96 · [p(1 - p)/(n/4)]1/2, where p is the proportion of tracking and n = 984.)
However, in the real world, the dynamic patterns in children’s dietary intake over time are more complex. It is not clear whether testing the difference between a randomly generated tracking result and our real world results would allow us to test in a meaningful way whether the tracking of extreme values is significant (47
). Moreover, from a public health point of view, if children with "extreme" eating patterns (e.g., a high fat diet) are likely to maintain such patterns over time, efforts should be made to help them modify their eating behaviors given the evidence of risks related to such dietary patterns.
Predictors of tracking of the six types of diet (see Table 1
for the definitions).
First we conducted logistic regression analyses using the whole sample to examine how those who maintained a certain dietary intake pattern (tracking group) were different from the others (seeTable 5
). This captured both predictors that related to subjects’ initial diet at baseline and those that affected the maintenance of the diet during the follow-up. Then, we conducted logistic regression analysis using data only for those who initially had a certain type of diet (e.g., a high VF diet), which accounted for about one fourth of the total sample (Table 6
, called "the subset-analysis"). The reference group, e.g., high VF diet, were those who initially had a high VF diet (in the upper quartile) but moved to another quartile at the end of the follow-up. The interpretation of the findings (OR) was more straightforward, for example, it shows why among those who initially had a high VF diet, some maintained the diet, but others did not. The concern, of course, is that the limited sample size for this analysis limits the covariates one can examine (43
).
|
|
and 6)
. Urban children were more likely to maintain high fat and high meat diets, but less likely to track a high VF diet. Family income was another important predictor of children’s dietary tracking patterns. High income children were more likely to track high fat and high meat diets, but less likely to maintain a high carbohydrate diet. Our analysis with the whole sample shows that mother’s education influenced children’s dietary fat, carbohydrate, VF and meat consumption (see Table 5
); however, the subset-analysis detected a significant effect only on VF (see Table 6
). Interestingly, among those who initially had a high energy intake, boys were more likely to maintain such a diet than girls [OR = 1.6 (1.1, 2.5)]. This might indicate some gender difference in body image and dieting practice among Chinese children. Predictors of the dynamics of high fat and high VF (vegetable and fruit) diets.
Using multinomial logistic models, we further studied the predictors of the dynamics of children’s fat and VF intake patterns (see Table 1
for the definitions). Consistent with what we found for the predictors of the above six special diets using ordinary logistic models, urban-rural residence, region, family income, mother’s education and baseline energy intake were important predictors of the dynamics of children’s dietary fat and VF intakes. Children who lived in urban areas had a higher baseline energy intake, and those whose mothers were better educated were more likely to reduce their fat intake to <30%E, whereas high income children were more likely to increase their fat intake to >30%E. Urban children and those who lived in the Coastal areas were more likely to track lower VF consumption (bottom quartile). In addition, children whose fathers had a medium-level occupation were more likely to maintain low or medium fat intake (%E 
30%) (P < 0.05); low paternal occupation predicted a lower possibility for children to change their fat intake (P < 0.05).
To better interpret the meaning of the OR achieved by the above multinomial logistic regression models, we calculated the predicted probabilities for individuals to have a certain type of dietary pattern (e.g., tracking low, tracking high, decrease and increase) when assuming that all of the subjects had a certain characteristic (e.g., all lived in rural areas) but holding all other variables such as age, sex, maternal education, family income and region constant (43
,44
). Considerable differences existed across different rural-urban, income, and education subgroups for dietary fat and for VF intake patterns (Figures 1
and 2).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
,28
,49
,50
). For example, per capita Gross Domestic Product has increased >10 times and per capita meat consumption has almost doubled in rural areas (27
). Nevertheless, we still found that a large proportion of children track their dietary intake patterns from childhood into adolescence. In general, the tracking patterns we observed were significant in statistical terms (e.g., correlation coefficients, coefficients and proportions). Nearly half of those who initially consumed a high fat, high carbohydrate, high vegetable and fruit or high meat diet still consumed such a diet 6 y later when they become adolescents. The proportions of children who remained in the same quartile over a 6-y period were significantly greater than those expected by chance (25%), and some were even greater than predicted given an overall positive association between repeated measures over time (e.g., a high VF diet).
Our results are consistent with findings from a few previous small-scale cohort studies conducted in industrialized countries (7
–12
). For example, the Framingham Children’s Study showed that tracking of dietary intakes begins at very young ages; the correlation coefficients in that study ranged from 0.35 to 0.63 (11
). Similarly, Deheeger et al. (51
) and Stein et al. (12
) reported tracking of macronutrient intakes among 112 young French children and a small cohort of predominantly Hispanic children, respectively. A study by Kelder et al. (8
) may be the largest study with the longest follow-up that has reported tracking of children’s diet, but it was focused on the number of healthy food choices. Our findings about tracking of vegetable and fruit consumption are consistent with a recent study by Resnicow et al. (10
). Of more interest, our results concerning the tracking of high fat and high meat diets are similar to findings by Clavien et al. (13
). They found that the types of diet linked with chronic diseases already prevail before pubertal maturation, and children’s dietary patterns change only marginally during pubertal development. Furthermore, the Amsterdam Growth and Health Longitudinal Study showed that dietary intake patterns track from adolescence into early adulthood (5
,34
). In summary, our findings and those of other studies all indicate that tracking of children’s dietary intake exists during childhood and adolescence in different societies, but the extent of tracking may vary in different populations due to social and economic differences among them, as well as differences in study designs.
Some of the dietary intake tracking patterns in our survey sample (e.g., carbohydrate and meat intakes) are stronger than in other studies from developed countries. This might relate to the large proportion of Chinese children who ate at home (e.g., >97% of dinners and 93% of lunches were consumed at home in our sample in 1997). When children and adolescents eat away from home, they are likely to have more and different food choices than at home. The shift to away from home food consumption in China is very slow. For instance, one study of >5900 Chinese adults found that the proportion of energy consumed at home declined from 92.5 in 1989 only to 90.8 in 1997 (32
). This paper focuses on examining tracking and change in dietary intakes among children in a society in which most meals are still household based. We do not attempt to clarify how much of these tracking patterns are unique to these age groups vs. those that are common to all household members. Most Chinese parents give their children the option to choose foods available at home. In addition, there is some evidence of the exertion of strong influence by Chinese children on their families’ food (49
).
A unique feature of this research is an attempt to understand the predictors of tracking and change in children’s dietary intake patterns over a 6-y period. We found that urban-rural residence, family income, mother’s education, region and energy intake at baseline were important predictors for tracking of dietary intake patterns. For example, urban children were more likely to maintain a high fat diet, but less likely to maintain high carbohydrate and high vegetable and fruit diets. Similarly, high income children were more likely to track high fat and high meat diets. These findings suggest that with economic development and improvement of people’s living standards in China, a transitional society, those who benefit the most and/or the most rapidly (e.g., urban and high income groups) may also be at an increased risk for nutritional problems related to chronic disease.
Another important predictor is mother’s education. Those children whose mothers had higher education levels were more likely to have a high fat diet but were less likely to maintain a high vegetable and fruit diet even when family income and urban residence were adjusted. Although these mothers were likely to have better access to the media and to health- and nutrition-related knowledge, their behavior indicates that they are not aware of the health concerns related to higher fat foods. These families might also have more family resources (e.g., better educated people were more likely to work in work units or government offices that could provide better welfare, although the salary might not be high), and thus could afford more expensive foods such as meats and cooking oil. Another possibility is that these women understood the need for energy-dense diets linked with high meat and fat intake and promoted that concern for growth and development over the concerns for obesity and diet-related noncommunicable disease. In summary, the association we observed probably suggests that mothers’ nutritional knowledge, health consciousness and exposure to the media play an important role in affecting their children’s dietary intake beyond the determining role of family resources and access to foods available to the community in developing countries undergoing rapid social and economic transitions.
Finally, this study has limitations. First, it is possible that three 24-h recalls do not measure individuals’ usual diet accurately, especially for children who may have a larger day-to-day variation than adults (36
). However, on the basis of the current literature (10
,12
) and our group’s previous work (52
,53
), we anticipate that stronger tracking patterns would be found if we could measure children’s usual diet more accurately or if intraindividual variation could be corrected, although the results are not likely to be changed dramatically. Unfortunately, it was not feasible for us to conduct such a sensitivity analysis because some of the children’s dietary intakes (e.g., fat consumption) were obtained with the supplement of household level 3-d food disappearance data. For example, the amount of cooking oil was divided among family members based on their meat and vegetable consumption (31
). Cooking oil represents the most important component of people’s fat consumption in the study population. Second, systematic loss to follow-up bias is a concern because it affects representativeness of the results or systematically precludes one subset of children and might bias the results. Losses that occurred in this study were due to the difficulties in reaching children in the households, loss of several villages and urban areas due to natural calamities, migration and the dropping of one of the eight provinces from the 1997 survey. We used the Heckman selection models to examine the magnitude of the bias for ordinary least-square models (44
,45
,54
). Our analysis suggests that correcting the selection bias does not change our estimates substantially. Third, it is possible that at least part of the tracking patterns we observed for some extreme dietary patterns such as high VF and high meat diets resulted from statistical properties. Usually when two repeated measures over time are correlated, individuals in the two ends of a distribution are more likely to keep their ranking position than others (46
–48
). It is of interest for future research to examine in an appropriate manner how biological, statistical and measurement factors may have influenced the tracking phenomena we and others observed. Finally, when studying the predictors of tracking, we could not examine the effects of other potential important factors such as children’s physical activity, knowledge and attitudes toward foods because these data were not collected at baseline.
In conclusion, our study suggests that even under rapidly changing conditions, children’s dietary intake patterns track from childhood into adolescence. A number of socioeconomic factors and family characteristics influence these patterns. Our findings provide some useful insights to guide those interested in promoting healthy eating behavior among children and adolescents. Efforts to promote a healthy diet should start during childhood.
|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
2 The study was supported in part by grants from the National Institutes of Health (R01-HD30880 and R01-HD38700) and by the Fogarty International Center (TW/HD00633). Funding for parts of the project design, data collection and computerization was provided by the CAPM, the UNC-CH and CPC and the NIH.
3 The China Health Nutrition Survey (CHNS) study is a collaborative research project between the Chinese Academy of Preventive Medicine (CAPM), directed by Zhai Fengying and a group from Carolina Population Center of the University of North Carolina at Chapel Hill (UNC-CH and CPC), directed by Barry M. Popkin.
5 Abbreviations used: BMR, basal metabolic rate; CHNS, the China Health and Nutrition Survey; 95%CI, 95% confidence interval; %E, percentage of total energy intake; OR, odds ratio; RDA, recommended dietary allowances; VF, vegetable and fruit; UNC-INFH, University of North Carolina and the Chinese Institute of Nutrition and Food Hygiene.
Manuscript received 10 September 2001. Initial review completed 16 October 2001. Revision accepted 19 December 2001.
|
LITERATURE CITED |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
1. Berenson, G. S., Srinivasan, S. R. & Nicklas, T. A. (1998) Atherosclerosis: a nutritional disease of childhood. Am. J. Cardiol. 82:22T-29T.[Medline]
2. McGill, H. C. (1997) Childhood nutrition and adult cardiovascular disease. Nutr. Rev. 55:S2-S11.[Medline]
3.
Dietz, W. H. (1994) Critical periods in childhood for the development of obesity. Am. J. Clin. Nutr. 59:955-959.
4.
Must, A. (1996) Morbidity and mortality associated with elevated body weight in children and adolescents. Am. J. Clin. Nutr. 63(suppl. 3):S445-S447.
5. Welten, D. C., Kemper, H. C., Post, G. B., Van Staveren, W. A. & Twisk, J. W. (1997) Longitudinal development and tracking of calcium and dairy intake from teenager to adult. Eur. J. Clin. Nutr. 51:612-618.[Medline]
6. Power, C., Lake, J. K. & Cole, T. J. (1997) Measurement and long-term health risks of child and adolescent fatness. Int. J. Obes. Relat. Metab. Disord. 21:507-526.[Medline]
7. Boulton, T. J., Magarey, A. M. & Cockington, R. A. (1995) Tracking of serum lipids and dietary energy, fat and calcium intake from 1 to 15 years. Acta Paediatr 84:1050-1055.[Medline]
8.
Kelder, S. H., Perry, C. L., Klepp, K. I. & Lytle, L. L. (1994) Longitudinal tracking of adolescent smoking, physical activity, and food choice behaviors. Am. J. Public Health 84:1121-1126.
9. Nicklas, T. A. (1995) Dietary studies of children: the Bogalusa Heart Study experience. J. Am. Diet. Assoc. 95:1127-1133.[Medline]
10. Resnicow, K., Smith, M., Baranowski, T., Baranowski, J., Vaughan, R. & Davis, M. (1998) 2-year tracking of children’s fruit and vegetable intake. J. Am. Diet. Assoc. 98:785-789.[Medline]
11.
Singer, M. R., Moore, L. L., Garrahie, E. J. & Ellison, R. C. (1995) The tracking of nutrient intake in young children: the Framingham Children’s Study. Am. J. Public Health 85:1673-1677.
12.
Stein, A. D., Shea, S., Basch, C. E., Contento, I. R. & Zybert, P. (1991) Variability and tracking of nutrient intakes of preschool children based on multiple administrations of the 24-hour dietary recall. Am. J. Epidemiol. 134:1427-1437.
13. Clavien, H., Theintz, G., Rizzoli, R. & Bonjour, J. P. (1996) Does puberty alter dietary habits in adolescents living in a western society?. J. Adolesc. Health 19:68-75.[Medline]
14. Mahan, K. L. & Escott-Stump, S. (1996) Krause’s Food, Nutrition, and Diet Therapy 9th ed. 1996 W. B. Saunders Philadelphia, PA. .
15. Trahms, C. M. & Pipes, P. L. (1997) Nutrition in Infancy and Childhood 1997 WCB/McGraw Hill New York, NY. .
16. Adams, L. B. (1997) An overview of adolescent eating behavior barriers to implementing dietary guidelines. Ann. N.Y. Acad. Sci. 817:36-48.[Medline]
17. American School Health Association (1989) Association for the advancement of Health Ed 1989 ucation and the Society for Public Health Education. The National Adolescent Student Health Survey A Report on the Health of America’s Youth. Third Party Publishing, Oakland, CA. .
18. Bull, N. L. (1988) Studies of the dietary habits, food consumption and nutrient intakes of adolescents and young adults. World Rev. Nutr. Diet. 57:24-74.[Medline]
19. Bull, N. L. (1992) Dietary habits, food consumption, and nutrient intake during adolescence. J. Adolesc. Health 13:384-388.[Medline]
20.
Cavadini, C., Siega-Riz, A.M. & Popkin, B.M. (2000) U.S. adolescent’s food intake trends from 1965–1996. Arch. Dis. Child. 83:18-24.
21. Moore, D. C. (1993) Body image and eating behavior in adolescents. J. Am. Coll. Nutr. 12:505-510.[Abstract]
22. Siega-Riz, A.M., Carson, T. & Popkin, B.M. (1997) Three squares or mostly snacks—what do teens really eat?. a sociodemographic study of meal patterns. J. Adolesc. Health 22:29-36.
23. World Health Organization (1998) Obesity: Preventing and Managing the Global Epidemic—Report of a WHO Consultation on Obesity 1998 WHO Geneva, Switzerland. .
24. Wang, Y., Monteiro, C. & Popkin, B. M. (2002) Obesity trends in older children and adolescents. Am. J. Clin. Nutr. in press.
25. U.S. Department of Health and Human Services (1992) The Healthy People 2000 National Health Promotion and Disease Prevention Objective 1992 Boston, MA .
26. United Nations (1999) Demographic Yearbook 1997 49th ed. 1999 United Nations New York, NY. .
27. China National Bureau of Statistics (1999) China Statistical Yearbook 1999 1999 Publishing House Beijing, China .
28. Ge, K. (1999) The Dietary and Nutritional Status of Chinese Population: 1992 National Nutrition Survey Children and Adolescents 2 People’s Medical Publishing House Beijing, China .
29. Popkin, B. M., Keyou, G., Zhai, F., Guo, X., Ma, H. & Zohoori, N. (1993) The nutrition transition in China: a cross-sectional analysis. Eur. J. Clin. Nutr. 47:333-346.[Medline]
30. Wang, Y., Popkin, B. & Zhai, F. (1998) The nutritional status and dietary pattern of Chinese adolescents, 1991 and 1993. Eur. J. Clin. Nutr. 52:908-916.[Medline]
31. Zhai, F., Guo, X., Popkin, B., Ma, L., Qang, Q., Yu, W., Jin, S. & Ge, K. (1996) The evaluation of the 24-hour individual recall method in China. Food Nutr. Bull 17:154-161.
32. Popkin, B. M., Lu, B. & Zhai, F. (2002) Understanding the nutrition transition: measuring rapid dietary: changes in transitional countries. Public Health Nutr. in press.
33. Institute of Nutrition and Food Hygiene, Chinese Academy of Preventive Medicine (1991) Food Composition Tables 1991 People’s Health Press Beijing, China. .
34. Kemper, H. C., Post, G. B. & Twisk, J. W. (1997) Rate of maturation during the teenage years: nutrient intake and physical activity between ages 12 and 22. Int. J. Sport Nutr. 7:229-240.[Medline]
35.
Willett, W. C., Howe, G. R. & Kushi, L. H. (1997) Adjustment for total energy intake in epidemiologic studies. Am. J. Clin. Nutr. 65(suppl. 4):S1220-S1228.
36. Willett, W. (1998) Nutritional Epidemiology 1998 Oxford University Press New York, NY. .
37. Goldberg, G. R., Black, A. E., Jebb, S. A., Cole, T. J., Murgatroyd, P. R., Coward, W. A. & Prentice, A. M. (1991) Critical evaluation of energy intake data using fundamental principles of energy physiology: 1. Derivation of cut-off limits to identify under-reporting. Eur. J. Clin. Nutr. 45:569-581.
38. Black, A. E., Goldberg, G. R., Jebb, S. A., Livingstone, M.B.E., Cole, T. J. & Prentice, A. M. (1991) Critical evaluation of energy intake data using fundamental principles of energy physiology: 2. Evaluating the results of published surveys. Eur. J. Clin. Nutr. 45:583-599.
39. FAO/WHO/UNU (1985) Energy and Protein Requirements. WHO Technical Report Series 724 1985 World Health Organization Geneva, Switzerland .
40. Landis, J. R. & Koch, G. G. (1977) The measurement of observer agreement for categorical data. Biometrics 33:159-174.[Medline]
41. Pagano, M. & Gauvreau, K. (1993) Principles of Biostatistics 1993:363-378 Duxbury Press Belmont, CA. .
42.
Twisk, J. W., Kemper, H. C. & Mellenbergh, G. J. (1994) Mathematical and analytical aspects of tracking. Epidemiol. Rev. 16:165-183.
43. Stokes, M. E., Davis, C. S. & Kock, G. G. (1995) Categorical Data Analysis Using the SAS System 1995 SAS Institute Inc Cary, NC. .
44. Stata Corporation (1999) Stata Statistical Software, Release 6.0 1999 Stata Corporation College Station, TX .
45. Heckman, J. (1979) Sample selection bias as a specification error. Econometrica 47:153-162.
46.
Lindsay, R., Hanson, R. L. & Knowler, W. (2001) Problems with tracking of body mass index during childhood and adolescence. Am. J. Clin. Nutr. 74:149.
47. Wang, Y., Popkin, B. M. & Wang, X. (2001) Request to reply: problems with tracking of body mass index during childhood and adolescence. Am. J. Clin. Nutr. 74:149-150.
48. Wang, Y. & Wang, X. (2001) Are individuals of extreme values expected to be more likely to maintain their ranking over time than others?. Am. J. Epidemiol. 153:S177(abs.).
49. Jing, J. (2000) Feeding China’s Little Emperors: Food, Children, and Social Change 2000 Stanford University Press Stanford, CA. .
50. Smith, C. J. (1993) (Over) eating success: the health consequences of the restoration of capitalism in rural China. Soc. Sci. Med. 37:761-770.
51. Deheeger, M., Akrout, M., Bellisle, F., Rossignol, C. & Rolland-Cachera, M. F. (1996) Individual patterns of food intake development in children: a 10 months to 8 years of age follow-up study of nutrition and growth. Physiol. Behav. 59:403-407.[Medline]
52. Paeratakul, S., Popkin, B. M., Kohlmeier, L., Hertz-Picciotto, I., Guo, X. & Edwards, L. (1998) Measurement error in dietary data: implications for the epidemiologic study of the diet-disease relationship. Eur. J. Clin. Nutr. 52:722-727.[Medline]
53. Richards, M. K. (2000) Environmental Predictors of Child Obesity: A Longitudinal Approach 2000 Doctoral thesis University of North Carolina at Chapel Hill, Chapel Hill, NC. .
54. Heckman, J. (1976) The common structure of statistical models of truncation, sample selection, and limited dependent variables and a simple estimator for such models. Ann. Econ. Soc. Meas. 5:475-492.
This article has been cited by other articles:
|
|
 |
|
  L. L. Moore, M. R. Singer, M. M. Qureshi, and M. L. Bradlee Dairy Intake and Anthropometric Measures of Body Fat among Children and Adolescents in NHANES J. Am. Coll. Nutr., December 1, 2008; 27(6): 702 - 710. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. B. Fein, J. Labiner-Wolfe, K. S. Scanlon, and L. M. Grummer-Strawn Selected Complementary Feeding Practices and Their Association With Maternal Education Pediatrics, October 1, 2008; 122(Supplement_2): S91 - S97. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Lim, W. Sohn, B. A. Burt, A. M. Sandretto, J. L. Kolker, T. A. Marshall, and A. I. Ismail Cariogenicity of Soft Drinks, Milk and Fruit Juice in Low-Income African-American Children: A Longitudinal Study J Am Dent Assoc, July 1, 2008; 139(7): 959 - 967. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Z. Zimmer, T. Kaneda, and L. Spess An Examination of Urban Versus Rural Mortality in China Using Community and Individual Data J. Gerontol. B. Psychol. Sci. Soc. Sci., September 1, 2007; 62(5): S349 - S357. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. R. Dore, L. S. Adair, and B. M. Popkin Low Income Russian Families Adopt Effective Behavioral Strategies to Maintain Dietary Stability in Times of Economic Crisis J. Nutr., November 1, 2003; 133(11): 3469 - 3475. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
