Patterns of Physical Activity, Sedentary Behavior and Diet in US Adolescents

Lack of physical activity (PA), excess sedentary behavior, and unhealthful diets contribute to adolescent obesity and early development of cardiovascular risk factors.^1–3 In addition, each of these behaviors has other health consequences. Child and adolescent physical activity has been related to adiposity, elements of Metabolic Syndrome, physical, mental and social health, and academic and cognitive performance.^1,2,4–6 Adolescent screen -based sedentary behavior (SB), particularly television viewing (TV), has been related to subsequent adiposity, risk for cardiovascular disease, metabolic syndrome, health complaints, physical aggression, and tobacco and alcohol use, and negatively related to quality of life, psychological well-being, and quality of family relationships.^1,5–9 A diet high in energy dense foods such as sugar -sweetened beverages, sweets, chips, and French fries has been positively associated with adiposity, waist circumference, total cholesterol, low density lipoproteins, diastolic blood pressure, and plasma triglycerides and glucose and negatively related to psychological functioning, while diets high in fiber from consumption of fruits and vegetables have been negatively related to adiposity and indicators of the Metabolic Syndrome.^10,11

Pediatric recommendations for reducing the risk of cardiovascular disease suggest that the optimum approach is a combination of increased PA, reduced SB, and improved diet quality.¹ Establishment of healthful patterns during childhood and adolescence is important because physical activity and dietary habits track during adolescence and from adolescence to adulthood.⁴

Although adolescent PA, SB, and diet may not be strongly related to each other, there is evidence that these behaviors are interrelated within individuals;⁹ that is, that there might be a limited number of individual patterns of adolescent PA, SB, and diet. In small regional samples, obesogenic behavior patterns combining PA, SB, diet, and weight consciousness have been identified in fourth-grade children¹² and combining PA and SB in adolescents.^13,14 However, to our knowledge, no studies have examined obesogenic behavior patterns in a nationally representative sample of US adolescents or examined how these patterns relate to physical and psychological health. Identifying these patterns could foster understanding of their development and maintenance, their broader health consequences, and potential avenues for intervention. The purpose of this study was to identify patterns in adolescents’ obesogenic behaviors, demographic differences across these patterns, and the relationship of these patterns to indicators of physical and psychological health.

Methods

Sample and Procedure

A nationally representative sample of US students in grades 6 to 10 was recruited from 230 schools in 39 states with census regions and grades as strata and classrooms as the primary sampling units. African-American and Hispanic students were oversampled to obtain better estimates for those groups. Participants completed the 2005/2006 Health Behavior in School-aged Children (HBSC)¹⁵ survey anonymously with a response rate of 87%. Youth assent and, depending on requirements of participating school districts, active or passive parental consent were obtained. The study protocol was reviewed and approved by the Institutional Review Board of the Eunice Kennedy Shriver National Institute of Child Health and Human Development.

Measures

Socio-demographic variables

Socio-demographic variables included gender, age, race/ethnicity (Caucasian, African-American, Hispanic, and others) and family affluence. Family socioeconomic status was estimated with the Family Affluence Scale (FAS): the adolescent having his/her own bedroom, frequency of family traveling vacations, and number of family computers and motor vehicles.¹⁶ The FAS produced a continuous scale from 0 to 9.

Physical activity (PA)

A definition of physical activity was provided, “any activity that usually increases your heart rate and makes you get out of breath some of the time,” and examples of activities preceded the question “How often over the past 7 days have you been physically active for a total of at least 60 minutes per day”. The measure has reasonable validity (r = .37 with 5-day accelerometer data) and reasonable accuracy for classifying adolescents meeting recommendations for 60 min/day of PA for at least 5 days/week (sensitivity = 83%).^17,18 PA was dichotomized: < 5 days/week; and ≥ 5 days/week, the threshold for decreasing odds of obesity in adolescents.¹⁹

Screen-based Sedentary Behaviors (SB)

Students were asked how many hours a day, in their free time, they usually: 1) watched television (including videos and DVDs); 2) played games on a computer or games console (Playstation, Xbox, GameCube, etc); and 3) used a computer for chatting on-line, internet, emailing, homework, etc. Separate questions were asked for weekday and weekend. Response categories ranged from “none at all” to “about 7 or more hours a day”. These HBSC items have acceptable test-retest reliability (ICCs ranging from .76 to .81) and validity (rs ranging from .36 to .54).²⁰ For each SB, a weighted average of weekdays and weekend days was dichotomized: > 2hr/day; and ≤ 2 hr/day, consistent with the recommendations of the American Academy of Pediatrics.²¹

Dietary intake

As part of a validated brief food frequency questionnaire,²² participants indicated how many times a week they usually ate fruits, vegetables, sweets (chocolates and candy), sweetened soft drinks, chips, and French fries. The response options were “never,” “less than once a week,” “once a week,” “2 to 4 days a week,” “5 to 6 days a week,” “once a day, every day” and “every day, more than once.” For each food category, a dichotomous variable was created: ≥ 1/day; and < 1/day.

Weight status

Adolescents reported their heights and weights. Using body mass index percentiles derived from the gender- and age-specific growth charts of the Centers for Disease Control and Prevention,²³ a four-category weight status variable was created: underweight or at risk for underweight (≤ 15^th percentile); normal weight (16^th to 84^th percentile); overweight or at risk for obesity (85^th to 94^th percentile); and obese (≥ 95^th percentile).

Weight loss behavior

Adolescents were asked whether they were currently on a diet or doing something else to lose weight. Four categories of weight loss behavior were generated from the four choices: “no, my weight is fine;” “no, but I should lose some weight;” “no, because I need to put on weight;” and, “yes.”

Body dissatisfaction

Five items from the body image subscale of the Body Investment Scale²⁴ (“I am frustrated with my physical appearance,” “I am satisfied with my appearance ,”“I hate my body,” “I feel comfortable with my body” and “I feel anger toward my body”) assessed body dissatisfaction on a 5-point scale (“strongly agree” to “do not agree at all”). A higher mean score across items indicated greater body dissatisfaction.

Physical Symptoms

Adolescents indicated how often in the last 6 months they experienced: (1) headache; (2) stomach-ache; (3) backache; and (4) feeling dizzy. Responses were coded from one to five: “rarely or never,” “about every month,” “more than once a week,” “about every week” and “about every day”(Cronbach’s alpha = .70). A higher mean score indicated more physical symptoms.²⁵

Depression

Participants indicated how often in the past 30 days they: (1) were very sad; (2) were grouchy or irritable, or in a bad mood; (3) felt hopeless about the future; (4) felt like not eating or eating more than usual; (5) slept a lot more or a lot less than usual; and (6) had difficulty concentrating on their school work.²⁶ Responses were coded one to five: “never,” “seldom,”“ sometimes,” “often” and “ always” (Cronbach’s alpha =.80). A higher mean score indicated greater depressive tendencies.

Life satisfaction

A ladder with rungs numbered vertically zero to ten was pictured and participants asked to choose a number with the top “10” for the best possible life and the bottom “0” for the worse possible life.²⁷

Overall health

A single item measured overall health, with response options coded from one to four: “poor,” “fair,” “good,” and “excellent.”²⁸

Data Analysis

Latent Class Analysis (LCA) identifies latent classes based on individuals’ patterns of responses to multiple categorical variables. The primary advantage of LCA over cluster analysis is that LCA is a model-based method and as such should generalize to other independent populations, thus making evaluation of classes in a cross-sectional sample appropriate and in a nationally representative sample even more robust.²⁹ Our application of LCA identifies the probability of individuals within a class reporting each health behavior, the prevalence of each pattern of health behaviors within a nationally representative sample, the characteristics that differentiate members across classes, and the health indicators associated with these patterns.

A series of LCA models were evaluated using Mplus version 5.1³⁰ with stratification, cluster and sampling weights to accommodate the complex sampling structure of the US HBSC data. To choose the optimal number of classes that best fit the data, LCA models with increasing numbers of classes were evaluated.

Model selection was based on conceptual implications, model parsimony and fit statistics, including loglikelihood, Akaike information criterion (AIC), Bayesian information criterion (BIC), sample-size adjusted Bayesian information criterion (ABIC), and average latent class probabilities for most likely membership (ACPs). In the current study, latent classes were defined by probabilities that individuals in each class met criteria for physical activity, television viewing, computer use, video gaming, and consumption of fruit, vegetables, sweets, sweetened soft drinks, chips and fries. After the optimal number of classes was selected, demographic differences across latent classes were examined by adding gender, grade, race/ethnicity, and FAS as covariates. Finally, the extracted latent classes were compared on physical and psychological variables, e.g., weight status, weight loss behavior, body dissatisfaction and depressive symptoms. Obesogenic behaviors and their correlates frequently differ by gender; therefore, analyses were also conducted separately by gender.

Results

Sample Characteristic

Among the 9,227 adolescents who completed the survey, 21 were excluded due to missing information on all PA, SB, and diet items, resulting in an analytic sample of 9,206 (mean age: 13.9 years), 48.4% males, 48.9% Caucasian Americans, 17.9% African-Americans, and 19.1% Hispanic Americans. Most (65.5%) children were in the normal weight classification with the balance being overweight (17.1%), obese (13.6%), or underweight (3.7%).

Descriptive Statistics: Prevalence Rates

Descriptive statistics for PA, SB and diet are reported in Table 1. Majorities of participants reported ≥ 5 days/week of PA but also > 2 hrs/day of TV. Boys reported significantly more PA, more video games, and less computer use than girls. Girls reported higher daily intake of sweets, but were less likely to consume sweetened drinks.

Table 1.

Descriptive Statistics for Latent Class Indicators By Gender

	Total N = 9,206	Male N = 4,455	Female N = 4,750	p-value (Gender)
Categorical Variables	% (SE)	% (SE)	% (SE)
PA (>=5 days/week)	51.7 (1.40)	60.3 (1.39)	43.4 (1.60)	<.0001
TV (>2 hrs/day)	54.8 (1.50)	53.5 (1.69)	56.1 (1.69)	.093
Video Game (>2 hrs/day)	21.3 (0.88)	29.9 (1.40)	13.1 (0.69)	<.0001
Computer (>2 hrs/day)	24.3 (1.15)	20.4 (1.21)	28.1 (1.46)	<.0001
Daily Food intake (>= once/day)
Fruits	38.7 (1.12)	37.3 (1.26)	40.1 (1.40)	.053
Vegetables	34.7 (0.96)	33.4 (1.46)	36.0 (1.04)	.105
Sweets	32.7 (1.21)	29.1 (1.20)	36.2 (1.52)	<.0001
Sweetened Soft Drink	34.9 (1.34)	36.3 (1.30)	33.5 (1.60)	.022
Chips	27.1 (1.53)	26.9 (1.40)	27.3 (1.86)	.773
French Fries	13.7 (0.79)	14.3 (0.95)	13.1 (0.90)	.245

Note. Percentages were weighted percentages by adjusting for the survey design including stratification, clustering and weighting. p-values are for test of gender differences.

Latent Classes

The model fit statistics for 1- to 5-class LCA models are provided in Table 2. Results showed that the loglikelihood values, as well as three information criteria indices (AIC, BIC and ABIC), decreased substantially for models with 1 to 3 classes and moderately for additional classes. The ACP, an indicator of the degree to which classes could be reliably distinguished from one another, was high for models with 2, 3, and 4 classes (>.80).³¹ With further consideration of conceptual implications and model parsimony, the 3-class model was selected as the best model and used in subsequent analyses for the full sample and for each gender.

Table 2.

Model Fit Statistics of the 1- to 5-class LCA models

	Number of Classes
All (N = 9,206)	1	2	3	4	5
Loglikelihood	−54721.1	−52011.1	−50979.0	−50766.8	−50591.4
Information Criteria
N of Free Parameters	10	21	32	43	54
Akaike (AIC) a	109462.2	104064.1	102022.0	101619.6	101290.7
Bayesian (BIC) b	109533.4	104213.8	102250.1	101926.1	101675.6
Sample-Size ABIC	109501.7	104147.1	102148.4	101789.5	101504.0
ACPs c	--	.893–.940	.827–.879	.801–.843	.652–.807
Male (N = 4,455)	1	2	3	4	5
Loglikelihood	−26403.9	−25176.1	−24676.3	−24553.8	−24449.9
Information Criteria
N of Free Parameters	10	21	32	43	54
Akaike (AIC) a	52827.8	50394.2	49416.5	49193.7	49007.7
Bayesian (BIC) b	52891.8	50528.6	49621.4	49468.9	49353.4
Sample-Size ABIC	52860.1	50461.9	49519.7	49332.3	49181.8
ACPs c	--	.893–.938	.835–.879	.811–.855	.653–.847
Female (N = 4,740)	1	2	3	4	5
Loglikelihood	−27839.2	−26326.6	−25776.4	−25645.4	−25543.9
Information Criteria
N of Free Parameters	10	21	32	43	54
Akaike (AIC) a	55698.4	52695.2	51616.7	51376.8	51195.8
Bayesian (BIC) b	55763.1	52830.9	51823.6	51654.7	51544.8
Sample-Size ABIC	55731.3	52764.2	51721.9	51518.1	51373.2
ACPs c	--	.893–.942	.843–.888	.877	.763–.861

Note.

^aAIC = Akaike’s Information Criteria;

^bBIC = Bayesian Information Criteria; Sample-Size ABIC = Sample-size adjusted BIC; ACPs = Average Latent Class Probabilities for Most Likely Latent Class Membership.

The three latent classes for the full sample are presented in Figure 1. The patterns in item probabilities of each PA, SB and diet item were quite similar for models of each gender (not shown).

Figure 1.

Item Probability: Involvement in Each Health Behavior for Each Latent Class

Note. The cut-off for each dietary behavior was >=once/day.

The first latent class represents approximately a fourth of respondents with a Healthful pattern relative to the other two classes. Class 1 had the highest proportion of members meeting the criterion for PA (63.9%), lowest proportions reporting > 2hr/day of TV (37.7%), video games (12.0%), or computer use (18.2%), the highest proportions consuming at least one serving of fruits (83.0%) or vegetables (79.3%) per day, and among the lowest proportions indicating intake of energy dense foods including sweets (22.0%), sweetened soft drinks (10.7%), chips (8.5%), and fries (2.8%).

Approximately one fourth of the participants were in Class 2, which was labeled Unhealthful. Across the three classes, these adolescents had the highest percentage exceeding 2hr/day TV (74.5%), video games (36.4%) and computer use (32.3%), and the highest proportions with daily intake of sweets (73.3%), sweetened soft drinks (77.8%), chips (80.9%), and fries (44.4%). However, relative to the other two classes, members of Class 2 had a moderate probability of meeting the criterion for PA (50.1%) or consuming fruits (45.5%) or vegetables (35.5%) daily.

Typical adolescents (47.2%) were in Class 3 and had a pattern of the lowest percentage meeting the criteria for PA (45.6%) or consuming fruits (10.1%) or vegetables (9.4%) and infrequently consumed sweets (16.3%), sweetened soft drinks (24.5%), chips (7.5%), or fries (2.7%). Members of this class were moderate relative to the other two classes with respect to TV (53.5%), video games (18.2%), and computer use (23.3%).

Socio-demographic differences in latent class membership were examined with Class 1: Healthful as the reference group (see Table 3). Due to missing information on socio-demographic variables, 223 (2.4%) subjects were excluded from this analysis. Compared to the Healthful pattern group, adolescents in the Unhealthful group were less likely to be Caucasian and more likely to have lower FAS scores. Adolescents in the Typical group were comparatively more likely to be males, older, and either African-American or Hispanic, and to have lower FAS scores. The socio-demographic differences were consistent across genders, with the exception of no significant racial/ethnic difference between the Typical and Healthful groups in the model with males only.

Table 3.

Results of LCA with Covariates: Demographic Differences

	All (N = 8,982)		Male (N = 4,323)		Female (N = 4,659)
	OR	[95% C.I.]	OR	[95% C.I.]	OR	[95% C.I.]
Class 2 (Unhealthful) a
Female b	0.87	[0.73–1.04]	--	--	--	--
Age	1.06	[0.96–1.17]	1.06	[0.95–1.19]	1.04	[0.91–1.19]
Race/Ethnicity b
African-American	10.5	[6.32–17.4]	6.84	[4.02–11.7]	15.7	[8.29–29.8]
Hispanic	2.96	[2.15–4.09]	2.28	[1.41–3.68]	3.71	[2.47–5.56]
Other	1.74	[1.23–2.46]	1.64	[1.09–2.47]	1.87	[1.19–2.92]
FASc	0.83	[0.77–0.89]	0.87	[0.78–0.97]	0.79	[0.72–0.87]
Class 3 (Typical) a
Female b	0.79	[0.63–0.99]	--	--	--	--
Age	1.18	[1.09–1.26]	1.20	[1.10–1.30]	1.14	[1.03–1.27]
Race/Ethnicity b
African-American	2.36	[1.56–3.57]	1.49	[0.92–2.39]	3.60	[1.88–6.89]
Hispanic	1.64	[1.22–2.23]	1.42	[0.99–2.03]	1.86	[1.15–2.99]
Other	1.02	[0.65–1.58]	0.82	[0.52–1.30]	1.29	[0.79–2.11]
FASc	0.83	[0.78–0.88]	0.90	[0.83–0.97]	0.76	[0.70–0.84]

Note. Significant odds ratios are in bold print.

^aThe latent class of Healthful was set as the reference group for the multinomial logistic regression model;

^bFor gender and race/ethnicity, the reference groups were male and Caucasian adolescents.

^cHigh Family Affluence Scale scores was the referent for FAS.

Comparing Classes on Health Indicators

Compared to the other two latent classes, adolescents demonstrating the Healthful pattern reported that they had fewer symptoms of depression, better life satisfaction, and better overall health (Table 4). Compared to other girls, girls in this class were more likely to be normal weight and less likely to be obese; they were less likely to indicate they should gain weight. However, boys in this class were more likely than other boys to indicate they were trying to lose weight.

Table 4.

Comparing Weight Status, Losing Weight Behaviors, Depression and Frustration of Physical Appearance across Latent Classes

Categorical Variables	All				Male				Female
	Overall	Class 1	Class 2	Class 3	Overall	Class 1	Class 2	Class 3	Overall	Class 1	Class 2	Class 3
	%	%	%	%	%	%	%	%	%	%	%	%
Weight Status
Underweight	3.7	3.9ab	4.9a	3.0b	4.3	4.3ab	5.8a	3.6b	3.2	3.6ab	4.0a	2.5b
Normal	65.5	67.9a	65.2ab	64.3b	61.9	62.9	62.2	61.2	69.0	72.2a	68.3b	67.5b
Overweight	17.1	15.8a	16.3ab	18.4b	18.3	17.5	17.3	19.3	16.0	14.3a	15.2ab	17.4b
Obese	13.6	12.4	13.6	14.4	15.4	15.3	14.7	15.9	11.9	9.9a	12.5b	12.7b
Losing Weight
No, my weight is fine	47.3	48.3a	50.6a	44.9b	54.2	53.0a	59.3b	52.2a	40.7	44.5a	42.4a	37.5b
No, but I should lose some weight	22.6	19.3a	21.6a	25.1b	16.7	13.6a	15.3a	19.1b	28.3	24.2a	27.8b	31.0b
No, because I need to put on weight	7.8	6.5a	10.1b	7.3a	10.8	11.0	11.8	10.3	4.9	2.5a	8.3c	4.2b
Yes	22.2	25.9a	17.7c	22.8b	18.2	22.4a	13.6c	18.4b	26.1	28.9a	21.5b	27.3a
Continuous Variables	Mean	Mean	Mean	Mean	Mean	Mean	Mean	Mean	Mean	Mean	Mean	Mean
Body Dissatisfaction (n=7,844)	2.20	2.14a	2.14a	2.27b	1.94	1.88a	1.87a	2.00b	2.45	2.37a	2.39a	2.53b
Physical Symptoms (n=9,101)	1.96	1.89a	2.11b	1.91a	1.78	1.71a	1.93b	1.75a	2.13	2.04a	2.29b	2.08a
Depression (n=9,058)	2.41	2.27a	2.56c	2.40b	2.20	2.11a	2.35c	2.17b	2.61	2.40a	2.76c	2.64b
Life Satisfaction (n=9,064)	7.37	7.67a	7.27b	7.25b	7.46	7.70a	7.38b	7.38b	7.28	7.67a	7.16b	7.12b
Overall Health (n=9,092)	2.95	3.09a	2.92b	2.88b	3.06	3.17a	3.07b	2.99c	2.84	3.03a	2.78b	2.78b

Note. Class 1: Healthful pattern; Class 2: Unhealthful pattern; Class 3: Typical pattern.

^a,b,cValues with same superscript were not significantly different from each other.

“Body dissatisfaction” was asked for all adolescents in grades 7 to 10 and half of adolescents in grade 6.

Compared to the other two classes, adolescents in the Unhealthful class were less likely to indicate they were currently trying to lose weight, with girls in this class more likely than other girls to report they were not trying to lose weight because they thought they needed to gain weight and boys more likely to indicating their weight was fine. Adolescents in this class also reported more frequent physical and depressive symptoms.

The Typical group reported the highest levels of body dissatisfaction, with Typical girls less likely to indicate their weight was fine and boys more likely than other boys to report that they should lose weight but were not currently trying. However, compared to the Unhealthful class, Typical adolescents were less likely to be underweight or report symptoms of depression.

Discussion

To our knowledge, this is the first application of latent class analysis to identify patterns in obesogenic behaviors in a nationally representative sample of US adolescents. The unique combination of obesogenic behaviors in each class has clear implications for cardiovascular health, each class is related to other indicators of physical and mental health, and each class suggests different approaches for public health efforts. This is a national sample and at a public health level, differences of 1.7% to 4.6% in prevalence of overweight/obesity across classes are quite significant.

Class 1: Healthful identified a pattern in approximately one fourth of the students who were more likely than the other classes to meet recommendations for PA and SB, had the highest frequencies of fruit and vegetables consumption, and low levels of consumption of sweets, sweetened drinks, chips, and fries. In both girls and boys, the Healthful pattern was reflected in other measures of health such as higher life satisfaction, lower depression, and higher overall health status. These findings are consistent with other studies showing a positive association of higher levels of PA, lower levels of SB, and a healthful diet with positive physical and psychological indicators.^4–12,32 Even though most of these adolescents met recommendations for daily PA, they did not meet nutritional guidelines with respect to fruit, vegetables and energy-dense foods.^3,19,21 One focus of future research could be to identify the social and environmental conditions that promote development and maintenance of this pattern. Interventions with this group should particularly promote improvements in diet and preventing the usual decline in PA during adolescence.

Adolescents in Class 2, the Unhealthful group, had essentially the opposite pattern of obesogenic behaviors as the Healthful group, with low PA, high SB, low fruit and vegetable intake and the highest intake of sweets, sweetened drinks, chips and fries. It is possible that some Unhealthful youth have adopted this pattern because they are actively trying to gain weight. If this is the case, this is not a healthful approach to gaining weight: consuming unhealthful quantities of energy dense foods, while not taking advantages of the health benefits associated with a more active lifestyle.^2–6,19,32 Future studies of this group could determine the causal direction of the links between this pattern of obesogenic behaviors and potentially maladaptive associations with symptoms of depression and somatic problems, and potential determinants of these relations. One potential hypothesis is that the high rate of physical symptoms contributes to a pattern of high sedentary behavior and symptoms of depression and low life satisfaction and perceived health. However, high rates of adolescent sedentary behaviors have been associated with a similar pattern of physical and psychological health.^5,6,8 Interventions at the school and clinic level could focus on decreasing the proportion of free time spent in sedentary activities and incorporating more moderate intensity physical activity into their lifestyle, including active transport to school. If weight gain is one of the motivations for some in this class, school and practice-based interventions might focus on the effect of physical activity on increased lean muscle mass and the disadvantages of energy dense foods for health in general.^3,9–11,32

Class 3 represents the largest group of US adolescents, provides evidence that the dominant lifestyle varies from recommendations regarding PA, SB and diet, and is associated with poorer physical and mental health. The pattern of weight status, weight loss behavior, and body dissatisfaction suggests that the Typical group may be older adolescents who are overweight/obese, dissatisfied with their body, and trying to lose weight by focusing exclusively on reducing dietary intake. That this is the most prevalent group and has the highest prevalence of overweight/obesity is a concern. Identification of successful interventions with this group might be a priority because of the public health implications. If these adolescents are motivated by a desire to manage their weight, they might benefit from interventions that focus on the weight-related benefits of daily physical activity and a diet with proportionately more fruits and vegetables. They might need to learn the dual role of caloric intake and physical activity in weight management.^2–4,8–11

Health disparities were evident across the three latent classes and the potential contribution of obesogenic environments should not be ignored. Adolescents with lower family affluence and minority adolescents were more likely to exhibit less healthful patterns of diet and physical activity. Potential socio-environmental contributors to these differences include neighborhood differences in access to recreational facilities and produce markets,^33,34 school differences in food services and policies supporting physical activity,³⁵ and family differences in availability of healthful versus unhealthful foods and support for physical activity.^36,37

It is difficult to compare these results to other studies using cluster analysis or LCA because of differences in the health behaviors included in analyses and the age of the samples.^13,14,38 In addition, the current study included a nationally representative sample of US students and therefore the classes should be more robust across samples and time compared to local samples. However, the classes identified in this national sample of adolescents are similar to three of the five classes identified by Huh et al.¹² in a regional sample of younger children: high PA and healthy diet; high SB and high fat/sugar diet but not overweight conscious; dieting without exercise but overweight conscious. Further work is necessary to confirm that these patterns are stable over time in US adolescents, that correlates of these patterns are consistent in other cultures, and whether membership in these classes predict subsequent health outcomes.

Because these data are cross-sectional, they cannot tell us about the development and determinants of these patterns. However, research identifying the age of onset of these patterns could facilitate early intervention. And if the relation of obesogenic behaviors with psychological health is causal, educating adolescents about these links might motivate development of healthier patterns. Further study of the characteristics of adolescents exhibiting each pattern, for example, the motivations for adolescents exhibiting each of these patterns and the potential influence of parents and peers on these patterns, could be informative. LCA suggests that individual patterns of health behaviors exist even though the behaviors may not be correlated. Thus, interventions might target determinants of these patterns even though the determinants of individual behaviors within a pattern might be unrelated.

The primary limitation of this study is the cross-sectional design which prohibits inference about causal relations between the obesogenic behaviors and physical and psychological correlates. Another limitation is the reliance on self-report for height and weight; however, studies have shown good correlations between self-reported and measured body mass index making it suitable for examining associations in population studies.^39,40 Strengths of the study also need to be considered including the application of LCA to a nationally representative sample of 6^th- through 10^th-grade US adolescents and the inclusion of multiple obesogenic behaviors in identifying the latent classes.

Conclusions

Three latent classes were identified in a nationally representative sample, with approximately three out of every four adolescents reporting two unhealthful dietary and activity patterns. One class had the optimum combination of these behaviors and good indicators of mental health. Two classes, both with poorer mental health, varied greatly in consumption of snacks but had lower rates of PA, SB and consumption of fruits and vegetables. The different patterns in these classes suggest that relative rates of PA, SB, and consumption of fruit and vegetables may drive differences in weight status and mental health more than consumption of energy dense snacks and drinks. These findings also suggest that interventions may need to be tailored to different audiences or require different social marketing strategies, combined with formative and efficacy research.