Longitudinal associations between key dietary behaviors and weight gain over time: Transitions through the adolescent years

Abstract

Previous studies have yielded inconsistent results in documenting the association between key dietary factors and adolescent weight change over time. The purpose of this study was to examine the extent to which changes in adolescent sugar-sweetened beverage, diet soda, breakfast and fast food consumption were associated with changes in BMI and percent body fat (PBF), both cross-sectionally and longitudinally. Our sample included 693 Minnesota adolescents followed over two years. Adjusting for physical activity, puberty, race, socio-economic status, age, and total energy intake, cross-sectional findings indicated that for both males and females, diet soda consumption was significantly and positively associated with BMI and PBF, and breakfast intake was significantly and negatively associated with BMI and PBF among girls. In longitudinal analyses, however, there were no significant associations after adjusting for the number of tests performed. This study adds to previous research through its methodological strengths, including adjustment for physical activity and energy intake assessed using state-of-the-art methods (i.e., accelerometers and 24-hour dietary recalls), as well as its evaluation of both BMI and PBF. Additional research is needed to better understand the complex constellation of factors that contribute to adolescent weight gain over time.

Introduction

Obesity is a major public health concern^{1, 2}. The transition through adolescence and into early adulthood is recognized as an influential age for excess weight gain, marked by poor dietary patterns and physical inactivity³. Adolescence is generally characterized as a period of growing independence, where individuals are increasingly beginning to make their own decisions about their day-to-day life, including what to eat. Recently, several national health organizations have identified a number of key dietary behaviors associated with excess weight gain among youth and adolescents, including frequent intake of sweetened beverages and fast food, and infrequent breakfast consumption⁴. A wide array of crosssectional studies have been published in this area, many of which support these dietary behaviors as important correlates of weight-related health⁵. However, longitudinal studies assessing the associations between these factors and weight gain have produced inconsistent results, and many have had significant methodological limitations⁶.

For example, a number of large, longitudinal adolescent cohort studies have examined the association between sugar-sweetened beverage (SSB) intake and body mass index (BMI) over time^7-14. Some of these studies have found no longitudinal association between SSB consumption and weight status^{8, 10, 14} while others have found a positive association^{7, 9, 11-13}. Importantly, not all studies have adjusted for potential confounders, such as physical activity ^{8, 9, 13}. In fact, none of these studies have adjusted for objectively measured physical activity; most relied on self-reported measures, which can be subject to high levels of error and bias¹⁵.

In addition, a number of these longitudinal adolescent cohort studies have also assessed the relationship between diet soda consumption and weight status; paradoxically, most findings suggest a positive association, indicating that more diet soda consumption is associated with greater BMI gain over time^{7, 8, 14}. However, in a number of these studies, these positive associations were not straightforward (i.e., significant associations were only observed for boys, but not girls, or associations were attenuated by adjusting for various covariates). In addition, Striegel-Moore and colleagues¹³ found that diet soda intake was not associated with weight status over time among 2,371 girls enrolled in the National Heart, Lung and Blood Institute Growth and Health Study. In contrast, a study by Ludwig et al11 among 548 Massachusetts adolescents found that change in diet soda intake over 19 months was inversely associated with obesity incidence (odds ratio: 0.44). Although most researchers have recognized the need to control for total energy intake in these types of analyses, a majority of previous studies in this area have used food frequency questionnaires to assess dietary intake^{7, 11, 14, 16}; food frequency questionnaires have been shown to provide substantially less accurate estimates of total energy intake than other methods of dietary assessment¹⁷.

Additional analyses from similar adolescent cohorts have examined other key dietary behaviors, such as breakfast consumption, in relation to weight status. Five studies have reported no significant association between the frequency of breakfast consumption and weight status in fully adjusted analyses^18-22. Three studies found an inverse association (i.e., less frequency breakfast intake being associated with greater weight gain)^{16, 23, 24}, though one study observed this association only among overweight participants²³. In addition, one longitudinal study of 159 first-year college students (ages 18-19 years) found a positive association between breakfast consumption and weight gain²⁵.

Finally, few adolescent cohort studies have examined longitudinal associations between fast food intake and weight change. Two studies reported positive associations, where greater fast food consumption was associated with greater weight gain^{26, 27}. Two other studies found a combination of positive and null effects among varying sub-groups of their samples and across different types of analyses^{16, 24}. None of these studies controlled for physical activity using objective measures, and several used measures of physical activity that have not been validated^{24, 26}.

Overall, this growing body of research presents inconsistencies in the relationships between key dietary factors and adolescent weight change over time, and additional research in this area is needed. The purpose of our current study was to examine the extent to which changes in SSB, diet soda, breakfast and fast food consumption were associated with changes in BMI in a cohort of nearly 700 adolescents over a two-year period. Given that previous studies in this area have primarily focused on the relationship between these key dietary factors and changes in BMI, we also sought to expand this focus by assessing the relationship between diet and objectively measured percent body fat (PBF). In addition, our aim was to quantify these associations independent of a number of important possible confounding factors, including objectively measured physical activity. Finally, although many longitudinal studies in this area have examined baseline dietary patterns as predictors of future weight status or weight change, we sought to examine the extent to which change in dietary patterns over time is associated with change in body mass. There is a subtle, yet valuable, distinction between these research questions that may have important implications for public health practice and clinical recommendations in addressing excess weight gain during adolescence.

Methods

The adolescents enrolled in this study were participants in two longitudinal cohort studies: 1) Identifying Determinants of Eating and Activity (IDEA) and 2) the Etiology of Childhood Obesity (ECHO). The baseline data used for these analyses were collected from the IDEA participants in 2006-7, with follow-up assessments in 2007-8 and 2008-9. Baseline data from ECHO participants were collected in 2007-8 with one follow-up in 2009-2010. Both studies were conducted within the seven-county metropolitan area of Minneapolis-St. Paul, Minnesota, and included identical measurement protocols.

For IDEA, 6^th through 11^th grade students and one parent or guardian were recruited via an existing cohort examining tobacco use among adolescents²⁸, an application list from the State department of motor vehicles, and a convenience sample from within the community. For ECHO, 6^th through 11^th grade students and one parent or guardian were recruited from the membership base of HealthPartners®, a large health maintenance organization in Minnesota. Recruitment for ECHO was designed to yield a racially/ethnically diverse sample with the intent to obtain a distribution of children and parents that matched national prevalence for healthy and unhealthy weight status. Combining the two samples, as was done in the present study, resulted in a larger and more diverse sample. An indicator variable representing the study (IDEA or ECHO) was included in all analytic models to account for any unmeasured confounding. All study protocols were approved by the University of Minnesota and Ohio State University Institutional Review Board.

Dependent variables

Body composition was measured by trained staff during the clinic visit. Height was measured without shoes using a Shorr Height Board to the nearest 0.1 cm (Shorr Productions, Olney, MD). Weight (to the nearest 0.1 kg) and percent body fat (to the nearest 0.1%) were assessed using a digital bioelectrical impedance scale (Tanita TBF-300A Body Composition Analyzer/Scale, Tanita Corporation, Tokyo, Japan).

Independent variables

Fast food intake was assessed by asking: 1) “In the past month … how many times did you buy food at a restaurant where food is ordered at a counter or at a drive-through window (there is no waiter/waitress)?” Numerous examples of fast food facility types were provided. For these items assessing beverage and fast food intake, nine response options ranged from “never or rarely” to “3 or more times per day.”

SSB consumption was assessed using four survey questions that asked about frequency of consuming regular soda (not including diet soda), sports drinks (e.g., Gatorade), other sweetened beverages (e.g., sweetened teas, juice drinks, lemonade), or coffee drinks (e.g., lattes, mochas, Frappuccinos, and Macchiatos, not including regular coffee). These items were summed to derive a composite score. In addition, one similarly structured question asked about “diet or sugar-free soda or pop.” The reliability and validity of the SSB and fast food survey items have been documented in previous research²⁹. These items generally yielded statistically significant test-retest reliability (0.63-0.82) and significant agreement with 24-hour recalls (0.19-0.38) for soft drinks and sports drinks.

To assess breakfast consumption, participants responded to a survey question asking how often they ate breakfast during a typical week. Eight response categories ranged from 0 to 7 days per week, adapted from previous epidemiologic research among young people¹⁶.

Covariates

Physical activity

The ActiGraph accelerometer, model 7164 (ActiGraph, LLC, Pensacola, FL) was used to collect seven days of physical activity data using 30-second epochs (data collection intervals). The monitor is an objective measure of physical activity and has been validated for use with children in laboratory and field settings^{30, 31}. At monitor distribution, trained research staff fit an elastic belt with an attached monitor to each adolescent, according to a standardized protocol. Participants were given written and verbal instructions on the use and care of the monitors and were instructed to wear the monitor during all waking hours except when swimming or bathing.

Accelerometry data were reduced using methods previously described^{32, 33}. Briefly, missing accelerometry data within participants’ 7-day record were replaced via imputation³⁴. On average, approximately 22 hours of data (about 13%) for the IDEA sample and 11 hours of data (about 18.5%) for the ECHO sample per adolescent were imputed over all 7 days of data collection. Participants were included who had at least 1 full day of data. The count threshold (counts/30 seconds) for MVPA was set at 1500 counts/30 seconds based on our previous work³². This count represented approximately 4.6 METs, which separates slow (<4.6 METs) and brisk (>4.6 METs) walking. Daily MET-weighted minutes of MVPA were calculated by summing METS for MVPA over the entire day. Total physical activity was defined as the sum of light, moderate, and vigorous activity.

Energy intake

Adolescents completed telephone-administered 24-hour dietary recalls. Trained and certified staff from the University of Minnesota Nutrition Coordination Center administered the recalls, using the Nutrition Data System-Research with an interactive, interview format with direct data entry linked to a nutrient database. Most participants completed three 24-hour recalls (two weekdays and one weekend day), though a limited number of participants completed only two.

Other covariates

Self-report demographic data were obtained at the initial study visit from adolescents and parents. Adolescents reported gender, age and race/ethnicity; parents reported if their child qualified for free or reduced priced lunch, and the highest level of education among the adults living in the household (college graduate, Y/N). Adolescents also completed the self-report Pubertal Development Scale (PDS)³⁵. The PDS is a five question summed score with good internal consistency (Cronbach’s α = 0.77) and reasonable associations with physician ratings (r=0.61-0.67)³⁵.

Analyses

Prior to analysis, gender was coded 1=male, 0=female. Race was coded 1=white, 0=other. Grade was coded 1=>=9th, 0=<9th. Parent’s education was coded 1=college graduate, 0=other. School lunch was coded 1= free and reduced lunch program, 0=other. Puberty, activity, and calories were modeled as continuous variables. Of 723 participants in the data set, participants were excluded who: (a) had less than one full day of accelerometry data (n=23), and/or (b) had less than two days of dietary recall data (n=7) at baseline, reducing the number included in the analyses to 693. Participants were included in the final analyses if they had data from at least two time points; given that some participants had data at two time points and others had data at three, sample sizes varied slightly across the study time points.

The exposure variables were servings per day of SSB and diet soda, and days per week purchasing fast food and eating breakfast. For each of the independent variables (i.e., beverages, breakfast, fast food), a mean across the two (ECHO) or three (IDEA) measurement visits was calculated; in addition, the deviation between the value observed at a given visit and that participant’s mean was calculated. The coefficient for the mean score estimated the cross-sectional difference in the outcome between groups of youth who differed by one unit of exposure. The coefficient for the deviation score estimated the longitudinal change in the outcome within youth who changed by one unit of exposure. We have used this decomposition scheme in previous studies related to obesity ^{36, 37}; there is a general discussion of this approach in a recent text on longitudinal data analysis ³⁸.

Analyses were conducted separately for males and females. To facilitate interpretation, all variables except the deviation scores were centered prior to analysis by subtracting the gender-specific mean from each observed value. That was not necessary for the deviation scores, as they had a mean of zero by definition.

Random coefficient models were used to examine the relationship between exposures and BMI and PBF; these models fit a random slope and intercept for each participant ^39-41. Consistent with the recommendations of Singer and Willett ⁴⁰, we used age as the index for time. For each dependent variable (BMI, PBF), we ran separate models for each exposure. In each model, we included age and the mean and deviation exposure variables. Those models provided a test of whether there were cross-sectional or longitudinal associations across the age range. We used empirical sandwich standard errors to accommodate the complex pattern of correlation in the data due to the nesting of youth within schools and neighborhoods and to the nesting of repeat observations within youth. We ran each model twice; in the first, we adjusted for race, grade, parent education, school lunch, puberty, activity measured at baseline, and study (ECHO vs IDEA). In the second, we included energy intake measured at baseline as an additional covariate.

Tables 2 and 3 report 64 tests, with 16 in each gender x design combination. As a result, we used an alpha of 0.05/16=0.003125 to indicate statistical significance.

Table 2.

Cross-sectional associations between key diet factors, BMI and percent body fat among adolescents.

		Model 1a			Model 2 b
Males		Coefficient	SE	p-value	Coefficient	SE	p-value
BMI	Sugar sweetened beverages (servings/day)	−0.22	0.28	0.437	−0.18	0.28	0.517
	Diet soda (servings/day)	1.63	0.76	0.032	1.63	0.76	0.032
	Fast food (purchases/ week)	0.16	0.39	0.683	0.17	0.40	0.668
	Breakfast (days/week)	−5.12	1.54	0.001	−5.08	1.53	0.001
PBF	Sugar sweetened beverages (servings/day)	−0.29	0.49	0.554	−0.04	0.47	0.927
	Diet soda (servings/day)	2.67	1.47	0.070	2.66	1.44	0.066
	Fast food (purchases/ week)	0.25	0.78	0.744	0.40	0.76	0.598
	Breakfast (days/week)	−9.52	2.91	0.001	−8.82	2.81	0.002
Females		Coefficient	SE	p-value	Coefficient	SE	p-value
BMI	Sugar sweetened beverages (servings/day)	0.33	0.54	0.542	0.40	0.54	0.456
	Diet soda (servings/day)	2.47	0.70	0.001	2.45	0.69	0.001
	Fast food (purchases/ week)	0.36	0.28	0.201	0.41	0.27	0.135
	Breakfast (days/week)	−4.44	1.62	0.007	−4.33	1.62	0.008
PBF	Sugar sweetened beverages (servings/day)	−0.23	0.69	0.739	0.002	0.69	0.998
	Diet soda (servings/day)	3.66	0.94	<0.001	3.64	0.91	<0.001
	Fast food (purchases/ week)	0.41	0.47	0.383	0.56	0.46	0.231
	Breakfast (days/week)	−7.65	2.25	0.001	−7.30	2.22	0.001

^aModel 1 was adjusted for physical activity, puberty, race, parental education, eligibility for free/reduced price lunch, age and study.

^bModel 2 adjusted for all of the covariates included in Model 1, as well as total energy intake (kcal/day).

Table 3.

Longitudinal associations between key diet factors, BMI and percent body fat among adolescents over time.

		Model 1a			Model 2 b
Males		Coefficient	SE	p-value	Coefficient	SE	p-value
BMI	Sugar sweetened beverages (servings/day)	0.25	0.10	0.012	0.27	0.10	0.008
	Dietsoda (servings/day)	−0.11	0.24	0.660	−0.09	0.24	0.722
	Fast food (purchases/week)	0.02	0.09	0.868	0.02	0.09	0.838
	Breakfast (days/week)	−0.21	0.48	0.653	−0.19	0.48	0.687
PBF	Sugar sweetened beverages (servings/day)	0.51	0.22	0.018	0.73	0.21	0.001
	Diet soda (servings/day)	−0.22	0.78	0.776	0.09	0.79	0.906
	Fast food (purchases/ week)	0.13	0.21	0.539	0.17	0.22	0.428
	Breakfast (days/week)	−1.80	1.23	0.143	−1.47	1.27	0.247
Females		Coefficient	SE	p-value	Coefficient	SE	p-value
BMI	Sugar sweetened beverages (servings/day)	−0.09	0.16	0.585	−0.05	0.17	0.746
	Diet soda (servings/day)	0.10	0.23	0.683	0.10	0.23	0.674
	Fast food (purchases/ week)	0.13	0.09	0.126	0.17	0.09	0.068
	Breakfast (days/week)	−0.31	0.45	0.491	−0.26	0.46	0.573
PBF	Sugar sweetened beverages (servings/day)	−0.06	0.33	0.861	0.04	0.35	0.908
	Diet soda (servings/day)	0.54	0.35	0.122	0.55	0.36	0.125
	Fast food (purchases/ week)	0.22	0.14	0.102	0.33	0.14	0.025
	Breakfast (days/week)	−0.38	0.86	0.660	−0.18	0.86	0.833

^aModel 1 was adjusted for physical activity, puberty, race, parental education, eligibility for free/reduced price lunch, age, and study.

^bModel 2 adjusted for all of the covariates included in Model 1, as well as total energy intake (kcal/day).

Descriptive statistics were prepared using SAS PROC MEANS and SAS PROC FREQ. Regression models were run in SAS PROC MIXED. All analyses were run in SAS Version 9.1 ⁴².

Results

The total baseline sample consisted of 723 participants from the IDEA and ECHO studies. The second measurement visit (n=332) included IDEA participants only. The sample at the two-year measurement visit was approximately 10% smaller than baseline, and included 648 IDEA and ECHO participants. Our final analytic sample for this analysis was 693. The average age of the sample at baseline was 14.6, and consisted of 49% males. At the first visit, participants reported an average daily consumption of 1.0 servings of SSB and 0.1 servings of diet soda. They reported 0.9 purchases of fast food and 5.6 breakfasts eaten in the past week. Average BMI was 22.0 kg/m², increasing to 23.3 kg/m² at the two-year visit. Average PBF was 21.4, increasing to 22.2 at the two-year visit. The average self-reported energy intake was 1977 kcal, increasing to 1995 at the two-year visit. The average minutes of daily physical activity was 310 at baseline, decreasing to 308 at the two-year visit.

Table 1 summarizes the characteristics of the sample at each measurement visit separately for males and females. Overall, males increased in BMI more rapidly than females. Males had higher self-reported energy intake and higher activity levels than girls. Girls had higher puberty scores than males. Male and female participants were predominantly white, with few participants in the free or reduced lunch program at school, and most of the participants had at least one parent who had graduated from college. For BMI, the median change from visit 1 to visit 3 was 1.4 kg/m2 (95% quantiles: −1.3 to 4.4). For dietary variables, median changes were: −0.02 servings/day for SSB (95% quantiles: −1.5 to 1.6), <0.01 servings/day for diet soda (95% quantiles: −0.2 to 0.4), <0.01 purchases/day for fast food (95% quantiles: −1.3 to 2.9), and 0 times/week for breakfast consumption (95% quantiles: −5 to 3).

Table 1.

Descriptive characteristics of the sample at each data collection period.

	Baseline, N=327		Two Year, N=276
Males	Mean	SD	Mean	SD
Age (years)	14.6	1.84	16.5	1.84
Body mass index (BMI) (kg/m2)	22.1	5.08	23.4	5.11
Percent body fat (PBF) (%)	16.2	9.58	16.3	8.98
Sugar sweetened beverages, servings/day	1.02	1.01	1.03	1.15
Diet soda, servings/day	0.16	0.49	0.15	0.49
Fast food, purchases/ week	0.91	0.88	1.19	1.25
Breakfast meals, days/week	0.91	0.20	0.83	0.27
Energy intake (kilocalories/day)	2196.2	673.4	2229.6	723.1
Physical activity (minutes/day)	320.4	74.7	304.6	64.0
Puberty	2.58	0.67	3.23	0.64
White/Caucasian (%)	86.5%		89.1%
Parent college graduate (%)	78.3%		81.9%
Free or reduced lunch-eligible (%)	10.4%		7.2%
	Baseline, N=339		Two Year, N=286
Females	Mean	SD	Mean	SD
Age (years)	14.6	1.83	16.6	1.86
Body mass index (BMI) (kg/m2)	21.9	4.91	22.8	4.55
Percent body fat (PBF) (%)	26.2	8.80	27.6	7.89
Sugar sweetened beverages, servings/day	0.70	0.84	0.65	0.77
Diet soda, servings/day	0.18	0.48	0.25	0.69
Fast food, purchases/ week	0.94	1.20	1.15	1.42
Breakfast meals, days/week	0.88	0.22	0.84	0.28
Energy intake (kilocalories/day)	1776.5	512.9	1767.3	494.4
Physical activity (minutes/day)	299.6	64.1	280.3	52.2
Puberty	3.17	0.68	3.61	0.46
White/Caucasian (%)	83.8%		85.0%
Parent college graduate (%)	72.9%		76.2%
Free or reduced lunch-eligible (%)	12.1%		9.8%

Table 2 summarizes the cross-sectional results of the regression analyses. Among boys and girls there were consistent and strong positive associations between self-reported diet soda consumption and both BMI and PBF. For example, among males the beta coefficient in the fully adjusted model (model 2) indicates that for each one serving per day increase in diet soda consumption, BMI and PBF increase by 4.9 and 6.9 units respectively. Among girls, consistent and strong inverse associations between self-reported breakfast and both BMI and PBF were observed. The associations were attenuated but remained statistically significant after adjustment for energy intake (model 2).

Table 3 summarizes the longitudinal results of the regression analyses. None of the associations were significant at the 0.003125 alpha level used to correct for the number of tests.

Conclusions

Findings from this cohort of adolescents yielded strong evidence for cross-sectional associations between diet soda consumption with weight status in both boys and girls. Specifically, youth who consumed diet soda were more likely to have a higher BMI and PBF compared to those who did not. In addition, girls who consumed breakfast more frequently were more likely to have a lower BMI and PBF compared to girls who did not; with a less conservative alpha level (0.05), we saw a similar pattern among boys. Interestingly, we did not observe cross-sectional associations between SSB intake or fast food consumption and either BMI or PBF, as have been identified in previous research.

Cross-sectional findings provide no insight into the temporality of these relationships; and reverse causality is a major concern. For example, the positive association found between diet soda consumption and BMI in the cross sectional analyses likely reflects the effect of excess body weight on eating habits (e.g. those struggling with their weight may begin drinking diet soda). Longitudinal studies provide better evidence to support causality. In our longitudinal analyses, we found no evidence of an association between our four dietary intake measures and BMI or PBF after taking into account the number of analyses performed. With a less conservative alpha level (0.05), there was evidence that more frequent breakfasts may be associated with lower PBF among boys and that more frequent fast food consumption may be associated with higher BMI among girls.

Previous research has yielded inconsistent evidence of longitudinal relationships between these key dietary factors and changes in adolescent weight status over time. Limitations in study designs and measurement tools may, in part, account for this wide range of inconsistent findings. For example, to our knowledge, there have been no adolescent cohort studies to date that have assessed these important dietary factors and weight change over time, independent of objectively-measured physical activity. It is imperative that these associations be examined independent of physical activity, due to the large body of literature supporting the notion that physical activity drives appetite and food consumption, perhaps particularly during these adolescent years, and that physical activity is also associated with weight status^{43, 44}. Accurately assessing physical activity in these adolescent cohort studies (for example, using objective assessment methods) may be of paramount importance. Furthermore, it may be important to also examine the influence of total energy intake in these associations. We presented our models using a two-stage approach, both with and without energy adjustment. Although we would expect overall energy intake to be on the causal pathway in some of these relationships (and thus would not want to control for it in our models), it has become standard practice to adjust for energy intake in these types of analyses as a means of adjusting for bias in food and nutrient intake reporting by weight status^{7-9, 11, 13, 18-20, 22, 23, 27}. For these reasons, it is important to obtain an accurate estimate of energy intake, using state-of-the-art methods, such as 24-hour dietary recalls.

Our study is also relatively unique in the research questions we have posed. We examined the extent to which change in dietary patterns over time is associated with change in body mass. In contrast, many longitudinal studies in this area have examined baseline dietary patterns as predictors of future weight status or weight change. Although documenting the relationship between early risk exposure and a later health outcome is an important research question, we also feel that examining the association between change in dietary factors and change in body mass is important in helping to understand the extent to which improvements in adolescent dietary intakes might result in body mass changes. In addition, we have examined the longitudinal associations between these dietary factors and PBF, and few studies to date have examined associations between adolescent dietary changes and changes in body composition over time. Although BMI is often used as a proxy for body fatness, it can be subject to error and misclassification of body fatness among individuals⁴⁵, and thus it is also important to assess these diet-weight relationships within the context of other, complimentary measures of body composition as well.

Given the strengths of our methods, it is important to consider possible reasons for the failure to replicate the cross-sectional findings in the cohort analyses. It is possible that our measures of dietary intake were inadequate; however, the significant associations observed in the cross-sectional analyses undermine that explanation. It is possible that the relationships are non-linear, and so missed in the linear models that we fit to the longitudinal data; that is a possibility, but we think it unlikely, given the short time period involved. Limited changes in diet in the cohort over the two-year follow-up period (limited range in exposure) could be another explanation for the null findings. Also, limited changes in body weight and PBF over the two-year period (outside of normal growth at this age) could be an issue. It is possible that the cohort may begin to experience more pronounced heterogeneity in these exposures and outcomes as participants begin to transition from adolescence to young adulthood. If so, longitudinal relationships may be more readily observed at that time; we hope to continue to follow the cohort so that we can address this possibility directly in future analyses.

The many challenges that researchers face in measuring dietary intake in these types of studies have been well-documented⁴⁶. Although food frequency questionnaires and 24-hour dietary recalls are the most commonly used tools to date for quantifying dietary intake in adolescents, they are subject to error and misreporting. In our study, we opted to utilize pre-tested and validated frequency-based measures to assess usual intake of key dietary factors over a period of one week to one month. This decision stemmed from our previous research, which indicated that three days of dietary intake data (as provided by the 24-hour recalls) may be insufficient in capturing episodic eating habits among many of these adolescents²⁹. However, given that estimates of total energy intake in particular have been shown to be more valid from 24-hour recalls than from food frequency questionnaires 46, we utilized dietary recalls to estimate energy intake as an additional covariate in our fully adjusted regression models as a means of adjusting for differential bias in reporting by weight status^{7-9, 11, 13, 18-20, 22, 23, 27}. In our study, additional adjustment for total calorie intake only slightly attenuated our significant cross-sectional findings, and did not appear to impact our longitudinal findings. Overall, it is very possible that the substantial error and bias associated with current self-reported dietary assessment tools may account some of the notable inconsistency reported in previous research.

The strengths of this study include the longitudinal cohort design, the substantial sample size and the ability to adjust for a number of important potential confounders, such as objectively measured physical activity. Despite the strengths of this work, however, our findings should be interpreted with several caveats in mind. For example, our sample was drawn from one geographic metropolitan region in the Midwestern United States, which may limit generalizability. Although there was some diversity within the sample, participants were primarily white with few coming from relatively low-socioeconomic backgrounds. It is possible that longitudinal dietary influences on weight change may be more readily apparent among more high-risk groups of adolescents (for example, those who are gaining excess weight at a more rapid pace). Finally, we conducted multiple testing of two dependent and four independent variables; therefore, individual p-values may need to be assessed in a more conservative manner, and robust patterns in findings should be highlighted, rather than individual associations.

To date, many adolescents and young adults fail to meet national dietary recommendations for health, and this may have important implications for a wide range of long-term chronic disease outcomes. In that adolescents and young adults are among the most frequent consumers of energy-dense food products, such as soda and fast food, and that they are also among the most heavily targeted age groups for food and beverage marketing, there is a critical need for clinical and public health efforts that target this age group. Overall, there is also an urgent need for a better understanding of the ways in which adolescent dietary patterns contribute to obesity and the avenues through which we can prevent excess weight gain over time. It is most likely that no single dietary factor may dramatically contribute alone to weight gain over time, but rather it is a constellation of factors that occur throughout childhood and adolescence that impact excess weight gain and long-term weight trajectories.