Enhancing Efficacy of a Dissonance-Based Obesity and Eating Disorder Prevention Program: Experimental Therapeutics

Abstract

Objective:

Test the hypothesis that the efficacy of a dissonance-based obesity/eating disorder prevention program, Project Health, could be enhanced by implementing it in single-sex groups and adding food response inhibition and attention training.

Method:

Using a 2×2 factorial design, young adults (N=261; M age=19.3, 79% female; 64% White) were randomized to (1) single-sex or (2) mixed-sex groups that completed food response inhibition and attention training or (3) single-sex or (4) mixed-sex groups that completed generic response inhibition and attention training with nonfood images. Pre-registered primary outcomes (body fat, BMI), eating disorder symptoms and other outcomes were assessed at pretest and posttest.

Results:

For one pre-registered primary outcome, body fat loss, there was a significant interaction between the two manipulated factors (d=−.28), as well as significant main effects for sex composition of groups (d=−.18) and food response inhibition and attention training (d=−0.17), with the largest body fat loss occurring for single-sex groups implemented with food response inhibition and attention training. Although the two manipulated factors did not significantly affect the other outcomes (including BMI, the other pre-registered primary outcome), there was a significant reduction in eating disorder symptoms across the conditions (within-participant d=−.78), converging with prior evidence that Project Health produced larger reductions in symptoms than educational control participants.

Conclusions:

Results suggest that implementing Project Health in single-sex groups with food response inhibition and attention training produced the largest body fat loss effects, as well as significant reductions eating disorder symptoms, suggesting that efforts to disseminate this prevention program are warranted.

Nearly 70% of US women and men are overweight or obese, resulting in 300,000 deaths and $150 billion in health-related expenses yearly (Flegal et al., 2012). Prevention may be a more effective way to combat obesity because the most common treatment, behavioral weight loss interventions, rarely results in lasting weight loss (Hartmann-Boyce et al., 2014). Yet, most prevention programs have not prevented future increases in weight and overweight/obesity onset (Plotnikoff et al., 2015).

Three obesity prevention programs have significantly reduced future weight gain in young adults (Allman-Farinelli et al., 2016; Bertz et al., 2015; Wing et al., 2016), but these interventions were very intensive, lasting from 8- to 28-months in duration, which would make broad implementation of these prevention programs difficult and expensive. In contrast, the 3-hour Healthy Weight obesity and eating disorder prevention program produced a 53% reduction in obesity onset and a 60% reduction in eating disorder onset over 3-year follow-up versus assessment-only controls for female adolescents with body image concerns and greater reductions in BMI gain and eating disorder symptoms through 3-year follow-up versus controls and two alternative interventions (Stice et al., 2008). It is also critical to prevent eating disorders because they increase risk for mortality (Arcelus, Mitchell, Wales, & Nielsen, 2011) and result in functional impairment for both women and men (Udo & Grilo, 2018) leading to calls for prevention programs that simultaneously target obesity and eating disorders (Neumark-Sztainer, 2005). In Healthy Weight, participants make small incremental healthy lifestyle changes to gradually balance caloric intake and expenditure; we encourage small gradual dietary changes to minimize metabolic slowing, which attenuates weight loss. Reaching energy balance theoretically improves body dissatisfaction and reduces eating disorder symptoms, based on evidence that elevated body weight is the strongest predictor of future increases in body dissatisfaction (Stice & Whitenton, 2002). The lifestyle change plan is participant-driven to promote internalization of the health goals. Healthy Weight is implemented in groups because it increases accountability and support for lifestyle change goals and is cost effective. Healthy Weight also produced greater reductions in negative affect compared to assessment-only controls that persisted through 2- and 3-year follow-up (Stice et al., 2008), which is important given the possibility that obesity prevention programs might contribute to shame and other types of negative affect. Compared to educational brochure controls, a 4-hour version of Healthy Weight resulted in significantly greater reductions in BMI gain and eating disorder symptoms in young women with weight concerns, prevented BMI gain through 1-year follow-up for initially overweight participants, and produced a 60% reduction in eating disorder onset over 2-year follow-up (Stice et al., 2012, 2013).

To improve the efficacy of Healthy Weight we added activities that promote cognitive dissonance about lifestyle behaviors that drive excess weight gain. Cognitive dissonance has been successfully used to prevent future eating disorder onset (Ghaderi et al., 2020; Stice et al., 2008; Stice et al., 2020). We created a 6-hour dissonance-based version of Healthy Weight, referred to as Project Health, that added verbal, written, and behavioral exercises to create dissonance regarding unhealthy lifestyle behaviors (e.g., discussing costs of obesity, an unhealthy diet, and sedentary behavior), which theoretically increase the likelihood that participants will align their attitudes with their perspectives assumed in the sessions, resulting in healthier lifestyle choices and less binge eating and compensatory weight control behaviors. Young adults at risk for weight gain randomized to Project Health showed smaller increases in BMI through 2-year follow-up versus participants who completed Healthy Weight and obesity education video controls, and a 41% and 42% reduction in overweight/obesity onset over 2-year follow-up relative to Healthy Weight participants and obesity education controls, respectively; they also showed a reduction in eating disorder symptoms and a marginal 60% reduction in future eating disorder onset relative to controls (Stice et al., 2018). This experimental therapeutics trial provided evidence that adding activities that induce dissonance regarding unhealthy lifestyle behaviors improved weight gain prevention effects relative to the intervention lacking these activities. We use the term experimental therapeutics to describe trials that manipulate only one factor to permit stronger inferences regarding the effects of that factor on change in outcomes (Stice et al., 2018).

In the present study we experimentally manipulated two factors theorized to improve the body fat gain prevention effects of Project Health. First, the fact that the original Healthy Weight intervention produced weight gain prevention effects when implemented in female-only groups (Stice et al., 2006, 2008, 2012, 2013), but not when implemented in mixed-sex groups (Stice et al., 2018), suggests that Project Health may likewise be more effective when implemented in single-sex groups. Several interventions have shown greater efficacy when implemented in female-only versus mixed-sex groups (Babinski et al., 2013; Hser et al., 2011; Prendergast et al., 2011), though research has not focused on the benefits of single-sex groups for males. Theoretically, the greater commonality of single-sex groups for both sexes promotes cohesion, trust, honesty, and willingness to disclose personal information (Lewis & Neighbors, 2004). Females report feeling safer and more willing to disclose sensitive material in single- versus mixed-sex substance abuse treatment groups (Greenfield et al., 2013). Further, conversations in mixed-sex dyads contain more interruptions by both sexes than conversations in same-sex dyads (Turner, 1995), which might reduce verbal participation in dissonance-induction discussions for both sexes. However, no trial has tested the effects of the sex composition of group-delivered obesity prevention programs. We hypothesized that Project Health will produce larger body fat gain prevention effects when implemented with single- versus mixed-sex groups for both females and males. We also hypothesized that greater participation in dissonance-inducing discussions, fewer interruptions, and greater group cohesion would mediate the effect of the sex composition of groups.

A second idea for improving the effectiveness of Project Health is to add food response inhibition and attention training. Greater responsivity of brain regions implicated in reward (striatum, orbitofrontal cortex [OFC]) to high-calorie food images (Demos et al., 2012; Stice et al., 2015a; Yokum et al., 2014; Yokum & Stice, 2019), greater attentional bias for high-calorie food (Calitri et al., 2010), and lower inhibitory control in response to high-calorie foods (Evans et al., 2012; Francis & Susman, 2009; Schlam et al., 2013) predicted future weight gain. Results imply that an intervention that reduces reward and attention region response to high-calorie foods and increases inhibitory control may decrease overeating. Go/no-go and stop-signal computer training wherein participants are cued to repeatedly respond behaviorally to low-calorie food or non-food images, and to repeatedly inhibit behavioral responses to high-calorie food images, produced weight loss compared to control training (Lawrence et al., 2015; Veling et al., 2014). Further, completing dot-probe tasks that train attention away from high-calorie foods and toward low-calorie foods reduces attentional bias for high-calorie foods and intake of such foods versus control training (Kakoschke et al., 2014; Kemps et al., 2014). Overweight or obese adults who completed a multifaceted response inhibition training with high-calorie foods and attend-away training from high-calorie foods, and response training with low-calorie foods and attend-to training with low-calorie foods showed greater body fat loss, and reduced fMRI-assessed reward region (putamen; mid insula) and attention region (inferior parietal lobe) response to, and palatability ratings and monetary valuation of, high-calorie foods than controls who completed the trainings with nonfood images (Stice et al., 2017). We hypothesized that adding food response inhibition and attention training to Project Health would result in larger body fat gain prevention effects versus the standard Project Health intervention paired with generic response and attention training with nonfood images. We also hypothesized that decreases in palatability ratings of, willingness to pay for, and attentional bias for high-calorie foods would mediate the effect of response training, based on the intervention effects from our last trial (Stice et al., 2017).

In sum, this experimental therapeutics trial tested whether implementing Project Health in single- versus mixed-sex groups and adding food response inhibition and attention training versus generic response inhibition and attention training would increase the body fat loss effects of this obesity/eating disorder prevention program. Thus, in this experimental therapeutics trial Project Health groups implemented in mixed-sex groups with generic response inhibition/attention training served as the baseline comparison condition, rather than some minimal-intervention control condition.

Methods

Participants and Procedures

Participants were 261 young women and men (M age = 19.3, SD = 0.8; M BMI = 24.5, SD = 2.6; 79% female; 29% Asian, 2% Native Americans, 7% Black, 2% Pacific Islander, 10% Hispanic, and 64% White). From October 2018 to March 2020 participants were recruited from 2 universities and surrounding areas in Oregon and Pennsylvania using mailings, flyers, and leaflets inviting young people aged 17–20 with weight concerns to participate in a trial of a healthy lifestyle intervention designed to improve weight concerns. We focused on this age group because this developmental period is marked by excess weight gain (Lowe et al., 2006). This is a critical developmental period because for many it is the first time they are solely responsible for choices regarding caloric intake and exercise. Weight concerns were required because they increase risk for future weight gain (Haines et al., 2007). We evaluated this prevention program when delivered by university staff who typically deliver clinical care (grad students in psychology clinics), because they are well positioned to provide interventions to students. Moreover, because 64% of late adolescents attend college (US Dept of Education, 2011), this is an efficient method of reaching a large population of late adolescents at risk for weight gain and eating disorders.

Potential participants completed an on-line screening that probed for weight concerns and times during which they could attend sessions. Informed written consent was obtained for this Oregon Research Institute Institutional Review Board approved study. Individuals were eligible if they were between 17–20 years of age, had a BMI greater than 20 and less than 30, and reported concern about their weight, defined as: (1) a response of at least moderate weight concerns to a question asking participants to rate the strength of their concerns on a 9-point scale (0 = none to 8 = extreme), and (2) an endorsement of the following question: “Do you believe there is room for improvement in your diet and exercise habits?” We excluded individuals with a BMI of less than 20 because we reasoned that they are at very low risk for future obesity onset and we excluded individuals with a BMI of greater than 30 because this was an obesity prevention trial. Exclusion criteria were a current diagnosis of anorexia nervosa, bulimia nervosa, or binge eating disorder. Subthreshold diagnosis was permissible and resulted in zero subthreshold anorexia nervosa, 12 (4.5%) subthreshold bulimia nervosa, 4 (1.5%) subthreshold binge eating disorder, and zero purging disorder cases. Figure 1 describes participant study flow. Participants completed interview, survey, and behavioral measurements at pretest and posttest (approximately 10 weeks later [Mean = 10.9, SD = 4.4]); they received $25 for completing each assessment. Participants were randomized on an individual level to Project Health delivered in single- versus mixed-sex groups and to response and attention training with food versus non-food images using an envelope system. Assessors were masked to condition.

Figure 1.

Participant flow through the study.

Interventions

Project Health.

This intervention consisted of 6 1-hour group sessions with 6–10 participants in which members committed to making participant-identified small, incremental reductions in dietary intake and increases in activity levels to bring caloric intake into balance with caloric expenditure, which is intended to reduce body dissatisfaction and unhealthy compensatory weight control behaviors. Participants were encouraged to support other participants in making their lifestyle improvements. Participants also completed verbal, written, and behavioral activities designed to induce dissonance about lifestyle behaviors that contribute to excess weight gain and promote lifestyle behaviors that contribute to a healthy weight, which should reduce binge eating. Each session included exercises in which participants discussed health, interpersonal, and societal costs of overeating high-calorie foods, sedentary behaviors, and obesity, as well as health, interpersonal, and societal benefits of consuming low-calorie foods, regular exercise, and a lean physique. It is important to note that we structured the discussions to avoid stigmatizing people with obesity or eating disorders (e.g., we asked group participants to avoid saying negative things about obese individuals). These activities were shared with the group and video-recorded to create accountability and maximize dissonance. Those assigned to the single-sex group attended groups with only males or females, based on reported sex at birth (including the facilitator); those assigned to the mixed-sex group attended groups with both males and females, including at least two participants of each sex (the facilitator could be of either sex). Six individuals reported non-cisgender identity (2 nonbinary and 4 transgender) and were assigned to mixed-sex groups. Fourteen graduate students in four health fields attended a 6-hour training and facilitated groups. If a participant missed a session, brief individual make-up sessions were conducted.

Sessions were video-recorded for fidelity and competence ratings by Drs. Rohde, Shaw, and Butryn using scales from a prior trial (Stice et al., 2018). The first 6 sessions delivered by each facilitator were reviewed, followed by 50% of sessions in remaining groups delivered by each facilitator. Raters independently coded a randomly selected 5% of reviewed sessions to obtain inter-rater agreement data. Checklists assessing the key components for the intervention (e.g., discussed health problems associated with obesity) were rated on a 100-point scale (1 = “No adherence; the section was skipped” to 100 = “Perfect; all material in the section was presented as written”; a score of 70 was “good”). Facilitator competence was rated with 12 items (e.g., leaders express ideas clearly and at an appropriate pace) using a 100-point scale (20 = “Poor; leaders are difficult to follow and session proceeds at an uncomfortable pace” to 100 = “Superior; leaders are unusually articulate and express ideas in way that all group members understand; perfect pace”; a score of 60 was considered “Good/average”). Supervision based on fidelity and competence ratings was provided via email before the next session.

Food Response Inhibition and Attention Training.

After each Project Health session, participants completed 25-minute trainings that included 5 minutes of dissonance-based motivational enhancement writing prompts, 5-minutes of stop-signal training, 5-minutes of go/no-go training, 5-minutes of dot probe training and 5-minutes of visual search training (2 hours of computer training in total). Participants first completed brief written activities to create dissonance about unhealthy lifestyle behaviors (3 prompts from a bank of questions on benefits of healthy lifestyle/costs of unhealthy lifestyle, health/fitness goal generation, and reframing/circumnavigating barriers). The stop-signal and go/no-go tasks involve being cued to repeatedly respond behaviorally with a button press to low-calorie food images and repeatedly inhibit a behavioral response to high-calorie food images. The dot-probe paradigm reinforces people for looking at the low-calorie food because that is where the probe appears 90% of the trials, thereby training attention to low-calorie foods and away from high-calorie foods. The visual-search task trains people to rapidly allocate their attention to the 1 low-calorie food within an array of high-calorie foods, training them to ignore the latter foods. During the intake assessment participants reported the types of foods they often overeat; the images of the high-calorie foods used in the response and attention trainings were selected from these categories (i.e., high-calorie food images were tailored to participants using the top 3 of 13 categories of high-calorie foods they ate most frequently and 40 images of those foods that they rated highest in palatability). Low-calorie food images contained a standardized set of vegetables and fruits (40 of 60 images rated highest in palatability). The generic response inhibition training intervention was identical to the food response training described above, but used images of birds and flowers, as done in our past trial (Stice et al., 2017). Participants were told that both interventions were designed to improve response inhibition, which should produce weight loss given that impulsivity increases risk for overeating, fostering credibility of the control intervention. Participants in the generic training condition also completed the motivational enhancement writing exercises.

Measures

When we proposed this trial and registered it at ClinicalTrials.gov, we stated that percent body fat and BMI were our primary outcomes. However, four trials have now found that percent body fat is a more sensitive outcome for weight gain prevention and weight loss trials (Ranucci et al., 2017; Stice et al., 2015; Stice et al., 2017; Williamson et al., 2012), which prompted us to declare percent body fat as our primary outcome. Preliminary analyses confirmed that percent body fat was more sensitive in detecting the intervention effects in this weight gain prevention trial than BMI; the hypothesized 3-way and 2-way interactions was not significant when BMI was analyzed (see Tables 1 and 2). Secondary outcomes included eating disorder symptoms, weight concerns, and depressive symptoms. Mediators included participation in dissonance-inducing discussions, number of interruptions, group cohesion, palatability ratings of high-calorie foods, willingness to pay for high-calorie, and attentional bias for high-calorie foods.

Table 1.

Descriptive Summary for Study Variables by Condition.

	Pretest		Posttest
	Mean	SD	Mean	SD
Single sex group with food response training (n = 54)
Percent body fat	26.84	7.33	22.03	7.22
BMI	23.78	2.63	23.57	2.68
Eating disorder symptoms	12.46	11.37	5.73	5.92
Weight concerns	3.33	1.21	2.58	1.32
Depressive symptoms	11.31	8.03	7.41	6.39
Palatability ratings for high calorie foods	5.45	1.38	4.68	1.54
Willingness to pay for high calories foods	4.41	1.35	3.98	1.77
Attention bias for high calorie foods	0.28	2.12	−0.47	1.97
Single sex group with generic response training (n = 61)
Percent body fat	32.83	6.74	32.71	6.57
BMI	24.64	2.51	24.34	2.33
Eating disorder symptoms	11.92	8.82	6.91	6.28
Weight concerns	3.63	0.97	2.71	1.37
Depressive symptoms	12.18	8.27	8.82	8.28
Palatability ratings for high calorie foods	5.36	1.62	4.28	1.53
Willingness to pay for high calories foods	4.46	1.73	3.86	1.69
Attention bias for high calorie foods	0.36	1.74	−0.36	1.51
Mixed sex group with food response training (n = 74)
Percent body fat	29.20	9.55	29.61	7.66
BMI	24.98	3.17	24.86	2.97
Eating disorder symptoms	12.56	11.60	6.87	6.06
Weight concerns	3.21	1.38	2.90	1.41
Depressive symptoms	10.86	8.51	8.82	8.16
Palatability ratings for high calorie foods	5.52	1.50	4.53	1.52
Willingness to pay for high calories foods	4.47	1.75	3.91	1.58
Attention bias for high calorie foods	0.13	1.90	0.09	1.87
Mixed sex group with generic response training (n = 72)
Percent body fat	27.72	9.83	28.51	10.02
BMI	24.40	3.09	24.15	2.79
Eating disorder symptoms	11.65	11.42	7.88	10.17
Weight concerns	3.32	1.24	2.51	1.37
Depressive symptoms	13.31	8.82	10.20	9.07
Palatability ratings for high calorie foods	5.45	1.46	5.21	2.75
Willingness to pay for high calories foods	4.18	1.68	4.31	1.77
Attention bias for high calorie foods	0.08	1.74	0.11	1.49

Table 2.

Outcome Effects for Sex and Inhibition/Attention Training Factors for Total Sample and Completers.

Effect / Outcome	Intent-To-Treat			High Compliance
Single vs. Mixed Sex Groups × Time	p-value	d	95% CI	p-value	d	95% CI
Percent body fat	.002	−.18	−.63, −.13	.002	−.25	−.89, −.20
BMI	.694	−.02	−.29, .20	.407	−.04	−.48, .20
Eating disorder symptoms	.867	.03	−.20, .29	.703	.07	−.28, .41
Weight concerns	.159	−.18	−.43, .05	.074	−.29	−.66, .03
Depressive symptoms	.475	−.15	−.35, .14	.254	−.19	−.54, .14
Food Response vs. Generic Response × Time
Percent body fat	.005	−.16	−.60, −.11	.003	−.24	−.86, −.17
BMI	.795	−.01	−.28, .21	.925	−.01	−.35, .32
Eating disorder symptoms	.836	−.03	−.27, .22	.854	−.03	−.37, .31
Weight concerns	.019	.29	.05, .54	.011	.41	.10, .78
Depressive symptoms	.782	.04	−.17, .31	.375	.14	−.18, .49

Body fat.

We used air displacement plethysmography (ADP) via the Bod Pod to assess percent body fat. ADP is preferable to DEXA because it is less expensive and does not involve radiation. ADP is preferable to bioimpedance measures because it is less sensitive to procedural variation. Body density is calculated as body mass divided by body volume; body density is used to calculate percent body fat. ADP percent body fat shows high test-retest reliability (r = .92–.99) and correlates with DEXA and hydrostatic weighing estimates (r = .98–.99), with ADP estimated percent body fat falling an average of only 1.7% different than DEXA estimates (Weyers et al., 2002). Participants were asked to not consume any food or beverages (other than water) for at least 3 hours, refrain from using nicotine for at least 3 hours, and refrain from vigorous exercise for at least 24 hours prior to each Bod Pod measurement. Participants were asked to bring a swimsuit or form-fitting athletic clothes for the assessment.

Eating disorder symptoms.

The semi-structured Eating Disorder Diagnostic Interview (EDDI; Stice et al., 2018) assessed DSM-5 eating disorder symptoms (frequency of binge eating, vomiting, laxative/diuretic use, fasting, and excessive exercise; and overvaluation of weight/shape, feeling fat, and fear of weight gain). Items assessing symptoms in the past month were summed to form a composite at the pretest and the posttest assessments. This composite has shown internal consistency (α = .92), inter-rater agreement (ICC r = .93), 1-week test-retest reliability (ICC r = .95), predictive validity, and sensitivity to detecting intervention effects (Stice et al., 2017; α = .85 at baseline).

Weight concerns.

The 7-item Weight Concerns scale from the Eating Disorder Examination Questionnaire (Fairburn & Beglin, 2008) assesses the importance of weight, dissatisfaction with weight, discomfort of seeing one’s body, desire to lose weight, and avoidance of body exposure. It has shown internal consistency (α = .82–.89), test-retest reliability (r = .90–.92), and convergent validity with interview-assessed weight concerns (r = .77; Rose et al., 2013; α = .87 at baseline).

Depressive symptoms.

The 21-item Beck Depression Index (Beck et al., 1988) assesses depression symptoms. It has shown internal consistency (α = .73–.95), test-retest reliability (r = .60–.90), and convergent validity with clinician ratings of depressive symptoms (M r = .75; Beck et al., 1988; α = .91 at baseline).

Verbal participation during group discussions.

We used video-recordings of Project Health sessions to code the amount of time each participant spoke during dissonance-induction discussions and the number of times each participant interrupted others. Dr. Shaw developed a manualized procedure for assigning low-inference (i.e., objective, easily determined) ratings for time spent talking and number of interruptions, based on previous coding schemes (Karakowsky, 2004; Turner, 1995). One randomly selected session from each of the groups was coded. Raters were trained to a kappa agreement of > .80 on a set of 5 randomly selected training sessions rated by Dr. Shaw.

Group cohesion.

Participants completed the 9-item Cohesion Scale from the Therapeutic Factors Inventory (Lese & MacNair-Semands, 2000) at posttest to assess whether participants perceived a greater sense of bonding, working toward common goals, and affiliation with the group in single- versus mixed-sex groups. Participants rated items on a 7-point response scale ranging from “not at all” to “extremely”. It has shown internally consistency (α = .94) and 1-week test-retest reliability (r = .93) with college students (Lese & MacNair-Semands, 2000; α = .91 at baseline).

Palatability & Monetary Valuation of Food Images.

Using established procedures (Stice et al., 2017) participants rated the palatability of 60 high-calorie and 60 low-calorie food images (rating each on a 9-point scale ranging from “unappetizing” to “extremely appetizing”) and indicated how much they would be willing to pay for the food (using a 10-point scale ranging from $1 to $10). As part of the baseline survey, participants rated the frequency of their consumption of broad food groups. For the high-calorie foods, we included three food groups in the high-calorie food categories that were consumed by participants; for low-calorie foods, all participants saw fruits and vegetables.

Dot Probe Attentional Bias Task.

In this computer-administered task a fixation cross was shown in the center of the screen to standardize gaze. Next, pairs of high-calorie and low-calorie foods were presented side by side for 500 ms. We used 26 images of high-calorie foods and 26 images of low-calorie foods used in the training tasks. A probe (*) appears where one of the images had been and participants indicate as quickly as possible with a key press whether the cue is on the left or right. In the critical trials, one high-calorie food is shown next to one low-calorie food (order counterbalanced) and the probe appears where one of the images had been on a 50:50 basis. Attentional bias for high-calorie foods is operationalized as a faster reaction time when the probe appears where high-calorie foods versus where low-calorie foods had been. The average reaction time to high-calorie foods was subtracted from the average reaction time to low-calorie foods to create the attentional bias score. A positive score indicates bias towards unhealthy foods and a negative score a bias towards healthy foods.

Data analyses.

We originally planned to power our study to detect small effects (d = .28) of the two manipulated factors on change in continuous outcomes (N = 450), as reflected in our pre-registration. However, the pandemic forced us to close recruitment for this study prematurely when we had an N = 261 and precluded in-person assessments, which included body fat for over 60% of these participants at the planned 6- and 12-month follow-ups. Given that with an N = 261, we had a power of .80 to detect moderate effects (d = .43), we tested for the effects of the two interventions on the continuous outcomes at posttest. Intent-to-treat (ITT) analyses of the mixed factorial design, including two between-subject factors (single [coded 1] vs. mixed [coded 0] sex groups; and food response inhibition and attention training [coded 1] vs. generic response and attention training [coded 0]); and a within-subject time factor (coded in months since the pretest assessment), were evaluated with fixed effects growth models using SAS 9.2 PROC MIXED. Although participants were nested within groups (Mean intraclass correlation coefficient = .076), with pretest and posttest data only, we were limited to specifying fixed-effects models. Thus, we could not model individual and group-level heterogeneity associated with change in the outcomes. Instead, we treated group as a fixed-effect and included it as a covariate in all models to address the nested nature of the data. Models were estimated using maximum likelihood, a preferred method of handling missing data. To evaluate the effectiveness of the treatments, we interpreted all interactions crossed with time. We first tested the 3-way interaction, which included variability in outcomes from pretest to posttest modeled as a function of the main effects of each factor, and all higher order interaction terms. Next, we tested, separately, growth models for each 2-way interaction which included each between-subject factor crossed with the within-subject time factor. Then, for significant 2-way interactions, we tested simple slopes for one between-subject factor, conditional on each level of the other between-subject factor. We also tested whether birth assigned sex moderated the intervention effects by adding the main effect of birth sex and all higher order interactions with each factor. To facilitate interpretation of the interactions we provide graphs of the interactions and effect sizes that are equivalent to the d statistic (Feingold, 2009).

As a preliminary step for tests of mediation, we evaluated the effects of our two manipulated factors on the hypothesized mediators. For single- versus mixed-sex group comparison we used t-tests to evaluate group differences in participant ratings of group cohesions and in-sessions ratings of total time participants spoke (interruptions occurred too infrequently to be examined). For the food response inhibition and attention training versus generic response inhibition and attention training comparison we used fixed effects growth models, described above, to evaluate group differences on change in palatability ratings, willingness to pay for, and attentional bias for high-calories foods. A null relation between our manipulated factors and mediators would signal lack of support for our hypothesized mediation models. A significant relation would warrant full investigation of the Baron and Kenny (1986) criteria for mediation that includes: (1) condition predicts change in the mediator; (2) condition predicts change in the outcome; (3) change in the mediator predicts change in the outcome; and (4) the condition effect on change in the outcome becomes significantly weaker when controlling for change in the mediator.

Results

Preliminary Analyses

Participants were randomized to 4 conditions: (1) single-sex groups that completed food response inhibition and attention training (n = 54; 100% female), (2) single-sex groups that completed generic response inhibition and attention training (n = 61; 100% female), (3) mixed-sex groups that completed food response inhibition and attention training (n = 74; 66% female), and (4) mixed-sex groups that completed generic response inhibition and attention training (n = 72; 61% female). Figure 1 reports the participant flow through this trial. Attrition from pretest to posttest was 26% and did not significantly differ by condition (χ²[3,261] = 0.51, p = .917). Rates of missing data were 0% to 9% at pretest and 26% to 41% at posttest. Failure to provide complete data was not significantly associated (at p <.05) with age, reported sex at birth, or body fat and eating disorder symptoms at pretest. Descriptive statistics for the study variables are reported in Table 1. Inspection of Q-Q plots, frequency distributions, and measures of skew and kurtosis indicated all measures approximated normal distributions with the exception of the eating disorder symptom score, which was normalized with a log transformation. Conditions did not differ on pretest outcome measures with the exception of percent body fat (F [3,254] = 5.64, p = .001). Single-sex group with generic response and attention training had significantly greater pretest percent body fat than the single-sex group with food response inhibition and attention training and the mixed-sex group with generic response and attention training. Thus, analyses control for pretest body fat. The mean fidelity rating on a 90-point scale (10 = “no adherence”, 60 “fair”, 100= “perfect adherence”) was 79.4 (SD = 7.3) and the mean competence rating on a 90-point scale (10 = “very poor”, 60 “good/average”, 100= “superior”) was 77.1 (SD = 8.5).

Three-Way Interactions

The 3-way interaction of single- versus mixed-sex group × food versus generic response inhibition and attention training × time was significant for change in percent body fat (estimate = −1.25, SE = 0.50, t = −2.49, p = .014, d = −.28 [95% CI = −0.39, −0.05]). Figure 2a, a graph of the 3-way interaction, shows participants in single-sex groups that completed food response inhibition and attention training and participants in single-sex groups that completed generic response and attention training showed significant pretest to posttest decrease in percent body fat. Participants in mixed-sex groups did not show significant change in percent body fat. Non-significant 3-way interactions were found for change in eating disorder symptoms, weight concerns, and depressive symptoms.

Figure 2.

All panels represent pretest to posttest change in percent body fat. Figure 2a represents the 3-way interaction of single versus mixed sex group × food versus generic response inhibition and attention training × time. Figure 2b represents the 2-way interaction of single versus mixed sex groups × time and figure 2c the 2-way interaction of food versus generic response inhibition and attention training × time. Figures 2d and 2e represent the simple slopes test of single versus mixed sex groups, at conditional levels of food response inhibition and attention training (2d) and generic response and attention training (2e). P-values (p) and effect sizes (d) represent the test of within-group change figure 2a and between-group change for figures 2b–2e.

Two-way Interactions and Simple Slopes

Next, we tested, separately, the 2-way interactions of each between-subject factor with time. Results for percent body fat showed that the single- versus mixed-sex group × time interaction (estimate = −0.78, SE = 0.26, t = −3.04, p = .003, d = −0.18 [95% CI = −0.62, −0.13]) and the food versus generic response inhibition and attention training × time interaction (estimate = −0.73, SE = 0.26, t = −2.86, p = .005, d = −0.17 [95% CI = −0.60, −0.11]) were significant. Figure 2b shows a greater reduction in percent body fat for the single-sex group relative to the mixed-sex group and Figure 2c shows a greater reduction in percent body fat for food versus generic response inhibition and attention training.

Results for the test of simple slopes for body fat change (Figure 2d and 2e) showed a significant single- vs. mixed-sex group × time interaction for participants who received food response inhibition and attention training (estimate = −1.54, SE = 0.39, t = −3.96, p <.001, d = −0.35 [95% CI = −1.07, −0.35]). However, a non-significant single- vs. mixed-sex group × time interaction was found for participants who received generic response inhibition and attention training (estimate = −0.29, SE = 0.32, t = −0.91, p =.366, d = −0.07 [95% CI = −0.50, 0.19]).

Results for change in weight concerns showed that the 2-way interaction of single- vs. mixed-sex group × time interaction (estimate = −0.12, SE = 0.08, t = −1.54, p =.125, d = −0.19 [95% CI = −0.43, 0.05]) was non-significant, but the 2-way interaction of food versus generic response inhibition and attention training × time interaction (estimate = 0.18, SE = 0.07, t = 2.39, p = .018, d = 0.29 [95% CI = 0.05, 0.54]) was significant. The significant 2-way interaction showed a greater reduction in weight concerns for the generic versus food response inhibition and attention training, which was contrary to hypotheses. Results of the simple slopes showed a non-significant food versus generic response inhibition and attention training × time interaction (estimate = 0.05, SE = 0.12, t = 0.42, p = .679, d = 0.10 [95% CI = −0.28, 0.44]) on weight concerns for participants in same-sex groups. However, a significant food versus generic response inhibition and attention training × time interaction for weight concerns for participants in mixed-sex groups (estimate = 0.26, SE = 0.09, t = 2.75, p = .007, d = 0.36 [95% CI = 0.13, 0.78]), which as noted, was contrary to expectations. Non-significant 2-way interactions were found for eating disorder symptoms and depressive symptoms.

Birth Sex Moderation Analysis

All tests of moderation were non-significant indicating that, for each outcome, change over time as a function of single- vs. mixed-sex groups or food versus generic response inhibition and attention training did not differ for females relative to males.

Complier Analysis

We examined 2-way interactions of each factor with time for participants with high levels of compliance, defined as attendance of 5 or 6 sessions (n = 139, 53%). Table 2 presents the significance and effect sizes; the complier sample showed greater reductions in outcomes relative to the ITT analyses. Notably, a 39% greater reduction in percent body fat (d =−.25 vs. −.18) was found for the complier sample compared to the ITT sample when testing the single- versus mixed-sex groups × time interaction. Similarly, a 41% greater reduction in percent body fat (d = −.24 vs. −.17) was found for the complier sample compared to the ITT sample when testing the food versus generic response inhibition and attention training × time interaction. Also of note was the 61% greater reduction in weight concerns (d = −.29 vs. −.18) for the complier sample compared to the ITT sample when testing the single- versus mixed-sex groups × time interaction.

Mediation Analyses

Non-significant differences were found by sex composition of the groups for cohesion (t [245] = −0.02, p = .980, d = −.01 [95% CI = −0.37, 0.36]). Similarly, non-significant differences were found by sex composition of groups for total time participants talked during sessions (t [33] = 1.23, p = .226, d = .42 [95% CI = −0.26, 1.12]). Non-significant differences were also found for food versus generic response inhibition and attention training for change in palatability ratings for high-calorie foods (estimate = −0.12, SE = 0.14, t [171] = −0.88, p =.383, d = −0.15 [95% CI = −0.47, 0.18]), willingness to pay for high-calorie foods (estimate = −0.06, SE = 0.12, t [156] = −0.54, p =.588, d = −0.07 [95% CI = −0.41, 0.24]), and attentional bias for high calorie foods (estimate = 0.01, SE = 0.19, t [139] = 0.04, p = .964, d = 0.01 [95% CI = −0.32, 0.33]). Given the lack of evidence for relations between the two manipulated factors and the hypothesized mediators, we did not conduct any further mediational analyses.

Discussion

The present study adds to the evidence base for Project Health obesity/eating disorder prevention program. We experimentally manipulated two factors theorized to improve the efficacy of Project Health: sex composition of the group and adding food response inhibition and attention training. Both hypotheses were supported with regard to change in body fat. Regarding group composition, implementing Project Health groups in single- versus mixed-sex groups produced larger reductions in percent body fat (d =−.18), consistent with prior research finding that prevention and treatment interventions are more efficacious when delivered to female-only versus mixed-sex groups (e.g., Babinski et al., 2013; Hser et al., 2011). This is an important finding because this is the first obesity prevention trial to address this question.

Regarding the second factor hypothesized to enhance effectiveness, adding food response inhibition and attention training to Project Health groups produced stronger body fat reduction effects than generic response inhibition and attention training (d = −0.17). This finding is consistent with prior research that has found that food response inhibition training produced weight loss (Lawrence et al., 2015; Veling et al., 2014) and that food response inhibition and attention training produced body fat loss (Stice et al., 2017), but is particularly notable given that the response training was only 2 hours in duration, which was less than the 3.3 hours used in our original trial (Stice et al., 2017). Further, we found a significant interaction between these two factors (d = −.28), indicating that adding food response inhibition and attention training to Project Health produced stronger body fat prevention effects when the intervention was delivered in single- versus mixed-sex groups. Table 1 reveals that for every hour of response inhibition/attention training completed, there was a 2.3% reduction in body fat for single-sex Project Health groups relative to single-sex groups that completed generic training.

Complier analyses suggested that effects were more pronounced for participants who attended 5 or more of the 6 Project Health sessions, which suggests that improving retention and acceptability might result in larger body fat prevention effects. To this end, we collected qualitative feedback from our participants and group leaders regarding what they liked about the intervention and areas for improvement. This iterative process should allow us to improve attendance, which might contribute to larger prevention effects.

The present study also adds to the growing literature supporting the argument for focusing on body fat versus BMI in trials of obesity prevention or treatment interventions. Four past trials likewise found that body fat is a more sensitive outcome than BMI (Ranucci et al., 2017; Stice et al., 2015; Stice et al., 2017; Williamson et al., 2012). As seen in Table 1, the mean BMI values suggest that participants in each of the 4 conditions showed a reduction in BMI from pre to post-intervention. In contrast, the body fat means suggest that only participants in single-sex conditions showed a reduction in body fat; participants in the mixed -sex conditions showed an increase in body fat. Theoretically, BMI is a less sensitive outcome because it does not distinguish between changes in fat mass versus changes in muscle mass and bone mass. This is key because participants were asked to increase exercise, which reduces fat mass, but increases muscle mass and bone mass (Hamilton et al., 2010; Liberman et al., 2017).

We did not find support for our hypotheses that greater participation in dissonance-inducing discussions and greater group cohesion would mediate the effect of the sex composition of groups or that greater decreases in palatability ratings of, willingness to pay for, and attentional bias for high-calorie foods would mediate the effect of response training. Regarding the first mediation hypothesis, it is important to note that time spent talking during dissonance-induction exercises was a group-level variable, and thus we had less power to detect effects than participant-level variables. It might be useful for future research to evaluate the factors that mediate the effects of these two factors in a more adequately powered study.

Results did not provide evidence that manipulating the sex composition of the groups or adding food response inhibition and attention training resulted in greater reductions in the secondary outcomes of eating disorder symptoms or depressive symptoms. With regard to the sex composition of the groups, the findings suggest that any effect on dissonance induction primarily impacted lifestyle choices that impacted body fat, but that dissonance induction did not affect eating disorder symptoms and depressive symptoms. It seems logical that the dissonance-induction activities, which focused on producing dissonance for unhealthy lifestyle behaviors would have only impacted body fat. With regard to the response inhibition training, the food response training was expected to primarily affect dietary intake, which is probably why this manipulation did not affect eating disorder symptoms and depression symptoms. Nonetheless, it is important to note that averaging across the four conditions in which participants completed Project Health, there was a significant reduction in eating disorder symptoms (p < .001, d = −.76). The average reduction in eating disorder symptoms was −5.3, which was greater than the average reduction in the Project Health condition (−4.3) and the educational control condition (−3.4) from our previously published trial (Stice et al., 2018). Thus, results from the present trial suggest the Project Health produced the expected reductions in eating disorder symptoms, converging with evidence that this prevention program has produced a 60% reduction in future onset of eating disorders compared to controls in three past trials (Stice et al., 2008, 2013, 2018), but that the reductions in eating disorder symptoms were not increased when the intervention was implemented in single-sex groups or in conjunction with food response inhibition and attention training.

Averaging across the four conditions in which participants completed Project Health, there was also a significant reduction in depressive symptoms (p < .01, d = −.34). The fact that participants who completed Project Health showed reductions in depressive symptoms is important because it is inconsistent with the possibility that obesity prevention programs increase negative affect, such as shame and guilt, and stigmatize heavier individuals. As noted, the Project Health intervention script expressly requests that group participants do not make negative comments about obese individuals during the sessions. Indeed, we were unable to locate any randomized trials that found that obesity prevention programs produced a significant increase in negative affect relative to control conditions.

Unexpectedly, there was a greater decrease in weight concerns for participants in the generic versus the food response inhibition and attention training condition, which suggests that repeated exposure to food images might temper improvement in weight concerns that Project Health otherwise produces. Exposure to the sight of palatable foods can stimulate appetite, cravings, and hedonic hunger (Ferriday & Brunstrom, 2011; Jansen et al., 2003). It might be useful to test if elevated food cue reactivity attenuates the extent of improvement that occurs in weight concerns among participants receiving food response inhibition and attention training.

Study limitations should be noted. The most significant limitation was that the COVID-19 pandemic prevented us from conducting in-person posttest assessments for numerous participants, which contributed to a higher attrition (26%) than the observed in our previous trial of Project Health (8%; Stice et al., 2018), and prevented us from conducting most of the planned 6-, 12-, and 24-month follow-up assessments, forcing us to focus on pretest-to-posttest effects. The pandemic also forced us to close recruitment prematurely, which reduced statistical power, particularly for outcomes that were coded at the group-level. It will be important for future trials to evaluate the long-term effects of Project Health. A second limitation is that the experimental therapeutics design did not include a minimal-intervention or placebo control condition, given that all participants completed Project Health. We made this decision to maximize power to detect effects of sex composition and of adding food response inhibition and attention training but that design decision limited some inferences that could be drawn. A third limitation was that males and Black participants were under-represented in the sample.

A number of future directions can be offered. First, it would be useful to conduct a randomized effectiveness trial of Project Health when implemented in single sex-groups and delivered in conjunction with the food response inhibition and attention training compared to a credible control condition, such as an obesity education control condition to characterize the net effects of this combined intervention. Second, body fat gain prevention effects were larger among participants who completed more of the intervention sessions, suggesting that it would be useful to determine how to increase engagement in the intervention, as it may improve intervention effects. Third, it will be important to evaluate methods of supporting broad implementation of this and other effective weight gain prevention programs given the health benefits of reducing excess weight gain.

In sum, the present study identified two methods for enhancing the effects for an evidence-based obesity/eating disorder prevention program. As hypothesized, focusing group membership to include young people of the same sex and augmenting the Project Health intervention with food response inhibition and attention retraining magnified reductions in body fat. Although the two factors that we manipulated only produced increased body fat gain prevention effects, participants in all conditions showed large reductions in eating disorder symptoms that were comparable to the reductions observed in a past trial that compared Project Health to an obesity education control condition (Stice et al., 2018). As such, this study adds to the encouraging evidence that Project Health produces reliable prevention effects for both body fat gain and eating disorder symptoms, suggesting that efforts to broadly disseminate this program are warranted.

Public Health Significance:

The present study adds to the evidence base for Project Health as an obesity and eating disorder prevention intervention. Implementing in single-sex groups and adding computer-based food response inhibition and attention training improved body fat loss effects, but not eating disorder symptom reduction effects, though participants who completed Project Health in this trial showed clinically meaningful reductions in eating disorder symptoms.