Clinical Psychology Review 87 (2021) 102046

Contents lists available at ScienceDirect

Clinical Psychology Review

journal homepage: www.elsevier.com/locate/clinpsychrev

A meta-analytic review of trials that tested whether eating disorder

prevention programs prevent eating disorder onset
Eric Stice a, *, Z. Ayotola Onipede a, C. Nathan Marti b
a
Stanford University, USA
b
University of Texas at Austin, USA

A R T I C L E I N F O A B S T R A C T

Keywords: This report provides a review of randomized controlled trials that tested whether an eating disorder prevention
Eating disorder program significantly reduced future onset of eating disorders, which is important because eating disorders are
Prevention programs common and result in marked functional impairment. We identified 15 trials involving 5080 participants (mean
Effect size moderators
ages ranging from 14.5 to 22.3) that reported 19 tests of whether selective eating disorder prevention programs
Meta-analysis
significantly reduced future onset of eating disorders relative to some type of minimal control condition or a
credible alternative intervention. Healthy lifestyle modification prevention programs, dissonance-based pre­
vention programs, and a self-esteem/self-efficacy prevention program significantly reduced future onset of eating
disorders, though the later was only evaluated in one trial. Psychoeducational, cognitive behavioral, behavioral
weight gain, interpersonal, and family-therapy-based prevention programs did not significantly reduce future
onset of eating disorders. The average prevention effect size was statistically significant (OR = 1.64, 95% CI =
[1.09, 2.46], t = 2.54, p = .020) and there was heterogeneity in effect sizes (Q [18] = 35.96, p = .007). Pre­
vention effects were significantly larger for trials that recruited participants with elevations on a single risk factor
versus with elevations in multiple risk factors and for healthy lifestyle modification prevention programs versus
cognitive behavioral prevention programs, though the remaining examined factors did not moderate intervention
effect sizes (e.g., risk of bias).
The fact that lifestyle modification and dissonance-based prevention programs significantly reduced future
onset of eating disorders in multiple trials, producing a 54% to 77% reduction in future eating disorder onset
implies that broadly implementing these prevention programs could reduce the population prevalence of eating
disorders.

1. Introduction functional impairment than other psychiatric disorders, including

schizophrenia, major depression, and bipolar disorder (Crow & Smiley,
Representative data from the US, Italy, and Australia suggest that 2010) and are the second leading cause of mental health disability for
13% of females and 6% of males experience anorexia nervosa, bulimia adolescent girls and young women (Begg et al., 2007). Unfortunately,
nervosa, binge eating disorder and other specified and unspecified 80% of individuals with eating disorders do not receive treatment
feeding and eating disorders, such as subthreshold levels of these dis­ (Swanson, Crow, le Grange, Swendsen, & Merikangas, 2011) and
orders and purging disorder (Allen, Byrne, Oddy, & Crosby, 2013; evidence-based treatments produce symptom remission for only
Dakanalis et al., 2017; Stice, Marti, & Rohde, 2013). For both females 30–40% of patients (Fairburn et al., 2015; McIntosh et al., 2016; Won­
and males eating disorders are characterized by functional impairment, derlich et al., 2014). Thus, broadly implementing effective prevention
poorer academic performance, emotional distress, lower mental health programs might be a more effective way to address this public health
quality of life, and risk for future obesity, depression, suicide, and problem. Prevention programs may be particularly effective for eating
mortality (Allen et al., 2013; Arcelus, Mitchell, Wales, & Nielsen, 2011; disorders because the peak period for onset occurs between the ages of
Claydon & Zullig, 2019; Stice, Marti, & Rohde, 2013). Indeed, eating 16 to 19 (Hudson, Hiripi, Pope, & Kessler, 2007; Stice, Marti, & Rohde,
disorders show stronger relations to mortality, suicide attempts, and 2013), suggesting that if effective prevention programs were broadly

* Corresponding author at: Psychiatry and Behavioral Sciences, 401 Quarry Road, Stanford, CA 94305-5719, USA.
E-mail address: [email protected] (E. Stice).

https://doi.org/10.1016/j.cpr.2021.102046
Received 23 July 2020; Received in revised form 11 February 2021; Accepted 5 May 2021
Available online 21 May 2021
0272-7358/© 2021 Elsevier Ltd. All rights reserved.
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

implemented during adolescence, they could reduce the population Santonastaso, Friederici, & Favaro, 1999). Risk factor findings are
prevalence of eating disorders. summarized in Table 1. (See Fig. 1.)
The goal of this report is to review randomized controlled trials that Only a few studies were powered to identify risk factors that pre­
tested whether an eating disorder prevention program significantly dicted future onset of specific eating disorders, which include anorexia
reduced future onset of eating disorders. To our knowledge, no previous nervosa, bulimia nervosa, binge eating disorder, and purging disorder.
review has summarized findings from trials that tested for such pro­ Low body mass index (BMI), weight suppression, weight and shape
phylactic prevention effects, potentially because few trials were overvaluation, fear of weight gain, and impaired psychosocial func­
designed to detect a reduction in future onset of eating disorders. These tioning predicted future onset of threshold or subthreshold anorexia
trials are difficult to conduct because it is necessary to (1) recruit a large nervosa (Stice, Desjardins, Rohde, & Shaw, in press; Stice, Gau, Rohde,
sample, (2) conduct diagnostic interviews with all participants with & Shaw, 2017; Stice, Rohde, Shaw, & Desjardins, 2020). It is critical to
assessors who are blinded to the condition to which participants were note that several variables that are often considered established risk
assigned, and (3) assess the cohort over a long follow-up period to factors for anorexia nervosa were found to not predict onset of this
permit a high enough incidence of eating disorder onset in the control eating disorder in these prospective studies, including early puberty,
condition to detect a significant reduction in the intervention condition. perceived pressure for thinness, thin-ideal internalization, body dissat­
Large sample sizes and long follow-ups are necessary for sufficient isfaction, dieting, fasting, excessive exercise, compensatory behaviors,
power because inferential tests for detecting intervention effects on binge eating, feeling fat, and social support deficits. It is tempting to
dichotomous outcomes, such as onset of eating disorders, have less attribute these null findings to insufficient power, but the prospective
power than inferential tests for detecting change in continuous out­ studies that generated these null findings identified several factors that
comes, such as an eating disorder symptom composite measure and did predict future onset of anorexia nervosa, suggesting adequate power.
because the low incidence of eating disorders reduces sensitivity (Cohen, An alternative possibility is that the measures used to assess those
1988). One meta-analysis sought to summarize the effects of eating constructs were not sufficiently reliable, but most of those variables
disorder prevention programs on future onset of eating disorders at have predicted future onset of other eating disorders (as reviewed
universities (Harrer et al., 2019; Harrer, Cuijpers, Furukawa, & Ebert, below). Thus, a more parsimonious interpretation of these null findings
2019), but did not include all studies involving college students that met is that these variables do not increase risk for anorexia nervosa.
the stated inclusion criteria, did not include several studies that included Perpetuating beliefs about putative risk factors for specific eating dis­
younger adolescents, and was not able to include several recently pub­ orders that have not received empirical support in prospective studies
lished trials.
We first review findings from prospective studies that have identified Table 1
risk factors that predicted future onset of eating disorders because this Eating disorder risk factors.
knowledge should inform the optimal content of prevention programs. Eating disorder Eating disorder type
We then describe the methods we used to conduct this literature review risk factors
Anorexia Bulimia Binge Purging Any
and the variables that were hypothesized to correlate with the magni­
nervosa nervosa eating disorder eating
tude of the effects of the prevention programs on eating disorder onset (i. disorder disorder
e., moderators). Next, we provide a narrative review of randomized
Social
trials that tested for a reduction in future onset of eating disorders, Social pressure x x
noting strengths and limitations of each trial. We then report the results for thinness
of the meta-analysis that characterized the average eating disorder Pursuit of the x x x x
prevention effect achieved in these trials and examined moderators that thin beauty
ideal
correlate with larger eating disorder prevention effects, which should
Social support x x
inform efforts to broadly implement the most effective prevention pro­ deficits
grams. Because the small number of identified trials limited power to Impaired x x x x
identify moderators that correlate with larger prevention effects, the psychosocial
functioning
moderator analyses should be considered preliminary. Last, we discuss
Psychological
implications of these findings for efforts to prevent eating disorders and Overvaluation x x x
important research directions. of weight and
shape
1.1. Risk factors that predict future onset of eating disorders Self- x
objectification
Body- x x x x
Several prospective studies have identified risk factors that predict dissatisfaction
future onset of any eating disorder. We focus on studies that used Feeling fat x x x
diagnostic interviews, as this is the gold-standard method of diagnosing Fear of weight x x x x
gain
eating disorders. We conducted a systematic review of several electronic
Negative affect x x x x
databases (e.g., PsycInfo, Web of Science, and MedLine) to identify pro­ Low x
spective studies that have sought to identify variables that predicted interoceptive
future onset of any eating disorder or one of the four specific types of awareness
eating disorders and reviewed findings from published reviews that Biological
Parental x
focused on this topic (e.g., Stice, 2016). All of the identified studies overweight
appeared to have had sufficient power based on the sample size and the Overeating x x x
incidence of eating disorder onset over follow-up, as each study was able Binge eating x x x
to identify baseline variables that did significantly predict future onset Dieting x x x x
Weight x x x
of eating disorders. Social pressure for thinness, pursuit of the thin
suppression
beauty ideal, self-objectification, body dissatisfaction, parental over­ Low BMI x
weight, dieting, negative affect, social support deficits, alcohol use, and Compensatory x x
female sex predicted future onset of any eating disorder (Allen, Byrne, behaviors
Oddy, Schmid, & Crosby, 2014; Dakanalis et al., 2017; Ghaderi & Scott, Alcohol use x x
Female sex x
2001; Jacobi et al., 2011; McKnight, 2003; Rohde, Stice, & Marti, 2015;

2
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

Records idenfied through database

Idenficaon searching (PsycInfo, Web of Science, Addional records idenfied
MedLine) through other sources
(n = 3314) (n = 12)

Records aer duplicates removed

(n = 2625)
Screening

Records screened Records excluded

(n = 2625) (n = 2413)

Full-text arcles assessed Full-text arcles excluded, with reasons

for eligibility (n = 197)
(n = 212) x No test for eang disorder onset
(n = 142)
Eligibility

x No diagnosc interview used to

measure eang disorder onset
Studies included in (n = 46)
qualitave synthesis x No RCT/control group (n = 9)
(n = 15)

Studies included in
Included

quantave synthesis
(meta-analysis)
(n = 15)

Fig. 1. Flow diagram.

has the potential to hinder needed etiologic studies and can result in weight gain, body dissatisfaction, feeling fat, dieting, weight suppres­
prevention and treatment interventions that are ineffective. sion, overeating, binge eating, compensatory behaviors, negative affect,
Social pressure for thinness, pursuit of the thin beauty ideal, weight and impaired social functioning predicted future onset of purging dis­
and shape overvaluation, fear of weight gain, body dissatisfaction, order (Stice et al., 2017; Stice et al., in press; Stice, Rohde, Shaw, &
feeling fat, overeating, binge eating, dieting, weight suppression, Desjardins, 2020). Social pressure to be thin, BMI, and exercise did not
compensatory behaviors, negative affect, social support deficits, predict future onset of purging disorder in these studies.
impaired psychosocial functioning, alcohol use, and low interoceptive Thus, emerging data suggest that the risk factors for the specific
awareness predicted future onset of threshold or subthreshold bulimia eating disorders are somewhat distinct. Nonetheless, pursuit of the thin
nervosa (Killen et al., 1996; Patton, Johnson-Sabine, Wood, Mann, & ideal, weight and shape overvaluation, body dissatisfaction, fear of
Wakeling, 1990; Patton, Selzer, Coffey, Carlin, & Wolfe, 1999; Stice, weight gain, dieting, weight suppression, feeling fat, overeating, binge
Davis, Miller, & Marti, 2008; Stice et al., in press; Stice et al., 2017; Stice, eating, negative affect, and impaired psychosocial functioning predicted
Marti, & Durant, 2011; Stice, Rohde, Shaw, & Desjardins, 2020). Again, future onset of eating disorders in at least three prospective studies or
it is important to note variables that did not predict future onset of have predicted future onset of at least three specific eating disorders.
bulimia nervosa in these studies, which included perfectionism, matu­ Results suggest that prevention programs that reduce risk factors with
rity fears, interpersonal distrust, BMI, peer dieting, exercise, and the greatest evidence-base, such as these, may prove effective in
parental separation. decreasing future eating disorder onset. However, these data also sug­
Thin-ideal internalization, weight and shape overvaluation, fear of gest that many prevention programs target variables that are not
weight gain, body dissatisfaction, feeling fat, dieting, overeating, binge established risk factors for anorexia nervosa and that it might be
eating, negative affect, and impaired psychosocial functioning predicted necessary to target different risk factors and at-risk populations to pre­
future onset of threshold or subthreshold binge eating disorder (Stice vent this particular eating disorder. Although targeting risk factors that
et al., 2017; Stice et al., in press; Stice, Rohde, Shaw, & Desjardins, have been found to predict onset of all four specific types of eating
2020). BMI, exercise, weight suppression, fasting, and compensatory disorders might be optimal for preventing all of these disorders, which
weight control behaviors did not predict future onset of binge eating include negative affect, impaired psychosocial functioning, weight and
disorder in these studies. shape overvaluation, and fear of weight gain, none of the prevention
Thin-ideal internalization, weight and shape overvaluation, fear of programs evaluated in trials that we identified target these risk factors.

3
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

2. Methods if at least one domain was considered “high risk” and another domain
was considered “some concerns”. Risk of Bias was assessed by the second
2.1. Search strategy author and carefully reviewed by the first author. When discrepancies
arose, the authors reviewed the published reports together and arrived
To identify relevant trials, the first two authors conducted a database at a consensus.
search to retrieve articles was performed on PsycInfo, Web of Science, and
MedLine for the years January 1980 – May 2020 using the following
keywords: eating disorder, anorexia, bulimia, binge eating, prevention, 2.4. Meta-analytic statistical methods
intervention, and onset. Further, two previously published systematic
reviews by Stice, Shaw, and Marti (2007) and Watson et al. (2016) were 2.4.1. Effect size estimation
examined for reports not captured in the database search. We also Effect sizes were based on odds ratios from the final assessment point
examined reference lists of relevant literature for additional reports and that were transformed into their natural log for analysis in the models
contacted active eating disorder prevention researchers for any relevant, described below. Odds ratios were coded so that a positive value reflects
unpublished data. Studies written in other languages would have been improvement in the outcome (i.e., interventions resulted in fewer eating
included, but no published or unpublished studies that met our inclusion disorders than control groups). Odds ratios of 1.68 (0.60), 3.47 (0.29),
criteria were found in languages other than English. and 6.71 (0.15) are benchmarks for small, medium, and large effect sizes
Studies were included in this review if they met the following respectively (Chen, Cohen, & Chen, 2010). Effect sizes and effect size
criteria: (1) the study aimed to evaluate an eating disorder prevention sampling variance estimates were computed using the R metafor pack­
program, (2) the program was evaluated in a randomized controlled trial age (Viechtbauer, 2010) which implements methods described in Lipsey
(RCT) that contained a minimal intervention control condition (e.g.,
assessment-only or waitlist control condition) or a credible alternative Table 3
intervention (e.g., expressive writing or a health education comparison Odds ratio and number needed to treat for prevention programs versus controls/
condition), and (3) eating disorder incidence over follow-up in each alternative interventions.
condition was assessed using diagnostic interviews. Authors (intervention vs. control/alternative Odds Number needed
A total of 3326 studies were identified from the literature search as intervention) and year Ratio to treat
potentially eligible for inclusion. The abstracts and measures sections of Favaro, Zanetti, Huon, Sanosanto (Intervention vs. 2.22 17.46
each article were screened, articles that met at least two points of control) 2005
eligibility (RCT and tested an eating disorder prevention program) were Jacobi, Volker, Trockel, & Taylor (Student bodies+ 7.92 16.00
read to further determine if all criteria were met. Studies that met the vs control) 2012
Stice, Rohde, Shaw, & Gau (Body project vs control) 1.38 61.77
outlined eligibility criteria were then reviewed to finalize the sample of
2011
studies. Stice, Rohde, Butyrn, Shaw, & Marti (Body project 1.08 108.34
vs control) 2015
2.2. Data extraction Stice, Rohde, Shaw, & Gau (Clinician-led Body 0.88 53.72
Project vs control) 2020
Stice, Rohde, Shaw, & Gau (Peer-led Body Project vs 2.36 10.85
Data was extracted from each report to describe study and sample control) 2020
characteristics (e.g., intervention theory, delivery modality, duration, Stice, Rohde, Shaw, & Gau (Ebodyproject vs 1.18 44.41
length of follow up). A coding manual was used for data extraction. Data control) 2020
was extracted and coded by the second author and carefully reviewed Stice, Rohde, Shaw, & Gau (Clinician-led Body 0.37 9.03
Project vs peer-led Body Project) 2020
and checked by the first author. When discrepancies arose in coding, the Stice, Rohde, Shaw, & Gau (Clinician-led Body 0.75 24.31
authors reviewed each study together and arrived at a consensus Project vs Ebodyproject) 2020
(consulting with the authors of the retrieved articles in some instances). Stice, Rohde, Shaw, & Gau (Peer-led Body Project vs 2.00 14.36
Inter-rater agreement was 100% after the consensus discussions. Ebodyproject) 2020
Jacobi et al. (Parents Act Now vs control) 2018 0.74 –
Martinsen et al. (Intervention vs control) 2014 22.80 24.00
2.3. Research quality assessment Stice, Marti, Spoor, Presnell, & Shaw (Body Project 2.50 11.05
vs control) 2008
The Cochrane Collaboration Risk of Bias criteria were used to eval­ Stice, Marti, Spoor, Presnell, & Shaw (Healthy 2.88 10.05
uate the risk of bias for each trial. The Risk of Bias criteria measures Weight vs control) 2008
Stice, Marti, Spoor, Presnell, & Shaw (Body Project 1.28 55.27
research quality in five domains: randomization process, deviations vs Expressive Writing) 2008
from intended interventions, missing outcome data, outcome measure­ Stice, Marti, Spoor, Presnell, & Shaw (Healthy 1.47 36.90
ment, and selective outcome reporting (Higgans et al., 2011). Within Weight vs Expressive Writing) 2008
these domains, specific factors such as randomization sequence, allo­ Stice, Marti, Spoor, Presnell, & Shaw (Body Project 0.87 111.00
vs Healthy Weight) 2008
cation concealment, blinding of assessor, and statistical analyses, are
Stice, Marti, Spoor, Presnell, & Shaw (Expressive 1.96 13.80
assessed. For each domain, the risk of bias was categorized as ‘high’, Writing vs control) 2008
‘low’, or ‘some concerns’. Under the ‘Deviations from Intended Inter­ Taylor et al. (Student Bodies vs control) 2006 0.69 20.06
vention’ domain, criteria related to blinding of participants and inter­ Taylor et al. (Image and Mood vs control) 2016 1.40 14.98
vention facilitators were not included, as this type of blinding was not Ghaderi, Stice, Andersson, Persson, & Allzén 4.73 7.93
(Virtual Body Project vs control) 2020
possible for the interventions included in this meta-analysis. Risk of Bias Tanofsky-Kraff et al. (Intervention vs control) 2014 4.27 16.33
in this domain was determined by use of appropriate statistical analyses Stice, Rohde, Shaw, & Marti (Healthy Weight vs 2.79 18.72
(i.e. intention-to-treat analyses). Intention-to-treat analyses was defined control) 2013
as the inclusion of all randomized participants in the final outcome Stice, Rohde, Shaw, & Gau (Project Health vs 0.96 1063.75
Healthy Weight) 2018
analysis. Missing outcome data were considered ‘low risk’ in cases when
Stice, Rohde, Shaw, & Gau (Healthy Weight vs 3.17 19.17
attrition was below 20%, participant flowchart provided clear reporting control) 2018
of drop-outs, and reasons for dropout were not related to the study Stice, Rohde, Shaw, & Gau (Project Health vs 3.06 19.52
condition to which they were randomized, or trials reported no signifi­ control) 2018
cant differences in baseline characteristics of study drop-outs versus Lowe, Stice, Butryn, Sawrer (intervention vs. 0.54 8.78
assessment control)
non-drop-outs. Studies were given a categorization as High Risk of Bias

4
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

and Wilson (2001). (See Table 3.) an unconditional model that provides an overall effect size. We used the
unconditional model to examine effect size heterogeneity using the Q
2.4.2. Intervention group contrasts statistic (Cochran, 1954) that evaluates the null hypotheses that effect
Effect sizes were computed for prevention programs versus minimal size variance is zero. In the event of significant heterogeneity, we next
intervention control conditions and for prevention programs versus examined moderators that could potentially explain variability in bi-
credible alternative interventions. In a preliminary analysis, we variate regression models. Intervention theory contained three levels
compared effect sizes for the interventions versus minimal intervention (cognitive behavioral, dissonance, and healthy lifestyle modification);
control conditions and interventions versus credible alternative in­ we dummy-coded for dissonance and healthy lifestyle modification,
terventions and did not find a significant difference. Thus, we limited the using cognitive behavioral programs as the reference group. Continuous
effect sizes included in the meta analytic models to prevention programs moderators were standardized (i.e., z score) prior to fitting models so
versus minimal intervention control conditions with the exception of that continuous moderators were on the same scale. Continuous mod­
two studies that did not have a minimal intervention control condition erators were evaluated in linear and quadratic models and, in the event
(Ghaderi, Stice, Andersson, Enö Persson, & Allzén, 2020; Tanofsky-Kraff of a significant quadratic effect, the quadratic model would be preferred;
et al., 2014) and thus included effect sizes based on prevention programs there were no significant quadratic effects and the continuous modera­
versus alternative interventions. tors were thus fit in linear models. Significant moderator effects were
probed by estimating marginal odds ratios (i.e., odds ratios predicted by
2.4.3. Operationalization and coding of effect size moderators the model). Categorical moderators were estimated at each level of the
Table 2 defines each moderator that we extracted from the studies moderator; continuous moderators were estimated at 1 SD above and
and characterizes the distribution of the moderators in the identified below the mean. In addition, we included all moderators in a multi­
studies. We attempted to code average attendance and homework variate model.
completion rates, but the identified studies did not report these data in a We assessed publication bias using Egger’s regression test for funnel
consistent form, preventing analyses of those moderators. There were plot asymmetry with effect sizes pooled at the study level, which is
missing data on several moderators and homework assigned had only preferable to visual inspection of funnel plots (Simmonds, 2015). We
one instance of one of the levels of that moderator, prompting us to omit used the sample size as the independent variable given limitations using
it as well. We hypothesized that prevention effects would be larger for standard errors for the natural log of the odds ratio due to its association
trials that evaluated longer versus shorter interventions and included with the natural log of the odds ratio (Peters, Sutton, Jones, Abrams, &
home exercises, as both should result in greater uptake of the concepts Rushton, 2006). The number needed to treat (NNT) was calculated using
and skills taught in the prevention program. We hypothesized that ef­ the NNT function in the dmeatar package (Harrer, Adam, et al., 2019;
fects would be larger for trials with longer follow-up periods, as this Harrer, Cuijpers, et al., 2019) which implements methods described in
should increase the incidence of eating disorder onset and thus power to Kraemer and Kupfer (2006). NNT reflects the number of participants
detect effects. Based on the pattern of findings revealed in the narrative who must complete an intervention to prevent the onset of one eating
review, we hypothesized that prevention effects would be larger for disorder (Kraemer & Kupfer, 2006).
trials that delivered the prevention program in-person versus via the
internet. We also hypothesized that prevention effects would be larger 2.5. Results of the literature search
for trials that recruited participants with elevations on a single risk
factor versus with elevations in multiple risk factors, as the greater Fifteen trials were identified that tested for prevention of future
heterogeneity in risk potentially introduced more noise to the data, onset of eating disorders. These 15 trials evaluated the prevention effects
which could reduce power. In addition, we hypothesized that prevention with 5080 participants from the United States and Europe. Thirteen
effects would be larger for dissonance-based and lifestyle modification- trials included only females; two trials included both males and females.
based prevention programs than for cognitive-behavioral prevention Studies focused primarily on adolescent and young adult populations
programs (referred to as intervention theory). Finally, we hypothesized with mean ages ranging from 14.5 to 22.3. Six of the 14 trials did not
that prevention effects would be smaller for trials with a lower risk of report racial/ethnic demographic data for participants, which limited
bias. (See Table 4.) the ability to understand potential differences in response to in­
terventions. Of the 9 trials that did report these demographics, the racial
2.4.4. Meta-analysis models distribution was: 65.2% white, 13.1% Hispanic/Latino, 10.4% Asian/
All effect size analyses were modeled using multilevel random- Pacific Islander, 5.4% Black or African-American, 2.6% other or more
effects models that were implemented using the R metafor package than one race, 1.6% American Indian/Alaskan Native, and 0.2% Native
(Viechtbauer, 2010). Models were fit using Restricted Maximum Like­ Hawaiian/Pacific Islander. This composition is a reasonable represen­
lihood (REML) which corrects downward bias in the maximum likeli­ tation of the college and high school populations from which the trials
hood estimates of the covariance matrix of the random effects when the sampled. No trial has tested whether an eating disorder prevention
number of studies is small as is the case in the present meta-analysis program produces similar reductions in eating disorder onset for
(Schwarzer, Carpenter, & Rücker, 2015). The Knapp and Hartung different ethnic groups.
(2003) method was used, which produces more accurate standard errors In the 15 identified trials, 19 contrasts tested whether an eating
and thus more accurate inferential statistics and confidence intervals disorder prevention program reduced future onset of eating disorders
(van Aert & Jackson, 2019) in addition to reducing the false positive rate relative to a comparison condition. We categorized prevention programs
(IntHout, Ioannidis, & Borm, 2014). as universal, selective, or indicated using the Institute of Medicine’s
The multilevel random-effect model is a variance-known multilevel levels of classification (Mrazek & Haggerty, 1994). An intervention was
model that accounts for non-independence in effect sizes from a com­ considered universal if it targeted the entire population, selective if it
mon source (Hox, 2010), which is the case in the present meta-analysis targeted a subgroup of the population at elevated risk (variables
wherein independent treatment groups are compared with a common considered risk factors are summarized in Table 1), and indicated if it
control group in some of the trials. The effect size variance is based on targeted participants with symptoms of an eating disorder below the
the effect size variance described above and variation in other model diagnostic threshold. All prevention programs were considered selective
parameters is estimated in a multilevel model framework. The covari­ based on the reported inclusion criteria, as none targeted the entire
ance of dependent effect sizes is unknown in meta-analysis but is population or required that all participants reported eating disorder
captured in overestimating study-level covariation (Van den Noortgate, symptoms. Three reports described their prevention programs as indi­
López-López, Marin-Martinez, & Sánchez-Meca, 2013). We initially fit cated, however upon examination of the intervention theory and

5
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

Table 2
Study characteristics.
Study Sample Sample Study conditions (1. Length of Outcome Intervention Intervention format Intervention theory Risk of
size (age, % intervention, 2. follow-up measures duration bias
female, % control) (months) (hours)
White)

Favaro et al. 138 M = 17.0 1. Psychoeducational 12 SCID 12 Classroom, in- Psychoeducation High
(2005) (SD = 1.1), intervention person
100%, N/A 2. Assessment only
control
Jacobi et al., 126 M = 22.3 1. Student bodies+ 6 EDE 5 Internet Cognitive High
2012 (SD = 2.9), 2. Waitlist control Behavioral
100%, N/A
Stice et al. 306 M = 15.7 1. Body project 36 EDDI 4 group based; in- Dissonance Based Low
(2011) (SD = 1.1), 2. Educational person
100%, 82% brochure
Stice, Rohde, 408 M = 21.6 1. Body project 36 EDDI 4 Group based; In- Dissonance Based Low
et al. (SD = 5.6), 2. Educational person
(2015) 100%, 58% brochure
Stice, Rohde, 680 M = 22.2 1. Clinician-led Body 48 EDDI 4 Group based; in- Dissonance Based Some
Shaw, and (SD = 7.1), Project person Concerns
Gau (2020) 100%, 60% 2. Educational video
Stice, Rohde, 680 M = 22.2 1. Peer-led Body 48 EDDI 4 Group based; in- Dissonance Based Some
Shaw, and (SD = 7.1), Project person Concerns
Gau (2020) 100%, 60% 2. Educational video
Stice, Rohde, 680 M = 22.2 1. Ebodyproject 48 EDDI 4 Self-guided, Internet Dissonance Based Some
Shaw, and (SD = 7.1), 2. Educational video Concerns
Gau (2020) 100%, 60%
Stice, Rohde, 680 M = 22.2 1. Clinician-led Body 48 EDDI 4 Group based; in- Dissonance Based Some
Shaw, and (SD = 7.1), Project person Concerns
Gau (2020) 100%, 60% 2. Peer-led Body
Project
Stice, Rohde, 680 M = 22.2 1. Clinician-led Body 48 EDDI 4 Group based; in- Dissonance Based Some
Shaw, and (SD = 7.1), Project person Concerns
Gau (2020) 100%, 60% 2. Ebodyproject
Stice, Rohde, 680 M = 22.2 1. Peer-led Body 48 EDDI 4 Group based; in- Dissonance Based Some
Shaw, and (SD = 7.1), Project person/ Internet Concerns
Gau (2020) 100%, 60% 2. Ebodyproject
Jacobi et al. 66 M = 13.8 1. Parents Act Now 12 EDE 999 Family based; Family Based High
(2018) (SD = 1.1), 2. Wait list control Internet Treatment
100%, N/A
Martinsen 577 M = 16.5 1. Intervention 9 EDE 6 Classroom, in- Self Esteem/Self High
et al. (SD = 0.3), 2. Assessment only person Efficacy
(2014) 33.7%, N/ control Enhancement
A
Stice et al. 481 M = 17.0 1. Body project 36 EDDI 3 Group based; In- Dissonance Based Low
(2008) (SD = 1.4), 2. Assessment only person
100%, 58% control
Stice et al. 481 M = 17.0 1. Healthy weight 36 EDDI 3 Group based; In- Lifestyle Low
(2008) (SD = 1.4), 2. Assessment only person Modification
100%, 58% control
Stice et al. 481 M = 17.0 1. Body project 36 EDDI 3 Group based; In- Dissonance Based Low
(2008) (SD = 1.4), 2. Expressive writing person
100%, 58%
Stice et al. 481 M = 17.0 1. Healthy weight 36 EDDI 3 Group based; In- Lifestyle Low
(2008) (SD = 1.4), 2. Expressive writing person Modification
100%, 58%
Stice et al. 481 M = 17.0 1. Body project 36 EDDI 3 Group based; In- Dissonance Based Low
(2008) (SD = 1.4), 2. Healthy weight person
100%, 58%
Stice et al. 481 M = 17.0 1. Expressive writing 36 EDDI 3 Group based; In- Psychoeducation Low
(2008) (SD = 1.4), 2. Assessment only person
100%, 58% control
Taylor et al. 480 M = 20.8 1. Student bodies 36 EDE 5 Internet Cognitive High
(2006) (SD = 2.6), 2. Waitlist control Behavioral
100%, 60%
Taylor et al. 206 M = 20 1. Image and mood 24 EDE 5 Internet Cognitive High
(2016) (SD = 1.8), 2. Waitlist control Behavioral
100%, 51%
Ghaderi et al. 443 M = 17.3 1. Virtual body project 24 EDDI 4 Group based; Virtual Dissonance Based Low
(2020) (SD = 1.4), 2. Expressive writing (via
100%, N/A teleconferencing)
Tanofsky- 113 M = 14.5 1. Interpersonal 12 EDE 20 Group based; In- Interpersonal High
Kraff et al. (SD = 1.7), psychotherapy person Psychotherapy
(2014) 100%, 2. Health education
56.6% program
398 24 EDDI 4 Low
(continued on next page)

6
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

Table 2 (continued )
Study Sample Sample Study conditions (1. Length of Outcome Intervention Intervention format Intervention theory Risk of
size (age, % intervention, 2. follow-up measures duration bias
female, % control) (months) (hours)
White)

Stice, Rohde, M = 18.4 1. Healthy weight Group based; In- Lifestyle

et al. (SD = 0.6), 2. Educational person Modification
(2013) 100%, 83% brochure
Stice et al. 364 M = 19.1 1. Project health 24 EDDI 6 Group based; In- Lifestyle Low
(2018) (SD = 1.2), 2. Healthy weight person Modification
72%, 68%
Stice et al. 364 M = 19.1 1. Healthy weight 24 EDDI 6 Group based; In- Lifestyle Low
(2018) (SD = 1.2), 2. Educational video person Modification
72%, 68%
Stice et al. 364 M = 19.1 1. Project health 24 EDDI 6 Group based; In- Lifestyle Low
(2018) (SD = 1.2), 2. Educational video person Modification
72%, 68%
Lowe et al. 294 M = 17.0 1. Intervention 24 EDE 6 Group based; In- Behavioral Therapy High
(2012) (SD = 1.1), 2. Assessment only person
100%, N/A control

Abbreviations: SCID, Structured Clinical Interview for DSM-IV Axis I Disorders; EDE, Eating Disorder Examination; EDDI, Eating Disorder Diagnostic Interview.

in which we describe each prevention program, the research design, the

Table 4
sample, and the analytic approach. We also attempted to gauge the
Operationalization and descriptive statistics for moderators.
acceptability of each prevention program. However, the papers did not
Moderator Coding description and criteria Descriptive provide parallel information about completion rates of the prevention
statistics
programs, which made it impossible to apply a single definition of
Length of follow-up Continuous variable reflecting length M = 27.32; SD = acceptability. We could only report the data provided in each trial about
of follow-up in months. 13.38; completion of the prevention program and comment on whether this
Range = (6, 48)
Length of Continuous variable reflecting the M = 5.72, SD =
suggested “good”, “reasonable”, “moderate”, or “low” acceptability. For
intervention length of the intervention in hours 4.10; studies that reported the percent of participants who completed all of the
Range = (3,20) intervention sessions, the most consistently reported completion statis­
Homework Dichotomous variable reflecting Yes = 16; tic, we characterized acceptability to be “good” if over 80% of the
whether intervention assigned No = 1
sample completed all sessions, “reasonable” if between 60% to 79%
homework. 1 = Yes; 0 = No
Implementation Dichotomous variable reflecting Internet = 6; completed all sessions, “moderate” if between 50% to 59% completed all
modality whether intervention was delivered In-person = 13 sessions, and “low” if less than 50% completed all sessions.
over the internet or in-person. 1 =
Internet; 0 = No
2.7. Psychoeducational prevention program
Single risk sample Dichotomous variable reflecting Yes = 13;
whether intervention recruitment No = 6
was based on a single risk factor (v. One trial tested whether a psychoeducational eating disorder pre­
multiple risk factors). 1 = Yes; 0 = No vention program reduced future onset of eating disorders during a 1-
Intervention theory Categorical variable reflecting Dissonance = 7; year follow-up compared to an assessment-only control condition
whether intervention was a Cognitive
dissonance-based, cognitive behavioral = 4;
(Favaro, Zanetti, Huon, & Santonastaso, 2005). This prevention pro­
behavioral, or a healthy lifestyle Healthy lifestyle gram focused on discouraging use of unhealthy weight control practices
modification modification = 4 (e.g., strict dieting) and the cultivation of skills to respond to the media’s
Low risk of bias Dichotomous variable reflecting Yes = 11; overemphasis on thinness, consistent with evidence that dietary re­
whether intervention was low or No = 8
striction and pursuit of the thin beauty ideal predict future onset of
moderate risk (v. high risk) 1 = Yes; 0
= No eating disorders. It also provided information about eating disorders and
risk factors that predict onset of eating disorders. This effectiveness trial
recruited 141 female adolescents from a vocational training school,
inclusion criteria it was determined that these programs did not fit the though only 129 provided complete data (mean age 17.0 SD 1.1). This
parameters of indicated prevention. With regard to the intervention trial did not take an intent-to-treat analytic approach wherein data from
theory of the prevention programs evaluated in these 19 contrasts, 1 was all participants randomly assigned to each condition are included in
psychoeducational, 1 was based on interpersonal psychotherapy, 1 was analyses of intervention effects (Little & Yau, 1996). Statisticians have
based on family treatment, 1 focused on promoting self-esteem and self- shown that using multiple imputation to impute missing data, allowing
efficacy, 1 focused on weight gain prevention using behavioral therapy an intent-to-treat analytic approach, produces more accurate and effi­
techniques from obesity treatments and principles from nutrition sci­ cient parameter estimates than alternative analytic approaches such as
ence, 3 promoted lifestyle refinements that bring caloric intake into analyzing only participants who provide complete data, making multi­
balance with caloric expenditure, 3 were based on cognitive behavioral ple imputation the recommended procedure for handling missing data
therapy, and 8 were dissonance-based prevention programs. These (Graham, 2009; Schafer & Graham, 2002). Investigators have imputed
programs were delivered as in-person group interventions (n = 13), the data used to make eating disorder diagnoses (e.g., Allen et al., 2014).
virtually delivered group interventions (n = 1), and internet-delivered Thus, the test of the intervention effects is probably overly liberal
interventions (n = 5). The interventions varied from 3 h to 20 h. compared to trials that used intent-to-treat analyses because the par­
ticipants who did not provide complete data may have not completed
2.6. Narrative review of the identified trials the prevention program. We considered this a selective prevention
program because it focused solely on females, and female sex has been
We provide a narrative review of the trials that we located that tested found to increase risk for onset of eating disorders (Allen et al., 2014).
whether a prevention program reduced future onset of eating disorders, The prevention program was delivered to groups of participants in 6 2-h

7
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

sessions. Participants completed diagnostic interviews that assessed any intervention condition versus 14% in the control condition.
threshold or subthreshold eating disorder with assessors who were A second efficacy trial tested whether Student Bodies reduced future
blinded to the condition of participants. In total, 5.3% of participants onset of eating disorders over a 6-month follow-up compared to an
who received the prevention program developed an eating disorder assessment-only control condition (Jacobi, Völker, Trockel, & Taylor,
during the 1-year follow-up versus 11.0% in the control condition. 2012). Participants were 126 young women (mean age = 22.3, SD = 2.9)
Although this represented a 52% reduction in future onset of eating who reported weight concerns, dieting, binge eating, or purging. Given
disorders, this difference was not significant (t = 1.17, n.s.), likely that weight concerns and dieting are not symptoms of eating disorders,
because of limited power. we classified this as a selective prevention program. Although 126
participants were randomized to condition, data from only 103 partic­
2.8. Cognitive-behavioral prevention programs ipants were included in analyses. Again, because the authors did not
include all of participants randomized to condition, results likely pro­
Three trials tested whether Student Bodies, a cognitive-behavioral vide an overly liberal test of intervention effects. Eating disorders were
eating disorder prevention program reduced future onset of eating dis­ assessed by diagnostic interviews and assessors were blinded to the
orders. The first was an efficacy trial that tested whether Student Bodies condition to which participants had been randomized. Participants
reduced future onset of eating disorders during a 1- to 3-year follow-up assigned to Student Bodies opened 67% of the pages and completed 38%
compared to an assessment-only control condition (Taylor et al., 2006). of the required postings (we note the completion rates reported in the
Student Bodies is delivered in 8 30-min Internet-delivered modules that articles, which differed across studies, but sometimes was not reported).
focus on reducing weight concerns using cognitive behavioral therapy Onset of eating disorders over follow-up was 7% in the control condition
and included moderated asynchronous on-line discussions between compared to 0% in the intervention condition. Although the authors did
group participants and a facilitator. Participants were young women not report a test of whether the incidence of eating disorder onset was
(mean age = 20.8; SD = 2.6) who reported weight concerns, fear of significantly lower in the intervention versus control condition, we
weight gain, and overvaluation of weight and shape with a BMI between conducted a Fisher’s exact test (because of the low incidence) that
18 and 32 at baseline, making this a selective prevention trial. Eating revealed that the reduction in future eating disorder onset was not sta­
disorders were assessed with diagnostic interviews by assessors who tistically significant (p = .109).
were blinded to the condition to which participants were assigned. In A third efficacy trial tested whether Student Bodies produced a
total, 480 participants were randomized to condition, but only data from reduction in future onset of any eating disorders over 2-year follow-up
421 were included in the analyses because the rest of the participants compared to an assessment-only control condition (Taylor et al.,
only provided baseline data (i.e., analyses were not conducted on an 2016). The intervention targeted individuals with weight concerns plus
intent-to-treat basis). Because the authors did not include all partici­ being teased about weight, having current or historical depression, or
pants randomized to condition, results likely provide an overly liberal reporting compensatory weight control behaviors, leading us to classify
test of intervention effects. Excluding participants who were random­ this as a selective prevention program. This version of Student Bodies was
ized to Student Bodies who did not start or complete the prevention adapted for use with this more selective population, resulting in a longer
program from the analyses of intervention effects violates the principles 10-week intervention. In this trial, Student Bodies also included monthly
of a randomized controlled trial in which all participants randomized to email messages to participants about the content of the intervention and
condition are retained in analyses (Shadish, Cook, & Campbell, 2002). reminders to use the skills from the intervention over a 9-month follow-
Although the authors did not provide the sample size for the survival up (making this nearly a 12-month intervention). In total 206 partici­
model and Dr. Taylor was unable to provide the sample size when pants were recruited at universities (mean age = 20.0, SD = 1.8) and
contacted, calculations suggest that only 261 participants were randomized to condition. However, only 185 participants were included
included. The authors state that they excluded individuals with a in the analyses, meaning analyses were not intent-to-treat and that the
threshold or subthreshold eating disorder at baseline and that 43 par­ test of intervention effects was likely overly liberal. Eating disorders
ticipants developed an eating disorder over follow-up. Fig. 2 from Taylor were assessed via diagnostic interviews by assessors who were blinded
et al. (2006) indicates that 19% of Student Bodies participants showed to condition. Of the participants assigned to Student Bodies 5% never
onset of an eating disorder versus 14% of controls. Proportionally this logged onto the program and another 10% dropped out after logging
means that 57.5% of the cases must have occurred in the Student Bodies into the program; of the remaining participants only 56% completed at
condition and 42.4% in the control condition (19% + 14% = 33%; 19%/ least half of the intervention sessions. These data also suggest low
33% = 57.5%; 14%/33% = 42.4%). If 43 participants showed onset of acceptability of this version of Student Bodies. By 2-year follow-up 24%
an eating disorder over follow-up, 25 of the 43 cases must have been in of intervention participants showed onset of an eating disorder
the Student Bodies condition and 18 in the control condition (43 × compared to 31% of the assessment-only control group; this difference
57.5% = 25 and 43 × 42.4% = 18). If 25 cases represent 19% of the was not statistically significant (Hazard Ratio [HR] = 0.73, p = .28).
participants in the Student Bodies condition, the cell size for that con­
dition must have been 132 (25/19% = 1.32 × 100 = 132). If 18 cases 2.9. Dissonance-based eating disorder prevention programs
represent 14% of the participants in the control condition, the cell size
for that condition must have been 129 (18/14% = 1.28 × 100 = 129). Five trials evaluated whether a dissonance-based eating disorder
These calculations indicate that the sample used in the survival model prevention program, referred to as the Body Project, reduced onset of
must have been 132 + 129 = 261. Given that Taylor et al. (2006) stated eating disorders. One efficacy trial tested whether Body Project produced
that 206 participants from the Student Bodies condition and 215 partic­ a reduction in future onset of eating disorders over 3-year follow-up
ipants from the control condition were included in analyses (206 + 215 relative to a credible alternative prevention program (Healthy Weight),
= 421), this means that data from only 261 of the 421 participants who a credible expressive writing alternative intervention, and an
were supposed to be included in the analyses were included and that 160 assessment-only control condition (Stice, Marti, Spoor, Presnell, &
participants (38%) must have been excluded from the survival model for Shaw, 2008). Participants in the group-delivered Body Project preven­
reasons not stated. Attrition from the prevention program (16%) was tion program engage in verbal, written, and behavioral exercises in
greater than attrition from the assessment-only control condition (9%). which they collectively explore the negative effects of pursuing the thin
Overall 11% of participants assigned to Student Bodies never logged into ideal, which putatively prompts participants to reduce their subscription
the program, suggesting low acceptability. There were no significant to this ideal because people are motivated to align their attitudes with
differences in onset of any eating disorder over the 1- to 3-year follow- their publicly displayed behavior. Theoretically this intervention re­
up. The incidence of eating disorder onset over follow-up was 19% in the duces pursuit of the thin ideal, which decreases body dissatisfaction,

8
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

dieting, and negative affect, which should reduce eating disorder peer educators or to Internet delivery, without loss of effectiveness in
symptoms and future eating disorder onset. Participants in the group- reducing future eating disorder onset over 4-year follow-up compared to
delivered Healthy Weight prevention program are encouraged to make an educational video control condition (Stice, Rohde, Shaw, & Gau,
gradual lifestyle refinements to bring energy intake into balance with 2020). Young women (N = 680; mean age 22.2) with body image con­
energy expenditure, based on the reasoning that this would reduce body cerns were randomly assigned to clinician-led Body Project groups, peer-
dissatisfaction, a key risk factor for eating disorders. Both prevention led Body Project groups, the internet-delivered eBody Project, or an
programs were described as body acceptance interventions and imple­ educational video control condition. Clinicians and peers delivered the
mented in three 1-h sessions by a clinician and an undergraduate peer standard 4 1-h session enhanced dissonance version of the Body Project.
educator. In the expressive writing condition participants write about The eBody Project contains activities designed to mirror the activities in
any emotionally important topic of their choosing in 3 individuals 45- the Body Project and is implemented without a moderator to reduce cost.
min sessions. Participants were 481 adolescent girls (mean age 17.0, Assessors were blinded to the condition of participants. Analyses were
SD 1.4) recruited from high schools (51%) and colleges who reported intent-to-treat. In clinician-led Body Project groups 47% of participants
body image concerns. Participants were randomized to condition. As­ completed all sessions, in peer-led Body Project groups 45% of partici­
sessors blinded to condition conducted diagnostic interviews over 3-year pants completed all sessions, and in the eBody Project condition 57% of
follow-up. Analyses included data from the full sample (i.e., were intent- participants completed all modules, suggesting lower acceptability of
to-treat), which means that those who were randomized to one of the the Body Project than in past trials. It was noteworthy that in this trial the
prevention programs but did not attend any sessions were included in completion rate was 11% higher for the eBody Project than for the in-
analyses, making this a conservative test of intervention effects. With person Body Project, replicating the evidence that the completion rate
regard to attendance, 91% of Body Project participants attended all 3 for the eBody Project was 10% higher than for the in-person Body Project
sessions, 91% of Healthy Weight participants attended all 3 sessions, and from an earlier trial (Stice, Rohde, Durant, & Shaw, 2012), as this pro­
98% of expressive writing controls completed all 3 writing sessions, vides evidence that digital eating disorder prevention programs can
suggesting good acceptability. At 3-year follow-up, onset of any eating have higher acceptability than in-person prevention programs. By 4-
disorder was 7% for Body Project participants, 6% for Healthy Weight year follow-up, eating disorder onset was 8.1% in the peer-led Body
participants, 9% for expressive writing controls, and 16% for Project condition, 19.3% in the clinician-led Body Project condition,
assessment-only controls. Eating disorder onset was significantly lower 15.5% in the eBody Project condition, and 17.6% in educational video
for Body Project participants and Healthy Weight participants relative to control participants. Eating disorder onset was significantly lower in the
assessment only controls (HR = 2.75, p = .022 and HR = 2.50, p = .018, peer-led Body Project condition than in the clinician-led Body Project
respectively), but not relative to expressive writing controls. As such, condition (HR = 2.53, p = .009) and the educational video control
Body Project produced a 60% reduction in future eating disorder onset condition (HR = 2.33, p = .020), and marginally lower than in the eBody
and Healthy Weight produced a 61% reduction in future eating disorder Project condition (HR = 2.04, p = .056). Thus, peer-led Body Project
onset. groups produced a 54% reduction in eating disorder onset compared to
A second trial tested whether the Body Project produced a reduction educational video controls, a 58% reduction in eating disorder onset
in future eating disorder onset over 3-year follow-up compared to an compared to clinician-led Body Project groups, and a 48% reduction in
educational brochure control condition (Stice, Rohde, Shaw, & Gau, eating disorder onset relative to the eBody Project.
2011). In this effectiveness trial high school clinicians recruited partic­ A fifth trial tested whether Body Project groups delivered virtually
ipants and delivered Body Project groups. To make it easier for clinicians over the internet by peer educators produced a reduction in future
to cover the material, the intervention was delivered in 4 1-h sessions in eating disorder onset over 2-year follow-up relative to an expressive
contrast to 3 1-h sessions as in previous trials, though the content was writing comparison condition (Ghaderi et al., 2020). In this effectiveness
similar. In total, 306 female high school students (mean age 15.7, SD = trial 443 female adolescents and young women (mean age = 17.2, SD =
1.1) who reported body image concerns were recruited and randomized. 1.5) with body image concerns were randomized to virtual Body Project
Participants completed diagnostic interviews at all assessments with groups, an expressive writing comparison condition, or a 6-month
assessors blinded to condition. Intent-to-treat analyses were used. Of the waitlist control condition (participants in the latter condition were not
306 participants, 78% attended all 4 sessions of the intervention, included in the onset analyses because they did not provide data for the
implying reasonable acceptability. By the end of the 3-year follow-up full 2-year follow-up). The standard Body Project was delivered virtually,
4.0% of Body Project participants showed onset of an eating disorders with only minimal modifications to improve cultural fit and virtual
compared to 5.5% in the educational brochure control condition, which implementation. Assessors were blinded to the condition of participants
was not a significant difference (p = .28). and analyses were intent-to-treat. In total, 72% of participants
A third trial tested whether the Body Project produced a reduction in completed at least half of the Body Project sessions. By the end of the 2-
future eating disorder onset over 3-year follow-up compared to an year follow-up eating disorder onset was 2.0% in the virtual Body Project
educational brochure control condition (Stice, Rohde, Butryn, Shaw, & condition and 8.8% in the expressive writing comparison condition, a
Marti, 2015). In this effectiveness trial college clinicians recruited par­ significant difference (HR = 3.85 p = .037). This translated into a 77%
ticipants and delivered Body Project groups. This trial evaluated a new reduction in future eating disorder onset.
enhanced-dissonance version of the Body Project that underscored the
voluntary commitment, increased accountability for counter-attitudinal 2.10. Lifestyle modification eating disorder prevention programs
exercises, and increased the level of effort required because these factors
increase dissonance induction. In total, 408 female college students Three trials evaluated an eating disorder prevention program that
(mean age 21.6, SD = 5.6) who reported body image concerns were encourages participants to make lasting lifestyle modifications that
recruited and randomized to condition. Assessors were blinded to the bring caloric intake into balance with caloric expenditure, which should
condition of participants. Among participants in the Body Project con­ reduce body dissatisfaction, a key risk factor for eating disorders. The
dition, 84% attended or made up all 4 sessions, suggesting good first efficacy trial, which was summarized previously because it also
acceptability. Analyses were intent-to-treat. By the end of the 3-year evaluated the effects of the Body Project (Stice et al., 2008) found that
follow-up 12.9% of Body Project participants showed onset of an Healthy Weight significantly reduced future onset of eating disorders by
eating disorders compared to 12.9% in the educational brochure control 61% relative to assessment-only controls. A second efficacy trial tested
condition, which was not a significant difference (p = .16). whether Healthy Weight reduced future onset of eating disorders over a
A fourth effectiveness trial tested whether implementation of the 2-year follow-up compared to an educational brochure control condition
Body Project could be task shifted from implementation by clinicians to (Stice, Rohde, Shaw, & Marti, 2013). Participants were 398 young

9
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

women and men (mean age = 18.4, SD = 0.6) who reported weight among female and male adolescents who were attending high schools
concerns and were randomly assigned to condition. Assessors who for athletes (Martinsen et al., 2014). This intervention primarily focused
conducted diagnostic interviews were blinded to condition. Analyses on improving self-esteem and self-efficacy and was delivered to students
were intent-to-treat. In total, 67% of Healthy Weight participants atten­ in 4 90-min sessions and in social media advertisements by respected
ded all four sessions, suggesting only moderate acceptability. By the end athletes. These intervention elements were supplemented with educa­
of the 2-year follow-up 3.4% of Healthy Weight participants showed tional seminars for coaches and school staff on puberty and disordered
onset of any eating disorder, compared to 8.5% in the educational eating. Overall, 55.5% of participants attended all intervention sessions,
brochure control condition, a significant difference (p = .039). This suggesting moderate acceptability. In total, 577 participants were ran­
translated into a 60% reduction in future eating disorder onset. These domized to condition, though it appears that only 439 participants were
findings are noteworthy because Healthy Weight was the first prevention included in the analyses (i.e., analyses were not intent-to-treat). Ran­
program to significantly reduce future eating disorder onset in two trials domized participants were selectively assessed at baseline, and only
and the only one to produce this effect in a sample containing both fe­ participants that provided data at follow-up were included in the ana­
male and male adolescents. lyses. Diagnostic interviews were used to assess eating disorders, though
A third trial tested whether a new dissonance-based version of assessors were not blinded to condition. By the end of the 1-year follow-
Healthy Weight (referred to as Project Health) reduced future onset of up, 0% of intervention participants showed onset of an eating disorder
eating disorders relative to the original Healthy Weight prevention pro­ compared to 4.2% of controls. Although the authors did not test whether
gram and an educational video control condition (Stice, Rohde, Shaw, & the incidence of eating disorder onset was lower in the intervention
Gau, 2018). In the group-based Project Health participants make the versus control condition for the full sample, we conducted a Fisher’s
small lasting lifestyle improvements to bring caloric intake into balance exact test that revealed that this effect was statistically significant (p =
with caloric expenditure, as in the original Healthy Weight prevention .001). This effect translates into a 100% reduction in future eating dis­
program, and also discuss the negative effects of obesity, overeating, and order onset.
a sedentary lifestyle and the benefits of a healthy body weight, eating a
healthy diet, and regular exercise. These latter activities theoretically 2.13. Interpersonal therapy-based eating disorder prevention program
create dissonance about engaging in behaviors that contribute to un­
healthy weight gain. In this effectiveness trial young women and men One efficacy trial tested whether an interpersonal psychotherapy
with weight concerns (N = 364) were randomized to the three condi­ intervention reduced future eating disorder onset compared to a health
tions and completed blinded diagnostic interviews assessing eating education intervention over 1-year follow-up (Tanofsky-Kraff et al.,
disorders over a 2-year follow-up. Analyses were intent-to-treat. In the 2014). Both interventions were group-based and lasted 20-h, which is a
Project Health condition 36% of participants attended all sessions and in positive design feature because only three other trials have compared
the Healthy Weight condition 48% of participants attended all sessions, two prevention programs that were equal in credibility and contact time.
suggesting low acceptability. By 2-year follow-up 3% of participants in Participants were 116 adolescent girls recruited from the community
Project Health, 3% of participants in Healthy Weight, and 8% of partici­ who were overweight and reported loss of control eating, making this a
pants in the educational control condition showed onset of an eating trial of selective prevention programs. However, the participant flow­
disorder. The 63% reduction in future eating disorder onset in the two chart noted that only 98 were included in the analyses, indicating that
eating disorder prevention programs relative to the control condition analyses were not intent-to-treat, despite the claims in the article. Data
was only marginal (HR = 2.77 p = .064). indicated 65% of interpersonal psychotherapy participants and 85% of
health education participants attended at least 80% of the sessions,
2.11. Behavioral therapy-based eating disorder & weight gain prevention suggesting moderate acceptability. By 12-month follow-up, 8.2% of
program participants in the health education intervention developed an eating
disorder compared to 2.0% participant in interpersonal psychotherapy
One unpublished efficacy trial tested whether a weight gain pre­ (a 76% reduction in eating disorder onset), though this difference was
vention program reduced future onset of eating disorders over 2-year not statistically significant (OR = 4.27, p = .20), likely due to limited
follow-up among female college students (Lowe, Stice, Butryn, & power caused by the small sample.
Sarwer, 2012). This prevention program used behavioral therapy tech­
niques from obesity treatments to reduce caloric intake and increase 2.14. Family-therapy-based eating disorder prevention program
exercise, principles from nutrition science that focused on reducing the
energy density of dietary intake, and persuasion principles from social One efficacy trial tested whether a parent-based internet-delivered
psychology to increase motivation for change. Participants in this pro­ intervention, Parents Act Now, reduced onset of anorexia nervosa over
gram were female college students (M age = 18.24, SD = 0.44) who were 12-month follow-up relative to an assessment-only control condition
at high-risk for weight gain and eating disorders by virtue of body (Jacobi et al., 2018). This intervention included activities from the first
dissatisfaction, elevated dieting, and elevated weight suppression, phase of family-therapy for anorexia nervosa developed by Lock and Le
making this a selective prevention program. Among participants Grange (2001), which primarily aims to have parents play the key role in
assigned to this prevention program, 26% attended all six sessions and re-feeding their underweight children. Female adolescents who reported
78.7% attended at least half of the sessions, suggesting low accept­ a low body weight, as well as weight concerns, drive for thinness,
ability. In total, 294 participants were randomized to the intervention or perfectionism, amenorrhea, excessive exercise, or family history of an
an assessment-only control condition. Analyses for this trial were intent- eating disorder were recruited along with their parents. It is critical to
to-treat. By 2-year follow-up, 18.1% of participants in the prevention note that only a low BMI has been shown to predict future onset of
program showed onset of an eating disorder, compared to 13.2% in the anorexia nervosa (Stice et al., 2017); none of the other factors (e.g., body
control condition, which was not a significant difference (OR = 0.69, dissatisfaction, perfectionism, pursuit of the thin ideal, excessive exer­
95% CI [0.74–2.83], p = .278). cise, amenorrhea or parental history of an eating disorder) have been
shown to predict future onset of anorexia nervosa in a prospective risk
2.12. Self-esteem/self-efficacy enhancement eating disorder prevention factor study. Female adolescents (mean age = 13.8, SD = 1.0) and their
program parents (N = 66) were recruited, but due to extreme attrition (59% by
12-month follow-up), this trial was terminated before the full sample
One effectiveness trial tested whether an intensive 1-year classroom was recruited. Unfortunately, 50% of participants assigned to the
delivered prevention program reduced future onset of eating disorders intervention dropped out of treatment, parents opened only 28% of the

10
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

program pages, and 29% of the parents never logged into the program, procedures; 26.7% of studies did not use blinded assessors, and 60% of
signaling very low acceptability of this intervention. Assessors who studies either did not report or did not perform allocation sequence
conducted diagnostic interviews were not blinded to intervention con­ concealment in their randomization procedures. Risk of bias findings are
dition of participants. However, results revealed that no participants in presented in Fig. 3.
either condition, including the assessment-only control condition,
showed onset of anorexia nervosa. The fact that none of the participants 3.2. Average effect size and effect size heterogeneity
in the control condition showed onset of anorexia nervosa is consistent
with the impression that this population was not at high-risk for this The literature search identified 15 published or unpublished reports
eating disorder, likely because they did not target youth with established from which we extracted data for 19 contrasts for prevention programs.
risk factors for anorexia nervosa. We requested data on whether par­ Table 1 provides an overview of the effect size from each prevention
ticipants showed future onset of any other threshold or subthreshold program. The average effect size was significantly different from zero
eating disorder, but the first author did not think that they collected this (OR = 1.64, 95% CI = [1.09, 2.46], 95% prediction interval = [0.50,
data. 5.32], t = 2.54, p = .020). The heterogeneity in effect sizes was signif­
In sum, the 15 trials that we located provided 19 tests of whether a icant (Q [18] = 35.98, p = .007). Egger’s regression test for funnel plot
prevention program produced a significant reduction in future eating asymmetry was not significant (t[13] = 0.04, p = .972), providing no
disorder onset over 1- to 4-year follow-up periods relative to either a evidence for a relation between effect size and sample size and thus no
minimal intervention control condition or another prevention program. evidence of publication bias.
Seven of these tests revealed significant reductions in eating disorder
onset over follow-up compared to some type of comparison condition. 3.3. Moderator analysis
The dissonance-based Body Project significantly reduced future onset of
eating disorders when implemented or co-implemented by peer educa­ To maximize what can be learned from this literature, we conducted
tors in three trials, but not when implemented solely by adult clinicians exploratory analyses of moderators. We restricted our focus to moder­
in three trials. This pattern of findings suggests that the intervention is ators that were available for at least 14 of the 19 contrasts and we
more credible if implemented by someone similar to the participants in excluded any moderators for which fewer than two studies provided
age and sex, implying that future implementation projects would be effect sizes for each level of the moderator (e.g., only one prevention
more effective in preventing eating disorder onset if peer educators program involved no homework). Correlations between the examined
deliver or co-deliver the Body Project. Further, the lifestyle modification- potential moderators are shown in Table 5. Results for the bivariate
based Healthy Weight prevention program significantly reduced future moderator analyses are displayed in Table 6. Intervention effects were
onset of eating disorders in two trials, and produced a marginal reduc­ significantly larger for trials that recruited participants with elevations
tion in eating disorder onset in a third trial. The new dissonance-based on only a single risk factor versus recruiting participants with elevations
Project Health also produced a marginal reduction in future eating dis­ in multiple risk factors (t = 2.14, p = .047); marginal odds were 2.00
order onset. The fact that these lifestyle modification interventions (95% CI = [1.31, 3.06]) and 0.95 (95% CI = [0.52, 1.73]) for single risk
produced significant or marginal reductions in future eating disorder factor and multiple risk factor trials, respectively. In the intervention
onset is encouraging, though acceptability of these interventions was theory model, healthy lifestyle modification programs exhibited signif­
only moderate. These two lifestyle modification prevention programs icantly larger eating disorder prevention effects than cognitive behav­
also significantly reduced excess weight gain, suggesting that these ioral programs (t = 2.32, p = .039); marginal odds were 1.15 (95% CI =
prevention programs are unique in preventing both eating disorder [0.64, 2.06]), 1.47 (95% CI = [>1.00, 2.15]), and 2.93 (95% CI = [1.52,
onset and obesity onset. The prevention program targeting self-esteem 5.67]) for cognitive behavioral, dissonance, and healthy lifestyle
and self-efficacy among high school athletes also significantly reduced modification programs, respectively. The average eating disorder pre­
onset of future eating disorders, suggesting that it would be useful to vention effect was not significantly larger for dissonance-based pre­
attempt to replicate those findings. In contrast, the cognitive-behavioral vention programs than for cognitive behavioral prevention programs.
Student Bodies eating disorder prevention program did not significantly None of the other examined moderators correlated with intervention
reduce onset of future eating disorders in the three trials included in this effect sizes, including length of follow-up, length of the prevention
review. Results imply that cognitive behavioral eating disorder pre­ program, implementation modality, and risk of bias. None of the mod­
vention programs are not effective in reducing future onset of eating erators showed significant unique effects when entered simultaneously
disorders, potentially because of the limited acceptability of this inter­ in a multivariate model.
vention. However, it is also possible that the Internet is not the ideal
medium for eating disorder prevention programs. Data likewise indi­
cated that psychoeducational, behavioral therapy-based, interpersonal-
therapy-based, and family-therapy-based prevention programs did not
reduce future onset of eating disorders.
Table 5
Correlation matrix of moderators.
3. Meta-analytic results
1. Length of follow-up in
months
3.1. Research quality assessment
2. Length of intervention in − 0.51
hours
Of the 15 studies included in this meta-analysis, 6 studies (40%) were 3. Implementation − 0.12 − 0.24
considered low risk of bias, 1 study (6.7%) had some concerns for bias, modality
and 8 studies (53.3%) were considered high risk of bias. Missing 4. Single risk sample 0.43 − 0.32 − 0.51
5. Dissonance intervention 0.66 − 0.39 − 0.09 0.47
outcome data, statistical analysis methods, and measurement of 6. Healthy lifestyle − 0.19 0.00 − 0.43 0.30 − 0.56
outcome introduced the most bias in the included studies. Nine studies modification
(60%) were high risk or had some concerns for bias regarding missing intervention
data and its impact on results. Six studies (40%) did not use intent-to- 7. Low bias risk 0.68 − 0.48 − 0.34 0.80 0.56 0.36
treat analyses when testing for intervention effects, likely contributing Note: p values are not provided as moderator values from the same study are not
to an overly liberal estimation of intervention effects. Other sources of independent; thus, the correlation matrix is intended for descriptive purposes
bias included non-blinded assessors and problems with randomization only.

11
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

Table 6
Moderator models for effect sizes from prevention programs versus control conditions.
Moderator Parameter Estimate SE t p σ between σ within I2 between [95% CI] I2 within [95% CI]

Length of follow-up in months Intercept 0.58 0.21 2.79 0.013 0.52 0.18 51.79 [0, 54.57] 5.95 [0, 34.35]
Length of follow-up − 0.40 0.26 − 1.55 0.139
Length of intervention in hours Intercept 0.53 0.21 2.60 0.019 0.50 0.23 50.44 [0, 54.95] 10.18 [0, 35.88]
Length of intervention 0.18 0.28 0.64 0.534
Implementation modality Intercept 0.56 0.22 2.49 0.023 0.44 0.31 37.03 [0, 50.08] 19.04 [0, 39.35]
Internet delivery − 0.18 0.35 − 0.52 0.611
Single risk factor sample Intercept − 0.05 0.28 − 0.17 0.865 0.34 0.21 29.89 [0, 49.85] 11.31 [0, 35.36]
Single risk factor 0.74 0.35 2.14 0.047
Intervention theory Intercept 0.14 0.27 0.51 0.619 <0.01 0.24 <0.01 [0, 43.24] 21.94 [0, 39.61]
Dissonance 0.24 0.32 0.76 0.461
Healthy lifestyle modification 0.94 0.40 2.32 0.039
Risk of bias Intercept 0.15 0.28 0.55 0.591 0.38 0.23 34.64 [0, 52.74] 12.02 [0, 34.99]
Low bias risk 0.50 0.36 1.40 0.179

Note: Confidence intervals computed using the Q-profile method (Viechtbauer, 2007).

4. Discussion effects were rated as low risk of bias, as this suggests that it is unlikely
that the conclusions are misleading because the significant prevention
The goal of this report was to characterize the results of randomized effects were driven by trials with a low risk of bias.
controlled trials that tested whether an eating disorder prevention The dissonance-based Body Project showed the most repeated effects
program significantly reduced future onset of eating disorders. Our of reduction of eating disorder onset, in that this prevention program (1)
literature search identified 15 trials that provided 19 tests of whether a significantly reduced future onset of eating disorders in 3 separate trials
prevention program reduced future onset of eating disorders relative to a conducted in 2 countries (Ghaderi et al., 2020; Stice et al., 2008; Stice,
minimal intervention control condition or a credible alternative inter­ Rohde, Shaw, & Gau, 2020), (2) was the only prevention program shown
vention. Results revealed that three prevention programs significantly to produce a significantly greater reduction in future onset of eating
reduced the future onset of eating disorders over follow-up in one or disorders relative to a credible alternative intervention, and did so in
more trials (Fig. 2). Risk for bias ratings for the trials ranged from low to two trials, and (3) produced prevention effects that persisted through 4-
high, with 8 trials being rated high risk for bias, one being rated as some year follow-up. Yet these results were nuanced in that these prevention
concerns, and 6 being rated as low risk for bias. It was reassuring that effects only emerged when this intervention was delivered or co-
five of the six trials that detected significant eating disorder prevention delivered by peer educators who were similar to participants in terms

Fig. 2. Forest plot of odds ratios with 95% confidence intervals. Point sizes are inverse-variance weighted to indicate the contribution of each individual study.
Asterisks indicate effect size contrast not included in the meta-analysis models.

12
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

Fig. 3. Risk of bias summary.

of sex and age. Three trials found that the Body Project significantly the prevention programs evaluated in the identified trials. Theoretically
reduced future onset of eating disorders when delivered or co-delivered the preventive effects of the Body Project emerged because this inter­
by peer educators and 3 trials found that the Body Project did not vention reduces pursuit of the thin beauty ideal, which has been shown
significantly reduce future onset of eating disorders when delivered by to increase risk for future eating disorder onset (Dakanalis et al., 2017;
clinicians. Indeed, the fact that in one trial, peer-led Body Project groups Rohde et al., 2015; Stice et al., 2017). Consistent with the intervention
produced a significantly greater reduction (OR = 0.37 [95% CI = theory, reductions in thin-ideal internalization mediate the effect of this
[0.18–0.79]) in future eating disorder onset relative to clinician-led prevention program on reduction in eating disorder symptoms (Seidel,
Body Project groups powerfully illustrates this point. We thus conduct­ Presnell, & Rosenfield, 2009; Stice et al., 2007), this prevention program
ed post hoc analyses that tested whether reductions in future onset of reduces brain reward region response to thin models (Stice, Yokum, &
eating disorders were significantly larger for Body Project groups Waters, 2015), as well as attentional-bias for thin models (Tobin, 2020)
implemented or co-implemented by peer educators versus clinicians. and implicitly-assessed valuation of the thin ideal (Kant, Wong-Chung,
Intervention effects were significantly larger when the Body Project was Evans, Stanton, & Boothroyd, 2019).
implemented by peer educators versus clinicians (t = 2.88, p = .045); The healthy lifestyle modification prevention program Healthy
marginal odds were 2.70 (95% CI = [1.29, 5.67]) and 1.02 (95% CI = Weight produced significant reductions in future onset of eating disor­
[0.57, 1.81])) for peer-led and clinician-led groups, respectively. The ders in two trials (Stice et al., 2008; Stice, Rohde, et al., 2013), though in
fact that the 2.70 odds ratio was not within the 95% confidence interval the third trial, reductions in future eating disorder onset were only
for the effects of cognitive behavioral prevention programs (0.64, 2.06) marginal (Stice et al., 2018), potentially because that trial included
also provides evidence that peer-implemented Body Project groups pro­ males, which lowered the incidence of eating disorder onset over follow-
duce significantly stronger prevention effects than cognitive behavioral up and thus sensitivity to detect preventive effects. The NNT effect sizes
programs. In addition, peer-delivered, but not clinician-delivered Body indicated that it would be necessary to implement this prevention pro­
Project groups produced a marginally greater reduction in future onset of gram to between 10 and 20 participants to prevention one case of an
eating disorders than the internet-delivered eBody Project. It was also eating disorder. Thus, the effect size for this prevention program was
noteworthy that virtually implemented Body Project groups produced a smaller than for the Body Project. These findings might be considered
significant reduction in eating disorder onset relative to a credible unexpected because many clinicians and researchers believe that dietary
alternative intervention, because virtual implementation could mark­ restriction increases risk for eating disorders (e.g., Fairburn, 1997;
edly expend the reach of this prevention program and allow easy Heatherton & Polivy, 1992; Neumark-Sztainer et al., 2006). However,
implementation to most communities because internet access is now this prevention program encourages lasting gradual reductions in caloric
widespread. The NNT effect sizes indicated that it would be necessary to intake and increases in physical activity to bring caloric intake into
implement this prevention program to 8 to 11 participants to prevent balance with caloric expenditure, rather than transient dietary restric­
onset of one case of an eating disorder, the largest effect sizes for any of tion. The evidence that a prevention program that reduces future weight

13
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

gain and onset of overweight and obesity significantly reduces onset of This intervention focused primarily on attaining a healthy weight via
eating disorders is also consistent with evidence that overeating in­ involvement of the family in feeding. However, this trial did find that the
creases risk for future onset of eating disorders (Stice et al., 2017). The family therapy prevention program produced significantly greater in­
fact that healthy lifestyle modification prevention programs reduce risk creases in BMI than observed in the control condition. This study’s lack
for both future onset of eating disorders and overweight/obesity should of significant reduction in anorexia nervosa onset may have occurred
be considered a positive feature given that overweight and obesity is the because of the low participant engagement, which forced the authors to
second leading cause of premature mortality in Western countries close recruitment before the planned sample was enrolled. Thus,
(Flegal, Carroll, Kit, & Ogden, 2012), markedly eclipsing the mortality although this prevention program attempted to target a unique popu­
produced by eating disorders (by a factor of 30). Nonetheless, this pre­ lation that is at high risk for anorexia nervosa, this population seemed to
vention program was somewhat less acceptable to participants than the have limited motivation to complete the intervention and the low
dissonance-based Body Project, based on the somewhat lower attendance acceptability of this prevention program would likely preclude broad
and completion data for the former prevention program. implementation of this intervention.
The other prevention program to significantly reduce future eating Results likewise revealed that the prevention program that used
disorder onset focused on increasing self-efficacy and self-esteem behavioral therapy principles from weight loss treatments and concepts
(Martinsen et al., 2014). The NNT for this prevention program, in­ from nutrition science did not significantly prevent eating disorders
dicates that 24 participants would need to receive this intervention in (Lowe et al., 2012). The low acceptability of this prevention program
order to prevent the emergence of one case of an eating disorder, which and the fact that this trial recruited participants based on multiple risk
is a smaller effect size than observed for the dissonance-based and life­ factors may have played a key role in the lack of preventive effects
style modification-based prevention programs. The theoretical un­ Finally, the psychoeducational program did not produce a statisti­
derpinnings of this prevention program are based on the idea that cally significant reduction in onset of eating disorders (Favaro et al.,
individuals must have the required personal skills, and degree of self- 2005). Although the incidence of eating disorders reported in the
efficacy to change behaviors (Bandura, 1986). However, low self- intervention condition were lower than that of the control condition, the
esteem and low self-efficacy have not been shown to predict future difference was not significant. These findings are in line with the con­
onset of eating disorders to our knowledge. Indeed, prior prevention clusions drawn from eating disorder prevention meta-analyses (Stice
programs that focused on increasing self-esteem have not produced et al., 2007; Yager & O’Dea, 2008) and other prevention areas that
significant reductions in eating disorder symptoms (O’Dea, 1995; Wade, psychoeducational interventions are not effective in producing behav­
Davidson, & O’Dea, 2003). Though the intervention evaluated by ioral change (e.g., Larimer & Cronce, 2002).
Martinsen et al. (2014) produced significant results, based on the NNT it
does not appear to reduce eating disorder onset at the same magnitude 4.1. Effects of the meta-analytic review
as the above described prevention programs. Moreover, that trial has a
high risk for bias, due to missing outcome data and the use of non- The average prevention effect size was statistically significant (OR =
blinded assessors. These aspects of bias have the potential to favor the 1.64, 95% CI = [1.09, 2.46], t = 2.54, p = .020) and there was hetero­
intervention in the results, so it is important to consider the results geneity in effect sizes (Q [18] = 35.96, p = .007). These results converge
within this context. Based on the data provided, the program did reduce with those from an earlier meta-analytic review with a narrower focus
eating disorder onset more effectively than some other prevention pro­ on trials conducted at universities (Harrer, Adam, et al., 2019; Harrer,
grams included in this analysis, however it would be useful to evaluate Cuijpers, et al., 2019).
this program in another trial with stronger methodological and statis­ Moderator analyses provided evidence that eating disorder preven­
tical rigor in future trials. tion effects were significantly larger for trials that recruited participants
The remaining eating disorder prevention programs were cognitive with elevations on a single risk factor versus those that recruited par­
behavioral therapy, interpersonal therapy, behavioral therapy, family ticipants based on elevations in multiple risk factors. Theoretically this is
therapy, and a psychoeducational intervention. None of these preven­ because the latter trials recruited more heterogeneous samples which
tion programs resulted in significant reductions of eating disorder onset. may have attenuated the ability to detect effects relative to trials
Somewhat surprisingly given the evidence-base for cognitive behavioral involving samples that were more homogeneous in terms of risk for
treatments for a variety of psychiatric disorders, the one prevention eating disorders. This finding implies that it might be best if future
program based on cognitive behavioral therapy, Student Bodies, did not eating disorder prevention trials recruit participants with elevations on
significantly reduce future onset of eating disorders in three trials only a single risk factor. Moderator analyses also revealed that healthy
(Jacobi et al., 2012; Taylor et al., 2006; Taylor et al., 2016). Although lifestyle modification eating disorder prevention programs produced
these findings may suggest that cognitive behavioral principles are not significantly larger eating disorder prevention effects than the cognitive
effective in reducing risk factors for eating disorders, it is also possible behavioral eating disorder prevention programs. Further, the post hoc
that delivery of eating disorder prevention programs over the internet is analyses provided evidence that the dissonance-based prevention pro­
less effective, potentially because of the relatively low acceptability of gram produced larger eating disorder prevention effects when imple­
this prevention program based on completion rates. mented by peer educators versus clinicians, and data also suggested the
The one trial that tested an interpersonal psychotherapy prevention peer-implemented dissonance-based prevention programs produce
program (Tanofsky-Kraff et al., 2014) did not find a significant reduction larger effects than cognitive behavioral prevention programs.
in eating disorder onset compared to the health education comparison Moderator analyses did not provide support for the other factors that
condition. In that trial, the interpersonal therapy intervention was tested were hypothesized to correlate with larger eating disorder prevention
against a health education comparison condition, making this a more effects, which included length of follow-up, length of the prevention
conservative test than provided by many of the other trials. Nonetheless, program, internet versus in-person implementation, and risk of bias of
the fact that there was a 76% greater reduction in future eating disorder the trials. However, the small number of trials included in this meta-
onset in the interpersonal therapy condition suggests that it might be analytic review limited the ability to detect the effects of moderators
useful to evaluate interpersonal therapy prevention programs in a larger in the univariate models. Further, single versus multiple risk samples
trial that has greater sensitivity to detecting significant prevention and lifestyle modification prevention programs versus cognitive
effects. behavioral prevention programs did not produce a unique effect when
The family therapy-based prevention program, Parents Act Now, also entered alongside the other examined moderators, though the collin­
did not produce a reduction in eating disorder onset, though that trial earity among moderators likely contributed to the null unique effect of
appears to have only assessed anorexia nervosa (Jacobi et al., 2018). the two significant moderators.

14
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

4.2. Limitations rates. Improving acceptability of eating disorder prevention programs

has the potential to increase the prevention effects of these
It is important to consider the limitations of this review when interventions.
interpreting the findings and our conclusions. First, we only identified
15 trials that could be included in our meta-analysis, which limits the 5. Conclusions
generalizability of our findings, as most eating disorder prevention
programs that have been developed were not evaluated in trials The fact that lifestyle modification and dissonance-based prevention
designed to detect the prevention of new onset of eating disorders. It is programs significantly reduced future onset of eating disorders in mul­
thus possible that other prevention programs might show efficacy in tiple trials, producing a 54% to 77% reduction in future eating disorder
reducing future onset of eating disorders if they were evaluated in onset implies that broadly implementing these prevention programs
rigorously-designed and adequately-powered trials. Second, the rela­ could reduce the population prevalence of eating disorders. The NNT
tively small number of trials that tested for prevention effects also effect sizes were between 8 and 11 for the dissonance-based prevention
limited the sensitivity to detecting moderators that correlate with larger program, and between 10 and 20 for the lifestyle modification preven­
prevention effects. Third, some studies did not confirm that alternative tion program, suggesting that the former prevention program is more
interventions were perceived as equally credible as the eating disorder efficacious. The fact that the dissonance-based eating disorder preven­
prevention program evaluated in the trials (Ghaderi et al., 2020; tion program produced a large reduction in eating disorder onset when
Tanofsky-Kraff et al., 2014), raising the possibility that the superior ef­ implemented virtually is an encouraging new finding, as this imple­
fects for the eating disorder prevention program produced in the former mentation method should facilitate broad implementation, including
trial were driven by greater perceived credibility of that intervention the possibility of implementing this prevention program internationally.
compared to the comparison intervention. Fourth, none of the trials that It is exciting that the field has produced several eating disorder pre­
we located were powered to test whether the eating disorder prevention vention programs that have shown efficacy and effectiveness in reducing
effects differed for various ethnic groups, genders, or individuals with future onset of eating disorders, as this may permit a reduction in the
different sexual orientations or identities. An important priority for population prevalence of these pernicious psychiatric disorders.
future studies will be to test whether these prevention programs are
similarly effective for various sub-populations that experience eating Author notes
disorders, though this would likely require over-sampling of participants
from each ethnic minority group in order to have sufficient sensitivity. Preparation of this manuscript was supported by research grants
Finally, many of the trials did not report sufficient information for MH/DK061957, MH070699, MH086582, MH097720, and MH111782
coding of important moderators, which prevented an extensive evalua­ from the National Institutes of Health.
tion of moderators in the meta-analysis. The authors have no financial conflicts of interest to disclose.
Thanks go to the authors who were kind enough to provide data or
4.3. Future directions effect sizes, or to conduct additional analyses when requested.

It will be critical for more researchers to conduct trials that are Declaration of Competing Interest
designed to detect whether prevention programs reduce future eating
disorder onset (i.e., enroll large samples, conduct blinded diagnostic The authors have no conflicts of interest to disclose.
interviews over at least a 2-year follow-up). Our literature review
revealed that only a small portion of prevention programs that have References
been developed have been evaluated in trials designed to detect true
prophylactic effects. Results suggest that it might be optimal to work van Aert, R., & Jackson, D. (2019). A new justification of the Hartung-Knapp method for
with peers to deliver the prevention programs and to target individuals random-effects meta-analysis based on weighted least squares regression. Research
Synthesis Methods, 10, 515–527. https://doi.org/10.1002/jrsm.1356.
who are at risk for eating disorders because of a single risk factor. It will Allen, K., Byrne, S., Oddy, H., & Crosby, R. (2013). Eating disorders in adolescents:
also be important to conduct large enough trials that permit analyses of Prevalence, stability, and psychosocial correlates in a population-based sample of
whether the prevention programs are equally effective in reducing male and female adolescents. Journal of Abnormal Psychology, 122, 720–732.
Allen, K., Byrne, S., Oddy, W., Schmid, U., & Crosby, R. (2014). Risk factors for binge
future eating disorder onset for various sub-groups, such as different
eating and pruging eating disorders: Differences based on age of onset. International
ethnic groups, different genders, and people with different sexual Journal of Eating Disorders, 47, 802–812.
identifications and orientations. Dissonance-based eating disorder pre­ Arcelus, J., Mitchell, A., Wales, J., & Nielsen, S. (2011). Mortality rates in patients with
vention programs have been found to produce similar reductions in risk anorexia nervosa and other eating disorders: A meta-analysis of 36 studies. Archives
of General Psychiatry, 68, 724–731. https://doi.org/10.1001/
factors and eating disorder symptoms for White, Black, Asians, and archgenpsychiatry.2011.74.
Hispanic participants, as well as for heterosexual and non-heterosexual Bandura, A. (1986). Social foundations of thought and action (pp. 23–28). Englewood
adolescent girls and young women (Rodriguez, Marchand, Ng, & Stice, Cliffs, NJ: Prentice-Hall.
Begg, S., Vos, T., Baker, B., Stevenson, C., Stanley, L., & Lopez, A. (2007). The burden of
2008; Shaw, Rohde, Desjardins, & Stice, 2020; Stice, Marti, & Cheng, disease and injury in Australia 2003. Canberra: Australian Institute of Health and
2014). However, no trials have tested whether other eating disorder Welfaire.
prevention effects are similar across various ethnic groups. Such Chen, H., Cohen, P., & Chen, S. (2010). How big is a big odds ratio? Interpreting the
magnitudes of odds ratios in epidemiological studies. Communications in Statistics, 39,
research should allow identification of prevention programs that are not 860–864.
effective for all sub-populations and may inform the design of preven­ Claydon, E., & Zullig, K. (2019). Eating disorders and academic performance among
tion programs that are more effective for certain sub-populations. It will college students. Journal of American College Health. https://doi.org/10.1080/
07448481.2018.1549556.
also be important to conduct research on how to effectively implement Cochran, W. (1954). The combination of estimates from different experiments.
prevention programs that have been shown to produce reliable re­ Biometrics, 10, 101–129. https://doi.org/10.2307/3001666.
ductions in future eating disorder onset on a broad scale. The evidence Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale, NJ:
Erlbaum.
that prevention programs can produce a large reduction in future eating
Crow, S., & Smiley, N. (2010). Costs and cost-effectiveness in eating disorders. In
disorder onset when implemented virtually via videoconferencing is W. S. Agras (Ed.), The Oxford handbook of eating disorders (pp. 480–485). Oxford:
encouraging because this delivery modality could vastly expand the Oxford University Press.
reach and accessibility of prevention programs. In addition, it will be Dakanalis, A., Clerici, M., Bartoli, F., Caslini, M., Crocamo, C., Riva, G., & Carrà, G.
(2017). Risk and maintenance factors for young women’s DSM-5 eating disorders.
important to determine how to improve the acceptability of the pre­ Archives of Women’s Mental Health, 20(6), 721–731. https://doi.org/10.1007/
vention programs, as they often had low attendance and completion s00737-017-0761-6.

15
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

Fairburn, C. (1997). Eating disorders. In D. Clark, & C. Fairburn (Eds.), Science and Neumark-Sztainer, D., Wall, M., Guo, J., Story, M., Haines, J., & Eisenberg, M. (2006).
practice of cognitive behaviour therapy (pp. 209–241). Oxford: Oxford University Press. Obesity, disordered eating, and eating disorders in a longitudinal study of
Fairburn, C., Bailey-Straebler, S., Basden, S., Doll, H., Jones, R., Murphy, R., et al. (2015). adolescents: How do dieters fare 5 years later? Journal of the American Dietetic
A transdiagnostic comparison of enhanced cognitive behaviour therapy (CBT-E) and Association, 106, 559–568.
interpersonal psychotherapy in the treatment of eating disorders. Behaviour Research O’Dea, J. (1995). Everybody’s different: A self-esteem program for young adolescents. Sudney
and Therapy, 70, 64–71. University Press.
Favaro, A., Zanetti, T., Huon, G., & Santonastaso, P. (2005). Engaging teachers in an Patton, G., Johnson-Sabine, E., Wood, K., Mann, A., & Wakeling, A. (1990). Abnormal
eating disorder prevention intervention. International Journal of Eating Disorders, 38, eating attitudes in London school girls- a prospective epidemiological study:
73–77. Outcome at twelve month follow-up. Psychological Medicine, 20, 383–394.
Flegal, K., Carroll, M., Kit, B., & Ogden, C. (2012). Prevalence of obesity and trends in the Patton, G., Selzer, R., Coffey, C., Carlin, J., & Wolfe, R. (1999). Onset of adolescent eating
distribution of body mass index among US adults, 1999–2010. Journal of the disorders: Population based cohort study over 3 years. BMJ., 318, 765–768.
American Medical Association [Internet], 307, 491. Peters, J., Sutton, A., Jones, D., Abrams, K., & Rushton, L. (2006). Comparison of two
Ghaderi, A., & Scott, B. (2001). Prevalence, incidence and prospective risk factors for methods to detect publication bias in meta-analysis. JAMA, 295, 676–680. https://
eating disorders. Acta Psychiatrica Scandinavica, 104, 122–130. doi.org/10.1001/jama.295.6.676.
Ghaderi, A., Stice, E., Andersson, G., Enö Persson, J., & Allzén, E. (2020). A randomised Rodriguez, R., Marchand, E., Ng, J., & Stice, E. (2008). Effects of a cognitive-dissonance-
controlled trial of the effectiveness of virtually delivered body project (vBP) groups based eating disorder prevention program are similar for Asian American, Hispanic,
to prevent eating disorders. Journal of Consulting and Clinical Psychology, 88, and White participants. International Journal of Eating Disorders, 41, 618–625.
643–656. Rohde, P., Stice, E., & Marti, N. (2015). Development and predictive effects of eating
Graham, J. (2009). Missing data analysis: Making it work in the real world. Annual disorder risk factors during adolescence: Implications for prevention efforts.
Review of Psychology, 60, 549–576. International Journal of Eating Disorders, 48, 187–198.
Harrer, M., Adam, S., Mag, E., Baumeister, H., Cuijpers, P., Bruffaerts, R., et al. (2019). Santonastaso, P., Friederici, S., & Favaro, A. (1999). Full and partial syndromes in eating
Prevention of eating disorders at universities: A systematic review and meta- disorders: A 1-year prospective study of risk factors among female students.
analysis. International Journal of Eating Disorders, 53, 813–833. Psychopathology, 32, 50–56.
Harrer, M., Cuijpers, P., Furukawa, T., & Ebert, D. D. (2019). dmetar: Companion R Schafer, J., & Graham, J. (2002). Missing data: Our view of the state of the art.
package for the guide ’Doing Meta-Analysis in R’. Retrieved from http://dmetar. Psychological Methods, 7, 147–177.
protectlab.org. Schwarzer, G., Carpenter, J., & Rücker, G. (2015). Meta-analysis with R. New York:
Heatherton, T., & Polivy, J. (1992). Chronic dieting and eating disorders: A spiral model. Springer.
In J. Crowther, D. Tennenbaum, S. Hobfold, & M. Parris (Eds.), The etiology of bulimia Seidel, A., Presnell, K., & Rosenfield, D. (2009). Mediators in the dissonance eating
nervosa: The individual and familial context (pp. 133–155). Washington, DC: disorder prevention program. Behaviour Research and Therapy, 47, 645–653.
Hemisphere. Shadish, W., Cook, T., & Campbell, D. (2002). Experimental and quasi-experimental designs
Higgans, J., Altman, D., Gotzsche, P., Moher, D., Oxman, A., Savovic, J., et al. (2011). for generalized causal inference. Boston, MA: Houghton Mifflin.
The Cochrane Collaboration’s tool for assessing risk of bias in randomized trials. The Shaw, H., Rohde, P., Desjardins, C., & Stice, E. (2020). Sexual orientation correlates with
British Medical Journal, 343, d5928. baseline characteristics but show no moderating effects of dissonance-based eating
Hox, J. (2010). Multilevel analysis: Techniques and applications (2nd ed.). New York, NY: disorder prevention programs for women. Body Image, 32, 94–102.
Routledge. Simmonds, M. (2015). Quantifying the risk of error when interpreting funnel plots.
Hudson, J., Hiripi, E., Pope, H., & Kessler, R. (2007). The prevalence and correlates of Systematic Reviews, 4. https://doi.org/10.1186/s13643-015-0004-8.
eating disorders in the National Comorbidity Survey Replication. Biological Stice, E. (2016). Interactive and meditational etiologic models of eating disorder onset:
Psychiatry, 61, 348–358. Evidence from prospective studies. Annual Review of Clinical Psychology, 12,
IntHout, J., Ioannidis, J., & Borm, G. (2014). The Hartung-Knapp-Sidik-Jonkman method 359–381.
for random effects meta-analysis is straightforward and considerably outperforms Stice, E., Davis, K., Miller, N., & Marti, C. N. (2008). Fasting increases risk for onset of
the standard DerSimonian-Laird method. BMC Medical Research Methodology, 14. binge eating and bulimic pathology: A 5-year prospective study. Journal of Abnormal
https://doi.org/10.1186/1471-2288-14-25. Psychology, 117, 941–946.
Jacobi, C., Fittig, E., Bryson, S. W., Wilfley, D., Kraemer, H., & Taylor, C. (2011). Who is Stice, E., Desjardins, C., Rohde, P., & Shaw, H. (in press). Sequencing of symptom
really at risk? Identifying risk factors for subthreshold and full syndrome eating emergence in anorexia nervosa, bulimia nervosa, binge eating disorder, and purging
disorders in a high-risk sample. Psychological Medicine, 41, 1939–1949. disorder and relations of prodromal symptoms to future onset of these disorders.
Jacobi, C., Hutter, K., Volker, U., Mobius, K., Richter, R., Trockel, M., et al. (2018). Journal of Abnormal Psychology.
Efficacy of a parent-based, indicated prevention for anorexia nervosa: Randomized Stice, E., Gau, J., Rohde, P., & Shaw, H. (2017). Risk factors that predict future onset of
controlled trial. Journal of Medical Internet Research, 20, e296. each DSM-5 eating disorder: Predictive specificity in high-risk adolescent females.
Jacobi, C., Völker, U., Trockel, M. T., & Taylor, C. B. (2012). Effects of an Internet-based Journal of Abnormal Psychology, 126, 38–51.
intervention for subthreshold eating disorders: A randomized controlled trial. Stice, E., Marti, C., & Cheng, Z. (2014). Effectiveness of a dissonance-based eating
Behaviour Research and Therapy, 50(2), 93–99. https://doi.org/10.1016/j. disorder prevention program for ethnic groups in two randomized controlled trials.
brat.2011.09.013. Behaviour Research and Therapy, 55, 54–64.
Kant, R., Wong-Chung, A., Evans, E., Stanton, E., & Boothroyd, L. (2019). The impact of a Stice, E., Marti, N., & Durant, S. (2011). Risk factors for onset of eating disorders:
dissonance-based eating disorder intervention on implicit attitudes to thinness in Evidence of multiple risk pathways from an 8-year prospective study. Behaviour
women of diverse sexual orientations. Frontiers in Psychology, 10, 2611. Research and Therapy, 49, 622–627.
Killen, J., Taylor, C., Hayward, C., Haydel, K., Wilson, D., Hammer, L., et al. (1996). Stice, E., Marti, C., & Rohde, P. (2013). Prevalence, incidence, impairment, and course of
Weight concerns influence the development of eating disorders: A 4-year prospective the proposed DSM-5 eating disorder diagnoses in an 8-year prospective community
study. Journal of Consulting and Clinical Psychology, 64, 936–940. study of young women. Journal of Abnormal Psychology, 122, 445–457.
Knapp, G., & Hartung, J. (2003). Improved tests for a random effects meta-regression Stice, E., Marti, C., Spoor, S., Presnell, K., & Shaw, H. (2008). Dissonance and healthy
with a single covariate. Statistics in Medicine, 17, 2693–2710. weight eating disorder prevention programs: Long-term effects from a randomized
Kraemer, H., & Kupfer, D. (2006). Size of treatment effects and their importance to efficacy trial. Journal of Consulting and Clinical Psychology, 76, 329–340.
clinical research and practice. Biological Psychiatry, 59, 990–996. Stice, E., Rohde, P., Butryn, M., Shaw, H., & Marti, C. (2015). Effectiveness trial of a
Larimer, M., & Cronce, J. (2002). Identification, prevention and treatment: A review of selective dissonance-based eating disorder prevention program with female college
individual-focused strategies to reduce problematic alcohol consumption by college students: Effects at 2-and 3-year follow-up. Behaviour Research and Therapy, 71,
students. Journal of Studies on Alcohol, Supplement(s14), 148–163. 20–26. https://doi.org/10.1016/j.brat.2015.05.012.
Lipsey, M., & Wilson, D. (2001). Practical meta-analysis. Thousand Oaks, CA: Sage. Stice, E., Rohde, P., Durant, S., & Shaw, H. (2012). A preliminary trial of a prototype
Little, R., & Yau, L. (1996). Intent-to-treat analysis of longitudinal studies with drop-outs. Internet dissonance-based eating disorder prevention program for young women
Biometrics, 52, 1324–1333. with body image concerns. Journal of Consulting and Clinical Psychology, 80, 907–916.
Lock, J., & Le Grange, D. (2001). Treatment manual for anorexia nervosa: A family-based Stice, E., Rohde, P., Shaw, H., & Desjardins, C. (2020). Weight suppression increases odds
approach. New York: Guilford. for future onset of anorexia nervosa, bulimia nervosa, and purging disorder, but not
Lowe, M., Stice, E., Butryn, M., & Sarwer, D. (2012). Experimental test of a weight gain binge eating disorder. American Journal of Clinical Nutrition, 112, 941–947.
prevention program among female college freshmen (Unpublished study). Stice, E., Rohde, P., Shaw, H., & Gau, J. (2011). An effectiveness trial of a selected
Martinsen, M., Bahr, R., Borresen, R., Holme, I., Pensgaard, A., & Sundgot-Borden, J. dissonance-based eating disorder prevention program for female high school
(2014). Preventing eating disorders among young elite athletes: A randomized students: Long-term effects. Journal of Consulting and Clinical Psychology, 79,
controlled trial. Medicine and Science in Sports & Exercise, 46, 435–447. https://doi. 500–508.
org/10.1249/MSS.0b013e3182a702fc. Stice, E., Rohde, P., Shaw, H., & Gau, J. (2018). An experimental therapeutics test of
McIntosh, V., Jordan, J., Carter, J., Luty, S., McKenzie, J., Bulik, C., et al. (2016). whether adding dissonance-induction activities improves the effectiveness of a
Psychotherapy for transdiagnostic binge eating: A randomized controlled trial of selected obesity and eating disorder prevention program. International Journal of
cognitive-behavioural therapy, appetite-focused cognitive-behavioural therapy, and Obesity, 42, 462–468.
scheme therapy. Psychiatric Research, 240, 412–420. Stice, E., Rohde, P., Shaw, H., & Gau, J. (2020). Clinician-led, peer-led, and internet-
McKnight, I. (2003). Risk factors for the onset of eating disorders in adolescent girls: delivered dissonance-based eating disorder prevention programs: Effectiveness of
Results of the McKnight longitudinal risk factor study. American Journal of these delivery modalities through 4-yr follow-up. Journal of Consulting and Clinical
Psychiatry, 160, 248–254. Psychology, 88, 481–494.
Mrazek, P., & Haggerty, R. (1994). Reducing risks for mental disorders: Frontiers for Stice, E., Rohde, P., Shaw, H., & Marti, C. (2013). Efficacy trial of a selective prevention
preventive intervention research. Washington, D.C.: National Academy Press. program targeting both eating disorders and obesity among female college students:

16
E. Stice et al. Clinical Psychology Review 87 (2021) 102046

1- and 2-year follow-up effects. Journal of Consulting and Clinical Psychology, 81, Viechtbauer, W. (2007). Confidence intervals for the amount of heterogeneity in meta-
183–189. analysis. Statistics in Medicine, 26, 37–52. https://doi.org/10.1002/sim.2514.
Stice, E., Shaw, H., & Marti, C. (2007). A meta-analytic review of eating disorder Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package.
prevention programs: Encouraging findings. Annual Review of Clinical Psychology, 3, Journal of Statistical Software, 36, 1–48. Retrieved from http://www.jstatsoft.org
233–257. /v36/i03/.
Stice, E., Yokum, S., & Waters, A. (2015). Dissonance-based eating disorder prevention Wade, T., Davidson, S., & O’Dea, J. (2003). A preliminary controlled evaluation of a
program reduces reward region response to thin models: How actions shape school-based media literacy program and self-esteem program for reducing eating
valuation. PLoS One, 10(12), Article e0144530. disorder risk factors. International Journal of Eating Disorders, 33, 371–383.
Swanson, S., Crow, S., le Grange, D., Swendsen, J., & Merikangas, K. (2011). Prevalence Watson, H., Joyce, T., French, E., Willan, V., Kane, R., Tanner-Smith, E., … Egan, S.
and correlates of eating disorders in adolescents: Results from the National (2016). Prevention of eating disorders: A systematic review of randomized,
Comorbidity Survey Replication Adolescent Supplement. Archives of General controlled trials. International Journal of Eating Disorders. https://doi.org/10.1002/
Psychiatry, 68, 714–723. eat.22577. Advance on-line publication.
Tanofsky-Kraff, M., Shomaker, L., Wilfley, D., Young, J., Sbrocco, T., Stephens, M., et al. Wonderlich, S., Peterson, C., Crosby, R., Smith, T., Klein, M., Mitchell, J., et al. (2014).
(2014). Targeted prevention of excess weight gain and eating disorders in high-risk A randomized controlled comparison of integrative cognitive-affective therapy
adolescent girls: A randomized controlled trial. American Journal of Clinical Nutrition, (ICAT) and enhanced cognitive-behavioral therapy (CBT-E) for bulimia nervosa.
100, 1010–1018. Psychological Medicine, 44, 543–553.
Taylor, C., Bryson, S., Luce, K., Cunning, D., Celio, A., Abascal, L., … Wilfley, D. (2006). Yager, Z., & O’Dea, J. (2008). Prevention programs for body image and eating disorders
Prevention of eating disorders in at-risk college-age women. Archives of General on university campuses: A review of large controlled interventions. Health Promotion
Psychiatry, 63, 881–888. International, 23, 173–189.
Taylor, C., Trockel, M., Cunning, D., Bailey, J., Aspen, V., Jacobi, C., … Wilfley, D.
(2016). Reducing eating disorder onset in a very high risk sample with significant
Eric Stice served as an assistant professor and associate professor at the University of
comorbid depression: A randomized controlled trial. Journal of Consulting & Clinical
Texas at Austin and as a Senior Research Scientist at Oregon Research Institute before
Psychology, 84(5), 402–414.
joining the faculty at Stanford University. His research focuses on identifying risk factors
Tobin, L. (2020). Dissonance-based eating disorder preventive intervention and attentional
that predict onset of eating disorders, obesity, substance abuse, and depression to advance
biases in body-dissatisfied university women: A cluster randomized controlled trial
knowledge regarding etiologic processes, including the use of functional neural imaging.
(unpublished doctoral thesis). Calgary, Alberta, Canada: University of Calgary.
He also designs, evaluates, and disseminates prevention and treatment interventions for
Van den Noortgate, W., López-López, J., Marin-Martinez, F., & Sánchez-Meca, J. (2013).
eating disorders, obesity, and depression. For instance, he developed a dissonance-based
Three-level meta-analysis of dependent effect sizes. Behavior Research Methods, 45,
eating disorder prevention program that has been implemented with over 6 million
576–594. https://doi.org/10.3758/s13428-012-0261-6.
young girls in 139 countries. He has published 306 articles in peer-reviewed journals.

17