Applying Multiple Statistical Methods to Derive an Index of Dietary Behaviors Most Related to Obesity

Applying Multiple Statistical Methods to Derive an Index of Dietary Behaviors Most Related to... Abstract To evaluate the success of dietary interventions, we need measures that are more easily assessed and that closely align with intervention messaging. An index of obesogenic dietary behaviors (e.g., consumption of fast food and soft drinks, low fruit and vegetable consumption, and task eating (eating while engaging in other activities)) may serve this purpose and could be derived via data-driven methods typically used to describe nutrient intake. We used behavioral and physical measurement (i.e., body mass index, waist circumference) data from a subset of 2 independent cross-sectional samples of employees enrolled in the Promoting Activity and Changes in Eating (PACE) Study (Seattle, Washington) who were selected at baseline (2005–2007) (n = 561) and during follow-up (2007–2009) (n = 155). Index derivation methods, including principal components regression, partial least squares regression, and reduced rank regression, were compared. The best-fitting index for predicting physical measurements included consumption of fast food, French fries, and soft drinks. In linear mixed models, each 1-quartile increase in index score was associated with a 5% higher baseline body mass index (ratio of geometric means = 1.053, 95% confidence interval: 1.031, 1.075) and an approximately 4% higher baseline waist circumference (ratio = 1.036, 95% confidence interval: 1.019, 1.054) after adjustment for covariates. Results were similar at follow-up before and after adjustment for baseline measures. This index may be useful in evaluating public health or clinic-based dietary interventions to reduce obesity, especially given the ubiquity of these behaviors in the general population. dietary behaviors, obesity, partial least squares, principal components regression, reduced rank regression Obesity continues to be a significant threat to public health, with over one-third of US adults affected (1). Alarmingly, obesity prevention efforts, including dietary modification, have largely been unsuccessful in maintaining long-term weight reduction (2). An understanding of what dietary interventions are most effective is urgently needed, yet summarizing related evidence is hampered by differences in study design, study populations included, and dietary exposures evaluated (2). Dietary exposure variables used in intervention studies have mostly been based on patterns of nutrient intake identified via a priori combinations based on dietary recommendations or data reduction techniques from large epidemiologic studies collecting dietary information (3, 4). Traditionally, empirical derivation of dietary indices has involved the use of principal components or factor analysis to select factors which maximize the variance of the exposure (i.e., nutrient) variables (5). Therefore, such indices may maximally describe a combination of nutrients, but they may not be those most associated with the outcome of interest. Reduced rank regression (RRR) has been introduced as a method of selecting factors based on maximization of the variance of response variables that are highly correlated with or proximal to disease outcomes (5). While factors identified using RRR are likely to be more predictive of outcome variables by virtue of the predictive modeling methodology (6), evaluation of RRR in nutritional epidemiology has been limited. Partial least squares (PLS) regression is a method whereby identified factors balance the maximization of variance explained in both predictors and responses, and it is essentially a compromise between principal components regression (PCR) and RRR (5). Direct assessment of energy and nutrient intakes via self-report has been shown to be biased for each of the commonly used comprehensive intake assessment methods (7). Assessing dietary and eating behaviors at the same time may provide additional ways to estimate obesogenic behavior and may provide insight into the way in which high-energy foods are consumed. Several studies have shown that measures of adiposity, including body mass index (BMI), are associated with discrete dietary behaviors (including intakes of fruits and vegetables (8), fast food (9), and sugar-sweetened drinks (10)) and eating behaviors (including dieting (11), emotional eating (12), and task eating (eating while engaging in other activities) (13)). No one behavior, however, stands out as the most sensitive indicator. Modeling patterns of eating behavior may provide a more comprehensive way of looking at how diet is associated with obesity and disease (3, 14–16). In addition, many nutrients and foods are highly correlated, which makes it difficult to tease out separate associations. Evaluation of behaviors may reduce the effects of multicollinearity to some extent by reducing the sheer number of dietary variables. Identification of key behaviors most related to obesity may also provide more easily understood dietary recommendations for the lay public, as well as more actionable points of population-level obesity prevention and related chronic-disease-prevention efforts. The current study provided us with an opportunity to create an index of dietary behaviors to predictively model BMI and waist circumference in a population of mostly white-collar working adults. Our objective in this study, therefore, was to derive and evaluate indices of dietary and eating behaviors using various statistical methods (i.e., PCR, RRR, and PLS regression) to identify the combination of variables most predictive of obesity in a nested cohort of middle-aged US working adults. This index may ultimately identify persons at greater risk of obesity in both clinical and community settings. METHODS Study population The Promoting Activity and Changes in Eating (PACE) Study was a large group-randomized weight-reduction intervention trial carried out among approximately 3,000 individuals at 34 work sites in the Seattle, Washington, metropolitan area. Eligible companies employed 40–350 workers, were identified using US Standard Industrial Classification 2-digit codes (17), and included industries that were characterized as manufacturing, transportation or utilities, personal services, household and miscellaneous services, and nonclassifiable establishments. Eligibility criteria included having a high proportion of sedentary employees, having a low turnover rate during the previous 2 years, and having a low proportion of non-English-speaking employees. A detailed description of the PACE Study has been published elsewhere (18). All employees at participating work sites with fewer than 150 employees and a random subsample of 125 employees at work sites with more than 150 employees were asked to complete a standard questionnaire assessing self-reported dietary and physical activity behaviors, height, weight, and demographic information at baseline (2005–2007). An independent sample of employees, derived in the same way, were invited to complete the follow-up questionnaire (2007–2009). Data were collected among 3,054 individuals within 34 work sites at baseline and among 2,398 individuals within 33 work sites at follow-up. At baseline, a random “intensive assessment” subset of employees within all 34 work sites was also invited to provide additional physical measurements, including measured height, weight, and waist circumference (N = 34 work sites; n = 622 participants). For analyses at baseline, there were 561 participants after exclusion of persons with missing data for BMI (n = 4), waist circumference (n = 19), dietary behaviors (n = 27), and covariates (n = 11). At the follow-up intensive assessment, priority was given to participants who provided these measurements at baseline and who also were part of the follow-up survey. An additional random sample of follow-up survey respondents were invited to complete the intensive assessment. A total of 155 and 156 individuals provided baseline and follow-up measurements, respectively, for both BMI and waist circumference. Response variables Body mass index At both baseline and follow-up, height and weight were assessed via self-report and were measured by trained study personnel using a stadiometer and scale, respectively. BMI was calculated as weight (kg) divided by the square of height (m2) using both self-reported and physically measured data for these analyses. Waist circumference Waist circumference was assessed by study personnel via physical measurement during the same intensive assessment within work sites at baseline and follow-up. Values are reported in centimeters. Exposure variables Fruit and vegetable consumption Increased consumption of fruits and vegetables is a behavior promoted heavily by public health professionals, as it has been found to be inversely associated with obesity (19). Components of fruit and vegetable consumption were assessed using the National Cancer Institute’s 7-item 5-A-Day fruit and vegetable assessment tool (20). The total numbers of servings of fruit and vegetables consumed per day were calculated for descriptive purposes, whereas individual assessment items were included in dietary index analyses. Weekly frequency of fast-food meal consumption Frequency of fast-food meal consumption was assessed using a single question similar to that used in other studies: “Thinking about how often you eat out, how many times in a week or month do you eat breakfast, lunch, or dinner in a place such as McDonald’s®, Burger King®, Wendy’s®, Arby’s®, Pizza Hut®, or Kentucky Fried Chicken®?” (21, 22). Responses were given as number of times per week or number of times per month. All responses were converted to number of times per week. Weekly frequency of soft-drink consumption Average weekly soft-drink intake was also assessed, with the question “How often do you drink soft drinks or soda pop (regular or diet)?” (23). Response options were “never,” “less than once a week,” “about once a week,” “2–5 times per week,” “about once a day,” and “2 or more times per day.” Task eating (eating while engaged in other activities) The way in which food is consumed has also been associated with overweight and obesity (11–13). The “task eating” construct connotes a level of distraction, or lack of eating awareness or mindfulness, which has also been linked to obesity (24, 25). Task eating was assessed via a single item: “How often do you eat food (meals or snacks) while doing another activity—for example, watching TV, working at a computer, reading, driving, or playing video games?” (13). Response options were presented on a 5-point Likert scale ranging from 1 (“never”) to 5 (“always”). Covariates Individual-level factors included in the models were: age, sex, race/ethnicity (where “other” included Native Alaskan/American Indian and Pacific Islander/Native Hawaiian groups), and education. To further adjust for lifestyle differences, models also included adjustment for type of occupation (manual (i.e., machine operators, mechanics/technicians, service workers, tradesmen, and laborers) or nonmanual) and leisure-time physical activity of at least 10 minutes’ duration, assessed via the Godin-Shephard Leisure-Time Physical Activity Questionnaire (26). The Godin-Shephard questionnaire estimates the frequency of exercise bouts (i.e., vigorous, moderate, and light), multiplies each bout by the corresponding metabolic equivalent of task (MET) values (i.e., 9, 5, and 3 METs), and then sums these components to create an intensity-weighted score (i.e., leisure index score) that corresponds to a weekly MET frequency (MET-hours/week) (26). Statistical analyses Creating the dietary index A random half of participants (the test set) was selected at baseline, and 3 statistical methods (i.e., PCR, RRR, and PLS regression) were employed to identify a dietary index of reported dietary and eating behaviors using methodology described by Hoffmann et al. (5). Selection of factors using PCR is based on maximizing the percentage of variation explained among exposure variables (a partial goal of PLS regression and not a goal of RRR). Conversely, the selection of factors using RRR in the predictive model is based on maximizing the percentage of variation explained among response variables (a partial goal of PLS regression and not a goal of PCR). The baseline dietary exposure variables included in factor analyses were: all 7 individual items in the National Cancer Institute’s 5-A-Day fruit and vegetable screener (servings/day), frequency of consumption of fast-food meals (times/week), frequency of consumption of soft drinks (times/week), and frequency of eating while engaging in other activities (task eating; never, seldom, sometimes, most of the time, or always). Because the distributions of dietary and eating behaviors did not vary substantially, we chose not to standardize or transform the variables before factor analyses. The response variables included in factor analyses were measured BMI and waist circumference, since both were strongly correlated (r = 0.85); both variables were log-transformed to account for skewness before factor analyses. The PCR, RRR, and PLS methods were implemented using the PROC PLS command within SAS for Windows, release 9.4 (SAS Institute, Inc., Cary, North Carolina), with cross-validation to identify the minimum number of extracted factors for all methods. Validating the dietary index Dietary index scores for each statistical method were calculated as the average of variables identified in each extracted factor in the test set of participants at baseline. Models for prediction of baseline BMI and waist circumference employing these factors as well as all dietary behavior variables were compared for goodness of fit using Akaike’s Information Criterion (AIC) in the complement of the test set (validation set) of participants, where models with the smallest AIC indicated the best fit (27). The factor with the smallest AIC in separate models predicting BMI and waist circumference was selected for use in subsequent predictive models in the validation set at baseline, and again at follow-up. Longitudinal analyses predicting change in BMI and waist circumference were conducted in the nested cohort of baseline participants who had follow-up data. All predictive models were tested for interaction by sex; fully adjusted predictive models included age, sex, race/ethnicity, education, type of occupation (manual/nonmanual), leisure-time physical activity, and intervention arm as covariates. Wald tests were used to generate P values. All regression analyses were conducted using Stata SE, version 13.0 (StataCorp LP, College Station, Texas). RESULTS Baseline demographic and covariate data among the total, test, and validation samples are summarized in Table 1. The mean age was approximately 44 years, and participants were primarily white, had more than a high school education, worked in white- or pink-collar jobs (i.e., nonmanual occupations), and were sufficiently active according to an established cutpoint (leisure index score ≥24) for the Godin-Shephard Leisure-Time Physical Activity Questionnaire (28). Continuous and categorical measures of baseline response and exposure variables are presented in Tables 2 and 3, respectively. On average, participants were overweight/obese (mean BMI = 29.1), did not consume the recommended number of servings of fruit and vegetables per day (mean = 3.27 (standard error, 0.09) servings/day), and ate fast-food meals approximately 4 times per week (mean = 3.94 (standard error, 0.22) servings/week) (Table 2). Only about 19% and 4% of participants reported never drinking soda and never eating while engaging in other activities, respectively (Table 3). Table 4 presents Pearson’s correlation coefficients for all dietary behaviors (i.e., exposures) among both the test and validation sets of participants. Moderate positive correlations were evident between intakes of fast food, French fries/fried potatoes, and soft drinks, as well as between intakes of fruit, green salad, and vegetables (not including potatoes). Table 1. Demographic Characteristicsa of Participants in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Demographic Characteristic Total (N = 34; n = 561)b Test Set (N = 34; n = 280) Validation Set (N = 34; n = 281) No. % No. % No. % Age, yearsc 44.1 (0.5) 43.9 (0.7) 44.4 (0.7) Sex  Male 230 41.0 116 41.4 114 40.6  Female 331 59.0 164 58.6 167 59.4 Race/ethnicity  White 456 81.3 222 79.3 234 83.3  Hispanic/Latino 35 6.2 17 6.0 18 6.4  Black 28 5.0 15 5.4 13 4.6  Asian 24 4.3 15 5.4 9 3.2  Other 18 3.2 11 3.9 7 2.5 Education  Less than HS, HS diploma, or GED certificate 65 11.6 31 11.1 34 12.1  Some college or technical college 229 40.8 106 37.9 123 43.8  College graduate 193 34.4 102 36.4 91 32.4  Postgraduate study 74 13.2 41 14.6 33 11.7 Manual occupation  No 478 85.2 241 86.1 237 84.3  Yes 83 14.8 39 13.9 44 15.7 Leisure-time PA, MET-hours/weekc,d 28.9 (1.2) 29.6 (1.6) 28.2 (1.2) Demographic Characteristic Total (N = 34; n = 561)b Test Set (N = 34; n = 280) Validation Set (N = 34; n = 281) No. % No. % No. % Age, yearsc 44.1 (0.5) 43.9 (0.7) 44.4 (0.7) Sex  Male 230 41.0 116 41.4 114 40.6  Female 331 59.0 164 58.6 167 59.4 Race/ethnicity  White 456 81.3 222 79.3 234 83.3  Hispanic/Latino 35 6.2 17 6.0 18 6.4  Black 28 5.0 15 5.4 13 4.6  Asian 24 4.3 15 5.4 9 3.2  Other 18 3.2 11 3.9 7 2.5 Education  Less than HS, HS diploma, or GED certificate 65 11.6 31 11.1 34 12.1  Some college or technical college 229 40.8 106 37.9 123 43.8  College graduate 193 34.4 102 36.4 91 32.4  Postgraduate study 74 13.2 41 14.6 33 11.7 Manual occupation  No 478 85.2 241 86.1 237 84.3  Yes 83 14.8 39 13.9 44 15.7 Leisure-time PA, MET-hours/weekc,d 28.9 (1.2) 29.6 (1.6) 28.2 (1.2) Abbreviations: GED, General Educational Development; HS, high school; MET, metabolic equivalent of task; PA, physical activity; PACE, Promoting Activity and Changes in Eating. a Values presented are averaged work-site means unless otherwise indicated. bN, number of work sites; n, number of participants. c Values are expressed as mean (standard error). d Leisure-time physical activity of at least 10 minutes’ duration was assessed via the Godin-Shephard Leisure-Time Physical Activity Questionnaire (26). The questionnaire estimates the frequency of exercise bouts (i.e., vigorous, moderate, and light), multiplies each bout by the corresponding MET values (i.e., 9, 5, and 3 METs), and then sums these components to create an intensity-weighted score (i.e., leisure index score) that corresponds to a weekly MET frequency (MET-hours/week) (26). Table 1. Demographic Characteristicsa of Participants in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Demographic Characteristic Total (N = 34; n = 561)b Test Set (N = 34; n = 280) Validation Set (N = 34; n = 281) No. % No. % No. % Age, yearsc 44.1 (0.5) 43.9 (0.7) 44.4 (0.7) Sex  Male 230 41.0 116 41.4 114 40.6  Female 331 59.0 164 58.6 167 59.4 Race/ethnicity  White 456 81.3 222 79.3 234 83.3  Hispanic/Latino 35 6.2 17 6.0 18 6.4  Black 28 5.0 15 5.4 13 4.6  Asian 24 4.3 15 5.4 9 3.2  Other 18 3.2 11 3.9 7 2.5 Education  Less than HS, HS diploma, or GED certificate 65 11.6 31 11.1 34 12.1  Some college or technical college 229 40.8 106 37.9 123 43.8  College graduate 193 34.4 102 36.4 91 32.4  Postgraduate study 74 13.2 41 14.6 33 11.7 Manual occupation  No 478 85.2 241 86.1 237 84.3  Yes 83 14.8 39 13.9 44 15.7 Leisure-time PA, MET-hours/weekc,d 28.9 (1.2) 29.6 (1.6) 28.2 (1.2) Demographic Characteristic Total (N = 34; n = 561)b Test Set (N = 34; n = 280) Validation Set (N = 34; n = 281) No. % No. % No. % Age, yearsc 44.1 (0.5) 43.9 (0.7) 44.4 (0.7) Sex  Male 230 41.0 116 41.4 114 40.6  Female 331 59.0 164 58.6 167 59.4 Race/ethnicity  White 456 81.3 222 79.3 234 83.3  Hispanic/Latino 35 6.2 17 6.0 18 6.4  Black 28 5.0 15 5.4 13 4.6  Asian 24 4.3 15 5.4 9 3.2  Other 18 3.2 11 3.9 7 2.5 Education  Less than HS, HS diploma, or GED certificate 65 11.6 31 11.1 34 12.1  Some college or technical college 229 40.8 106 37.9 123 43.8  College graduate 193 34.4 102 36.4 91 32.4  Postgraduate study 74 13.2 41 14.6 33 11.7 Manual occupation  No 478 85.2 241 86.1 237 84.3  Yes 83 14.8 39 13.9 44 15.7 Leisure-time PA, MET-hours/weekc,d 28.9 (1.2) 29.6 (1.6) 28.2 (1.2) Abbreviations: GED, General Educational Development; HS, high school; MET, metabolic equivalent of task; PA, physical activity; PACE, Promoting Activity and Changes in Eating. a Values presented are averaged work-site means unless otherwise indicated. bN, number of work sites; n, number of participants. c Values are expressed as mean (standard error). d Leisure-time physical activity of at least 10 minutes’ duration was assessed via the Godin-Shephard Leisure-Time Physical Activity Questionnaire (26). The questionnaire estimates the frequency of exercise bouts (i.e., vigorous, moderate, and light), multiplies each bout by the corresponding MET values (i.e., 9, 5, and 3 METs), and then sums these components to create an intensity-weighted score (i.e., leisure index score) that corresponds to a weekly MET frequency (MET-hours/week) (26). Table 2. Mean Valuesa (and Standard Errors) for Continuous Measures of Obesity and Dietary Behaviors in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Measure of Obesity or Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) Continuous Measures of Obesity Physical measurements  Weight, kg 84.5 (0.9) 84.3 (1.3) 84.7 (1.2)  BMIb 29.1 (0.3) 29.0 (0.4) 29.2 (0.4)  Waist circumference, cm 90.5 (0.7) 90.4 (1.0) 90.6 (0.9) Self-reported measures  Weight, kg 82.4 (0.9) 82.3 (1.3) 82.5 (1.2)  BMI 28.1 (0.4) 28.1 (0.4) 28.2 (0.4) Continuous Dietary Behaviors Abbreviated FFQ intake, servings/day  100% fruit juice 0.29 (0.02) 0.33 (0.03) 0.25 (0.03)  Other fruit juicec 0.21 (0.03) 0.22 (0.03) 0.20 (0.03)  Fruit 1.05 (0.06) 1.11 (0.06) 0.99 (0.08)  Green salad 0.51 (0.02) 0.51 (0.03) 0.50 (0.03)  French fries/fried potatoes 0.12 (0.01) 0.12 (0.01) 0.12 (0.01)  Other potatoes 0.14 (0.01) 0.15 (0.01) 0.14 (0.01)  Other vegetablesd 1.09 (0.06) 1.06 (0.07) 1.11 (0.09) Fast-food consumption, times/week 3.94 (0.22) 3.95 (0.32) 3.94 (0.32) Measure of Obesity or Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) Continuous Measures of Obesity Physical measurements  Weight, kg 84.5 (0.9) 84.3 (1.3) 84.7 (1.2)  BMIb 29.1 (0.3) 29.0 (0.4) 29.2 (0.4)  Waist circumference, cm 90.5 (0.7) 90.4 (1.0) 90.6 (0.9) Self-reported measures  Weight, kg 82.4 (0.9) 82.3 (1.3) 82.5 (1.2)  BMI 28.1 (0.4) 28.1 (0.4) 28.2 (0.4) Continuous Dietary Behaviors Abbreviated FFQ intake, servings/day  100% fruit juice 0.29 (0.02) 0.33 (0.03) 0.25 (0.03)  Other fruit juicec 0.21 (0.03) 0.22 (0.03) 0.20 (0.03)  Fruit 1.05 (0.06) 1.11 (0.06) 0.99 (0.08)  Green salad 0.51 (0.02) 0.51 (0.03) 0.50 (0.03)  French fries/fried potatoes 0.12 (0.01) 0.12 (0.01) 0.12 (0.01)  Other potatoes 0.14 (0.01) 0.15 (0.01) 0.14 (0.01)  Other vegetablesd 1.09 (0.06) 1.06 (0.07) 1.11 (0.09) Fast-food consumption, times/week 3.94 (0.22) 3.95 (0.32) 3.94 (0.32) Abbreviations: BMI, body mass index; FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a Averaged work-site means. b Weight (kg)/height (m)2. c Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). d Does not include green salad or potatoes. Table 2. Mean Valuesa (and Standard Errors) for Continuous Measures of Obesity and Dietary Behaviors in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Measure of Obesity or Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) Continuous Measures of Obesity Physical measurements  Weight, kg 84.5 (0.9) 84.3 (1.3) 84.7 (1.2)  BMIb 29.1 (0.3) 29.0 (0.4) 29.2 (0.4)  Waist circumference, cm 90.5 (0.7) 90.4 (1.0) 90.6 (0.9) Self-reported measures  Weight, kg 82.4 (0.9) 82.3 (1.3) 82.5 (1.2)  BMI 28.1 (0.4) 28.1 (0.4) 28.2 (0.4) Continuous Dietary Behaviors Abbreviated FFQ intake, servings/day  100% fruit juice 0.29 (0.02) 0.33 (0.03) 0.25 (0.03)  Other fruit juicec 0.21 (0.03) 0.22 (0.03) 0.20 (0.03)  Fruit 1.05 (0.06) 1.11 (0.06) 0.99 (0.08)  Green salad 0.51 (0.02) 0.51 (0.03) 0.50 (0.03)  French fries/fried potatoes 0.12 (0.01) 0.12 (0.01) 0.12 (0.01)  Other potatoes 0.14 (0.01) 0.15 (0.01) 0.14 (0.01)  Other vegetablesd 1.09 (0.06) 1.06 (0.07) 1.11 (0.09) Fast-food consumption, times/week 3.94 (0.22) 3.95 (0.32) 3.94 (0.32) Measure of Obesity or Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) Continuous Measures of Obesity Physical measurements  Weight, kg 84.5 (0.9) 84.3 (1.3) 84.7 (1.2)  BMIb 29.1 (0.3) 29.0 (0.4) 29.2 (0.4)  Waist circumference, cm 90.5 (0.7) 90.4 (1.0) 90.6 (0.9) Self-reported measures  Weight, kg 82.4 (0.9) 82.3 (1.3) 82.5 (1.2)  BMI 28.1 (0.4) 28.1 (0.4) 28.2 (0.4) Continuous Dietary Behaviors Abbreviated FFQ intake, servings/day  100% fruit juice 0.29 (0.02) 0.33 (0.03) 0.25 (0.03)  Other fruit juicec 0.21 (0.03) 0.22 (0.03) 0.20 (0.03)  Fruit 1.05 (0.06) 1.11 (0.06) 0.99 (0.08)  Green salad 0.51 (0.02) 0.51 (0.03) 0.50 (0.03)  French fries/fried potatoes 0.12 (0.01) 0.12 (0.01) 0.12 (0.01)  Other potatoes 0.14 (0.01) 0.15 (0.01) 0.14 (0.01)  Other vegetablesd 1.09 (0.06) 1.06 (0.07) 1.11 (0.09) Fast-food consumption, times/week 3.94 (0.22) 3.95 (0.32) 3.94 (0.32) Abbreviations: BMI, body mass index; FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a Averaged work-site means. b Weight (kg)/height (m)2. c Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). d Does not include green salad or potatoes. Table 3. Distribution of Categorical Dietary Behaviors in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Measure of Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) No.a % No. % No. % Soft-drink consumption, times/week  0 107 19.0 53 18.9 54 19.2  0.5 129 23.0 65 23.3 64 22.8  1 76 13.6 37 13.2 39 13.9  3.5 106 18.9 53 18.9 53 18.9  7 79 14.1 39 13.9 40 14.2  14 64 11.4 33 11.8 31 11.0 Task eatingb  Never 22 3.9 8 2.9 14 5.0  Seldom 88 15.7 50 17.9 38 13.5  Sometimes 241 43.0 126 45.0 115 40.9  Most of the time 187 33.3 90 32.1 97 34.5  Always 23 4.1 6 2.1 17 6.1 Measure of Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) No.a % No. % No. % Soft-drink consumption, times/week  0 107 19.0 53 18.9 54 19.2  0.5 129 23.0 65 23.3 64 22.8  1 76 13.6 37 13.2 39 13.9  3.5 106 18.9 53 18.9 53 18.9  7 79 14.1 39 13.9 40 14.2  14 64 11.4 33 11.8 31 11.0 Task eatingb  Never 22 3.9 8 2.9 14 5.0  Seldom 88 15.7 50 17.9 38 13.5  Sometimes 241 43.0 126 45.0 115 40.9  Most of the time 187 33.3 90 32.1 97 34.5  Always 23 4.1 6 2.1 17 6.1 Abbreviation: PACE, Promoting Activity and Changes in Eating. a Number of participants. b Eating while engaging in other activities. Table 3. Distribution of Categorical Dietary Behaviors in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Measure of Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) No.a % No. % No. % Soft-drink consumption, times/week  0 107 19.0 53 18.9 54 19.2  0.5 129 23.0 65 23.3 64 22.8  1 76 13.6 37 13.2 39 13.9  3.5 106 18.9 53 18.9 53 18.9  7 79 14.1 39 13.9 40 14.2  14 64 11.4 33 11.8 31 11.0 Task eatingb  Never 22 3.9 8 2.9 14 5.0  Seldom 88 15.7 50 17.9 38 13.5  Sometimes 241 43.0 126 45.0 115 40.9  Most of the time 187 33.3 90 32.1 97 34.5  Always 23 4.1 6 2.1 17 6.1 Measure of Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) No.a % No. % No. % Soft-drink consumption, times/week  0 107 19.0 53 18.9 54 19.2  0.5 129 23.0 65 23.3 64 22.8  1 76 13.6 37 13.2 39 13.9  3.5 106 18.9 53 18.9 53 18.9  7 79 14.1 39 13.9 40 14.2  14 64 11.4 33 11.8 31 11.0 Task eatingb  Never 22 3.9 8 2.9 14 5.0  Seldom 88 15.7 50 17.9 38 13.5  Sometimes 241 43.0 126 45.0 115 40.9  Most of the time 187 33.3 90 32.1 97 34.5  Always 23 4.1 6 2.1 17 6.1 Abbreviation: PACE, Promoting Activity and Changes in Eating. a Number of participants. b Eating while engaging in other activities. Table 4. Pearson’s Correlation Coefficients for Correlations Between Dietary Behaviors Among Baseline Test and Validation Sets of Participants From the Intensive Assessment Subsample,a PACE Study, 2005–2007 Dietary Behavior Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingb 100% Fruit Juice Other Fruit Juicec Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablesd Abbreviated FFQ intake, servings/day  100% fruit juice 0.18 0.16 0.08 −0.03 0.08 0.05 −0.08 −0.08 −0.08  Other fruit juicec 0.12 0.07 0.02 −0.05 0.11 −0.01 −0.01 0.00 0.10  Fruit 0.08 −0.09 0.31e −0.17 0.01 0.45e −0.24e −0.13 −0.08  Green salad 0.09 −0.08 0.23e −0.10 0.04 0.35e −0.16 −0.03 −0.09  French fries/fried potatoes 0.07 −0.02 −0.20e −0.18 0.21 −0.16 0.62e 0.17 0.13  Other potatoes 0.17 0.02 0.05 0.01 0.07 0.08 0.06 0.08 0.01  Other vegetablesd 0.08 −0.06 0.45e 0.27e −0.20e 0.03 −0.21 −0.07 0.00 Fast-food consumption, times/week 0.02 0.06 −0.17 −0.22 0.44e 0.08 −0.21 0.25e 0.12 Soft-drink consumption, times/week −0.05 −0.03 −0.18 −0.05 0.24e 0.08 −0.13 0.30e 0.10 Task eatingb −0.11 −0.14 −0.02 −0.08 0.13 −0.07 −0.04 0.07 0.15 Dietary Behavior Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingb 100% Fruit Juice Other Fruit Juicec Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablesd Abbreviated FFQ intake, servings/day  100% fruit juice 0.18 0.16 0.08 −0.03 0.08 0.05 −0.08 −0.08 −0.08  Other fruit juicec 0.12 0.07 0.02 −0.05 0.11 −0.01 −0.01 0.00 0.10  Fruit 0.08 −0.09 0.31e −0.17 0.01 0.45e −0.24e −0.13 −0.08  Green salad 0.09 −0.08 0.23e −0.10 0.04 0.35e −0.16 −0.03 −0.09  French fries/fried potatoes 0.07 −0.02 −0.20e −0.18 0.21 −0.16 0.62e 0.17 0.13  Other potatoes 0.17 0.02 0.05 0.01 0.07 0.08 0.06 0.08 0.01  Other vegetablesd 0.08 −0.06 0.45e 0.27e −0.20e 0.03 −0.21 −0.07 0.00 Fast-food consumption, times/week 0.02 0.06 −0.17 −0.22 0.44e 0.08 −0.21 0.25e 0.12 Soft-drink consumption, times/week −0.05 −0.03 −0.18 −0.05 0.24e 0.08 −0.13 0.30e 0.10 Task eatingb −0.11 −0.14 −0.02 −0.08 0.13 −0.07 −0.04 0.07 0.15 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a Correlation coefficients for the validation set (n = 281) are presented above the diagonal, and those for the test set (n = 280) are presented below the diagonal. b Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). c Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). d Does not include green salad or potatoes. e Correlation coefficient differed significantly from zero (P < 0.001). Table 4. Pearson’s Correlation Coefficients for Correlations Between Dietary Behaviors Among Baseline Test and Validation Sets of Participants From the Intensive Assessment Subsample,a PACE Study, 2005–2007 Dietary Behavior Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingb 100% Fruit Juice Other Fruit Juicec Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablesd Abbreviated FFQ intake, servings/day  100% fruit juice 0.18 0.16 0.08 −0.03 0.08 0.05 −0.08 −0.08 −0.08  Other fruit juicec 0.12 0.07 0.02 −0.05 0.11 −0.01 −0.01 0.00 0.10  Fruit 0.08 −0.09 0.31e −0.17 0.01 0.45e −0.24e −0.13 −0.08  Green salad 0.09 −0.08 0.23e −0.10 0.04 0.35e −0.16 −0.03 −0.09  French fries/fried potatoes 0.07 −0.02 −0.20e −0.18 0.21 −0.16 0.62e 0.17 0.13  Other potatoes 0.17 0.02 0.05 0.01 0.07 0.08 0.06 0.08 0.01  Other vegetablesd 0.08 −0.06 0.45e 0.27e −0.20e 0.03 −0.21 −0.07 0.00 Fast-food consumption, times/week 0.02 0.06 −0.17 −0.22 0.44e 0.08 −0.21 0.25e 0.12 Soft-drink consumption, times/week −0.05 −0.03 −0.18 −0.05 0.24e 0.08 −0.13 0.30e 0.10 Task eatingb −0.11 −0.14 −0.02 −0.08 0.13 −0.07 −0.04 0.07 0.15 Dietary Behavior Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingb 100% Fruit Juice Other Fruit Juicec Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablesd Abbreviated FFQ intake, servings/day  100% fruit juice 0.18 0.16 0.08 −0.03 0.08 0.05 −0.08 −0.08 −0.08  Other fruit juicec 0.12 0.07 0.02 −0.05 0.11 −0.01 −0.01 0.00 0.10  Fruit 0.08 −0.09 0.31e −0.17 0.01 0.45e −0.24e −0.13 −0.08  Green salad 0.09 −0.08 0.23e −0.10 0.04 0.35e −0.16 −0.03 −0.09  French fries/fried potatoes 0.07 −0.02 −0.20e −0.18 0.21 −0.16 0.62e 0.17 0.13  Other potatoes 0.17 0.02 0.05 0.01 0.07 0.08 0.06 0.08 0.01  Other vegetablesd 0.08 −0.06 0.45e 0.27e −0.20e 0.03 −0.21 −0.07 0.00 Fast-food consumption, times/week 0.02 0.06 −0.17 −0.22 0.44e 0.08 −0.21 0.25e 0.12 Soft-drink consumption, times/week −0.05 −0.03 −0.18 −0.05 0.24e 0.08 −0.13 0.30e 0.10 Task eatingb −0.11 −0.14 −0.02 −0.08 0.13 −0.07 −0.04 0.07 0.15 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a Correlation coefficients for the validation set (n = 281) are presented above the diagonal, and those for the test set (n = 280) are presented below the diagonal. b Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). c Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). d Does not include green salad or potatoes. e Correlation coefficient differed significantly from zero (P < 0.001). The statistical methods PCR, PLS regression, and RRR were employed to explain variation in measures of obesity (i.e., responses) and dietary behavior variables (i.e., exposures) for a test set of individuals who did not having missing information for any of these variables at baseline. Cross-validation was employed, and a single factor solution was identified for each method. As expected, the factor selected via the PCR method explained the highest proportion of variance among exposure variables, whereas the factor selected by the RRR method explained the highest proportion of variance among response variables (Table 5). The proportions of variance explained among exposures and responses by the PLS factor were a compromise between the other two methods. Differences in proportions of variance explained between methods, however, were small. Model effect loadings for each dietary behavior variable that were greater than the absolute value of 0.4 are presented for each factor derived from various statistical methods in Table 6. Notably, the RRR-derived factor and PLS-derived factors included the same dietary behavior variables (i.e., intake of fast-food meals, soda, and French fries/fried potatoes). The proposed simple index is the average of the weekly frequency of each of these 3 behaviors. Table 5. Percentage of Variation in Dietary Behaviors and Obesity Measures Explained by Factors Extracted Using Different Statistical Methods in a Test Set of Participants From the Intensive Assessment Subsample,a PACE Study, 2005–2007 Measure of Obesity or Dietary Behaviorb Index Derivation Method Principal Components Regression Partial Least Squares Regression Reduced Rank Regression Measure of obesity  Body mass indexc 10.60 12.56 13.85  Waist circumference, cm 13.25 16.68 18.19  Total (both measures of obesity) 11.93 14.62 16.02 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice 1.45 0.43 0.75   Other fruit juiced 0.69 1.25 1.91   Fruit 39.21 23.07 6.68   Green salad 27.50 24.53 22.08   French fries/fried potatoes 40.31 42.09 30.07   Other potatoes 0.01 3.00 7.67   Other vegetablese 40.94 31.01 25.72  Fast-food consumption, times/week 43.06 53.12 51.95  Soft-drink consumption, times/week 24.22 33.94 41.56  Task eatingf 5.02 3.10 1.05  Total (all dietary behaviors) 22.24 21.55 18.94 Measure of Obesity or Dietary Behaviorb Index Derivation Method Principal Components Regression Partial Least Squares Regression Reduced Rank Regression Measure of obesity  Body mass indexc 10.60 12.56 13.85  Waist circumference, cm 13.25 16.68 18.19  Total (both measures of obesity) 11.93 14.62 16.02 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice 1.45 0.43 0.75   Other fruit juiced 0.69 1.25 1.91   Fruit 39.21 23.07 6.68   Green salad 27.50 24.53 22.08   French fries/fried potatoes 40.31 42.09 30.07   Other potatoes 0.01 3.00 7.67   Other vegetablese 40.94 31.01 25.72  Fast-food consumption, times/week 43.06 53.12 51.95  Soft-drink consumption, times/week 24.22 33.94 41.56  Task eatingf 5.02 3.10 1.05  Total (all dietary behaviors) 22.24 21.55 18.94 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a The sample comprised 561 participants after exclusions; the test set was a random half (n = 280) of these participants. b Variables were log-transformed. c Weight (kg)/height (m)2. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. f Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). Table 5. Percentage of Variation in Dietary Behaviors and Obesity Measures Explained by Factors Extracted Using Different Statistical Methods in a Test Set of Participants From the Intensive Assessment Subsample,a PACE Study, 2005–2007 Measure of Obesity or Dietary Behaviorb Index Derivation Method Principal Components Regression Partial Least Squares Regression Reduced Rank Regression Measure of obesity  Body mass indexc 10.60 12.56 13.85  Waist circumference, cm 13.25 16.68 18.19  Total (both measures of obesity) 11.93 14.62 16.02 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice 1.45 0.43 0.75   Other fruit juiced 0.69 1.25 1.91   Fruit 39.21 23.07 6.68   Green salad 27.50 24.53 22.08   French fries/fried potatoes 40.31 42.09 30.07   Other potatoes 0.01 3.00 7.67   Other vegetablese 40.94 31.01 25.72  Fast-food consumption, times/week 43.06 53.12 51.95  Soft-drink consumption, times/week 24.22 33.94 41.56  Task eatingf 5.02 3.10 1.05  Total (all dietary behaviors) 22.24 21.55 18.94 Measure of Obesity or Dietary Behaviorb Index Derivation Method Principal Components Regression Partial Least Squares Regression Reduced Rank Regression Measure of obesity  Body mass indexc 10.60 12.56 13.85  Waist circumference, cm 13.25 16.68 18.19  Total (both measures of obesity) 11.93 14.62 16.02 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice 1.45 0.43 0.75   Other fruit juiced 0.69 1.25 1.91   Fruit 39.21 23.07 6.68   Green salad 27.50 24.53 22.08   French fries/fried potatoes 40.31 42.09 30.07   Other potatoes 0.01 3.00 7.67   Other vegetablese 40.94 31.01 25.72  Fast-food consumption, times/week 43.06 53.12 51.95  Soft-drink consumption, times/week 24.22 33.94 41.56  Task eatingf 5.02 3.10 1.05  Total (all dietary behaviors) 22.24 21.55 18.94 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a The sample comprised 561 participants after exclusions; the test set was a random half (n = 280) of these participants. b Variables were log-transformed. c Weight (kg)/height (m)2. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. f Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). Table 6. Model Effect Loadingsa of Dietary Behavior Variables for Factors Using Different Statistical Methods in a Test Set of Participants From the Intensive Assessment Subsampleb at Baseline, PACE Study, 2005–2007 Index Derivation Method Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingc 100% Fruit Juice Other Fruit Juiced Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablese PCR −0.4199 0.4257 −0.4290 0.4400 PLS regression 0.4419 0.4964 0.3968 RRR 0.3984 0.5237 0.4684 Index Derivation Method Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingc 100% Fruit Juice Other Fruit Juiced Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablese PCR −0.4199 0.4257 −0.4290 0.4400 PLS regression 0.4419 0.4964 0.3968 RRR 0.3984 0.5237 0.4684 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating; PCR, principal components regression; PLS, partial least squares; RRR, reduced rank regression. a Values less than |0.40| to 2 significant digits are not shown. b The sample comprised 561 participants after exclusions; the test set was a random half (n = 280) of these participants. c Eating while engaging in other activities. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. Table 6. Model Effect Loadingsa of Dietary Behavior Variables for Factors Using Different Statistical Methods in a Test Set of Participants From the Intensive Assessment Subsampleb at Baseline, PACE Study, 2005–2007 Index Derivation Method Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingc 100% Fruit Juice Other Fruit Juiced Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablese PCR −0.4199 0.4257 −0.4290 0.4400 PLS regression 0.4419 0.4964 0.3968 RRR 0.3984 0.5237 0.4684 Index Derivation Method Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingc 100% Fruit Juice Other Fruit Juiced Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablese PCR −0.4199 0.4257 −0.4290 0.4400 PLS regression 0.4419 0.4964 0.3968 RRR 0.3984 0.5237 0.4684 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating; PCR, principal components regression; PLS, partial least squares; RRR, reduced rank regression. a Values less than |0.40| to 2 significant digits are not shown. b The sample comprised 561 participants after exclusions; the test set was a random half (n = 280) of these participants. c Eating while engaging in other activities. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. We next performed confirmatory analyses within a validation sample where each index, created as the average of behavioral variables with loadings greater than 0.40 in the test set, was used separately to estimate BMI and waist circumference at baseline. Using the same observations within the validation set, AICs were generated for model comparisons to determine the best fit. The difference in AICs (ΔAIC) can be calculated by subtracting the minimum AIC value of all candidate models in the comparison set from each model AIC (27). Larger ΔAIC values (i.e., >10) indicate no support for the hypothesis that the model in question is the best-fitting model in the candidate set, while smaller ΔAIC values (i.e., <2) indicate strong support for the hypothesis that the model in question may also be the best-fitting model in the candidate set (27). As was suggested by the slightly higher explained variance among response variables, the model including the RRR/PLS-derived index had a substantially better goodness of fit when singly estimating BMI and waist circumference than the model including the PCR-derived index (Table 7). However, the fit of the model including the single fast-food variable was virtually indistinguishable from that of the model including the RRR/PLS-derived simple index. Table 7. Model Comparison of Dietary Behaviors and Indices Derived From Multiple Statistical Methods to Predict Baseline Response Variables in a Validation Set of Participants (n = 281) From the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Modeled Exposure AICa BMIb,c ΔAICBMI WCb, cm ΔAICWC Index derivation method  RRR/PLS regression −96.1588 Referent −223.7590 Referent  PCR −82.0188 14.1 −213.4112 10.3 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice −75.3211 20.8 −205.3858 18.4   Other fruit juiced −68.7927 27.4 −200.8212 22.9   Fruit −70.3788 25.8 −202.9168 20.8   Green salad −68.3496 27.8 −199.0820 24.7   French fries/fried potatoes −83.4799 12.7 −216.8664 6.9   Other potatoes −67.0699 29.1 −198.6125 25.1   Other vegetablese −69.9879 26.2 −201.5253 22.2  NCI fruit and vegetable intake, servings/dayf −74.6735 21.5 −206.0221 17.7  Fast-food consumption, times/week −94.7259 1.4 −221.7040 2.1  Soft-drink consumption, times/week −74.2355 21.9 −204.9203 21.8  Task eatingg −70.0333 26.1 −199.9747 23.8  All dietary behaviorsh −90.7357 5.4 −217.8623 5.9 Modeled Exposure AICa BMIb,c ΔAICBMI WCb, cm ΔAICWC Index derivation method  RRR/PLS regression −96.1588 Referent −223.7590 Referent  PCR −82.0188 14.1 −213.4112 10.3 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice −75.3211 20.8 −205.3858 18.4   Other fruit juiced −68.7927 27.4 −200.8212 22.9   Fruit −70.3788 25.8 −202.9168 20.8   Green salad −68.3496 27.8 −199.0820 24.7   French fries/fried potatoes −83.4799 12.7 −216.8664 6.9   Other potatoes −67.0699 29.1 −198.6125 25.1   Other vegetablese −69.9879 26.2 −201.5253 22.2  NCI fruit and vegetable intake, servings/dayf −74.6735 21.5 −206.0221 17.7  Fast-food consumption, times/week −94.7259 1.4 −221.7040 2.1  Soft-drink consumption, times/week −74.2355 21.9 −204.9203 21.8  Task eatingg −70.0333 26.1 −199.9747 23.8  All dietary behaviorsh −90.7357 5.4 −217.8623 5.9 Abbreviations: AIC, Akaike’s Information Criterion; ΔAIC, change in AIC; BMI, body mass index; FFQ, food frequency questionnaire; NCI, National Cancer Institute; PACE, Promoting Activity and Changes in Eating; PCR, principal components regression; PLS, partial least squares; RRR, reduced rank regression; WC, waist circumference. a AIC is a measure of model fit for nonnested models (using the same sample) employing likelihood maximization; smaller (e.g., more negative) values indicate a better fit. b BMI and WC were log-transformed in analyses; models adjusted for age, sex, and intervention arm. c Weight (kg)/height (m)2. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. f Includes 6 items from the abbreviated FFQ (i.e., excluding French fries/fried potatoes). g Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). h Includes 7 items from the abbreviated FFQ, plus fast-food meals, soft drinks, and task eating. Table 7. Model Comparison of Dietary Behaviors and Indices Derived From Multiple Statistical Methods to Predict Baseline Response Variables in a Validation Set of Participants (n = 281) From the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Modeled Exposure AICa BMIb,c ΔAICBMI WCb, cm ΔAICWC Index derivation method  RRR/PLS regression −96.1588 Referent −223.7590 Referent  PCR −82.0188 14.1 −213.4112 10.3 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice −75.3211 20.8 −205.3858 18.4   Other fruit juiced −68.7927 27.4 −200.8212 22.9   Fruit −70.3788 25.8 −202.9168 20.8   Green salad −68.3496 27.8 −199.0820 24.7   French fries/fried potatoes −83.4799 12.7 −216.8664 6.9   Other potatoes −67.0699 29.1 −198.6125 25.1   Other vegetablese −69.9879 26.2 −201.5253 22.2  NCI fruit and vegetable intake, servings/dayf −74.6735 21.5 −206.0221 17.7  Fast-food consumption, times/week −94.7259 1.4 −221.7040 2.1  Soft-drink consumption, times/week −74.2355 21.9 −204.9203 21.8  Task eatingg −70.0333 26.1 −199.9747 23.8  All dietary behaviorsh −90.7357 5.4 −217.8623 5.9 Modeled Exposure AICa BMIb,c ΔAICBMI WCb, cm ΔAICWC Index derivation method  RRR/PLS regression −96.1588 Referent −223.7590 Referent  PCR −82.0188 14.1 −213.4112 10.3 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice −75.3211 20.8 −205.3858 18.4   Other fruit juiced −68.7927 27.4 −200.8212 22.9   Fruit −70.3788 25.8 −202.9168 20.8   Green salad −68.3496 27.8 −199.0820 24.7   French fries/fried potatoes −83.4799 12.7 −216.8664 6.9   Other potatoes −67.0699 29.1 −198.6125 25.1   Other vegetablese −69.9879 26.2 −201.5253 22.2  NCI fruit and vegetable intake, servings/dayf −74.6735 21.5 −206.0221 17.7  Fast-food consumption, times/week −94.7259 1.4 −221.7040 2.1  Soft-drink consumption, times/week −74.2355 21.9 −204.9203 21.8  Task eatingg −70.0333 26.1 −199.9747 23.8  All dietary behaviorsh −90.7357 5.4 −217.8623 5.9 Abbreviations: AIC, Akaike’s Information Criterion; ΔAIC, change in AIC; BMI, body mass index; FFQ, food frequency questionnaire; NCI, National Cancer Institute; PACE, Promoting Activity and Changes in Eating; PCR, principal components regression; PLS, partial least squares; RRR, reduced rank regression; WC, waist circumference. a AIC is a measure of model fit for nonnested models (using the same sample) employing likelihood maximization; smaller (e.g., more negative) values indicate a better fit. b BMI and WC were log-transformed in analyses; models adjusted for age, sex, and intervention arm. c Weight (kg)/height (m)2. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. f Includes 6 items from the abbreviated FFQ (i.e., excluding French fries/fried potatoes). g Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). h Includes 7 items from the abbreviated FFQ, plus fast-food meals, soft drinks, and task eating. To further evaluate whether the dietary index was predictive of BMI and waist circumference, we used linear regression models to estimate predicted mean values and 95% confidence intervals at baseline (in the validation set only) and follow-up (in the nested cohort of participants with both baseline and follow-up data). Figure 1 presents cross-sectional associations between quartiles of the RRR/PLS-derived dietary index and baseline and follow-up BMI, while Figure 2 presents cross-sectional associations with baseline and follow-up measures of waist circumference. A 1-quartile higher dietary index score was associated with a statistically significant 5% higher BMI (ratio of geometric mean values per 1-quartile increase in dietary index = 1.053, 95% confidence interval (CI): 1.031, 1.075) and a 4% higher waist circumference (ratio = 1.036, 95% CI: 1.019, 1.054) at baseline. At follow-up, a 1-quartile higher score was associated with a statistically significant 6% higher BMI (ratio = 1.058, 95% CI: 1.029, 1.088) and a 5% higher waist circumference (ratio = 1.052, 95% CI: 1.032, 1.073). To determine whether the dietary index was predictive of obesity measures at follow-up after adjustment for corresponding baseline obesity measures, we used linear regression models to estimate ratios of geometric means per quartile increase in dietary index score (Table 8). Higher baseline dietary index score was associated with higher self-reported BMI and height-adjusted weight as well as measured height-adjusted weight and waist circumference at follow-up. Associations with measured BMI at follow-up were attenuated and rendered marginally nonsignificant after adjustment for baseline BMI. Figure 1. View largeDownload slide Quartiles (Qs) of reduced rank regression (RRR) dietary index score (DIS) according to measured body mass index (BMI; weight (kg)/height (m)2) at baseline (2005–2007) and follow-up (2007–2009) in the intensive assessment subsample of the PACE Study. Predicted mean values were estimated by means of linear mixed models that adjusted for age, sex, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity; BMI was log-transformed for analyses, and geometric mean values are presented. Analyses at baseline included a validation set of participants with nonmissing data at baseline (N = 34 work sites; n = 281 participants); analyses at follow-up included all participants with nonmissing data at baseline and follow-up (N = 27; n = 155). Results of tests for linear trend (Wald test) across quartiles of dietary index score were significant (P < 0.0001). Bars, 95% confidence intervals (CIs). PACE, Promoting Activity and Changes in Eating. Figure 1. View largeDownload slide Quartiles (Qs) of reduced rank regression (RRR) dietary index score (DIS) according to measured body mass index (BMI; weight (kg)/height (m)2) at baseline (2005–2007) and follow-up (2007–2009) in the intensive assessment subsample of the PACE Study. Predicted mean values were estimated by means of linear mixed models that adjusted for age, sex, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity; BMI was log-transformed for analyses, and geometric mean values are presented. Analyses at baseline included a validation set of participants with nonmissing data at baseline (N = 34 work sites; n = 281 participants); analyses at follow-up included all participants with nonmissing data at baseline and follow-up (N = 27; n = 155). Results of tests for linear trend (Wald test) across quartiles of dietary index score were significant (P < 0.0001). Bars, 95% confidence intervals (CIs). PACE, Promoting Activity and Changes in Eating. Figure 2. View largeDownload slide Quartiles (Qs) of reduced rank regression (RRR) dietary index score (DIS) according to measured waist circumference (WC; cm) at baseline (2005–2007) and follow-up (2007–2009) in the intensive assessment subsample of the PACE Study. Predicted mean values were estimated by means of linear mixed models that adjusted for age, sex, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity; waist circumference was log-transformed for analyses, and geometric mean values are presented. Analyses at baseline included a validation set of participants with nonmissing data at baseline (N = 34 work sites; n = 281 participants); analyses at follow-up included all participants with nonmissing data at baseline and follow-up (N = 27; n = 156). Results of tests for linear trend (Wald test) across quartiles of dietary index score were significant (P < 0.0001). Bars, 95% confidence intervals (CIs). PACE, Promoting Activity and Changes in Eating. Figure 2. View largeDownload slide Quartiles (Qs) of reduced rank regression (RRR) dietary index score (DIS) according to measured waist circumference (WC; cm) at baseline (2005–2007) and follow-up (2007–2009) in the intensive assessment subsample of the PACE Study. Predicted mean values were estimated by means of linear mixed models that adjusted for age, sex, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity; waist circumference was log-transformed for analyses, and geometric mean values are presented. Analyses at baseline included a validation set of participants with nonmissing data at baseline (N = 34 work sites; n = 281 participants); analyses at follow-up included all participants with nonmissing data at baseline and follow-up (N = 27; n = 156). Results of tests for linear trend (Wald test) across quartiles of dietary index score were significant (P < 0.0001). Bars, 95% confidence intervals (CIs). PACE, Promoting Activity and Changes in Eating. Table 8. Associations Between Quartiles of RRR-Derived Dietary Index Scores and Baseline-Adjusted Self-Reported and Objective Measures of Obesity at Follow-up Among Participants in the Intensive Assessment Subsample, PACE Study, 2007–2009a Measure of Obesity Model 1b Model 2c Ratio of GM Values 95% CI P for Trendd Ratio of GM Values 95% CI P for Trendd Body mass indexe  Self-reported 1.011 1.003, 1.019 0.007 1.012 1.004, 1.020 0.004  Measured 1.011 0.999, 1.022 0.075 1.011 0.999, 1.024 0.068 Weight, kgf  Self-reported 1.010 1.003, 1.018 0.009 1.011 1.003, 1.019 0.007  Measured 1.011 1.002, 1.020 0.021 1.012 1.003, 1.022 0.009 Waist circumference, cm 1.016 1.006, 1.026 0.001 1.017 1.007, 1.027 0.001 Measure of Obesity Model 1b Model 2c Ratio of GM Values 95% CI P for Trendd Ratio of GM Values 95% CI P for Trendd Body mass indexe  Self-reported 1.011 1.003, 1.019 0.007 1.012 1.004, 1.020 0.004  Measured 1.011 0.999, 1.022 0.075 1.011 0.999, 1.024 0.068 Weight, kgf  Self-reported 1.010 1.003, 1.018 0.009 1.011 1.003, 1.019 0.007  Measured 1.011 1.002, 1.020 0.021 1.012 1.003, 1.022 0.009 Waist circumference, cm 1.016 1.006, 1.026 0.001 1.017 1.007, 1.027 0.001 Abbreviations: GM, geometric mean; PACE, Promoting Activity and Changes in Eating; RRR, reduced rank regression. a The analysis included participants with measured body mass index (n = 155) and waist circumference (n = 156) at both baseline and follow-up. All dependent variables were log-transformed for analyses, and the ratio of geometric mean values per 1-quartile increase in dietary index is presented. b Results were adjusted for age, sex, the baseline measure, and intervention arm. c Results were adjusted for age, sex, the baseline measure, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity. d Wald test for linear trend with RRR-derived dietary index scores modeled as quartiles. e Weight (kg)/height (m)2. f Also adjusted for baseline height. Table 8. Associations Between Quartiles of RRR-Derived Dietary Index Scores and Baseline-Adjusted Self-Reported and Objective Measures of Obesity at Follow-up Among Participants in the Intensive Assessment Subsample, PACE Study, 2007–2009a Measure of Obesity Model 1b Model 2c Ratio of GM Values 95% CI P for Trendd Ratio of GM Values 95% CI P for Trendd Body mass indexe  Self-reported 1.011 1.003, 1.019 0.007 1.012 1.004, 1.020 0.004  Measured 1.011 0.999, 1.022 0.075 1.011 0.999, 1.024 0.068 Weight, kgf  Self-reported 1.010 1.003, 1.018 0.009 1.011 1.003, 1.019 0.007  Measured 1.011 1.002, 1.020 0.021 1.012 1.003, 1.022 0.009 Waist circumference, cm 1.016 1.006, 1.026 0.001 1.017 1.007, 1.027 0.001 Measure of Obesity Model 1b Model 2c Ratio of GM Values 95% CI P for Trendd Ratio of GM Values 95% CI P for Trendd Body mass indexe  Self-reported 1.011 1.003, 1.019 0.007 1.012 1.004, 1.020 0.004  Measured 1.011 0.999, 1.022 0.075 1.011 0.999, 1.024 0.068 Weight, kgf  Self-reported 1.010 1.003, 1.018 0.009 1.011 1.003, 1.019 0.007  Measured 1.011 1.002, 1.020 0.021 1.012 1.003, 1.022 0.009 Waist circumference, cm 1.016 1.006, 1.026 0.001 1.017 1.007, 1.027 0.001 Abbreviations: GM, geometric mean; PACE, Promoting Activity and Changes in Eating; RRR, reduced rank regression. a The analysis included participants with measured body mass index (n = 155) and waist circumference (n = 156) at both baseline and follow-up. All dependent variables were log-transformed for analyses, and the ratio of geometric mean values per 1-quartile increase in dietary index is presented. b Results were adjusted for age, sex, the baseline measure, and intervention arm. c Results were adjusted for age, sex, the baseline measure, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity. d Wald test for linear trend with RRR-derived dietary index scores modeled as quartiles. e Weight (kg)/height (m)2. f Also adjusted for baseline height. DISCUSSION We identified a linear combination of 3 obesogenic behaviors that formed an index associated with obesity. The dietary index was a simple average of the weekly frequency of consumption of 3 items: French fries, soft drinks, and fast-food meals. The index was found to explain the most variability in obesity (BMI and waist circumference) compared with the principal components method. This study builds on methods proposed by Hoffmann et al. (5) by extending them to dietary behaviors, beyond intake of specific food groups. This unique approach addressed several issues in the current body of nutritional epidemiology literature. First, dietary behaviors most related to 2 measures of obesity were identified. Second, a simple index of multiple dietary behaviors was identified which may prove to be a more reliable indicator of obesity risk than a method reliant on caloric estimation. A method using traditional dietary assessment methods and reliant on caloric estimation is subject to bias in reported energy and nutrient values (7). Our identified index was based on a single-factor solution using RRR and PLS methods that identified intake of fast food, French fries/fried potatoes, and soda as dietary behaviors that explained the most variation in both BMI and waist circumference, simultaneously, in a test set of participants. Furthermore, this index performed best in predictive models of measured BMI and waist circumference in a validation set of participants in comparison with other statistical methods based on principal components analysis. Specifically, a higher dietary index score was associated with higher measured BMI and waist circumference among adults at both baseline (in a validation set) and follow-up (among all participants with follow-up data). We found that this index was also associated with higher baseline-adjusted waist circumference after 2 years of follow-up. Our findings are somewhat consistent with those of studies individually linking intakes of fast food (9) and sugar-sweetened drinks (10) with obesity risk among adults. In a review which included 3 studies of fast-food intake and change in weight or BMI among adults, 2 found significant associations (9). Similarly to our study, the null study on BMI by Jeffery et al. (29) only assessed change over a relatively short time period (i.e., 1 year), whereas the 2 studies with positive findings evaluated change over a longer time period (i.e., 3–15 years) (23, 30). The 2 positive studies were also conducted among either younger individuals (30) or women only (23). While we did not find a significant baseline-adjusted association between the dietary behavior index and BMI at 2-year follow-up, we did see a significant association with baseline-adjusted waist circumference and height-adjusted weight. As of the date of this writing, we had not found other studies evaluating longitudinal associations between fast-food intake and waist circumference among adults. In a review which included 12 studies (7 prospective cohort studies and 5 clinical trials) of sweetened-drink intake and change in weight or BMI among adults, most found significant associations (10). Meta-analysis of prospective studies found that each serving/day increase in sweetened-drink intake was associated with an additional weight gain of 0.22 kg over 1 year (10). These findings were supported by a meta-analysis among trials which found a significant difference in change in body weight between intervention and control arms (10). Again, as of the date of this writing, we have not found other studies evaluating longitudinal associations between intake of sugar-sweetened drinks or nondiet soda and waist circumference among adults. On the other hand, Fowler et al. (31) recently reported a significant association between diet soda intake and increased waist circumference over a median follow-up time of 9 years. Strengths of this prospective study include use of a rigorous statistical approach to identify factors within a test set for prediction in a validation set of participants, as well as physical measurement of height, weight, and waist circumference. We also adjusted the model results for individual-level covariates while accounting for variation at the work-site level. Limitations include the small size of the nested cohort, which resulted in lower statistical power and which may not have been fully representative of the larger baseline sample. We also used a relatively small number of dietary behavior variables to generate indices. While we did choose a priori dietary behavior variables that have been associated with obesity in numerous studies, it is possible that we omitted some salient variables from our analyses. Despite these limitations, we were still able to detect significant associations with future waist circumference. This may be important, given that waist circumference has been shown to be more strongly associated with obesity-related disease risk (32). We must also acknowledge that the performance of the model for our dietary index could not be distinguished from that of a model including only our single fast-food variable. This may not be of concern, however, given that this variable also typically reflects intake of French fries and soft drinks as part of the frequency of fast-food meal consumption. Finally, while our methods do support the robustness of our findings, they may have limited generalizability given our sample of mostly white working adults. In summary, we have successfully identified a dietary index of 3 behaviors consistently associated with 2 measures of obesity. It is a simple average of the weekly frequency of consumption of French fries, soft drinks, and fast-food meals. Given the ubiquity of these dietary behaviors in the general population, the index may be useful in evaluating dietary interventions designed to reduce obesity risk. ACKNOWLEDGMENTS Author affiliations: Department of Psychosocial and Community Health, School of Nursing, University of Washington, Seattle, Washington (Wendy E. Barrington); Department of Epidemiology, School of Public Health, University of Washington, Seattle, Washington (Wendy E. Barrington, Shirley A. A. Beresford); and Cancer Prevention Program, Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, Washington (Wendy E. Barrington, Shirley A. A. Beresford). The Promoting Activity and Changes in Eating (PACE) Study was funded by the National Heart, Lung, and Blood Institute (grant R01 HL079491). We thank Sonia Bishop for her continued coordination of PACE Study efforts and Dale McLerran for his help with data management and his guidance in the planning of statistical analyses. Conflict of interest: none declared. Abbreviations AIC Akaike’s Information Criterion BMI body mass index CI confidence interval MET metabolic equivalent of task PACE Promoting Activity and Changes in Eating PCR principal components regression PLS partial least squares RRR reduced rank regression REFERENCES 1 Flegal KM , Kruszon-Moran D , Carroll MD , et al. . Trends in obesity among adults in the United States, 2005 to 2014 . JAMA . 2016 ; 315 ( 21 ): 2284 – 2291 . Google Scholar CrossRef Search ADS PubMed 2 Lemmens VE , Oenema A , Klepp KI , et al. . A systematic review of the evidence regarding efficacy of obesity prevention interventions among adults . Obes Rev . 2008 ; 9 ( 5 ): 446 – 455 . Google Scholar CrossRef Search ADS PubMed 3 Hu FB . Dietary pattern analysis: a new direction in nutritional epidemiology . Curr Opin Lipidol . 2002 ; 13 ( 1 ): 3 – 9 . Google Scholar CrossRef Search ADS PubMed 4 Kant AK . Dietary patterns and health outcomes . J Am Diet Assoc . 2004 ; 104 ( 4 ): 615 – 635 . Google Scholar CrossRef Search ADS PubMed 5 Hoffmann K , Schulze MB , Schienkiewitz A , et al. . Application of a new statistical method to derive dietary patterns in nutritional epidemiology . Am J Epidemiol . 2004 ; 159 ( 10 ): 935 – 944 . Google Scholar CrossRef Search ADS PubMed 6 Shmueli G . To explain or to predict? Stat Sci . 2010 ; 25 ( 3 ): 289 – 310 . Google Scholar CrossRef Search ADS 7 Neuhouser ML , Tinker L , Shaw PA , et al. . Use of recovery biomarkers to calibrate nutrient consumption self-reports in the Women’s Health Initiative . Am J Epidemiol . 2008 ; 167 ( 10 ): 1247 – 1259 . Google Scholar CrossRef Search ADS PubMed 8 Ledoux TA , Hingle MD , Baranowski T . Relationship of fruit and vegetable intake with adiposity: a systematic review . Obes Rev . 2011 ; 12 ( 5 ): e143 – e150 . Google Scholar CrossRef Search ADS PubMed 9 Bezerra IN , Curioni C , Sichieri R . Association between eating out of home and body weight . Nutr Rev . 2012 ; 70 ( 2 ): 65 – 79 . Google Scholar CrossRef Search ADS PubMed 10 Malik VS , Pan A , Willett WC , et al. . Sugar-sweetened beverages and weight gain in children and adults: a systematic review and meta-analysis . Am J Clin Nutr . 2013 ; 98 ( 4 ): 1084 – 1102 . Google Scholar CrossRef Search ADS PubMed 11 Lowe MR , Doshi SD , Katterman SN , et al. . Dieting and restrained eating as prospective predictors of weight gain . Front Psychol . 2013 ; 4 : 577 . Google Scholar CrossRef Search ADS PubMed 12 Scherwitz L , Kesten D . Seven eating styles linked to overeating, overweight, and obesity . Explore (NY) . 2005 ; 1 ( 5 ): 342 – 359 . Google Scholar CrossRef Search ADS PubMed 13 Liebman M , Pelican S , Moore SA , et al. . Dietary intake, eating behavior, and physical activity-related determinants of high body mass index in rural communities in Wyoming, Montana, and Idaho . Int J Obes Relat Metab Disord . 2003 ; 27 ( 6 ): 684 – 692 . Google Scholar CrossRef Search ADS PubMed 14 Kant AK . Dietary patterns and health outcomes . J Am Diet Assoc . 2004 ; 104 ( 4 ): 615 – 635 . Google Scholar CrossRef Search ADS PubMed 15 Kant AK . Dietary patterns: biomarkers and chronic disease risk . Appl Physiol Nutr Metab . 2010 ; 35 ( 2 ): 199 – 206 . Google Scholar CrossRef Search ADS PubMed 16 Bhupathiraju SN , Tucker KL . Coronary heart disease prevention: nutrients, foods, and dietary patterns . Clin Chim Acta . 2011 ; 412 ( 17-18 ): 1493 – 1514 . Google Scholar CrossRef Search ADS PubMed 17 Occupational Safety and Health Administration, US Department of Labor . Standard Industrial Classification (SIC) System Search. https://www.osha.gov/pls/imis/sicsearch.html. Accessed April 6, 2018. 18 Beresford SA , Locke E , Bishop S , et al. . Work site study Promoting Activity and Changes in Eating (PACE): design and baseline results . Obesity (Silver Spring) . 2007 ; 15 ( suppl 1 ): 4S – 15S . Google Scholar CrossRef Search ADS PubMed 19 Thompson FE , Midthune D , Subar AF , et al. . Dietary intake estimates in the National Health Interview Survey, 2000: methodology, results, and interpretation . J Am Diet Assoc . 2005 ; 105 ( 3 ): 352 – 363 . Google Scholar CrossRef Search ADS PubMed 20 Thompson FE , Byers T . Dietary assessment resource manual . J Nutr . 1994 ; 124 ( 11 suppl ): 2245S – 2317S . Google Scholar PubMed 21 Dave JM , An LC , Jeffery RW . Relationship of attitudes toward fast food and frequency of fast-food intake in adults . Obesity (Silver Spring) . 2009 ; 17 ( 6 ): 1164 – 1170 . Google Scholar PubMed 22 Rosenheck R . Fast food consumption and increased caloric intake: a systematic review of a trajectory towards weight gain and obesity risk . Obes Rev . 2008 ; 9 ( 6 ): 535 – 547 . Google Scholar CrossRef Search ADS PubMed 23 French SA , Harnack L , Jeffery RW . Fast food restaurant use among women in the Pound of Prevention study: dietary, behavioral and demographic correlates . Int J Obes Relat Metab Disord . 2000 ; 24 ( 10 ): 1353 – 1359 . Google Scholar CrossRef Search ADS PubMed 24 Framson C , Kristal A , Schenk JM , et al. . Development and validation of the mindful eating questionnaire . J Am Diet Assoc . 2009 ; 109 ( 8 ): 1439 – 1444 . Google Scholar CrossRef Search ADS PubMed 25 Wansink B . Environmental factors that increase the food intake and consumption volume of unknowing consumers . Annu Rev Nutr . 2004 ; 24 : 455 – 479 . Google Scholar CrossRef Search ADS PubMed 26 Godin G , Shephard RJ . A simple method to assess exercise behavior in the community . Can J Appl Sport Sci . 1985 ; 10 ( 3 ): 141 – 146 . Google Scholar PubMed 27 Burnham KP , Anderson DR . Multimodel inference: understanding AIC and BIC in model selection . Sociol Methods Res . 2004 ; 33 ( 2 ): 261 – 304 . Google Scholar CrossRef Search ADS 28 Amireault S , Godin G . The Godin-Shephard Leisure-Time Physical Activity Questionnaire: validity evidence supporting its use for classifying healthy adults into active and insufficiently active categories . Percept Mot Skills . 2015 ; 120 ( 2 ): 604 – 622 . Google Scholar CrossRef Search ADS PubMed 29 Jeffery RW , French SA . Epidemic obesity in the United States: are fast foods and television viewing contributing? Am J Public Health . 1998 ; 88 ( 2 ): 277 – 280 . Google Scholar CrossRef Search ADS PubMed 30 Pereira MA , Kartashov AI , Ebbeling CB , et al. . Fast-food habits, weight gain, and insulin resistance (the CARDIA study): 15-year prospective analysis . Lancet . 2005 ; 365 ( 9453 ): 36 – 42 . Google Scholar CrossRef Search ADS PubMed 31 Fowler SP , Williams K , Hazuda HP . Diet soda intake is associated with long-term increases in waist circumference in a biethnic cohort of older adults: the San Antonio Longitudinal Study of Aging . J Am Geriatr Soc . 2015 ; 63 ( 4 ): 708 – 715 . Google Scholar CrossRef Search ADS PubMed 32 Janssen I , Katzmarzyk PT , Ross R . Waist circumference and not body mass index explains obesity-related health risk . Am J Clin Nutr . 2004 ; 79 ( 3 ): 379 – 384 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png American Journal of Epidemiology Oxford University Press

Applying Multiple Statistical Methods to Derive an Index of Dietary Behaviors Most Related to Obesity

American Journal of Epidemiology , Volume Advance Article (7) – Mar 12, 2018

Loading next page...
 
/lp/oxford-university-press/applying-multiple-statistical-methods-to-derive-an-index-of-dietary-u2GKn0CH0S
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
0002-9262
eISSN
1476-6256
D.O.I.
10.1093/aje/kwx370
Publisher site
See Article on Publisher Site

Abstract

Abstract To evaluate the success of dietary interventions, we need measures that are more easily assessed and that closely align with intervention messaging. An index of obesogenic dietary behaviors (e.g., consumption of fast food and soft drinks, low fruit and vegetable consumption, and task eating (eating while engaging in other activities)) may serve this purpose and could be derived via data-driven methods typically used to describe nutrient intake. We used behavioral and physical measurement (i.e., body mass index, waist circumference) data from a subset of 2 independent cross-sectional samples of employees enrolled in the Promoting Activity and Changes in Eating (PACE) Study (Seattle, Washington) who were selected at baseline (2005–2007) (n = 561) and during follow-up (2007–2009) (n = 155). Index derivation methods, including principal components regression, partial least squares regression, and reduced rank regression, were compared. The best-fitting index for predicting physical measurements included consumption of fast food, French fries, and soft drinks. In linear mixed models, each 1-quartile increase in index score was associated with a 5% higher baseline body mass index (ratio of geometric means = 1.053, 95% confidence interval: 1.031, 1.075) and an approximately 4% higher baseline waist circumference (ratio = 1.036, 95% confidence interval: 1.019, 1.054) after adjustment for covariates. Results were similar at follow-up before and after adjustment for baseline measures. This index may be useful in evaluating public health or clinic-based dietary interventions to reduce obesity, especially given the ubiquity of these behaviors in the general population. dietary behaviors, obesity, partial least squares, principal components regression, reduced rank regression Obesity continues to be a significant threat to public health, with over one-third of US adults affected (1). Alarmingly, obesity prevention efforts, including dietary modification, have largely been unsuccessful in maintaining long-term weight reduction (2). An understanding of what dietary interventions are most effective is urgently needed, yet summarizing related evidence is hampered by differences in study design, study populations included, and dietary exposures evaluated (2). Dietary exposure variables used in intervention studies have mostly been based on patterns of nutrient intake identified via a priori combinations based on dietary recommendations or data reduction techniques from large epidemiologic studies collecting dietary information (3, 4). Traditionally, empirical derivation of dietary indices has involved the use of principal components or factor analysis to select factors which maximize the variance of the exposure (i.e., nutrient) variables (5). Therefore, such indices may maximally describe a combination of nutrients, but they may not be those most associated with the outcome of interest. Reduced rank regression (RRR) has been introduced as a method of selecting factors based on maximization of the variance of response variables that are highly correlated with or proximal to disease outcomes (5). While factors identified using RRR are likely to be more predictive of outcome variables by virtue of the predictive modeling methodology (6), evaluation of RRR in nutritional epidemiology has been limited. Partial least squares (PLS) regression is a method whereby identified factors balance the maximization of variance explained in both predictors and responses, and it is essentially a compromise between principal components regression (PCR) and RRR (5). Direct assessment of energy and nutrient intakes via self-report has been shown to be biased for each of the commonly used comprehensive intake assessment methods (7). Assessing dietary and eating behaviors at the same time may provide additional ways to estimate obesogenic behavior and may provide insight into the way in which high-energy foods are consumed. Several studies have shown that measures of adiposity, including body mass index (BMI), are associated with discrete dietary behaviors (including intakes of fruits and vegetables (8), fast food (9), and sugar-sweetened drinks (10)) and eating behaviors (including dieting (11), emotional eating (12), and task eating (eating while engaging in other activities) (13)). No one behavior, however, stands out as the most sensitive indicator. Modeling patterns of eating behavior may provide a more comprehensive way of looking at how diet is associated with obesity and disease (3, 14–16). In addition, many nutrients and foods are highly correlated, which makes it difficult to tease out separate associations. Evaluation of behaviors may reduce the effects of multicollinearity to some extent by reducing the sheer number of dietary variables. Identification of key behaviors most related to obesity may also provide more easily understood dietary recommendations for the lay public, as well as more actionable points of population-level obesity prevention and related chronic-disease-prevention efforts. The current study provided us with an opportunity to create an index of dietary behaviors to predictively model BMI and waist circumference in a population of mostly white-collar working adults. Our objective in this study, therefore, was to derive and evaluate indices of dietary and eating behaviors using various statistical methods (i.e., PCR, RRR, and PLS regression) to identify the combination of variables most predictive of obesity in a nested cohort of middle-aged US working adults. This index may ultimately identify persons at greater risk of obesity in both clinical and community settings. METHODS Study population The Promoting Activity and Changes in Eating (PACE) Study was a large group-randomized weight-reduction intervention trial carried out among approximately 3,000 individuals at 34 work sites in the Seattle, Washington, metropolitan area. Eligible companies employed 40–350 workers, were identified using US Standard Industrial Classification 2-digit codes (17), and included industries that were characterized as manufacturing, transportation or utilities, personal services, household and miscellaneous services, and nonclassifiable establishments. Eligibility criteria included having a high proportion of sedentary employees, having a low turnover rate during the previous 2 years, and having a low proportion of non-English-speaking employees. A detailed description of the PACE Study has been published elsewhere (18). All employees at participating work sites with fewer than 150 employees and a random subsample of 125 employees at work sites with more than 150 employees were asked to complete a standard questionnaire assessing self-reported dietary and physical activity behaviors, height, weight, and demographic information at baseline (2005–2007). An independent sample of employees, derived in the same way, were invited to complete the follow-up questionnaire (2007–2009). Data were collected among 3,054 individuals within 34 work sites at baseline and among 2,398 individuals within 33 work sites at follow-up. At baseline, a random “intensive assessment” subset of employees within all 34 work sites was also invited to provide additional physical measurements, including measured height, weight, and waist circumference (N = 34 work sites; n = 622 participants). For analyses at baseline, there were 561 participants after exclusion of persons with missing data for BMI (n = 4), waist circumference (n = 19), dietary behaviors (n = 27), and covariates (n = 11). At the follow-up intensive assessment, priority was given to participants who provided these measurements at baseline and who also were part of the follow-up survey. An additional random sample of follow-up survey respondents were invited to complete the intensive assessment. A total of 155 and 156 individuals provided baseline and follow-up measurements, respectively, for both BMI and waist circumference. Response variables Body mass index At both baseline and follow-up, height and weight were assessed via self-report and were measured by trained study personnel using a stadiometer and scale, respectively. BMI was calculated as weight (kg) divided by the square of height (m2) using both self-reported and physically measured data for these analyses. Waist circumference Waist circumference was assessed by study personnel via physical measurement during the same intensive assessment within work sites at baseline and follow-up. Values are reported in centimeters. Exposure variables Fruit and vegetable consumption Increased consumption of fruits and vegetables is a behavior promoted heavily by public health professionals, as it has been found to be inversely associated with obesity (19). Components of fruit and vegetable consumption were assessed using the National Cancer Institute’s 7-item 5-A-Day fruit and vegetable assessment tool (20). The total numbers of servings of fruit and vegetables consumed per day were calculated for descriptive purposes, whereas individual assessment items were included in dietary index analyses. Weekly frequency of fast-food meal consumption Frequency of fast-food meal consumption was assessed using a single question similar to that used in other studies: “Thinking about how often you eat out, how many times in a week or month do you eat breakfast, lunch, or dinner in a place such as McDonald’s®, Burger King®, Wendy’s®, Arby’s®, Pizza Hut®, or Kentucky Fried Chicken®?” (21, 22). Responses were given as number of times per week or number of times per month. All responses were converted to number of times per week. Weekly frequency of soft-drink consumption Average weekly soft-drink intake was also assessed, with the question “How often do you drink soft drinks or soda pop (regular or diet)?” (23). Response options were “never,” “less than once a week,” “about once a week,” “2–5 times per week,” “about once a day,” and “2 or more times per day.” Task eating (eating while engaged in other activities) The way in which food is consumed has also been associated with overweight and obesity (11–13). The “task eating” construct connotes a level of distraction, or lack of eating awareness or mindfulness, which has also been linked to obesity (24, 25). Task eating was assessed via a single item: “How often do you eat food (meals or snacks) while doing another activity—for example, watching TV, working at a computer, reading, driving, or playing video games?” (13). Response options were presented on a 5-point Likert scale ranging from 1 (“never”) to 5 (“always”). Covariates Individual-level factors included in the models were: age, sex, race/ethnicity (where “other” included Native Alaskan/American Indian and Pacific Islander/Native Hawaiian groups), and education. To further adjust for lifestyle differences, models also included adjustment for type of occupation (manual (i.e., machine operators, mechanics/technicians, service workers, tradesmen, and laborers) or nonmanual) and leisure-time physical activity of at least 10 minutes’ duration, assessed via the Godin-Shephard Leisure-Time Physical Activity Questionnaire (26). The Godin-Shephard questionnaire estimates the frequency of exercise bouts (i.e., vigorous, moderate, and light), multiplies each bout by the corresponding metabolic equivalent of task (MET) values (i.e., 9, 5, and 3 METs), and then sums these components to create an intensity-weighted score (i.e., leisure index score) that corresponds to a weekly MET frequency (MET-hours/week) (26). Statistical analyses Creating the dietary index A random half of participants (the test set) was selected at baseline, and 3 statistical methods (i.e., PCR, RRR, and PLS regression) were employed to identify a dietary index of reported dietary and eating behaviors using methodology described by Hoffmann et al. (5). Selection of factors using PCR is based on maximizing the percentage of variation explained among exposure variables (a partial goal of PLS regression and not a goal of RRR). Conversely, the selection of factors using RRR in the predictive model is based on maximizing the percentage of variation explained among response variables (a partial goal of PLS regression and not a goal of PCR). The baseline dietary exposure variables included in factor analyses were: all 7 individual items in the National Cancer Institute’s 5-A-Day fruit and vegetable screener (servings/day), frequency of consumption of fast-food meals (times/week), frequency of consumption of soft drinks (times/week), and frequency of eating while engaging in other activities (task eating; never, seldom, sometimes, most of the time, or always). Because the distributions of dietary and eating behaviors did not vary substantially, we chose not to standardize or transform the variables before factor analyses. The response variables included in factor analyses were measured BMI and waist circumference, since both were strongly correlated (r = 0.85); both variables were log-transformed to account for skewness before factor analyses. The PCR, RRR, and PLS methods were implemented using the PROC PLS command within SAS for Windows, release 9.4 (SAS Institute, Inc., Cary, North Carolina), with cross-validation to identify the minimum number of extracted factors for all methods. Validating the dietary index Dietary index scores for each statistical method were calculated as the average of variables identified in each extracted factor in the test set of participants at baseline. Models for prediction of baseline BMI and waist circumference employing these factors as well as all dietary behavior variables were compared for goodness of fit using Akaike’s Information Criterion (AIC) in the complement of the test set (validation set) of participants, where models with the smallest AIC indicated the best fit (27). The factor with the smallest AIC in separate models predicting BMI and waist circumference was selected for use in subsequent predictive models in the validation set at baseline, and again at follow-up. Longitudinal analyses predicting change in BMI and waist circumference were conducted in the nested cohort of baseline participants who had follow-up data. All predictive models were tested for interaction by sex; fully adjusted predictive models included age, sex, race/ethnicity, education, type of occupation (manual/nonmanual), leisure-time physical activity, and intervention arm as covariates. Wald tests were used to generate P values. All regression analyses were conducted using Stata SE, version 13.0 (StataCorp LP, College Station, Texas). RESULTS Baseline demographic and covariate data among the total, test, and validation samples are summarized in Table 1. The mean age was approximately 44 years, and participants were primarily white, had more than a high school education, worked in white- or pink-collar jobs (i.e., nonmanual occupations), and were sufficiently active according to an established cutpoint (leisure index score ≥24) for the Godin-Shephard Leisure-Time Physical Activity Questionnaire (28). Continuous and categorical measures of baseline response and exposure variables are presented in Tables 2 and 3, respectively. On average, participants were overweight/obese (mean BMI = 29.1), did not consume the recommended number of servings of fruit and vegetables per day (mean = 3.27 (standard error, 0.09) servings/day), and ate fast-food meals approximately 4 times per week (mean = 3.94 (standard error, 0.22) servings/week) (Table 2). Only about 19% and 4% of participants reported never drinking soda and never eating while engaging in other activities, respectively (Table 3). Table 4 presents Pearson’s correlation coefficients for all dietary behaviors (i.e., exposures) among both the test and validation sets of participants. Moderate positive correlations were evident between intakes of fast food, French fries/fried potatoes, and soft drinks, as well as between intakes of fruit, green salad, and vegetables (not including potatoes). Table 1. Demographic Characteristicsa of Participants in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Demographic Characteristic Total (N = 34; n = 561)b Test Set (N = 34; n = 280) Validation Set (N = 34; n = 281) No. % No. % No. % Age, yearsc 44.1 (0.5) 43.9 (0.7) 44.4 (0.7) Sex  Male 230 41.0 116 41.4 114 40.6  Female 331 59.0 164 58.6 167 59.4 Race/ethnicity  White 456 81.3 222 79.3 234 83.3  Hispanic/Latino 35 6.2 17 6.0 18 6.4  Black 28 5.0 15 5.4 13 4.6  Asian 24 4.3 15 5.4 9 3.2  Other 18 3.2 11 3.9 7 2.5 Education  Less than HS, HS diploma, or GED certificate 65 11.6 31 11.1 34 12.1  Some college or technical college 229 40.8 106 37.9 123 43.8  College graduate 193 34.4 102 36.4 91 32.4  Postgraduate study 74 13.2 41 14.6 33 11.7 Manual occupation  No 478 85.2 241 86.1 237 84.3  Yes 83 14.8 39 13.9 44 15.7 Leisure-time PA, MET-hours/weekc,d 28.9 (1.2) 29.6 (1.6) 28.2 (1.2) Demographic Characteristic Total (N = 34; n = 561)b Test Set (N = 34; n = 280) Validation Set (N = 34; n = 281) No. % No. % No. % Age, yearsc 44.1 (0.5) 43.9 (0.7) 44.4 (0.7) Sex  Male 230 41.0 116 41.4 114 40.6  Female 331 59.0 164 58.6 167 59.4 Race/ethnicity  White 456 81.3 222 79.3 234 83.3  Hispanic/Latino 35 6.2 17 6.0 18 6.4  Black 28 5.0 15 5.4 13 4.6  Asian 24 4.3 15 5.4 9 3.2  Other 18 3.2 11 3.9 7 2.5 Education  Less than HS, HS diploma, or GED certificate 65 11.6 31 11.1 34 12.1  Some college or technical college 229 40.8 106 37.9 123 43.8  College graduate 193 34.4 102 36.4 91 32.4  Postgraduate study 74 13.2 41 14.6 33 11.7 Manual occupation  No 478 85.2 241 86.1 237 84.3  Yes 83 14.8 39 13.9 44 15.7 Leisure-time PA, MET-hours/weekc,d 28.9 (1.2) 29.6 (1.6) 28.2 (1.2) Abbreviations: GED, General Educational Development; HS, high school; MET, metabolic equivalent of task; PA, physical activity; PACE, Promoting Activity and Changes in Eating. a Values presented are averaged work-site means unless otherwise indicated. bN, number of work sites; n, number of participants. c Values are expressed as mean (standard error). d Leisure-time physical activity of at least 10 minutes’ duration was assessed via the Godin-Shephard Leisure-Time Physical Activity Questionnaire (26). The questionnaire estimates the frequency of exercise bouts (i.e., vigorous, moderate, and light), multiplies each bout by the corresponding MET values (i.e., 9, 5, and 3 METs), and then sums these components to create an intensity-weighted score (i.e., leisure index score) that corresponds to a weekly MET frequency (MET-hours/week) (26). Table 1. Demographic Characteristicsa of Participants in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Demographic Characteristic Total (N = 34; n = 561)b Test Set (N = 34; n = 280) Validation Set (N = 34; n = 281) No. % No. % No. % Age, yearsc 44.1 (0.5) 43.9 (0.7) 44.4 (0.7) Sex  Male 230 41.0 116 41.4 114 40.6  Female 331 59.0 164 58.6 167 59.4 Race/ethnicity  White 456 81.3 222 79.3 234 83.3  Hispanic/Latino 35 6.2 17 6.0 18 6.4  Black 28 5.0 15 5.4 13 4.6  Asian 24 4.3 15 5.4 9 3.2  Other 18 3.2 11 3.9 7 2.5 Education  Less than HS, HS diploma, or GED certificate 65 11.6 31 11.1 34 12.1  Some college or technical college 229 40.8 106 37.9 123 43.8  College graduate 193 34.4 102 36.4 91 32.4  Postgraduate study 74 13.2 41 14.6 33 11.7 Manual occupation  No 478 85.2 241 86.1 237 84.3  Yes 83 14.8 39 13.9 44 15.7 Leisure-time PA, MET-hours/weekc,d 28.9 (1.2) 29.6 (1.6) 28.2 (1.2) Demographic Characteristic Total (N = 34; n = 561)b Test Set (N = 34; n = 280) Validation Set (N = 34; n = 281) No. % No. % No. % Age, yearsc 44.1 (0.5) 43.9 (0.7) 44.4 (0.7) Sex  Male 230 41.0 116 41.4 114 40.6  Female 331 59.0 164 58.6 167 59.4 Race/ethnicity  White 456 81.3 222 79.3 234 83.3  Hispanic/Latino 35 6.2 17 6.0 18 6.4  Black 28 5.0 15 5.4 13 4.6  Asian 24 4.3 15 5.4 9 3.2  Other 18 3.2 11 3.9 7 2.5 Education  Less than HS, HS diploma, or GED certificate 65 11.6 31 11.1 34 12.1  Some college or technical college 229 40.8 106 37.9 123 43.8  College graduate 193 34.4 102 36.4 91 32.4  Postgraduate study 74 13.2 41 14.6 33 11.7 Manual occupation  No 478 85.2 241 86.1 237 84.3  Yes 83 14.8 39 13.9 44 15.7 Leisure-time PA, MET-hours/weekc,d 28.9 (1.2) 29.6 (1.6) 28.2 (1.2) Abbreviations: GED, General Educational Development; HS, high school; MET, metabolic equivalent of task; PA, physical activity; PACE, Promoting Activity and Changes in Eating. a Values presented are averaged work-site means unless otherwise indicated. bN, number of work sites; n, number of participants. c Values are expressed as mean (standard error). d Leisure-time physical activity of at least 10 minutes’ duration was assessed via the Godin-Shephard Leisure-Time Physical Activity Questionnaire (26). The questionnaire estimates the frequency of exercise bouts (i.e., vigorous, moderate, and light), multiplies each bout by the corresponding MET values (i.e., 9, 5, and 3 METs), and then sums these components to create an intensity-weighted score (i.e., leisure index score) that corresponds to a weekly MET frequency (MET-hours/week) (26). Table 2. Mean Valuesa (and Standard Errors) for Continuous Measures of Obesity and Dietary Behaviors in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Measure of Obesity or Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) Continuous Measures of Obesity Physical measurements  Weight, kg 84.5 (0.9) 84.3 (1.3) 84.7 (1.2)  BMIb 29.1 (0.3) 29.0 (0.4) 29.2 (0.4)  Waist circumference, cm 90.5 (0.7) 90.4 (1.0) 90.6 (0.9) Self-reported measures  Weight, kg 82.4 (0.9) 82.3 (1.3) 82.5 (1.2)  BMI 28.1 (0.4) 28.1 (0.4) 28.2 (0.4) Continuous Dietary Behaviors Abbreviated FFQ intake, servings/day  100% fruit juice 0.29 (0.02) 0.33 (0.03) 0.25 (0.03)  Other fruit juicec 0.21 (0.03) 0.22 (0.03) 0.20 (0.03)  Fruit 1.05 (0.06) 1.11 (0.06) 0.99 (0.08)  Green salad 0.51 (0.02) 0.51 (0.03) 0.50 (0.03)  French fries/fried potatoes 0.12 (0.01) 0.12 (0.01) 0.12 (0.01)  Other potatoes 0.14 (0.01) 0.15 (0.01) 0.14 (0.01)  Other vegetablesd 1.09 (0.06) 1.06 (0.07) 1.11 (0.09) Fast-food consumption, times/week 3.94 (0.22) 3.95 (0.32) 3.94 (0.32) Measure of Obesity or Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) Continuous Measures of Obesity Physical measurements  Weight, kg 84.5 (0.9) 84.3 (1.3) 84.7 (1.2)  BMIb 29.1 (0.3) 29.0 (0.4) 29.2 (0.4)  Waist circumference, cm 90.5 (0.7) 90.4 (1.0) 90.6 (0.9) Self-reported measures  Weight, kg 82.4 (0.9) 82.3 (1.3) 82.5 (1.2)  BMI 28.1 (0.4) 28.1 (0.4) 28.2 (0.4) Continuous Dietary Behaviors Abbreviated FFQ intake, servings/day  100% fruit juice 0.29 (0.02) 0.33 (0.03) 0.25 (0.03)  Other fruit juicec 0.21 (0.03) 0.22 (0.03) 0.20 (0.03)  Fruit 1.05 (0.06) 1.11 (0.06) 0.99 (0.08)  Green salad 0.51 (0.02) 0.51 (0.03) 0.50 (0.03)  French fries/fried potatoes 0.12 (0.01) 0.12 (0.01) 0.12 (0.01)  Other potatoes 0.14 (0.01) 0.15 (0.01) 0.14 (0.01)  Other vegetablesd 1.09 (0.06) 1.06 (0.07) 1.11 (0.09) Fast-food consumption, times/week 3.94 (0.22) 3.95 (0.32) 3.94 (0.32) Abbreviations: BMI, body mass index; FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a Averaged work-site means. b Weight (kg)/height (m)2. c Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). d Does not include green salad or potatoes. Table 2. Mean Valuesa (and Standard Errors) for Continuous Measures of Obesity and Dietary Behaviors in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Measure of Obesity or Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) Continuous Measures of Obesity Physical measurements  Weight, kg 84.5 (0.9) 84.3 (1.3) 84.7 (1.2)  BMIb 29.1 (0.3) 29.0 (0.4) 29.2 (0.4)  Waist circumference, cm 90.5 (0.7) 90.4 (1.0) 90.6 (0.9) Self-reported measures  Weight, kg 82.4 (0.9) 82.3 (1.3) 82.5 (1.2)  BMI 28.1 (0.4) 28.1 (0.4) 28.2 (0.4) Continuous Dietary Behaviors Abbreviated FFQ intake, servings/day  100% fruit juice 0.29 (0.02) 0.33 (0.03) 0.25 (0.03)  Other fruit juicec 0.21 (0.03) 0.22 (0.03) 0.20 (0.03)  Fruit 1.05 (0.06) 1.11 (0.06) 0.99 (0.08)  Green salad 0.51 (0.02) 0.51 (0.03) 0.50 (0.03)  French fries/fried potatoes 0.12 (0.01) 0.12 (0.01) 0.12 (0.01)  Other potatoes 0.14 (0.01) 0.15 (0.01) 0.14 (0.01)  Other vegetablesd 1.09 (0.06) 1.06 (0.07) 1.11 (0.09) Fast-food consumption, times/week 3.94 (0.22) 3.95 (0.32) 3.94 (0.32) Measure of Obesity or Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) Continuous Measures of Obesity Physical measurements  Weight, kg 84.5 (0.9) 84.3 (1.3) 84.7 (1.2)  BMIb 29.1 (0.3) 29.0 (0.4) 29.2 (0.4)  Waist circumference, cm 90.5 (0.7) 90.4 (1.0) 90.6 (0.9) Self-reported measures  Weight, kg 82.4 (0.9) 82.3 (1.3) 82.5 (1.2)  BMI 28.1 (0.4) 28.1 (0.4) 28.2 (0.4) Continuous Dietary Behaviors Abbreviated FFQ intake, servings/day  100% fruit juice 0.29 (0.02) 0.33 (0.03) 0.25 (0.03)  Other fruit juicec 0.21 (0.03) 0.22 (0.03) 0.20 (0.03)  Fruit 1.05 (0.06) 1.11 (0.06) 0.99 (0.08)  Green salad 0.51 (0.02) 0.51 (0.03) 0.50 (0.03)  French fries/fried potatoes 0.12 (0.01) 0.12 (0.01) 0.12 (0.01)  Other potatoes 0.14 (0.01) 0.15 (0.01) 0.14 (0.01)  Other vegetablesd 1.09 (0.06) 1.06 (0.07) 1.11 (0.09) Fast-food consumption, times/week 3.94 (0.22) 3.95 (0.32) 3.94 (0.32) Abbreviations: BMI, body mass index; FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a Averaged work-site means. b Weight (kg)/height (m)2. c Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). d Does not include green salad or potatoes. Table 3. Distribution of Categorical Dietary Behaviors in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Measure of Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) No.a % No. % No. % Soft-drink consumption, times/week  0 107 19.0 53 18.9 54 19.2  0.5 129 23.0 65 23.3 64 22.8  1 76 13.6 37 13.2 39 13.9  3.5 106 18.9 53 18.9 53 18.9  7 79 14.1 39 13.9 40 14.2  14 64 11.4 33 11.8 31 11.0 Task eatingb  Never 22 3.9 8 2.9 14 5.0  Seldom 88 15.7 50 17.9 38 13.5  Sometimes 241 43.0 126 45.0 115 40.9  Most of the time 187 33.3 90 32.1 97 34.5  Always 23 4.1 6 2.1 17 6.1 Measure of Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) No.a % No. % No. % Soft-drink consumption, times/week  0 107 19.0 53 18.9 54 19.2  0.5 129 23.0 65 23.3 64 22.8  1 76 13.6 37 13.2 39 13.9  3.5 106 18.9 53 18.9 53 18.9  7 79 14.1 39 13.9 40 14.2  14 64 11.4 33 11.8 31 11.0 Task eatingb  Never 22 3.9 8 2.9 14 5.0  Seldom 88 15.7 50 17.9 38 13.5  Sometimes 241 43.0 126 45.0 115 40.9  Most of the time 187 33.3 90 32.1 97 34.5  Always 23 4.1 6 2.1 17 6.1 Abbreviation: PACE, Promoting Activity and Changes in Eating. a Number of participants. b Eating while engaging in other activities. Table 3. Distribution of Categorical Dietary Behaviors in the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Measure of Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) No.a % No. % No. % Soft-drink consumption, times/week  0 107 19.0 53 18.9 54 19.2  0.5 129 23.0 65 23.3 64 22.8  1 76 13.6 37 13.2 39 13.9  3.5 106 18.9 53 18.9 53 18.9  7 79 14.1 39 13.9 40 14.2  14 64 11.4 33 11.8 31 11.0 Task eatingb  Never 22 3.9 8 2.9 14 5.0  Seldom 88 15.7 50 17.9 38 13.5  Sometimes 241 43.0 126 45.0 115 40.9  Most of the time 187 33.3 90 32.1 97 34.5  Always 23 4.1 6 2.1 17 6.1 Measure of Dietary Behavior Total (n = 561) Test Set (n = 280) Validation Set (n = 281) No.a % No. % No. % Soft-drink consumption, times/week  0 107 19.0 53 18.9 54 19.2  0.5 129 23.0 65 23.3 64 22.8  1 76 13.6 37 13.2 39 13.9  3.5 106 18.9 53 18.9 53 18.9  7 79 14.1 39 13.9 40 14.2  14 64 11.4 33 11.8 31 11.0 Task eatingb  Never 22 3.9 8 2.9 14 5.0  Seldom 88 15.7 50 17.9 38 13.5  Sometimes 241 43.0 126 45.0 115 40.9  Most of the time 187 33.3 90 32.1 97 34.5  Always 23 4.1 6 2.1 17 6.1 Abbreviation: PACE, Promoting Activity and Changes in Eating. a Number of participants. b Eating while engaging in other activities. Table 4. Pearson’s Correlation Coefficients for Correlations Between Dietary Behaviors Among Baseline Test and Validation Sets of Participants From the Intensive Assessment Subsample,a PACE Study, 2005–2007 Dietary Behavior Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingb 100% Fruit Juice Other Fruit Juicec Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablesd Abbreviated FFQ intake, servings/day  100% fruit juice 0.18 0.16 0.08 −0.03 0.08 0.05 −0.08 −0.08 −0.08  Other fruit juicec 0.12 0.07 0.02 −0.05 0.11 −0.01 −0.01 0.00 0.10  Fruit 0.08 −0.09 0.31e −0.17 0.01 0.45e −0.24e −0.13 −0.08  Green salad 0.09 −0.08 0.23e −0.10 0.04 0.35e −0.16 −0.03 −0.09  French fries/fried potatoes 0.07 −0.02 −0.20e −0.18 0.21 −0.16 0.62e 0.17 0.13  Other potatoes 0.17 0.02 0.05 0.01 0.07 0.08 0.06 0.08 0.01  Other vegetablesd 0.08 −0.06 0.45e 0.27e −0.20e 0.03 −0.21 −0.07 0.00 Fast-food consumption, times/week 0.02 0.06 −0.17 −0.22 0.44e 0.08 −0.21 0.25e 0.12 Soft-drink consumption, times/week −0.05 −0.03 −0.18 −0.05 0.24e 0.08 −0.13 0.30e 0.10 Task eatingb −0.11 −0.14 −0.02 −0.08 0.13 −0.07 −0.04 0.07 0.15 Dietary Behavior Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingb 100% Fruit Juice Other Fruit Juicec Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablesd Abbreviated FFQ intake, servings/day  100% fruit juice 0.18 0.16 0.08 −0.03 0.08 0.05 −0.08 −0.08 −0.08  Other fruit juicec 0.12 0.07 0.02 −0.05 0.11 −0.01 −0.01 0.00 0.10  Fruit 0.08 −0.09 0.31e −0.17 0.01 0.45e −0.24e −0.13 −0.08  Green salad 0.09 −0.08 0.23e −0.10 0.04 0.35e −0.16 −0.03 −0.09  French fries/fried potatoes 0.07 −0.02 −0.20e −0.18 0.21 −0.16 0.62e 0.17 0.13  Other potatoes 0.17 0.02 0.05 0.01 0.07 0.08 0.06 0.08 0.01  Other vegetablesd 0.08 −0.06 0.45e 0.27e −0.20e 0.03 −0.21 −0.07 0.00 Fast-food consumption, times/week 0.02 0.06 −0.17 −0.22 0.44e 0.08 −0.21 0.25e 0.12 Soft-drink consumption, times/week −0.05 −0.03 −0.18 −0.05 0.24e 0.08 −0.13 0.30e 0.10 Task eatingb −0.11 −0.14 −0.02 −0.08 0.13 −0.07 −0.04 0.07 0.15 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a Correlation coefficients for the validation set (n = 281) are presented above the diagonal, and those for the test set (n = 280) are presented below the diagonal. b Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). c Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). d Does not include green salad or potatoes. e Correlation coefficient differed significantly from zero (P < 0.001). Table 4. Pearson’s Correlation Coefficients for Correlations Between Dietary Behaviors Among Baseline Test and Validation Sets of Participants From the Intensive Assessment Subsample,a PACE Study, 2005–2007 Dietary Behavior Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingb 100% Fruit Juice Other Fruit Juicec Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablesd Abbreviated FFQ intake, servings/day  100% fruit juice 0.18 0.16 0.08 −0.03 0.08 0.05 −0.08 −0.08 −0.08  Other fruit juicec 0.12 0.07 0.02 −0.05 0.11 −0.01 −0.01 0.00 0.10  Fruit 0.08 −0.09 0.31e −0.17 0.01 0.45e −0.24e −0.13 −0.08  Green salad 0.09 −0.08 0.23e −0.10 0.04 0.35e −0.16 −0.03 −0.09  French fries/fried potatoes 0.07 −0.02 −0.20e −0.18 0.21 −0.16 0.62e 0.17 0.13  Other potatoes 0.17 0.02 0.05 0.01 0.07 0.08 0.06 0.08 0.01  Other vegetablesd 0.08 −0.06 0.45e 0.27e −0.20e 0.03 −0.21 −0.07 0.00 Fast-food consumption, times/week 0.02 0.06 −0.17 −0.22 0.44e 0.08 −0.21 0.25e 0.12 Soft-drink consumption, times/week −0.05 −0.03 −0.18 −0.05 0.24e 0.08 −0.13 0.30e 0.10 Task eatingb −0.11 −0.14 −0.02 −0.08 0.13 −0.07 −0.04 0.07 0.15 Dietary Behavior Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingb 100% Fruit Juice Other Fruit Juicec Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablesd Abbreviated FFQ intake, servings/day  100% fruit juice 0.18 0.16 0.08 −0.03 0.08 0.05 −0.08 −0.08 −0.08  Other fruit juicec 0.12 0.07 0.02 −0.05 0.11 −0.01 −0.01 0.00 0.10  Fruit 0.08 −0.09 0.31e −0.17 0.01 0.45e −0.24e −0.13 −0.08  Green salad 0.09 −0.08 0.23e −0.10 0.04 0.35e −0.16 −0.03 −0.09  French fries/fried potatoes 0.07 −0.02 −0.20e −0.18 0.21 −0.16 0.62e 0.17 0.13  Other potatoes 0.17 0.02 0.05 0.01 0.07 0.08 0.06 0.08 0.01  Other vegetablesd 0.08 −0.06 0.45e 0.27e −0.20e 0.03 −0.21 −0.07 0.00 Fast-food consumption, times/week 0.02 0.06 −0.17 −0.22 0.44e 0.08 −0.21 0.25e 0.12 Soft-drink consumption, times/week −0.05 −0.03 −0.18 −0.05 0.24e 0.08 −0.13 0.30e 0.10 Task eatingb −0.11 −0.14 −0.02 −0.08 0.13 −0.07 −0.04 0.07 0.15 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a Correlation coefficients for the validation set (n = 281) are presented above the diagonal, and those for the test set (n = 280) are presented below the diagonal. b Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). c Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). d Does not include green salad or potatoes. e Correlation coefficient differed significantly from zero (P < 0.001). The statistical methods PCR, PLS regression, and RRR were employed to explain variation in measures of obesity (i.e., responses) and dietary behavior variables (i.e., exposures) for a test set of individuals who did not having missing information for any of these variables at baseline. Cross-validation was employed, and a single factor solution was identified for each method. As expected, the factor selected via the PCR method explained the highest proportion of variance among exposure variables, whereas the factor selected by the RRR method explained the highest proportion of variance among response variables (Table 5). The proportions of variance explained among exposures and responses by the PLS factor were a compromise between the other two methods. Differences in proportions of variance explained between methods, however, were small. Model effect loadings for each dietary behavior variable that were greater than the absolute value of 0.4 are presented for each factor derived from various statistical methods in Table 6. Notably, the RRR-derived factor and PLS-derived factors included the same dietary behavior variables (i.e., intake of fast-food meals, soda, and French fries/fried potatoes). The proposed simple index is the average of the weekly frequency of each of these 3 behaviors. Table 5. Percentage of Variation in Dietary Behaviors and Obesity Measures Explained by Factors Extracted Using Different Statistical Methods in a Test Set of Participants From the Intensive Assessment Subsample,a PACE Study, 2005–2007 Measure of Obesity or Dietary Behaviorb Index Derivation Method Principal Components Regression Partial Least Squares Regression Reduced Rank Regression Measure of obesity  Body mass indexc 10.60 12.56 13.85  Waist circumference, cm 13.25 16.68 18.19  Total (both measures of obesity) 11.93 14.62 16.02 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice 1.45 0.43 0.75   Other fruit juiced 0.69 1.25 1.91   Fruit 39.21 23.07 6.68   Green salad 27.50 24.53 22.08   French fries/fried potatoes 40.31 42.09 30.07   Other potatoes 0.01 3.00 7.67   Other vegetablese 40.94 31.01 25.72  Fast-food consumption, times/week 43.06 53.12 51.95  Soft-drink consumption, times/week 24.22 33.94 41.56  Task eatingf 5.02 3.10 1.05  Total (all dietary behaviors) 22.24 21.55 18.94 Measure of Obesity or Dietary Behaviorb Index Derivation Method Principal Components Regression Partial Least Squares Regression Reduced Rank Regression Measure of obesity  Body mass indexc 10.60 12.56 13.85  Waist circumference, cm 13.25 16.68 18.19  Total (both measures of obesity) 11.93 14.62 16.02 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice 1.45 0.43 0.75   Other fruit juiced 0.69 1.25 1.91   Fruit 39.21 23.07 6.68   Green salad 27.50 24.53 22.08   French fries/fried potatoes 40.31 42.09 30.07   Other potatoes 0.01 3.00 7.67   Other vegetablese 40.94 31.01 25.72  Fast-food consumption, times/week 43.06 53.12 51.95  Soft-drink consumption, times/week 24.22 33.94 41.56  Task eatingf 5.02 3.10 1.05  Total (all dietary behaviors) 22.24 21.55 18.94 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a The sample comprised 561 participants after exclusions; the test set was a random half (n = 280) of these participants. b Variables were log-transformed. c Weight (kg)/height (m)2. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. f Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). Table 5. Percentage of Variation in Dietary Behaviors and Obesity Measures Explained by Factors Extracted Using Different Statistical Methods in a Test Set of Participants From the Intensive Assessment Subsample,a PACE Study, 2005–2007 Measure of Obesity or Dietary Behaviorb Index Derivation Method Principal Components Regression Partial Least Squares Regression Reduced Rank Regression Measure of obesity  Body mass indexc 10.60 12.56 13.85  Waist circumference, cm 13.25 16.68 18.19  Total (both measures of obesity) 11.93 14.62 16.02 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice 1.45 0.43 0.75   Other fruit juiced 0.69 1.25 1.91   Fruit 39.21 23.07 6.68   Green salad 27.50 24.53 22.08   French fries/fried potatoes 40.31 42.09 30.07   Other potatoes 0.01 3.00 7.67   Other vegetablese 40.94 31.01 25.72  Fast-food consumption, times/week 43.06 53.12 51.95  Soft-drink consumption, times/week 24.22 33.94 41.56  Task eatingf 5.02 3.10 1.05  Total (all dietary behaviors) 22.24 21.55 18.94 Measure of Obesity or Dietary Behaviorb Index Derivation Method Principal Components Regression Partial Least Squares Regression Reduced Rank Regression Measure of obesity  Body mass indexc 10.60 12.56 13.85  Waist circumference, cm 13.25 16.68 18.19  Total (both measures of obesity) 11.93 14.62 16.02 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice 1.45 0.43 0.75   Other fruit juiced 0.69 1.25 1.91   Fruit 39.21 23.07 6.68   Green salad 27.50 24.53 22.08   French fries/fried potatoes 40.31 42.09 30.07   Other potatoes 0.01 3.00 7.67   Other vegetablese 40.94 31.01 25.72  Fast-food consumption, times/week 43.06 53.12 51.95  Soft-drink consumption, times/week 24.22 33.94 41.56  Task eatingf 5.02 3.10 1.05  Total (all dietary behaviors) 22.24 21.55 18.94 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating. a The sample comprised 561 participants after exclusions; the test set was a random half (n = 280) of these participants. b Variables were log-transformed. c Weight (kg)/height (m)2. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. f Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). Table 6. Model Effect Loadingsa of Dietary Behavior Variables for Factors Using Different Statistical Methods in a Test Set of Participants From the Intensive Assessment Subsampleb at Baseline, PACE Study, 2005–2007 Index Derivation Method Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingc 100% Fruit Juice Other Fruit Juiced Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablese PCR −0.4199 0.4257 −0.4290 0.4400 PLS regression 0.4419 0.4964 0.3968 RRR 0.3984 0.5237 0.4684 Index Derivation Method Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingc 100% Fruit Juice Other Fruit Juiced Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablese PCR −0.4199 0.4257 −0.4290 0.4400 PLS regression 0.4419 0.4964 0.3968 RRR 0.3984 0.5237 0.4684 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating; PCR, principal components regression; PLS, partial least squares; RRR, reduced rank regression. a Values less than |0.40| to 2 significant digits are not shown. b The sample comprised 561 participants after exclusions; the test set was a random half (n = 280) of these participants. c Eating while engaging in other activities. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. Table 6. Model Effect Loadingsa of Dietary Behavior Variables for Factors Using Different Statistical Methods in a Test Set of Participants From the Intensive Assessment Subsampleb at Baseline, PACE Study, 2005–2007 Index Derivation Method Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingc 100% Fruit Juice Other Fruit Juiced Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablese PCR −0.4199 0.4257 −0.4290 0.4400 PLS regression 0.4419 0.4964 0.3968 RRR 0.3984 0.5237 0.4684 Index Derivation Method Dietary Behavior Abbreviated FFQ Intake, servings/day Fast-Food Consumption, times/week Soft-Drink Consumption, times/week Task Eatingc 100% Fruit Juice Other Fruit Juiced Fruit Green Salad French Fries/Fried Potatoes Other Potatoes Other Vegetablese PCR −0.4199 0.4257 −0.4290 0.4400 PLS regression 0.4419 0.4964 0.3968 RRR 0.3984 0.5237 0.4684 Abbreviations: FFQ, food frequency questionnaire; PACE, Promoting Activity and Changes in Eating; PCR, principal components regression; PLS, partial least squares; RRR, reduced rank regression. a Values less than |0.40| to 2 significant digits are not shown. b The sample comprised 561 participants after exclusions; the test set was a random half (n = 280) of these participants. c Eating while engaging in other activities. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. We next performed confirmatory analyses within a validation sample where each index, created as the average of behavioral variables with loadings greater than 0.40 in the test set, was used separately to estimate BMI and waist circumference at baseline. Using the same observations within the validation set, AICs were generated for model comparisons to determine the best fit. The difference in AICs (ΔAIC) can be calculated by subtracting the minimum AIC value of all candidate models in the comparison set from each model AIC (27). Larger ΔAIC values (i.e., >10) indicate no support for the hypothesis that the model in question is the best-fitting model in the candidate set, while smaller ΔAIC values (i.e., <2) indicate strong support for the hypothesis that the model in question may also be the best-fitting model in the candidate set (27). As was suggested by the slightly higher explained variance among response variables, the model including the RRR/PLS-derived index had a substantially better goodness of fit when singly estimating BMI and waist circumference than the model including the PCR-derived index (Table 7). However, the fit of the model including the single fast-food variable was virtually indistinguishable from that of the model including the RRR/PLS-derived simple index. Table 7. Model Comparison of Dietary Behaviors and Indices Derived From Multiple Statistical Methods to Predict Baseline Response Variables in a Validation Set of Participants (n = 281) From the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Modeled Exposure AICa BMIb,c ΔAICBMI WCb, cm ΔAICWC Index derivation method  RRR/PLS regression −96.1588 Referent −223.7590 Referent  PCR −82.0188 14.1 −213.4112 10.3 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice −75.3211 20.8 −205.3858 18.4   Other fruit juiced −68.7927 27.4 −200.8212 22.9   Fruit −70.3788 25.8 −202.9168 20.8   Green salad −68.3496 27.8 −199.0820 24.7   French fries/fried potatoes −83.4799 12.7 −216.8664 6.9   Other potatoes −67.0699 29.1 −198.6125 25.1   Other vegetablese −69.9879 26.2 −201.5253 22.2  NCI fruit and vegetable intake, servings/dayf −74.6735 21.5 −206.0221 17.7  Fast-food consumption, times/week −94.7259 1.4 −221.7040 2.1  Soft-drink consumption, times/week −74.2355 21.9 −204.9203 21.8  Task eatingg −70.0333 26.1 −199.9747 23.8  All dietary behaviorsh −90.7357 5.4 −217.8623 5.9 Modeled Exposure AICa BMIb,c ΔAICBMI WCb, cm ΔAICWC Index derivation method  RRR/PLS regression −96.1588 Referent −223.7590 Referent  PCR −82.0188 14.1 −213.4112 10.3 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice −75.3211 20.8 −205.3858 18.4   Other fruit juiced −68.7927 27.4 −200.8212 22.9   Fruit −70.3788 25.8 −202.9168 20.8   Green salad −68.3496 27.8 −199.0820 24.7   French fries/fried potatoes −83.4799 12.7 −216.8664 6.9   Other potatoes −67.0699 29.1 −198.6125 25.1   Other vegetablese −69.9879 26.2 −201.5253 22.2  NCI fruit and vegetable intake, servings/dayf −74.6735 21.5 −206.0221 17.7  Fast-food consumption, times/week −94.7259 1.4 −221.7040 2.1  Soft-drink consumption, times/week −74.2355 21.9 −204.9203 21.8  Task eatingg −70.0333 26.1 −199.9747 23.8  All dietary behaviorsh −90.7357 5.4 −217.8623 5.9 Abbreviations: AIC, Akaike’s Information Criterion; ΔAIC, change in AIC; BMI, body mass index; FFQ, food frequency questionnaire; NCI, National Cancer Institute; PACE, Promoting Activity and Changes in Eating; PCR, principal components regression; PLS, partial least squares; RRR, reduced rank regression; WC, waist circumference. a AIC is a measure of model fit for nonnested models (using the same sample) employing likelihood maximization; smaller (e.g., more negative) values indicate a better fit. b BMI and WC were log-transformed in analyses; models adjusted for age, sex, and intervention arm. c Weight (kg)/height (m)2. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. f Includes 6 items from the abbreviated FFQ (i.e., excluding French fries/fried potatoes). g Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). h Includes 7 items from the abbreviated FFQ, plus fast-food meals, soft drinks, and task eating. Table 7. Model Comparison of Dietary Behaviors and Indices Derived From Multiple Statistical Methods to Predict Baseline Response Variables in a Validation Set of Participants (n = 281) From the Intensive Assessment Subsample at Baseline, PACE Study, 2005–2007 Modeled Exposure AICa BMIb,c ΔAICBMI WCb, cm ΔAICWC Index derivation method  RRR/PLS regression −96.1588 Referent −223.7590 Referent  PCR −82.0188 14.1 −213.4112 10.3 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice −75.3211 20.8 −205.3858 18.4   Other fruit juiced −68.7927 27.4 −200.8212 22.9   Fruit −70.3788 25.8 −202.9168 20.8   Green salad −68.3496 27.8 −199.0820 24.7   French fries/fried potatoes −83.4799 12.7 −216.8664 6.9   Other potatoes −67.0699 29.1 −198.6125 25.1   Other vegetablese −69.9879 26.2 −201.5253 22.2  NCI fruit and vegetable intake, servings/dayf −74.6735 21.5 −206.0221 17.7  Fast-food consumption, times/week −94.7259 1.4 −221.7040 2.1  Soft-drink consumption, times/week −74.2355 21.9 −204.9203 21.8  Task eatingg −70.0333 26.1 −199.9747 23.8  All dietary behaviorsh −90.7357 5.4 −217.8623 5.9 Modeled Exposure AICa BMIb,c ΔAICBMI WCb, cm ΔAICWC Index derivation method  RRR/PLS regression −96.1588 Referent −223.7590 Referent  PCR −82.0188 14.1 −213.4112 10.3 Dietary behavior  Abbreviated FFQ intake, servings/day   100% fruit juice −75.3211 20.8 −205.3858 18.4   Other fruit juiced −68.7927 27.4 −200.8212 22.9   Fruit −70.3788 25.8 −202.9168 20.8   Green salad −68.3496 27.8 −199.0820 24.7   French fries/fried potatoes −83.4799 12.7 −216.8664 6.9   Other potatoes −67.0699 29.1 −198.6125 25.1   Other vegetablese −69.9879 26.2 −201.5253 22.2  NCI fruit and vegetable intake, servings/dayf −74.6735 21.5 −206.0221 17.7  Fast-food consumption, times/week −94.7259 1.4 −221.7040 2.1  Soft-drink consumption, times/week −74.2355 21.9 −204.9203 21.8  Task eatingg −70.0333 26.1 −199.9747 23.8  All dietary behaviorsh −90.7357 5.4 −217.8623 5.9 Abbreviations: AIC, Akaike’s Information Criterion; ΔAIC, change in AIC; BMI, body mass index; FFQ, food frequency questionnaire; NCI, National Cancer Institute; PACE, Promoting Activity and Changes in Eating; PCR, principal components regression; PLS, partial least squares; RRR, reduced rank regression; WC, waist circumference. a AIC is a measure of model fit for nonnested models (using the same sample) employing likelihood maximization; smaller (e.g., more negative) values indicate a better fit. b BMI and WC were log-transformed in analyses; models adjusted for age, sex, and intervention arm. c Weight (kg)/height (m)2. d Does not include juice drinks or Kool-Aid (Kraft Foods Group, Inc., Northfield, Illinois). e Does not include green salad or potatoes. f Includes 6 items from the abbreviated FFQ (i.e., excluding French fries/fried potatoes). g Eating while engaging in other activities (5-point Likert scale: 1 = never, 2 = seldom, 3 = sometimes, 4 = most of the time, 5 = always). h Includes 7 items from the abbreviated FFQ, plus fast-food meals, soft drinks, and task eating. To further evaluate whether the dietary index was predictive of BMI and waist circumference, we used linear regression models to estimate predicted mean values and 95% confidence intervals at baseline (in the validation set only) and follow-up (in the nested cohort of participants with both baseline and follow-up data). Figure 1 presents cross-sectional associations between quartiles of the RRR/PLS-derived dietary index and baseline and follow-up BMI, while Figure 2 presents cross-sectional associations with baseline and follow-up measures of waist circumference. A 1-quartile higher dietary index score was associated with a statistically significant 5% higher BMI (ratio of geometric mean values per 1-quartile increase in dietary index = 1.053, 95% confidence interval (CI): 1.031, 1.075) and a 4% higher waist circumference (ratio = 1.036, 95% CI: 1.019, 1.054) at baseline. At follow-up, a 1-quartile higher score was associated with a statistically significant 6% higher BMI (ratio = 1.058, 95% CI: 1.029, 1.088) and a 5% higher waist circumference (ratio = 1.052, 95% CI: 1.032, 1.073). To determine whether the dietary index was predictive of obesity measures at follow-up after adjustment for corresponding baseline obesity measures, we used linear regression models to estimate ratios of geometric means per quartile increase in dietary index score (Table 8). Higher baseline dietary index score was associated with higher self-reported BMI and height-adjusted weight as well as measured height-adjusted weight and waist circumference at follow-up. Associations with measured BMI at follow-up were attenuated and rendered marginally nonsignificant after adjustment for baseline BMI. Figure 1. View largeDownload slide Quartiles (Qs) of reduced rank regression (RRR) dietary index score (DIS) according to measured body mass index (BMI; weight (kg)/height (m)2) at baseline (2005–2007) and follow-up (2007–2009) in the intensive assessment subsample of the PACE Study. Predicted mean values were estimated by means of linear mixed models that adjusted for age, sex, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity; BMI was log-transformed for analyses, and geometric mean values are presented. Analyses at baseline included a validation set of participants with nonmissing data at baseline (N = 34 work sites; n = 281 participants); analyses at follow-up included all participants with nonmissing data at baseline and follow-up (N = 27; n = 155). Results of tests for linear trend (Wald test) across quartiles of dietary index score were significant (P < 0.0001). Bars, 95% confidence intervals (CIs). PACE, Promoting Activity and Changes in Eating. Figure 1. View largeDownload slide Quartiles (Qs) of reduced rank regression (RRR) dietary index score (DIS) according to measured body mass index (BMI; weight (kg)/height (m)2) at baseline (2005–2007) and follow-up (2007–2009) in the intensive assessment subsample of the PACE Study. Predicted mean values were estimated by means of linear mixed models that adjusted for age, sex, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity; BMI was log-transformed for analyses, and geometric mean values are presented. Analyses at baseline included a validation set of participants with nonmissing data at baseline (N = 34 work sites; n = 281 participants); analyses at follow-up included all participants with nonmissing data at baseline and follow-up (N = 27; n = 155). Results of tests for linear trend (Wald test) across quartiles of dietary index score were significant (P < 0.0001). Bars, 95% confidence intervals (CIs). PACE, Promoting Activity and Changes in Eating. Figure 2. View largeDownload slide Quartiles (Qs) of reduced rank regression (RRR) dietary index score (DIS) according to measured waist circumference (WC; cm) at baseline (2005–2007) and follow-up (2007–2009) in the intensive assessment subsample of the PACE Study. Predicted mean values were estimated by means of linear mixed models that adjusted for age, sex, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity; waist circumference was log-transformed for analyses, and geometric mean values are presented. Analyses at baseline included a validation set of participants with nonmissing data at baseline (N = 34 work sites; n = 281 participants); analyses at follow-up included all participants with nonmissing data at baseline and follow-up (N = 27; n = 156). Results of tests for linear trend (Wald test) across quartiles of dietary index score were significant (P < 0.0001). Bars, 95% confidence intervals (CIs). PACE, Promoting Activity and Changes in Eating. Figure 2. View largeDownload slide Quartiles (Qs) of reduced rank regression (RRR) dietary index score (DIS) according to measured waist circumference (WC; cm) at baseline (2005–2007) and follow-up (2007–2009) in the intensive assessment subsample of the PACE Study. Predicted mean values were estimated by means of linear mixed models that adjusted for age, sex, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity; waist circumference was log-transformed for analyses, and geometric mean values are presented. Analyses at baseline included a validation set of participants with nonmissing data at baseline (N = 34 work sites; n = 281 participants); analyses at follow-up included all participants with nonmissing data at baseline and follow-up (N = 27; n = 156). Results of tests for linear trend (Wald test) across quartiles of dietary index score were significant (P < 0.0001). Bars, 95% confidence intervals (CIs). PACE, Promoting Activity and Changes in Eating. Table 8. Associations Between Quartiles of RRR-Derived Dietary Index Scores and Baseline-Adjusted Self-Reported and Objective Measures of Obesity at Follow-up Among Participants in the Intensive Assessment Subsample, PACE Study, 2007–2009a Measure of Obesity Model 1b Model 2c Ratio of GM Values 95% CI P for Trendd Ratio of GM Values 95% CI P for Trendd Body mass indexe  Self-reported 1.011 1.003, 1.019 0.007 1.012 1.004, 1.020 0.004  Measured 1.011 0.999, 1.022 0.075 1.011 0.999, 1.024 0.068 Weight, kgf  Self-reported 1.010 1.003, 1.018 0.009 1.011 1.003, 1.019 0.007  Measured 1.011 1.002, 1.020 0.021 1.012 1.003, 1.022 0.009 Waist circumference, cm 1.016 1.006, 1.026 0.001 1.017 1.007, 1.027 0.001 Measure of Obesity Model 1b Model 2c Ratio of GM Values 95% CI P for Trendd Ratio of GM Values 95% CI P for Trendd Body mass indexe  Self-reported 1.011 1.003, 1.019 0.007 1.012 1.004, 1.020 0.004  Measured 1.011 0.999, 1.022 0.075 1.011 0.999, 1.024 0.068 Weight, kgf  Self-reported 1.010 1.003, 1.018 0.009 1.011 1.003, 1.019 0.007  Measured 1.011 1.002, 1.020 0.021 1.012 1.003, 1.022 0.009 Waist circumference, cm 1.016 1.006, 1.026 0.001 1.017 1.007, 1.027 0.001 Abbreviations: GM, geometric mean; PACE, Promoting Activity and Changes in Eating; RRR, reduced rank regression. a The analysis included participants with measured body mass index (n = 155) and waist circumference (n = 156) at both baseline and follow-up. All dependent variables were log-transformed for analyses, and the ratio of geometric mean values per 1-quartile increase in dietary index is presented. b Results were adjusted for age, sex, the baseline measure, and intervention arm. c Results were adjusted for age, sex, the baseline measure, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity. d Wald test for linear trend with RRR-derived dietary index scores modeled as quartiles. e Weight (kg)/height (m)2. f Also adjusted for baseline height. Table 8. Associations Between Quartiles of RRR-Derived Dietary Index Scores and Baseline-Adjusted Self-Reported and Objective Measures of Obesity at Follow-up Among Participants in the Intensive Assessment Subsample, PACE Study, 2007–2009a Measure of Obesity Model 1b Model 2c Ratio of GM Values 95% CI P for Trendd Ratio of GM Values 95% CI P for Trendd Body mass indexe  Self-reported 1.011 1.003, 1.019 0.007 1.012 1.004, 1.020 0.004  Measured 1.011 0.999, 1.022 0.075 1.011 0.999, 1.024 0.068 Weight, kgf  Self-reported 1.010 1.003, 1.018 0.009 1.011 1.003, 1.019 0.007  Measured 1.011 1.002, 1.020 0.021 1.012 1.003, 1.022 0.009 Waist circumference, cm 1.016 1.006, 1.026 0.001 1.017 1.007, 1.027 0.001 Measure of Obesity Model 1b Model 2c Ratio of GM Values 95% CI P for Trendd Ratio of GM Values 95% CI P for Trendd Body mass indexe  Self-reported 1.011 1.003, 1.019 0.007 1.012 1.004, 1.020 0.004  Measured 1.011 0.999, 1.022 0.075 1.011 0.999, 1.024 0.068 Weight, kgf  Self-reported 1.010 1.003, 1.018 0.009 1.011 1.003, 1.019 0.007  Measured 1.011 1.002, 1.020 0.021 1.012 1.003, 1.022 0.009 Waist circumference, cm 1.016 1.006, 1.026 0.001 1.017 1.007, 1.027 0.001 Abbreviations: GM, geometric mean; PACE, Promoting Activity and Changes in Eating; RRR, reduced rank regression. a The analysis included participants with measured body mass index (n = 155) and waist circumference (n = 156) at both baseline and follow-up. All dependent variables were log-transformed for analyses, and the ratio of geometric mean values per 1-quartile increase in dietary index is presented. b Results were adjusted for age, sex, the baseline measure, and intervention arm. c Results were adjusted for age, sex, the baseline measure, intervention arm, race/ethnicity, education, type of occupation (manual/nonmanual), and physical activity. d Wald test for linear trend with RRR-derived dietary index scores modeled as quartiles. e Weight (kg)/height (m)2. f Also adjusted for baseline height. DISCUSSION We identified a linear combination of 3 obesogenic behaviors that formed an index associated with obesity. The dietary index was a simple average of the weekly frequency of consumption of 3 items: French fries, soft drinks, and fast-food meals. The index was found to explain the most variability in obesity (BMI and waist circumference) compared with the principal components method. This study builds on methods proposed by Hoffmann et al. (5) by extending them to dietary behaviors, beyond intake of specific food groups. This unique approach addressed several issues in the current body of nutritional epidemiology literature. First, dietary behaviors most related to 2 measures of obesity were identified. Second, a simple index of multiple dietary behaviors was identified which may prove to be a more reliable indicator of obesity risk than a method reliant on caloric estimation. A method using traditional dietary assessment methods and reliant on caloric estimation is subject to bias in reported energy and nutrient values (7). Our identified index was based on a single-factor solution using RRR and PLS methods that identified intake of fast food, French fries/fried potatoes, and soda as dietary behaviors that explained the most variation in both BMI and waist circumference, simultaneously, in a test set of participants. Furthermore, this index performed best in predictive models of measured BMI and waist circumference in a validation set of participants in comparison with other statistical methods based on principal components analysis. Specifically, a higher dietary index score was associated with higher measured BMI and waist circumference among adults at both baseline (in a validation set) and follow-up (among all participants with follow-up data). We found that this index was also associated with higher baseline-adjusted waist circumference after 2 years of follow-up. Our findings are somewhat consistent with those of studies individually linking intakes of fast food (9) and sugar-sweetened drinks (10) with obesity risk among adults. In a review which included 3 studies of fast-food intake and change in weight or BMI among adults, 2 found significant associations (9). Similarly to our study, the null study on BMI by Jeffery et al. (29) only assessed change over a relatively short time period (i.e., 1 year), whereas the 2 studies with positive findings evaluated change over a longer time period (i.e., 3–15 years) (23, 30). The 2 positive studies were also conducted among either younger individuals (30) or women only (23). While we did not find a significant baseline-adjusted association between the dietary behavior index and BMI at 2-year follow-up, we did see a significant association with baseline-adjusted waist circumference and height-adjusted weight. As of the date of this writing, we had not found other studies evaluating longitudinal associations between fast-food intake and waist circumference among adults. In a review which included 12 studies (7 prospective cohort studies and 5 clinical trials) of sweetened-drink intake and change in weight or BMI among adults, most found significant associations (10). Meta-analysis of prospective studies found that each serving/day increase in sweetened-drink intake was associated with an additional weight gain of 0.22 kg over 1 year (10). These findings were supported by a meta-analysis among trials which found a significant difference in change in body weight between intervention and control arms (10). Again, as of the date of this writing, we have not found other studies evaluating longitudinal associations between intake of sugar-sweetened drinks or nondiet soda and waist circumference among adults. On the other hand, Fowler et al. (31) recently reported a significant association between diet soda intake and increased waist circumference over a median follow-up time of 9 years. Strengths of this prospective study include use of a rigorous statistical approach to identify factors within a test set for prediction in a validation set of participants, as well as physical measurement of height, weight, and waist circumference. We also adjusted the model results for individual-level covariates while accounting for variation at the work-site level. Limitations include the small size of the nested cohort, which resulted in lower statistical power and which may not have been fully representative of the larger baseline sample. We also used a relatively small number of dietary behavior variables to generate indices. While we did choose a priori dietary behavior variables that have been associated with obesity in numerous studies, it is possible that we omitted some salient variables from our analyses. Despite these limitations, we were still able to detect significant associations with future waist circumference. This may be important, given that waist circumference has been shown to be more strongly associated with obesity-related disease risk (32). We must also acknowledge that the performance of the model for our dietary index could not be distinguished from that of a model including only our single fast-food variable. This may not be of concern, however, given that this variable also typically reflects intake of French fries and soft drinks as part of the frequency of fast-food meal consumption. Finally, while our methods do support the robustness of our findings, they may have limited generalizability given our sample of mostly white working adults. In summary, we have successfully identified a dietary index of 3 behaviors consistently associated with 2 measures of obesity. It is a simple average of the weekly frequency of consumption of French fries, soft drinks, and fast-food meals. Given the ubiquity of these dietary behaviors in the general population, the index may be useful in evaluating dietary interventions designed to reduce obesity risk. ACKNOWLEDGMENTS Author affiliations: Department of Psychosocial and Community Health, School of Nursing, University of Washington, Seattle, Washington (Wendy E. Barrington); Department of Epidemiology, School of Public Health, University of Washington, Seattle, Washington (Wendy E. Barrington, Shirley A. A. Beresford); and Cancer Prevention Program, Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, Washington (Wendy E. Barrington, Shirley A. A. Beresford). The Promoting Activity and Changes in Eating (PACE) Study was funded by the National Heart, Lung, and Blood Institute (grant R01 HL079491). We thank Sonia Bishop for her continued coordination of PACE Study efforts and Dale McLerran for his help with data management and his guidance in the planning of statistical analyses. Conflict of interest: none declared. Abbreviations AIC Akaike’s Information Criterion BMI body mass index CI confidence interval MET metabolic equivalent of task PACE Promoting Activity and Changes in Eating PCR principal components regression PLS partial least squares RRR reduced rank regression REFERENCES 1 Flegal KM , Kruszon-Moran D , Carroll MD , et al. . Trends in obesity among adults in the United States, 2005 to 2014 . JAMA . 2016 ; 315 ( 21 ): 2284 – 2291 . Google Scholar CrossRef Search ADS PubMed 2 Lemmens VE , Oenema A , Klepp KI , et al. . A systematic review of the evidence regarding efficacy of obesity prevention interventions among adults . Obes Rev . 2008 ; 9 ( 5 ): 446 – 455 . Google Scholar CrossRef Search ADS PubMed 3 Hu FB . Dietary pattern analysis: a new direction in nutritional epidemiology . Curr Opin Lipidol . 2002 ; 13 ( 1 ): 3 – 9 . Google Scholar CrossRef Search ADS PubMed 4 Kant AK . Dietary patterns and health outcomes . J Am Diet Assoc . 2004 ; 104 ( 4 ): 615 – 635 . Google Scholar CrossRef Search ADS PubMed 5 Hoffmann K , Schulze MB , Schienkiewitz A , et al. . Application of a new statistical method to derive dietary patterns in nutritional epidemiology . Am J Epidemiol . 2004 ; 159 ( 10 ): 935 – 944 . Google Scholar CrossRef Search ADS PubMed 6 Shmueli G . To explain or to predict? Stat Sci . 2010 ; 25 ( 3 ): 289 – 310 . Google Scholar CrossRef Search ADS 7 Neuhouser ML , Tinker L , Shaw PA , et al. . Use of recovery biomarkers to calibrate nutrient consumption self-reports in the Women’s Health Initiative . Am J Epidemiol . 2008 ; 167 ( 10 ): 1247 – 1259 . Google Scholar CrossRef Search ADS PubMed 8 Ledoux TA , Hingle MD , Baranowski T . Relationship of fruit and vegetable intake with adiposity: a systematic review . Obes Rev . 2011 ; 12 ( 5 ): e143 – e150 . Google Scholar CrossRef Search ADS PubMed 9 Bezerra IN , Curioni C , Sichieri R . Association between eating out of home and body weight . Nutr Rev . 2012 ; 70 ( 2 ): 65 – 79 . Google Scholar CrossRef Search ADS PubMed 10 Malik VS , Pan A , Willett WC , et al. . Sugar-sweetened beverages and weight gain in children and adults: a systematic review and meta-analysis . Am J Clin Nutr . 2013 ; 98 ( 4 ): 1084 – 1102 . Google Scholar CrossRef Search ADS PubMed 11 Lowe MR , Doshi SD , Katterman SN , et al. . Dieting and restrained eating as prospective predictors of weight gain . Front Psychol . 2013 ; 4 : 577 . Google Scholar CrossRef Search ADS PubMed 12 Scherwitz L , Kesten D . Seven eating styles linked to overeating, overweight, and obesity . Explore (NY) . 2005 ; 1 ( 5 ): 342 – 359 . Google Scholar CrossRef Search ADS PubMed 13 Liebman M , Pelican S , Moore SA , et al. . Dietary intake, eating behavior, and physical activity-related determinants of high body mass index in rural communities in Wyoming, Montana, and Idaho . Int J Obes Relat Metab Disord . 2003 ; 27 ( 6 ): 684 – 692 . Google Scholar CrossRef Search ADS PubMed 14 Kant AK . Dietary patterns and health outcomes . J Am Diet Assoc . 2004 ; 104 ( 4 ): 615 – 635 . Google Scholar CrossRef Search ADS PubMed 15 Kant AK . Dietary patterns: biomarkers and chronic disease risk . Appl Physiol Nutr Metab . 2010 ; 35 ( 2 ): 199 – 206 . Google Scholar CrossRef Search ADS PubMed 16 Bhupathiraju SN , Tucker KL . Coronary heart disease prevention: nutrients, foods, and dietary patterns . Clin Chim Acta . 2011 ; 412 ( 17-18 ): 1493 – 1514 . Google Scholar CrossRef Search ADS PubMed 17 Occupational Safety and Health Administration, US Department of Labor . Standard Industrial Classification (SIC) System Search. https://www.osha.gov/pls/imis/sicsearch.html. Accessed April 6, 2018. 18 Beresford SA , Locke E , Bishop S , et al. . Work site study Promoting Activity and Changes in Eating (PACE): design and baseline results . Obesity (Silver Spring) . 2007 ; 15 ( suppl 1 ): 4S – 15S . Google Scholar CrossRef Search ADS PubMed 19 Thompson FE , Midthune D , Subar AF , et al. . Dietary intake estimates in the National Health Interview Survey, 2000: methodology, results, and interpretation . J Am Diet Assoc . 2005 ; 105 ( 3 ): 352 – 363 . Google Scholar CrossRef Search ADS PubMed 20 Thompson FE , Byers T . Dietary assessment resource manual . J Nutr . 1994 ; 124 ( 11 suppl ): 2245S – 2317S . Google Scholar PubMed 21 Dave JM , An LC , Jeffery RW . Relationship of attitudes toward fast food and frequency of fast-food intake in adults . Obesity (Silver Spring) . 2009 ; 17 ( 6 ): 1164 – 1170 . Google Scholar PubMed 22 Rosenheck R . Fast food consumption and increased caloric intake: a systematic review of a trajectory towards weight gain and obesity risk . Obes Rev . 2008 ; 9 ( 6 ): 535 – 547 . Google Scholar CrossRef Search ADS PubMed 23 French SA , Harnack L , Jeffery RW . Fast food restaurant use among women in the Pound of Prevention study: dietary, behavioral and demographic correlates . Int J Obes Relat Metab Disord . 2000 ; 24 ( 10 ): 1353 – 1359 . Google Scholar CrossRef Search ADS PubMed 24 Framson C , Kristal A , Schenk JM , et al. . Development and validation of the mindful eating questionnaire . J Am Diet Assoc . 2009 ; 109 ( 8 ): 1439 – 1444 . Google Scholar CrossRef Search ADS PubMed 25 Wansink B . Environmental factors that increase the food intake and consumption volume of unknowing consumers . Annu Rev Nutr . 2004 ; 24 : 455 – 479 . Google Scholar CrossRef Search ADS PubMed 26 Godin G , Shephard RJ . A simple method to assess exercise behavior in the community . Can J Appl Sport Sci . 1985 ; 10 ( 3 ): 141 – 146 . Google Scholar PubMed 27 Burnham KP , Anderson DR . Multimodel inference: understanding AIC and BIC in model selection . Sociol Methods Res . 2004 ; 33 ( 2 ): 261 – 304 . Google Scholar CrossRef Search ADS 28 Amireault S , Godin G . The Godin-Shephard Leisure-Time Physical Activity Questionnaire: validity evidence supporting its use for classifying healthy adults into active and insufficiently active categories . Percept Mot Skills . 2015 ; 120 ( 2 ): 604 – 622 . Google Scholar CrossRef Search ADS PubMed 29 Jeffery RW , French SA . Epidemic obesity in the United States: are fast foods and television viewing contributing? Am J Public Health . 1998 ; 88 ( 2 ): 277 – 280 . Google Scholar CrossRef Search ADS PubMed 30 Pereira MA , Kartashov AI , Ebbeling CB , et al. . Fast-food habits, weight gain, and insulin resistance (the CARDIA study): 15-year prospective analysis . Lancet . 2005 ; 365 ( 9453 ): 36 – 42 . Google Scholar CrossRef Search ADS PubMed 31 Fowler SP , Williams K , Hazuda HP . Diet soda intake is associated with long-term increases in waist circumference in a biethnic cohort of older adults: the San Antonio Longitudinal Study of Aging . J Am Geriatr Soc . 2015 ; 63 ( 4 ): 708 – 715 . Google Scholar CrossRef Search ADS PubMed 32 Janssen I , Katzmarzyk PT , Ross R . Waist circumference and not body mass index explains obesity-related health risk . Am J Clin Nutr . 2004 ; 79 ( 3 ): 379 – 384 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

American Journal of EpidemiologyOxford University Press

Published: Mar 12, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off