STRESS AND PSYCHOLOGICAL CONSTRUCTS RELATED TO EATING BEHAVIOR ARE ASSOCIATED WITH ANTHROPOMETRY AND BODY COMPOSITION IN YOUNG ADULTS

Abstract

Background

The transition to college is associated with weight gain, but the relation between eating behavior indicators and anthropometric outcomes during this period remains unclear.

Objective

Evaluate sex differences in stress, emotional eating, tendency to overeat, and restrained eating behavior, and determine whether the psycho-behavioral constructs assessed immediately prior to starting college are associated with anthropometry and adiposity at the start of college, and with first-semester weight gain.

Methods

A prospective study administered the Three Factor Eating Questionnaire (TFEQ), Satter Eating Competence Inventory, and Perceived Stress Scale (PSS) to 264 participants one month before college. Body composition was assessed via dual energy X-ray absorptiometry (DXA) at the start of college, and anthropometry (weight, height, waist circumference[WC]) was collected at the beginning and end of the first semester. Ordinary least squares regression tested the cross-sectional association of baseline psychological and behavioral scales with baseline DXA and anthropometry, and the longitudinal association with change in anthropometry.

Results

Among 264 participants, 91% (241) had baseline data, and 66% (173) completed follow-up. In sex-adjusted linear regression models, baseline TFEQ disinhibited and emotional (DE;EE) eating sub-scales were positively associated with baseline weight (P=0.003;DE, P=0.014;EE), body mass index (BMI, P=0.002;DE, P=0.001;EE), WC (P=0.004;DE, P=0.006;EE) and DXA fat mass index (P=0.023;DE, P=0.014;EE). Baseline PSS was positively associated with subsequent changes in weight and WC among males only (weight P_interaction=0.0268;WC P_interaction=0.0017).

Conclusion

College freshmen with questionnaire scores indicating a greater tendency to overeat in response to external cues and emotions tended to have greater weight, BMI, and WC at the start of college. Males with higher perceived stress at college entrance subsequently gained significantly more weight in the first semester, but no such relation was evident in females.

Introduction

Behavioral Risk Factor Surveillance System (BRFSS) data report that approximately 42% of young adults aged 18-24 years are overweight or obese¹. The transition from high school to college is a critical period for establishing health-related behaviors, and about 18 million students attend American colleges and universities annually as undergraduates². The tendency to gain weight during college is well-documented^3–6; about 60–75% of first-year students gain an average of 1.4 to 1.8 kg^6–8, and weight gain early in college is associated with weight gain over all years⁹. First year weight gain leads to an average increase of about 1% in total body adiposity over the college experience⁵. Limiting adiposity gain, even in young adults with normal BMI, may be important to prevent adverse changes in cardio-metabolic biomarkers¹⁰.

Previous studies documented multiple determinants of freshman weight gain, including environmental and lifestyle factors such as access to on-campus dining facilities^{11, 12}, physical activity¹³, alcohol consumption^{14, 15}, snacking habits¹², dieting behavior¹⁶, sex^{7, 8, 17–22}, race/ethnicity²³, and residence type^{24, 25}. Psychosocial factors may also play a role in freshman weight gain; prior cross-sectional studies in freshman identified associations between psychological constructs, including dietary restraint and the behavioral tendency to overeat due to emotional cues or stress, and freshman weight gain^26–28. Previous qualitative research of the psycho-behavioral factors related to eating in college students reported perceived stress and emotional eating to be associated with overeating among women, while boredom and anxiety were key factors associated with overeating in men²⁹. Past studies took widely varying approaches to measuring psychological and behavioral indicators (choice of constructs, how constructs were assessed) and to the analytic approach (descriptive versus hypothesis-driven). Inferences from these studies are limited because they often focused on a single sex^{26, 30}, included only a small sample^31–33, and/or used only self-reported anthropometry data^{17, 28, 34–36}.

Given the transition to college is characterized by changing attitudes and interpersonal influences⁴, we hypothesized that higher stress, less competent eating, greater disinhibited eating, and greater emotional eating would each be associated with subsequent greater weight gain over the first semester of college. Since some psycho-behavioral attributes vary by sex, we also investigated whether the associations between psycho-behavioral constructs and weight change differed by sex. We also estimated the cross-sectional association of the psycho-behavioral constructs with body composition and anthropometry at study baseline with the objective of providing comparisons to past studies and improving on existing findings through dual energy X-ray absorptiometry measurements of body habitus.

Methods

Study Design and Participants

We conducted a longitudinal prospective cohort study over the first year of college (July 2011-June 2012) in participants at a large university in the northeastern United States. 1001 students ≥ 18 years of age were identified via stratified random sampling and invited to participate via email. Approximately half (N ∼ 500) accessed online materials describing the study, and a further half (N = 264) enrolled and participated in the baseline data collection. All participants provided written informed consent and the study methods were approved by the Cornell University Institutional Review Board for Human Participants Research.

Physical Measures

Trained personnel measured anthropometry (weight, height, waist circumference [WC], hip circumference) within the first three days on campus (late August) and at the end of the first semester (early December). Before each data collection, anthropometrists were trained to meet proficiency standards for accuracy and reliability³⁷ such that mean circumference measurements collected by every anthropometrist from several test subjects were <0.5 cm from the mean of measurements collected by the lead expert anthropometrist at the training held before each data collection. Training continued until proficiency was established. Measurements were collected using calibrated devices, participants wore minimal, light clothing at each visit, and data comprised the average of two repeated measurements of height (stadiometer; Shorr Productions, Olney, MD) and weight (digital scale; Seca, Chino, CA), and three repeated measurements of waist and hip circumferences (steel measuring tape; Lufkin, Apex, NC). At the beginning of the semester, a licensed radiologic technician conducted whole body scans using dual energy X-ray absorptiometry (DXA, QDR4500 fan beam densitometer, Hologic Inc., Bedford, MA, pooled precision 1.16%) to determine baseline total body fat percent (BF%), truncal fat percent (TF%), and fat mass index [FMI, the ratio of fat mass (kg) to height (m²)].

Physical Activity and Dietary Indicators

Within approximately one month prior to arriving on campus, all participants completed self-administered electronic questionnaires to assess physical activity (Global Physical Activity Questionnaire, GPAQ³⁸) and usual diet (Diet History Questionnaire, DHQII³⁹). The GPAQ assessed the intensity, duration, and frequency of physical activity during work, recreation, and transportation in a recent typical week and was administered at the beginning and end of the semester. Self-reported physical activity was assigned a metabolic equivalent (MET) value⁴⁰ per level of intensity to obtain a composite metric of physical activity, reflecting both duration and intensity of physical activity. 1 MET corresponded to energy expenditure at rest, 4-6.9 METs to moderate intensity physical activity, and ≥7 METs to vigorous activity. MET hours/day in each activity and MET hours/day at each level of intensity were calculated.

The DHQII food frequency questionnaire (FFQ) assessed usual consumption of 134 food and beverage items, based on standard portion sizes, over the prior year. In a separate questionnaire, also administered within the month prior to the start of the academic term, participants self-reported the typical number of daily meals and snacks they consumed.

In addition to self-reported dietary intake data, objective assessments of dining hall use were collected throughout the semester. Freshmen lived on-campus and were obligated by the University to enroll in a meal plan for the on-campus dining halls. To enter these all-you-care-to-eat dining environments, the student’s ID card was ‘swiped’ through an electronic card reader, yielding data on card swipes during the semester (e.g. average count of dining hall visits per day and the total cumulative count of dining hall visits during the semester) and the total number of days the ID card was swiped at least once during the fall semester, as objective indicators of eating behavior.

Psycho-behavioral Indicators

Psychological constructs related to eating were also assessed by questionnaires within one month prior to the start of classes and at end of the first semester including the Satter Eating Competence Inventory⁴¹, Perceived Stress Scale⁴², and the Three Factor Eating Questionnaire⁴³. The total score for the Satter Eating Competence Inventory (ecSI), a 16-item validated questionnaire^{41, 44} with excellent test-retest reliability⁴⁵, conveys the degree to which a respondent has a positive attitude and flexible approach toward feeding him- or herself healthfully. The ecSI score ranges from 0 to 48 with a score of ≥32 defined as ‘eating competence’^{41, 46}. Eating competence includes questions about internal regulation cues, contextual skills for procuring nutritious foods, acceptance of a variety of foods, and the level of comfort with, and enjoyment from, eating.

The Perceived Stress Scale (PSS) is a well-established scale that measures the subjective report of stressfulness of life situations in the previous month. The PSS, a 10-item questionnaire with Likert-scale responses and a score range of 0 to 40, has high validity and reliability^{42, 47, 48}. The Three Factor Eating Questionnaire (TFEQ), initially developed in the 1980s⁴⁹ and subsequently truncated and re-validated^{43, 50, 51}, was administered to assess three psychological domains related to eating behavior: restraint, disinhibited eating (DE), and emotional eating (EE). Restraint reflects the personal choice to limit food intake to control weight, DE reflects the tendency to overeat or lose control of how much is consumed, and EE is the degree to which emotions drive consumption. The 18 questions of the TFEQ are not evenly distributed across the three domains; in order to facilitate comparison between domains, raw scores were transformed using the established “half-scale” method⁵¹ to yield values for each domain ranging from 0 to 100, where higher score indicates a greater degree of restraint, DE or EE. Total TFEQ score, a variable reflecting a composite of restraint, disinhibition, and emotional eating, was the sum of the raw domain scores for each participant (maximum possible score 72).

Statistical Analyses

In bivariate analyses, differences in participant characteristics between groups were evaluated (chi-square test and Student’s t-test), and nonparametric tests were used when underlying statistical assumptions were not met. Multivariable ordinary least squares regression analyses tested the association between each psychological construct and the anthropometric outcomes, testing for effect modification by sex in each model. In models where the outcome was change in weight, BMI, or WC, the model was also adjusted for baseline weight, BMI, or WC, respectively. The analytic sample for the multivariable modeling in this study is limited to those with complete anthropometry follow-up (N = 173), unless otherwise noted. All analyses were completed in SAS 9.4 (Cary, NC), all tests were two-tailed, and the threshold for statistical significance was set at P < 0.05.

Results

Sample Characteristics

The study enrolled 264 participants; 91.3% (N = 241) completed baseline anthropometry assessment and were included in the cross-sectional analyses of baseline data. The mean age was 18.1 years (SD 0.3) and the sample was representative of the incoming class of 2015 on the distribution of sex, college of matriculation, and country of origin (domestic vs. international). Participants who completed anthropometry follow-up (66%, N = 173, mean follow-up 14.1 weeks [SD 1.1]) were included in the longitudinal analyses. Most characteristics were similar between participants with complete and incomplete follow-up (Supplemental Table 1S), although dropouts were somewhat lower on eating competence and somewhat higher on weight, BMI, and WC at study baseline.

Cross-Sectional Associations of Psychological Constructs with Body Habitus

Several psychological constructs related to eating differed by sex (Table 1). The TFEQ domain scores for restraint and emotional eating and the total TFEQ composite score were higher in females than males at baseline and at the end of the first semester. Eating competence was higher in males than females at the beginning of the study, but the difference by sex was not statistically significant at follow-up. About half of males and females were classified as eating competent (ecSI scores ≥32) at both time points, and there was no significant difference by sex. Males and females scored similarly in perceived stress just prior to starting college, but at follow-up females reported significantly greater stress than their male counterparts. Sex differences in anthropometrics and body composition were evident and anticipated due to biological differences, but energy expenditure indicators did not differ by sex at either time point.

Table 1.

Participant characteristics at the beginning and end of the first semester, by sex, for first year students

Characteristics	Baseline N=241			End of Fall Semester N=173§
	Females N=125	Males N=116		Females N=93	Males N=80
	Mean (SD)	Mean (SD)	P	Mean (SD)	Mean (SD)	P
Psycho-behavioral Constructs
Total TFEQ†	40.2 (7.4)	35.2 (7.7)	<0.0001	40.8 (7.2)	34.8 (7.6)	<0.0001
Cognitive Restraint†	46.5 (21.8)	31.3 (23.4)	<0.0001	48.4 (20.6)	33.9 (20.7)	<0.0001
Disinhibited Eating†	37.8 (18.6)	34.6 (18.3)	0.2221	38.5 (16.1)	32.0 (16.4)	0.0147
Emotional Eating†	39.5 (24.3)	25.2 (25.0)	<0.0001	42.0 (24.5)	22.9 (22.4)	<0.0001
Eating Competence	31.5 (7.2)	33.8 (6.3)	0.0408	32.2 (7.0)	33.3 (6.6)	0.3164
Perceived Stress	15.1 (5.5)	14.2 (6.7)	0.2106	17.9 (6.4)	14.7 (7.8)	0.0080
Anthropometry
Weight (kg)	58.2 (9.8)	70.0 (11.6)	<0.0001	59.4 (9.6)	71.6 (10.4)	<0.0001
Body mass index (kg/m2)	21.5 (3.0)	22.4 (3.1)	0.0057	22.1 (3.0)	22.9 (2.7)	0.0201
Waist circumference (cm)	69.9 (7.0)	76.6 (8.1)	<0.0001	71.4 (7.2)	78.5 (7.1)	<0.0001
Hip circumference (cm)	96.2 (7.3)	95.8 (7.4)	0.6663	97.5 (6.9)	96.6 (6.3)	0.3260
Body Composition (DXA)
Total body fat (%)	26.0 (5.5)	15.7 (5.2)	<0.0001
Truncal body fat (%)	21.5 (6.4)	13.3 (5.9)	<0.0001
Fat mass index (fat mass, kg/ht, m2)	5.8 (2.0)	3.7 (1.8)	<0.0001
Energy Expenditure
Total physical activity (MET·hr/d)‡	9.8 (10.1)	10.4 (11.0)	0.7803	11.7 (9.4)	9.6 (7.1)	0.1986
Sedentary time (hr/d)	6.2 (2.9)	6.5 (3.0)	0.3668	7.1 (3.4)	7.4 (3.7)	0.6292
Moderate intensity PA (hr/wk)	6.8 (9.6)	6.4 (8.2)	0.8427	4.9 (5.9)	3.6 (4.6)	0.0890
Vigorous intensity PA (hr/wk)	3.9 (5.4)	4.9 (6.7)	0.2668	3.2 (5.5)	2.7 (4.2)	0.9532
Dietary Intake
Self-reported total meals/d	3.0 (0.6)	2.9 (0.6)	0.2464	2.6 (0.7)	2.6 (0.6)	0.9869
Self-reported total snacks/d	2.5 (1.4)	2.3 (1.7)	0.1851	2.4 (1.3)	1.7 (1.0)	0.0028
Usual energy intake (kcal/d)	1797 (727)	2305 (751)	<0.0001
Energy from carbohydrate (%)	48.6 (7.2)	49.1 (7.6)	0.7872
Energy from protein (%)	16.2 (3.1)	16.1 (3.1)	0.8151
Energy from fat (%)	32.6 (6.5)	32.2 (6.1)	0.5193
Fall Dining Hall Use (semester cumulative)
Visits per day (average swipes/day)				1.5 (0.3)	1.7 (0.3)	0.0024
Total dining hall visits (total swipe count)				130.3 (35.3)	150.5 (39.5)	0.0006
Total days of dining hall use (count of days swiped)				83.5 (11.8)	88.5 (11.6)	0.0003

Sex differences tested by Student’s t-test for variables meeting test assumptions, otherwise, Wilcoxon Signed Rank test used; all tests two-sided

^§N=68 participants lost to follow-up (baseline data but no anthropometry data at end of fall semester)

^†Total TFEQ score is the sum of the three domain raw scores, and ranges from 0-76. In contrast to the total TFEQ score, domain scores were transformed and each domain ranges from 0-100 Shaded cells indicate variables were not measured at time-point

At study baseline the TFEQ domain scores for DE and EE and the total TFEQ composite score were positively and statistically significantly associated with anthropometry and body composition, after adjusting for sex (Table 2, parts A and B). For example, given the beta coefficient for the association of DE behavior with body weight (0.14; Table 2), a one standard deviation (1 SD = 18.3) higher DE score was associated with 2.6 kg higher body weight. Disinhibited and emotional eating were consistently positively associated with multiple indicators of body habitus including weight, BMI, WC, and DXA-derived fat mass index (FMI). Effect modification by sex was tested for all variables, and there was statistically significant effect modification of the stress – anthropometry associations. Stress was consistently inversely associated with baseline weight and WC in females; conversely, in males, the stress – body habitus association was consistently positive.

Table 2.

Least squares regression model estimates of the cross-sectional association of psychological factors with anthropometry and adiposity (N=173)^*

A. Anthropometric outcomes
Baseline:	β	Baseline weight 95% CI	P	β	Baseline BMI 95% CI	P	β	Baseline WC 95% CI	P
Total TFEQ†	0.24	0.01, 0.46	0.042	0.08	0.01, 0.14	0.017	0.14	−0.02, 0.30	0.086
Cognitive Restraint	−0.04	−0.11, 0.02	0.206	−0.01	−0.03, 0.01	0.270	−0.04	−0.09, 0.00	0.077
Disinhibited Eating	0.14	0.05, 0.23	0.003	0.04	0.02, 0.07	0.002	0.10	0.03, 0.16	0.004
Emotional Eating	0.08	0.02, 0.14	0.014	0.03	0.01, 0.05	0.001	0.06	0.02, 0.11	0.006
Eating Competence	−0.09	−0.32, 0.14	0.451	M −0.12 F 0.05	−0.23, −0.01 −0.04, 0.14	0.028 0.265	−0.03	−0.20, 0.14	0.733
Perceived Stress**	M 0.34 F −0.54	−0.01, 0.69 −0.93, −0.16	0.061 0.007	M 0.09 F −0.10	−0.01, 0.19 −0.22, 0.01	0.091 0.081	M 0.21 F −0.32	−0.04, 0.47 −0.60, −0.04	0.102 0.029

B. Adiposity measured by DXA Outcomes (N=173)
Baseline:	β	Baseline BF% 95% CI	P	β	Baseline FMI 95% CI	P
Total TFEQ†	0.06	−0.05,0.18	0.282	0.04	0.00,0.08	0.076
Cognitive Restraint	0.00	−0.04,0.03	0.800	−0.01	−0.02,0.01	0.415
Disinhibited Eating	0.03	−0.02,0.08	0.242	0.02	0.00,0.04	0.023
Emotional Eating	0.02	−0.01,0.06	0.155	0.01	0.00,0.03	0.014
Eating Competence	−0.03	−0.15,0.09	0.618	−0.01	−0.05,0.03	0.734
Perceived Stress**	0.05	−0.09,0.19	0.491	0.01	−0.04,0.06	0.718

Abbreviations: BMI, body mass index; WC, waist circumference, TFEQ, Three Factor Eating Questionnaire; M, male; F, female

^*N = 173 with complete anthropometry data; all models adjusted sex

^**The P-values for the sex by perceived stress interaction terms were 0.0012, 0.0157, and 0.0068 for weight, BMI and WC, respectively

^†Total TFEQ score is the sum of the three domain raw scores, and ranges from 0-76. In contrast to the total TFEQ score, domain scores were transformed and each domain ranges from 0-100

Abbreviations: BMI, body mass index; WC, waist circumference, TFEQ, Three Factor Eating Questionnaire; M, male; F, female,

^**The P-values for the sex by perceived stress interaction terms were 0.3878 and 0.1099 and for BF% and FMI, respectively

^†Total TFEQ score is the sum of the three domain raw scores, and ranges from 0-76. In contrast to the total TFEQ score, domain scores were transformed and each domain ranges from 0-100

We explored the association of variables related to energy balance, including dietary intake, physical activity and dining hall use, with anthropometry and body composition at the study baseline. Physical activity, particularly vigorous physical activity, was consistently inversely associated with DXA-derived indicators of body adiposity, but no associations were detected with anthropometric measures (Supplemental Table 2S).

Longitudinal Association of Psychological Constructs with Change in Body Habitus

There were no statistically significant associations of eating competence or TFEQ scores with subsequent change in weight, BMI, or WC in models adjusted for sex and baseline anthropometry (Table 3). However, perceived stress at study baseline was associated with change in all three anthropometric outcomes, with statistically significant effect modification by sex. Thus, in males, a 1 SD higher baseline stress score (1 SD = 6.7 points in males) was associated with a 0.8 kg greater change in weight, a 0.3 kg/m² greater change in BMI, and a 1.1 cm greater change in WC over the first semester, compared to their counterparts with lower stress. Conversely, in females, stress was inversely associated with change in anthropometry (all beta coefficients were negative), but the associations did not reach statistical significance thresholds.

Table 3.

Association of baseline psycho-behavioral constructs with longitudinal change in anthropometry the first semester, adjusted for sex and baseline anthropometry, and estimated in least squares regression models (N=173)

Characteristics Baseline:	β	Weight Change 95% CI	P	β	BMI Change 95% CI	P	β	WC Change 95% CI	P
Total TFEQ†	−0.021	−0.076, 0.035	0.463	−0.008	−0.027, 0.012	0.436	−0.019	−0.082, 0.044	0.557
Cognitive Restraint	−0.007	−0.023, 0.010	0.421	−0.002	−0.008, 0.004	0.439	−0.017	−0.035, 0.002	0.083
Disinhibited Eating	−0.005	−0.028, 0.018	0.692	−0.003	−0.011, 0.005	0.519	0.003	−0.023, 0.029	0.832
Emotional Eating	−0.001	−0.017, 0.014	0.870	0.001	−0.005, 0.006	0.840	0.007	−0.011, 0.025	0.437
Eating Competence	−0.007	−0.062, 0.048	0.811	0.000	−0.020, 0.019	0.988	−0.004	−0.066, 0.059	0.913
Perceived Stress*	M 0.118 F −0.028	0.034, 0.202 −0.123, 0.066	0.007 0.554	M 0.039 F −0.005	0.009, 0.069 −0.038, 0.028	0.011 0.777	M 0.162 F −0.073	0.066, 0.257 −0.178, 0.033	0.001 0.181

Abbreviations: BMI, body mass index; WC, waist circumference, TFEQ, Three Factor Eating Questionnaire; M, male; F, female,

^*Sex-stratified coefficients are shown for perceived stress, and the P-values for the sex by perceived stress interaction terms were 0.0268, 0.0572 and 0.0017 for weight, BMI and WC, respectively

^†Total TFEQ score is the sum of the three domain raw scores, and ranges from 0-76. In contrast to the total TFEQ score, domain scores were transformed and each domain ranges from 0-100

Other factors related to energy balance, including physical activity and diet, were investigated as predictors of longitudinal change in anthropometry. The frequency of eating in the ‘all-you-care-to-eat’ on-campus dining halls, measured as the total number of swipes in the semester, was associated with positive changes in weight (P = 0.002) and WC (P = 0.022) (Table 4). Thus, participants who were 1 SD higher on total swipes (1 SD = 37.5 swipes) gained about 0.6 kg more weight, increased 0.2 kg/m² more on BMI, and increased 0.5 cm more on WC compared to their counterparts with lower total swipes.

Table 4.

Estimates from least squares regression modeling of the association of baseline energy balance factors with longitudinal anthropometric change in the first semester, adjusted for sex and baseline anthropometry (N=173)

Characteristics Baseline:	β	Weight Change 95% CI	P	β	BMI Change 95% CI	P	β	WC Change 95% CI	P
Total physical activity (MET·hr/d)‡	0.011	−0.025, 0.047	0.541	0.003	−0.009, 0.016	0.596	0.015	−0.026, 0.056	0.478
Sedentary time (hr/d)	0.004	−0.121, 0.129	0.952	0.007	−0.037, 0.050	0.767	−0.007	−0.148, 0.133	0.919
Moderate intensity PA (hrs/wk)	−0.008	−0.051, 0.035	0.723	−0.005	−0.020, 0.011	0.556	−0.007	−0.056, 0.042	0.770
Vigorous intensity PA (hrs/wk)	0.029	−0.031, 0.089	0.339	0.011	−0.010, 0.032	0.295	0.043	−0.024, 0.111	0.210
Usual energy intake (kcal/d)	0.000	−0.001, 0.001	0.910	0.000	0.000, 0.000	0.923	0.000	−0.001, 0.001	0.782
Energy from carbohydrate (%)	0.021	−0.036, 0.078	0.475	0.008	−0.012, 0.028	0.435	0.001	−0.066, 0.068	0.981
Energy from protein (%)	0.020	−0.113, 0.153	0.766	0.015	−0.03, 0.062	0.539	0.048	−0.108, 0.204	0.549
Energy from fat (%)	−0.031	−0.098, 0.036	0.360	−0.013	−0.036, 0.011	0.289	−0.011	−0.089, 0.068	0.792
Self-reported total meals/d	0.503	−0.076, 1.081	0.091	0.161	−0.045, 0.367	0.127	0.365	−0.299, 1.028	0.283
Self-reported total snacks/d	−0.048	−0.283, 0.186	0.687	−0.019	−0.103, 0.064	0.652	−0.028	−0.297, 0.241	0.840
Visits per day (mean swipes/day)	1.82	0.52, 3.13	0.007	0.58	0.12, 1.04	0.014	1.43	−0.07, 2.93	0.063
Total dining hall visits (total swipes)	0.02	0.01, 0.03	0.002	0.01	0.002, 0.01	0.004	0.01	0.002, 0.02	0.022
Total days of dining hall use (count of days swiped)	0.05	0.02, 0.08	0.004	0.02	0.01, 0.03	0.006	0.04	0.009, 0.080	0.016

Abbreviations: BMI, body mass index; WC, waist circumference,

We quantified changes in the psycho-behavioral construct scores, anthropometry, physical activity, and self-reported eating frequency observed during the first semester of college and tested the means against zero in sex-stratified analyses (Supplemental Table 3S). On average, both sexes experienced a significant increase in restraint and a reduction in the number of meals typically consumed per day. Females reported a significant increase in perceived stress, while the mean change in perceived stress over the course of the semester was not significantly different from zero among males. Mean changes in the psycho-behavioral construct scores were generally not highly correlated with concurrent changes in anthropometry (Supplemental Table 4S).

Discussion

Weight gain during the college years is common^{3, 6, 52}, tends to occur relatively rapidly early in college⁷, and excess weight gained sets a trajectory that continues into adulthood⁵². The primary outcome in this study, first semester weight gain, accounts for most of the weight gained during the first year of college; moreover, weight gain in the first year is greater than the average annual weight gain during the subsequent years of college^{5, 52}. Understanding the extrinsic and intrinsic factors that affect energy balance leading to weight and adiposity gain is important to guide the development of prevention strategies. In males only, stress scores were stable over the semester and greater stress at study baseline was associated with greater subsequent increases in all three anthropometric indicators. In both males and females, the frequency of dining hall use was positively associated with subsequent increases in weight, BMI and WC. At the baseline of the study, there were consistently positive cross-sectional associations of disinhibited and emotional eating with both anthropometry and DXA-derived measures of adiposity.

Boyce et al.³¹ reported a positive association of perceived stress with weight gain in freshmen with BMI >25 kg/m² and a negative association in freshmen with low/normal BMI. We found no evidence of a baseline BMI – stress interaction in models of weight change, but there was evidence for a differential stress – body habitus association by sex. Stress may contribute to weight gain differently in males compared to females because of differential effects of stress on dietary intake and/or on internal hormonal regulators of energy utilization. In a recent study of stress and dietary behaviors in college freshmen⁵³, Papier et al. reported sex differences in the stress – dietary pattern association; greater stress was a stronger predictor of unhealthy food intake in males compared to females. In addition, there is some evidence of sex differences in the hormonal stress response, as indicated by increased circulating free cortisol, a biomarker of activation of the hypothalamic-pituitary-adrenal axis⁵⁴; prior studies reported that psychological stress from achievement challenges (i.e., exams) induced a greater cortisol response in young adult males compared to females, while females had a greater cortisol response to social rejection challenges compared to males^{55, 56}. Both of these reports are consistent with our finding that stress led to weight gain in males only.

There was little association between restraint or overeating due to external or emotional cues and subsequent changes in weight or adiposity. These findings agree with several prior studies of freshmen/sophomores^{30, 34} including a study of first-year university students in the United Kingdom (UK)⁵⁷, which reported no association between restraint, disinhibition, or emotional eating and body weight change over 3 or 12 months. A study of adolescents and young adults, which measured adiposity change as sum of skin folds, also reported no association between the TFEQ restraint score and adiposity change⁵⁸. Two prior studies reported higher restraint was associated with weight gain when models considered on-campus vs. off-campus residence⁵⁹ and self-esteem³³, neither of which could be considered in our study. When adiposity change was assessed with bioelectric impedance, baseline disinhibited eating score was associated with change in FMI over 3 and 12 months (β = 0.29, P < 0.0001; β = 0.28, P < 0.01 at 3 and 12 months, respectively)⁵⁷. We found no evidence of a disinhibited eating score – FMI association in models of change in FMI over 8 months (β = −0.006, P = 0.059, data not shown); the change in FMI over 3 months was not measured in our study. The difference in findings may be explained by differences in study population characteristics (including differences in the score range on disinhibited eating), study design, or by differences in the assessment of FMI (DXA versus bioelectrical impedance).

We found consistently positive cross-sectional associations at study baseline between the disinhibited and emotional eating scores and weight, WC, BMI, and FMI, but there was no evidence for longitudinal associations. Possible explanations for this discrepancy include: 1) no true association exists, 2) the instruments assessing the psychological factors are not sensitive enough to capture the behaviors or attitudes that drive gain in weight and adiposity, or 3) the cross-sectional findings are explained by reverse causality, such that higher weight, BMI, WC leads to more emotional and disinhibited eating. It is not possible to determine which explanation is correct in this study, but further research using longitudinal designs with repeated exposure and outcome assessments, including more sensitive indicators, would help to clarify the direction of the association.

The contribution of psychological constructs to anthropometric outcomes, and the differences by sex, may vary by age and/or cultural context. A large French cohort study reported a significant and positive cross-sectional TFEQ – BMI association in adults aged 31-67 years, but not in youth aged 14-24 years⁵¹. The Quebec Family Study showed differences in TFEQ scores by sex in adults, no sex differences in adolescents, and positive associations of TFEQ restraint and disinhibited eating scores with BMI in adolescents^{60, 61}.

This study was limited by a lack of information on background socioeconomic status, such as parental income, which meant that this variable could not be considered as a potential confounder in these analyses. Although socioeconomic data are unavailable, we do not expect significant confounding because all students eat at central dining halls and Cornell’s need-blind admission provides financial aid to students in order to attend. The study did not collect data on race/ethnicity from participants. However, the sample of 1001 incoming students invited to participate was selected via stratified random sampling from the full incoming class of college freshman and, ultimately, sample participants had exactly the same distribution on gender, proportion international, and proportion in each college as the full incoming class. It is reasonable to assume that the sample is also similar to the full class in terms of race/ethnicity, thus we expect that about 20% of sample would self-identify as under-represented minorities. Finally, although there were no statistically significant differences in the baseline characteristics of the analytic sample and participants who were lost to follow-up (Supplementary Table 1S), those who were lost to follow-up had marginally lower eating competence, were higher on measures of anthropometry and adiposity, and had lower self-reported caloric intake. We do not expect these small differences to lead to bias in the estimates of associations. Finally, the external validity or generalizability of these findings may be limited to college populations with similar demographics.

The strengths of this study include availability of anthropometry and body composition data collected by trained staff using an established protocol. The DXA-derived measures of adiposity at study baseline allowed a rigorous and novel estimation of cross-sectional psychological construct – body habitus associations. For example, in contrast to BMI, FMI is a height-scaled index of adiposity that is not confounded by lean body mass. An additional strength was the evaluation of effect modification by sex in a sample representative of the incoming class. A further strength of this study was the scope, quality and variety of the indicators assessed, including objective electronic card swipe records to provide data on the frequency of visits to the all-you-care-to-eat dining environments on campus.

In conclusion, psychological constructs, including eating competence, restrained eating, disinhibited eating, emotional eating, and perceived stress were associated with anthropometry and adiposity in college freshmen at the beginning of the academic year. Perceived stress was associated with subsequent changes in anthropometry in males only, such that greater stress at the start of college predicted greater increases in weight, BMI and WC. The sex-specific association between stress and weight gain may be related to a greater stress response to academic and/or other challenges in males compared to females and/or to a sex-specific influence of stress on food intake, energy balance factors, and/or metabolism.