Risk Factors that Predict Future Onset of Each DSM-5 Eating Disorder: Predictive Specificity in High-Risk Adolescent Females

Abstract

Objective

Because no single report has examined risk factors that predict future onset each type of eating disorder and core symptom dimensions that crosscut disorders, we addressed these aims to advance knowledge regarding risk factor specificity.

Method

Data from 3 prevention trials that targeted young women with body dissatisfaction (N=1272; M age 18.5, SD 4.2) and collected annual diagnostic interview data over 3-year follow-up were combined to identify predictors of subthreshold/threshold anorexia nervosa (AN), bulimia nervosa (BN), binge eating disorder (BED), and purging disorder (PD).

Results

Negative affect and functional impairment predicted onset of all eating disorders. Thin-ideal internalization, body dissatisfaction, dieting, overeating, and mental health care predicted onset of subthreshold/threshold BN, BED, and PD; positive thinness expectations, denial of cost of pursuing the thin ideal, and fasting predicted onset of 2 of these 3 disorders. Similar risk factors predicted core eating disorder symptom onset. Low BMI and dieting specifically predicted onset of subthreshold/threshold AN or low BMI. Only a subset of factors showed unique predictive effects in multivariate models, likely due to moderate correlations between the risk factors (M r = .14).

Conclusions

Results provide support for the theory that pursuit of the thin ideal and the resulting body dissatisfaction, dieting, and unhealthy weight control behaviors increase risk for binge/purge spectrum eating disorders, but suggest that youth who are inherently lean, rather than purposely pursuing the thin ideal, are at risk for AN. Impaired interpersonal functioning and negative affect are transdiagnostic risk factors, suggesting these factors should be targeted in prevention programs.

General Scientific Summary

Findings suggest that pursuit of the thin beauty ideal and the resulting body dissatisfaction and weight control behaviors increase risk for developing eating disorders involving binge eating and unhealthy weight control behaviors, but that inherently lean young women, rather than those purposely pursuing the thin ideal, are at risk for eating disorders involving a dangerously low body weight. Findings also suggest that negative affect and impaired interpersonal functioning are transdiagnostic factors that increase risk for the full spectrum of eating disorders.

Eating disorders affect 13% of females (Allen, Byrne, Oddy, & Crosby, 2013; Stice, Marti, & Rohde, 2013a) and are marked by chronicity, relapse, distress, functional impairment, and risk for future obesity, depression, suicide, substance abuse, and mortality (Arcelus, Mitchell, Wales, & Nielsen, 2011; Swanson, Crow, Le Grange, Swendsen, & Merikangas, 2011). However, few studies have evaluated risk factors that predict future onset of each DSM-5 (American Psychiatric Association, 2013) eating disorder using validated diagnostic interviews, which is vital because this design provides definitive evidence that the risk factors predate onset of the disorder (Kraemer, Stice, Kazdin, & Kupfer, 2001). Although prospective studies have predicted future increases in eating disorder symptom composite measures (e.g., Stice, 2001; Vohs Bardone, Joiner, Abramson, & Heatherton, 2001), they provide less definitive evidence of temporal precedence and can be challenging to interpret because symptoms often decrease over time and it is not clear that the models are predicting emergence of clinically significant eating pathology. Studies have also predicted future onset of individual eating disorder symptoms (e.g., Pearson, Combs, Zapolski, & Smith, 2012; Goldschmidt, Wall, Loth, Le Grange, & Neumark-Sztainer, 2012), which provides clear evidence of temporal precedence, but the clinical significance of individual eating disorder symptoms has not been well established. Another limitation of this literature is that many studies have not used diagnostic interviews, which is concerning because many researchers view diagnostic interviews as the gold standard. An improved understanding of risk factors that predict onset of each DSM-5 eating disorder will inform etiologic theory for each disorder, inform content for prevention programs tailored for the various eating disorders, which would be indicated if they have distinct risk factors, determine which high-risk subgroups to target with selective prevention programs for each disorder, and potentially identify transdiagnostic risk factors.

Only one study examined factors that predicted future onset of interview-assessed subthreshold/threshold anorexia nervosa (sAN/AN) among participants confirmed to be disorder-free at baseline. Low body mass index (BMI) and low dieting during early adolescence predicted sAN/AN over 5-year follow-up in a community sample of females who were in early adolescence at baseline, but early puberty, pressure for thinness, thin-ideal internalization, body dissatisfaction, negative affect, or social support deficits did not (Stice & Bohon, 2013). Findings from risk factor studies are summarized in Table 1.

Table 1.

Summary of Predictors for Specific Subthreshold/Threshold Eating Disorders from Previous Trials in which Participants were Confirmed to be Disorder-free at Baseline.

Potential Risk Factor	AN			BN			BED			PD
	+	0	-	+	0	-	+	0	-	+	0	-
Pressure for thinness		A		CEF			F				F
Thin-ideal internalization		AG		EG	F		G	F		FG
Thinness expectancies		G		G			G			G
Denial of costs of pursuing thin ideal		G		G				G		G
Body dissatisfaction		AG		CFG	E		G	F		FG
BMI			AG	E	CDG			G			G
Weight concerns				C
Dieting		G	A	BCDEG	F		G	F		FG
Weight control behaviors		G			G			G			G
Overeating		G		G			G			G
Daily exercise					D
Excessive exercise		G			G			G		G
Fasting		G		EG				G		G
Peer dieting					D
Negative affect		AG		EG	F		G	F		G	F
Negative affectivity				C
Depressive symptoms					F						F
Social support deficits		A		E
Early puberty		A		E
Ineffectiveness				C
Alcohol use				C
Low interoceptive awareness				C
Perfectionism					C
Maturity fears					C
Interpersonal distrust					C
Psychiatric problems				D
Mental health care		G		G			G			G
Functional impairment	G			G			G
Parental separation					D

Note. AN = anorexia nervosa; BN = bulimia nervosa, BED = binge eating disorder; PD = purging disorder. The + column contains studies finding a significant positive relation for the risk factor; the 0 column contains studies finding a non-significant relation; and the – column contains studies finding a significant negative relation for the risk factor. The letters refer to the study that reported each association: A = Stice & Bohon (2013); B = Patton, Johnson-Sabine, Wood, Mann, & Wakeling (1990); C = Killen et al. (1996); D = Patton, Selzer, Coffey, Carlin, & Wolfe (1999); E = Stice, Davis, Miller, & Marti (2008a); F = Stice, Marti, & Durant (2011a); G = Present Study.

Five studies examined factors that predicted future onset of interview-assessed subthreshold/threshold bulimia nervosa (sBN/BN), all of which involved community samples of adolescent girls. Dieting predicted onset of sBN/BN onset over 1-year follow-up (Patton, Johnson-Sabine, Wood, Mann, & Wakeling, 1990). Weight concerns, drive for thinness, body dissatisfaction, ineffectiveness, negative affectivity, dieting, and alcohol use, and low interoceptive awareness predicted onset of sBN/BN over 4-year follow-up, but perfectionism, maturity fears, interpersonal distrust, and BMI did not (Killen et al., 1996). Dieting and psychiatric problems predicted BN onset over 3-year follow-up, but peer dieting, BMI, daily exercise, and parental separation did not (Patton, Selzer, Coffey, Carlin, & Wolfe, 1999). Elevated BMI, social pressure for thinness, thin-ideal internalization, dieting, fasting, negative affect, social support deficits, and early puberty predicted onset of sBN/BN over 5-year follow-up, but body dissatisfaction did not (Stice, Davis, Miller, & Marti, 2008a; Stice & Bohon, 2013). Social pressure to be thin and body dissatisfaction predicted onset of sBN/BN over a longer 7-year follow-up in the same sample, but thin-ideal internalization, dieting, negative affect, or depressive symptoms did not (Stice, Marti, & Durant, 2011a).

Only one study each examined factors that predicted future onset of subthreshold/threshold binge eating disorder (sBED/BED) and purging disorder (PD), both of which used the same community sample of females who were in early adolescence at baseline. Social pressure for thinness predicted onset of sBED/BED over 7-year follow-up, but thin-ideal internalization, body dissatisfaction, dieting, or negative affect did not (Stice et al., 2011a). Thin-ideal internalization, body dissatisfaction, and dieting predicted PD onset over 7-year follow-up, but social pressure to be thin, negative affect, and depressive symptoms did not (Stice et al., 2011a).

Thus, no single report has investigated risk factors that predict future onset of each major DSM-5 eating disorder, which limits knowledge about risk factor specificity. Indeed, only four prospective studies have contributed to this literature, underscoring the need to expand this evidence-base. Equally striking, only one prospective study investigated risk factors that predict onset of AN, BED, and PD. The low incidence of certain eating disorders, particularly AN, has contributed to this lacuna. For instance, one of the largest prospective studies did not identify a single participant who showed AN onset (McKnight Investigators, 2003). One solution is to use a high-risk design wherein participants at elevated risk for mental health disorders are followed. Prospective high-risk designs have been used to elucidate risk factors for a variety of disorders, such as depression (LeMoult, Ordaz, Kircanski, Singh, & Gotlib, 2015), anxiety disorders (Schmidt, Lerew, & Jackson, 1999), substance abuse (Chassin, Pitts, DeLucia, & Todd, 1999), and psychosis (Cannon et al., 2008). One previous study used this approach for eating disorders (Jacobi et al., 2011), though not all diagnoses were confirmed by interview and the small sample size precluded the ability to examine risk factors for each type of eating disorder. Accordingly, the present study combined data from three large randomized eating disorder prevention trials that used parallel recruitment and assessment procedures, which increased sensitivity to detecting risk factors that predict future onset of eating pathology. The goal was to test whether a set of factors assessed at baseline predicted future onset of eating disorders and the core eating disorder symptom dimensions that crosscut different diagnostic categories over 3-year follow-up.

We investigated a range of factors that have been theorized to increase risk for eating disorders in several univariate and multivariate etiologic models (e.g., Hohlstein, Smith, & Atlas, 1998; Lavender et al., 2016; McCarthy 1990; Stice, 2001). For instance, the dual pathway model posits that elevated cultural pressure for thinness and internalization of the thin ideal promotes body dissatisfaction, which in turn contributes to unhealthy weight control behaviors (e.g., meal skipping) and negative affect, which both increase risk for emergence of binge eating and compensatory weight control behaviors (Stice, 2001). We hypothesized that baseline elevations in thin-ideal internalization, positive expectances regarding thinness, denial of costs of pursuing the thin ideal, and body dissatisfaction would increase risk for onset of disorders marked by use of weight loss behaviors, including AN, BN, and PD. We further hypothesized that baseline elevations in weight control behaviors, dieting, and negative affect would increase risk for onset of disorders characterized by binge eating, including AN, BN, and BED. We hypothesized that functional impairment and mental health care, which may represent proxy measures of general psychopathology, may be transdiagnostic risk factors that increase risk for any eating disorder. Based on past findings, we also hypothesized that a low BMI at baseline would increase risk for onset of AN, but a high BMI at baseline would increase risk for BN and BED, as it might be a proxy measure for a tendency toward dysregulated eating.

We also investigated the predictive effects of prodromal eating disorder symptoms, based on evidence that fasting predicted BN (Stice et al., 2008a). Specifically, we tested the hypotheses that overeating, fasting, and excessive exercise would predict onset of syndromal eating disorders. Such analyses are useful because it is possible that the emergence of one of these prodromal symptoms reliably predicts emergence of eating disorders, serving as a “gateway” to the full syndrome. However, it is important to consider the partial criterion confounding when interpreting any predictive effects of prodromal symptoms.

We estimated models that predicted onset of subthrehold or threshold eating disorders, as well as onset of threshold eating disorders, because individuals with threshold versus subthreshold levels of BN and BED showed similar functional impairment, emotional distress, and treatment care (Stice, Marti, Shaw, & Jaconis, 2009). This is also important because many individuals who seek treatment meet criteria for subthreshold rather than threshold eating disorders (Eddy, Celio, Hoste, Herzog, & Le Grange, 2008). A related lacuna is that only four studies reported descriptive data about progression from subthreshold to threshold levels of an eating disorder, which should also advance knowledge of etiologic processes; an average of 4% of young women with subthreshold AN progressed to threshold AN during follow-up, 25% of young women with subthreshold BN progressed to threshold BN during follow-up, and 21% of young woman with subthreshold BED progressed to threshold BED over follow-up (Lewinsohn et al., 2000; Patton et al., 1990; Stice et al., 2009, 2013). However, because those studies had small numbers of participants with subthreshold eating disorders, we also reported diagnostic progression for each eating disorder in the present study.

Finally, scholars have called for research on behavioral and cognitive symptom dimensions that crosscut psychiatric diagnoses (Sanislow et al., 2010), consistent with the Research Domain Criteria (RDoC) perspective. Studies using a wide variety of analytic approaches conducted over the last two decades suggest that eating pathology occurs along a continuum (Luo, Donnellan, Burt, & Klump, in press; Olatunji et al., 2012; Stice, Killen, Hayward, & Taylor, 1998; Tylka & Subich, 2003). Accordingly, we also investigated baseline factors that predicted onset of core symptom dimensions that crosscut eating disorders, including binge eating, compensatory behaviors, overvaluation of weight and shape, fear of weight gain, and low body weight.

Methods

Participants and Procedures

We combined data from one efficacy trial (Trial 1; Stice, Marti, Spoor, Presnell, & Shaw, 2008b) and two effectiveness trials (Trial 2; Stice, Rohde, Shaw, & Gau, 2011b; Trial 3; Stice, Rohde, Butryn, Shaw, & Marti, 2015) resulting in a sample of 1,272 participants.

Design of Randomized Prevention Trials

Mailings and fliers recruited female students for trials evaluating body acceptance interventions at high schools (Trial 1 and 2) and colleges (Trial 1 and 3). There was only one inclusion criterion: that participants answer affirmatively when asked if they had body image concerns during a phone screen. Regarding exclusion criteria, we sought to exclude participants who met criteria for DSM-IV (American Psychiatric Association, 2000) AN or BN, though analyses of the baseline diagnostic interview data indicated that we were not entirely successful (see below). Informed consent was obtained from participants (and parents if minors). Trial 1 participants were randomized to the Body Project eating disorder prevention program, Healthy Weight eating disorder prevention program, an expressive writing intervention, or assessment-only control condition. Trial 2 and 3 participants were randomized to the Body Project or educational brochure control condition. Participants completed surveys and interviews at baseline and at 1-, 6-, 12-, 24-, and 36-month follow-up. Additional details can be found in Stice et al. (2008b, 2011b, 2015).

Measures

Eating pathology

The semi-structured Eating Disorder Diagnostic Interview (EDDI; Stice et al., 2013a) assessed eating disorder symptoms over the past 3 months at baseline and since previous interview at follow-ups on a month-by-month basis. DSM-5 criteria for eating disorders, as operationalized in Stice et al. (2013a), were used (see Table 2 for definitions). However, we operationalized subthreshold AN following Stice et al. (2009) because it represents a subthreshold version that more closely parallels AN (overweight and obese participants can be diagnosed with atypical AN per DSM-5) and because we did not collect data on weight loss history, making it impossible to diagnose atypical AN. We also predicted onset of recurrent (2 or more episodes) binge eating, recurrent compensatory behaviors, and diagnostic threshold levels of overvaluation of weight/shape and fear of weight gain/obesity. With regard to prodromal symptoms, Overeating included counts of eating rapidly, eating until uncomfortably full, eating large amounts when not hungry, eating alone due to embarrassment, feeling depressed or guilty after overeating, and feeling upset that you could not control overeating, Fasting reflected the number of times per week in past three months two meals in a row were skipped to “as a means of controlling your shape or weight” and Excessive exercise reflected the number of times per week in the past 3 months of vigorous exercise for more than an hour or moderate exercise for more than 2 hours was used “to compensate for overeating”. EDDI eating disorder diagnoses have shown 1-week test-retest reliability (κ = .79) and inter-rater agreement (κ = .75) for DSM-5 eating disorders, sensitivity to intervention effects, and participants with versus without EDDI-diagnosed eating disorders show greater functional impairment, emotional distress, and treatment care (Stice et al., 2008b, 2009, 2013a, 2013b).

Table 2.

Diagnostic Criteria for DSM-5 Eating Disorders

Anorexia nervosa	Body mass index (BMI) less than 85% of the median expected for age and gender Definite fear of weight gain more than 75% of the days for at least 3 months Weight and shape were 1 of the main aspects of self-evaluation
Bulimia nervosa	At least four uncontrollable binge-eating episodes per month for at least 3 months At least 4 compensatory behavior episodes per month for at least 3 months Weight and shape were definitely 1 of the main aspects of self-evaluation
Binge eating disorder	At least four uncontrollable binge-eating episodes/days per month for at least 3 months
	Less than 1 compensatory behavior on average per month during this period Marked distress about binge eating Binge eating characterized by 3 or more of the following: rapid eating; eating until uncomfortably full; eating large amounts when not physically hungry; eating alone because of embarrassment; feeling disgusted, depressed, or guilty after overeating
Subthreshold anorexia nervosa	BMI between 90% and 85% of that expected for age and gender Definite fear of weight gain more than 25% of the days for at least 3 months Weight and shape were 1 of the main aspects of self-evaluation
Subthreshold bulimia nervosa	At least 2 uncontrollable binge-eating episodes per month for at least 3 months or at least 6 episodes over a shorter period At least 2 compensatory behavior episodes (i.e., self-induced vomiting, laxatives or diuretic use, fasting, and excessive exercise to compensate for overeating) per month for at least 3 months or at least 6 episodes over a shorter period Weight and shape were definitely 1 of the main aspects of self-evaluation
Subthreshold binge eating disorder	At least 2 uncontrollable binge-eating episodes/days per month for at least 3 months or at least 6 episodes over a shorter period Less than 1 compensatory behavior on average per month during this period Marked distress about binge eating Binge eating characterized by 3 or more of the following: rapid eating; eating until uncomfortably full; eating large amounts when not physically hungry; eating alone because of embarrassment; feeling disgusted, depressed, or guilty after overeating
Purging disorder	At least 4 episodes of self-induced vomiting or diuretic/laxative use for weight control purposes per month for at least 3 months Less than 1 uncontrollable binge-eating episode on average per month during this period Weight and shape were 1 of the main aspects of self-evaluation
Subthreshold purging disorder	At least 2 episodes of self-induced vomiting or diuretic/laxative use for weight control purposes per month for at least 3 months or at least 6 episodes over a shorter period Less than 1 uncontrollable binge-eating episode on average per month during this period Weight and shape were 1 of the main aspects of self-evaluation

Note. Anorexia nervosa took diagnostic precedence over bulimia nervosa and binge eating disorder.

Thin ideal internalization

The Ideal-Body Stereotype Scale–Revised assessed thin-ideal internalization (Stice et al., 2008b). It has shown internal consistency (α = .91), 2-week test–retest reliability (r = .80), predictive validity for bulimic symptom onset, and sensitivity to intervention effects (Stice et al., 2008b).

Positive thinness expectancies

The Thinness Expectancy scale (Hohlstein et al., 1998) assessed expected social and psychological benefits of thinness. It has shown internal consistency (α = .86) and predictive validity for body dissatisfaction onset (Stice & Whitenton, 2002).

Denial of costs of pursuing thin ideal

We created 5 items assessing denial of costs associated with pursuing the thin ideal (sample item: The risks associated with severe dieting are overrated). This scale has shown internal consistency (α = .82), 1-week test-retest reliability (r = .87), and predictive validity for eating disorder onset (Stice, Rohde, Gau, & Shaw, 2012).

Body dissatisfaction

The Body Dissatisfaction Scale (Stice et al., 2001) assessed dissatisfaction with 9 body parts. It has shown internal consistency (α = .94), 3-week test–retest reliability (r = .90), predictive validity for eating disorder onset, and sensitivity to intervention effects (Stice et al., 2008b, 2011a).

Weight control behaviors

We used the Weight Control scale to assess reported dietary and exercise behaviors used to control weight (sample item: I have eaten more fruits and vegetables). This scale showed internally consistent (α = .82), 4-week test-retest (r = .63), and sensitivity to intervention effects (Stice, Presnell, Gau, & Shaw, 2007).

Dieting

The Dutch Restrained Eating Scale (van Strien, Frijters, van Staveren, Defares, & Deurenberg, 1986) assesses the frequency of various dieting behaviors. It has shown internal consistency (α = .95), 2-week test–retest reliability (r = .82), convergent validity with self-reported (but not objectively measured) caloric intake, predictive validity for bulimic symptoms, and sensitivity to intervention effects (Stice et al., 2008b; van Strien et al., 1986).

Negative affect

Different measures of negative affect were used in each trial and were standardized to permit analyses of the combined data set. In Trial 1 negative affect was assessed with the sadness, guilt, and fear/anxiety subscales from the Positive Affect and Negative Affect Scale-Revised (PANAS-X; Watson & Clark, 1992), which has shown internal consistency (α = .95), 3-week test-retest reliability (r = .78), predictive validity for bulimic symptom onset, and sensitivity to detecting intervention effects (Stice et al., 2008b). In Trial 2 negative affect was assessed with the Center for Epidemiologic Studies-Depression Scale (CESD; Radloff, 1977), which has shown internal consistency (α = .74 – .91), temporal reliability (2- to 8-week test-retest r = .51 – .59), and convergent validity with clinician ratings of depressive symptoms (M r = .88; Andrews, Lewinsohn, Hops, & Roberts, 1993; Roberts, Lewinsohn, & Seeley, 1991). In Trial 3 negative affect was assessed with the 21-item Beck Depression Inventory (Beck et al., 1988), which has shown internal consistency (α = .73-.95), 1-week test-retest reliability (r = .93), convergent validity with clinician ratings of depressive symptoms, and sensitivity to intervention effects (M r = .75; Beck et al., 1988; Stice et al., 2015).

Functional impairment

Impairment in the family, peer group, romantic, and school domains was measured with 17 items from the Social Adjustment Scale-Self Report for Youth (Weismann & Bothwell, 1976). The original scale has shown convergent validity with clinician and collateral ratings (M r = .72) and sensitivity to detecting treatment effects (Weismann & Bothwell, 1976). The 17-item version has shown internal consistency (α = .77), 1-week test–retest reliability (r = .83) and sensitivity to intervention effects (Stice et al., 2008b).

Mental health care

Use of mental health services was operationalized as the frequency of visits to mental health care providers. Receipt of care in the last 6-months was coded “1” and no health care coded “0”. This item showed 1-year test-retest reliability (r = .89) and differentiate individuals with versus without eating disorders (Stice et al., 2008b; Stice et al., 2013a).

Body mass

The BMI (kg/m²; Pietrobelli et al., 1998) was examined as a baseline predictor of eating disorder onset and used to diagnose AN. Height was measured to the nearest mm using portable stadiometers. Weight was assessed to the nearest 0.1 kg using digital scales with participants wearing light indoor clothing without shoes or coats. Age- and sex-adjusted BMI centiles were used to determine whether participants were underweight for AN. Height and weight were measured twice at each assessment, to reduce errors, and averaged. BMI has shown convergent validity (r = .80 – .90) with direct measures of body fat (Pietrobelli et al., 1998).

Statistical Methods

Cox proportional hazard models estimated with SAS PHREG (SAS Institute Inc., 2011) tested which baseline risk factors predict onset of (a) subthreshold/threshold eating disorders, (b) threshold eating disorders, and (c) eating disorder symptoms. Participants who met criteria for the outcomes at baseline were excluded from the Cox models. Each risk factor was first tested in separate models and then significant predictors were entered simultaneously to assess unique effects. All risk factors were converted to z-scores so hazard ratios could be interpreted as the difference in risk at each time point associated with a one standard deviation change in the risk factor. To isolate risk factor effects from intervention effects, dummy coded vectors of intervention condition were included as covariates. The Breslow (1974) method was used to handle tied event times in the hazard models. A transform of the Martingale residuals evaluated the proportional hazards assumption. If this assumption was not met models were re-estimated with covariate interactions with time included as predictors. Missing onset data were accounted for with right-censoring in the hazard models and 20 imputed data sets were used to account for missing risk factor data. Sequential regression multiple imputation (SRMI; van Burren, 2007) was used to impute data sets using IVEware software V0.2 (Raghunathan, Solenberger, & Van Hoewyk, 2002). Model parameters and standard errors were combined following Rubin (1987) as implemented in SAS PROC MIANALYZE (SAS Institute Inc., 2011).

Results

Preliminary Analyses

Table 3 provides demographic data for the present sample and compares it to a representative sample of age-matched young women recruited from middle schools on demographic factors and study variables that were assessed in both studies. The mean body dissatisfaction score from our high-risk sample was 0.6 standard deviation (SD) higher than the mean body dissatisfaction score from a representative sample of adolescent girls (Rohde et al., 2015), mean thin-ideal internalization score was 0.6 SD higher, mean dieting score was 0.7 SD higher, mean overeating score was 1.2 SDs higher, mean fasting score was 0.1 SD higher, mean excessive exercise score was 0.1 SD higher, and mean functional impairment was 0.2 SD lower. Data suggest that the sample used in the present study was at moderately higher risk, suggesting that results may generalize to the broader population of adolescent girls and young women. A related question is whether there was a restriction in range for the risk factors, which could reduce sensitivity. Table 3 indicates that although some variables had a smaller SD in the present high-risk sample compared to the representative community-recruited sample, this was not the case for all variables examined. Moreover, the fact that such a high percentage of risk factors exhibited significant predictive effects (see below) implies we had sufficient sensitivity.

Table 3.

Demographic Data from a Representative Sample of Age-Matched Community-Recruited Young Women and the Present High-Risk Sample.

	Community Sample		High Risk Sample
Demographic Characteristics
Age [Mean, (SD)]	18.5	(1.4)	18.5	(4.2)
Race and Ethnicity [n, %]
American Indian or Alaskan Native	4	0.8	28	2.2
Asian	8	1.6	130	10.3
Black or African American	36	7.3	58	4.6
Native Hawaiian or other Pacific Islander	0	0.0	9	0.7
Hispanic	90	18.3	141	11.2
Caucasian	336	68.2	848	67.3
Other or mixed heritage	19	3.9	46	3.7
Parental education [n, %]
High school graduate or less	67	14.1	137	11.0
Some college	71	15.0	223	17.8
College graduate	212	44.8	465	37.2
Advanced degree	123	26.0	425	34.0
Baseline Risk Factors [Mean, (SD)]
Thin ideal internalization	3.30	(0.71)	3.69	(0.53)
Thinness expectancy	2.64	(0.99)	3.23	(0.78)
Denial of thin-ideal costs	NA	NA	2.26	(0.70)
Body dissatisfaction	2.79	(0.96)	3.40	(0.79)
Weight control behaviors	NA	NA	2.89	(0.70)
Dieting	2.07	(0.88)	2.66	(0.90)
Overeating	0.14	(0.74)	1.01	(2.00)
Fasting	0.62	(4.41)	1.08	(3.66)
Excessive exercise	0.70	(3.22)	1.11	(3.35)
Functional impairment	2.46	0.54	2.34	(0.53)
Mental health care	NA	NA	0.23	(0.42)
Body Mass Index	22.10	(4.58)	24.1	(5.11)

NA = not available.

Attrition for diagnostic data, which are the only data examined from each follow-up assessment, was 2% at posttest, 7% a 6-months, 6% at 1-year, 7% at 2-year, and 14% at 3-year follow-up. All but one case completed the full baseline assessment and 99% completed at least one follow-up (M = 5.3 of 6 assessments, SD = 0.9). Incidence rates for subthreshold and threshold levels of each disorder and core symptom dimensions over follow-up are shown in Table 4. Incidence rates were 9% for any threshold level disorder, 15% for any subthreshold disorder, and 52% for any core symptom dimension.

Table 4.

Rates of DSM-5 Eating Disorders and Symptoms

	Study Prevalence		Cumulative 3-year Incidence
	n	%	n	%
Full-syndrome disorders
Anorexia nervosa	9	0.7	7	0.6
Bulimia nervosa	77	6.0	47	3.8
Binge eating disorder	69	5.5	51	4.1
Purging disorder	53	4.2	41	3.3
Subthreshold eating disorders
Anorexia nervosa	25	2.0	21	1.7
Bulimia nervosa	115	9.1	87	7.0
Binge eating	71	5.6	60	4.8
Purging disorder	115	9.1	65	5.2
Eating disorder symptoms
Recurrent binge eating	367	29.0	177	16.5
Recurrent compensatory behaviors	722	32.9	266	32.9
Low body weight	65	5.1	33	2.7
Overvaluation of weight/shape	888	70.2	305	44.7
Fear of gaining weight	381	30.1	165	15.7

Note. Eating disorder classifications are not mutually exclusive across disorders. Study prevalence reflects the number of participants who met criteria at baseline and those who showed onset during the follow-up period. Cumulative incidence reflects participants who showed onset during the follow-up period, excluding those who met criteria at baseline.

Rates of diagnostic progression, which is defined as a subthreshold level disorder that escalates to threshold level at a later date, was 5% for AN, 40% for BN, 46% for BED, and 33% for PD. Participants who only experienced a subthreshold eating disorder after experiencing a full threshold eating disorder are not included in these analyses.

We tested whether participants with subthrehold versus threshold levels of each eating disorder differed significantly in terms of negative affect and functional impairment, as few studies have tested whether these groups show similar impairment. Negative affect and functional impairment scores reported closest to onset of each disorder were compared. Negative affect scores between subthreshold and threshold cases did not significantly differ (all p-values > .214) and were associated with small effect sizes differences for BN (d = .07), BED (d = 0.24), and PD (d = .01). Negative affect scores for subthreshold AN (Mean = 0.10, SD = 1.07) versus threshold AN (Mean = 1.30, SD = 1.13) significantly differed (t-value = 2.54, p-value = .018, d = 1.11). Functional impairment scores between subthreshold and threshold cases did not significantly differ (all p-values > .134) and were associated with small effect size differences for AN, BN, BED, and PD (Cohen’s d’s .24, .27, .11, and .10, respectively).

Correlations between baseline risk factors are shown in Table 5. The average correlation among risk factors was r = .14 (range = .01 – .54).

Table 5.

Pearson Correlations for Baseline Risk Factors

		1	2	3	4	5	6	7	8	9	10	11	12	13
1.	Thin ideal internalization	1.00
2.	Thinness expectancy	0.31	1.00
3.	Denial of thin-ideal costs	0.14	0.41	1.00
4.	Body dissatisfaction	0.27	0.48	0.23	1.00
5.	Weight control behaviors	0.10	0.13	0.08	0.10	1.00
6.	Dieting	0.31	0.36	0.29	0.37	0.47	1.00
7.	Negative affect	0.24	0.33	0.24	0.42	0.06	0.30	1.00
8.	Overeating	0.14	0.31	0.15	0.27	0.06	0.21	0.35	1.00
9.	Fasting	0.15	0.22	0.33	0.23	0.17	0.43	0.25	0.10	1.00
10.	Excessive exercise	0.08	0.14	0.20	0.14	0.21	0.37	0.15	0.08	0.50	1.00
11.	Functional impairment	0.07	0.22	0.14	0.29	-0.01	0.15	0.54	0.23	0.18	0.12	1.00
12.	Mental health care	0.03	0.13	0.08	0.16	0.07	0.10	0.26	0.13	0.15	0.07	0.18	1.00
13.	Body Mass Index	-0.04	0.19	0.03	0.34	0.08	0.10	0.07	0.07	0.05	0.01	0.01	0.04	1.00

Univariate Predictors of Specific Eating Disorders

We first tested in Cox proportional hazard moderation models whether the relation between the risk factors targeted in the Body Project prevention program (thin-ideal internalization, thinness expectancies, denial of thin-ideal costs) showed significantly different relations with eating disorder onset among Body Project versus control participants. This was necessary because an assumption of combining the data from these trials is that condition did not moderate the predictive effects of the risk factors targeted in the prevention programs. We likewise tested whether the relation between the risk factor targeted in the Healthy Weight prevention program (weight control behaviors) showed significantly different relations with eating disorder onset among Healthy Weight versus control participants. No significant interactive effects were found (all p-values > .098).

Table 6 reports results from the univariate Cox proportional hazard models predicting onset of future subthreshold or threshold eating disorders. Functional impairment and lower BMI at baseline predicted future onset of sAN/AN. Greater thin-ideal internalization, thinness expectancy, denial of costs of pursuing the thin ideal, body dissatisfaction, dieting, negative affect, overeating, fasting, functional impairment, and mental health care predicted future onset of sBN/BN. Greater thin-ideal internalization, thinness expectancy, body dissatisfaction, dieting, negative affect, overeating, functional impairment, and mental health care predicted future onset of sBED/BED. Greater thin-ideal internalization, thinness expectancies, denial of costs of pursuing the thin ideal, body dissatisfaction, dieting, negative affect, overeating, fasting, excessive exercise, functional impairment, and mental health care predicted future onset of sPD/PD.

Table 6.

Results from Univariate Cox Regression Models Predicting Future Onset of Subthreshold/Threshold Eating Disorders.

	Anorexia Nervosa				Bulimia Nervosa
		95% CI				95% CI
	HR	LB	UB	p-value	HR	LB	UB	p-value
Thin ideal internalization	1.38	0.91	2.09	.1270	1.69	1.30	2.18	<.0001
Thinness expectancy	1.04	0.69	1.58	.8434	1.73	1.30	2.31	.0002
Denial costs	1.09	0.72	1.65	.6807	1.40	1.10	1.78	.0067
Body dissatisfaction	1.11	0.75	1.65	.5928	1.94	1.51	2.51	<.0001
Weight control behaviors	1.15	0.73	1.82	.5546	1.29	0.96	1.73	.0919
Dieting	1.15	0.78	1.69	.4776	1.89	1.49	2.39	<.0001
Negative affect	1.27	0.89	1.80	.5337	1.79	1.48	2.15	<.0001
Overeating	1.12	0.79	1.58	.6537	2.08	1.74	2.48	<.0001
Fasting	0.90	0.57	1.42	.2837	1.23	1.06	1.43	.0074
Excessive exercise	1.15	0.89	1.49	.2835	1.13	0.95	1.35	.1749
Functional impairment	1.71	1.18	2.49	.0047	1.42	1.12	1.80	.0034
Mental health care	0.84	0.54	1.31	.4448	1.49	1.22	1.82	<.0001
Body Mass Index	0.05	0.02	0.13	<.0001	1.08	0.86	1.34	.5157
	Binge Eating Disorder				Purging Disorder
Thin ideal internalization	1.61	1.25	2.07	.0002	1.81	1.38	2.38	<.0001
Thinness expectancy	1.36	1.03	1.78	.0280	2.01	1.45	2.77	<.0001
Denial costs	1.20	0.94	1.54	.1432	1.66	1.28	2.17	.0002
Body dissatisfaction	2.02	1.57	2.60	<.0001	2.29	1.73	3.02	<.0001
Weight control behaviors	1.05	0.80	1.38	.7153	1.27	0.93	1.74	.1309
Dieting	1.58	1.25	1.99	.0001	2.74	2.08	3.61	<.0001
Negative affect	1.65	1.37	1.99	<.0001	1.70	1.40	2.07	<.0001
Overeating	2.84	2.08	3.88	<.0001	1.64	1.36	1.98	<.0001
Fasting	1.08	0.90	1.30	.4008	1.28	1.11	1.47	.0006
Excessive exercise	1.13	0.96	1.33	.1533	1.21	1.04	1.42	.0173
Functional impairment	1.56	1.26	1.94	<.0001	1.52	1.20	1.93	.0006
Mental health care	1.37	1.12	1.68	.0018	1.31	1.05	1.62	.0153
Body Mass Index	1.12	0.91	1.39	.2830	1.09	0.86	1.37	.4710

Notes. Bolded entries significant at p<.05. CI = confidence interval, HR = hazard ratio, LB = lower bound, UB = upper bound.

Table 7 reports results from univariate Cox proportional hazard models were also used to predict future onset of threshold eating disorders. Significant baseline risk factors were similar to those predicting onset of subthreshold/threshold eating disorders. Differences were that negative affect was a predictor for onset of AN (p = .014), but not sAN/AN (p = .534); denial costs was a predictor for onset BED (p = .007), but not sBED/BED (p = .143); thinness expectancy was a predictor for onset of sBED/BED (p = .028), but not BED (p = .056); and fasting and excessive exercise were predictors for onset of sPD/PD (p’s = .001 and .017, respectively), but not for PD (p’s = .078 and .706). No differences in predictors for sBN/BN versus BN were found.

Table 7.

Results from Univariate Cox Regression Models Predicting Future Onset of Threshold Eating Disorder.

	Anorexia Nervosa				Bulimia Nervosa
		95% CI				95% CI
	HR	LB	UB	p-value	HR	LB	UB	p-value
Thin ideal internalization	1.47	0.66	3.28	.3512	1.81	1.31	2.51	.0003
Thinness expectancy	1.54	0.63	3.76	.3374	1.75	1.22	2.53	.0028
Denial of thin-ideal costs	1.31	0.60	2.86	.5011	1.60	1.20	2.14	.0016
Body dissatisfaction	1.57	0.70	3.50	.2716	2.10	1.53	2.88	<.0001
Weight control behaviors	1.48	0.61	3.56	.3830	1.16	0.81	1.66	.4234
Dieting	2.14	0.98	4.68	.0560	1.66	1.24	2.23	.0006
Negative affect	2.06	1.15	3.68	.0144	1.85	1.47	2.32	<.0001
Overeating	1.50	0.83	2.69	.1765	2.34	1.88	2.93	<.0001
Fasting	0.92	0.40	2.13	.8512	1.35	1.17	1.57	<.0001
Excessive exercise	1.32	0.94	1.85	.1126	1.17	0.95	1.43	.1361
Functional impairment	2.17	1.08	4.33	.0292	1.73	1.31	2.30	.0001
Mental health care	0.80	0.33	1.94	.6166	1.61	1.26	2.06	.0001
Body Mass Index	0.16	0.03	0.75	.0201	0.96	0.71	1.29	.7651
	Binge Eating Disorder				Purging Disorder
Thin ideal internalization	1.52	1.13	2.06	.0062	1.92	1.36	2.71	.0002
Thinness expectancy	1.36	0.99	1.86	.0595	2.00	1.31	3.06	.0016
Denial of thin-ideal costs	1.47	1.11	1.94	.0068	1.66	1.18	2.34	.0038
Body dissatisfaction	2.13	1.57	2.89	<.0001	2.36	1.66	3.36	<.0001
Weight control behaviors	0.99	0.73	1.35	.9703	1.38	0.95	2.01	.0922
Dieting	1.51	1.14	2.00	.0038	2.35	1.68	3.29	<.0001
Negative affect	1.80	1.44	2.25	<.0001	1.72	1.34	2.21	<.0001
Overeating	3.38	2.25	5.10	<.0001	1.54	1.21	1.96	.0004
Fasting	1.16	0.95	1.41	.1390	1.20	0.98	1.46	.0788
Excessive exercise	1.16	0.97	1.40	.1026	1.05	0.80	1.38	.7062
Functional impairment	1.92	1.48	2.47	<.0001	1.45	1.08	1.95	.0140
Mental health care	1.48	1.17	1.87	.0012	1.36	1.04	1.78	.0227
Body Mass Index	1.17	0.92	1.50	.1937	1.10	0.83	1.46	.5163

Notes. Bolded entries significant at p<.05. CI = confidence interval, HR = hazard ratio, LB = lower bound, UB = upper bound.

Table 8 shows results from the univariate Cox proportional hazard models predicting future onset of core eating disorder symptoms. Greater thin-ideal internalization, thinness expectancy, denial costs of thin ideal pursuit, body dissatisfaction, dieting, negative affect, overeating, excessive exercise, functional impairment, and mental health care predicted binge eating onset. Similar risk factors predicted compensatory behavior onset, with the exception that the effect for denial of the costs of thin ideal pursuit was non-significant and weight control behaviors was significant. Similar risk factors also predicted onset of overvaluation of weight or shape, except that thin ideal internalization, denial of costs of thin ideal pursuit, and mental health care were non-significant, and fasting was significant. Finally, similar risk factors predicted onset of fear of gaining weight, except fasting was significant. Unlike the other symptoms, current BMI was only assessed at each assessment, not on a month-by-month basis over follow-up. Thus, logistic regression models were used to assess whether baseline risk factors predicted onset of low BMI at any follow-up assessment. Lower baseline dieting scores (OR = 0.67, 95% CI [0.47, 0.97], p = .033) and lower BMI (OR = 0.03, 95% CI [0.01, 0.10], p < .0001) predicted low BMI onset.

Table 8.

Results from Univariate Cox Regression Models Predicting Onset of Core Eating Disorder Symptoms that Crosscut Eating Disorder Diagnoses.

	Recurrent Binge Eating				Recurrent Compensatory Behaviors
		95% CI				95% CI
	HR	LB	UB	p-value	HR	LB	UB	p-value
Thin ideal internalization	1.34	1.14	1.57	.0003	1.53	1.26	1.86	<.0001
Thinness expectancy	1.21	1.01	1.46	.0368	1.46	1.16	1.84	.0015
Denial costs	1.19	1.01	1.41	.0410	1.24	1.00	1.54	.0514
Body dissatisfaction	1.39	1.19	1.62	<.0001	1.39	1.14	1.70	.0010
Weight control behaviors	1.10	0.93	1.31	.2577	1.20	1.05	1.37	.0078
Dieting	1.49	1.28	1.72	<.0001	2.08	1.71	2.54	<.0001
Negative affect	1.50	1.32	1.71	<.0001	1.41	1.25	1.58	<.0001
Overeating	1.70	1.47	1.98	<.0001	1.65	1.39	1.97	<.0001
Fasting	1.12	1.00	1.27	.0548	10.2	0.16	674.7	.2769
Excessive exercise	1.16	1.04	1.29	.0099	31.4	2.08	475.0	.0131
Functional impairment	1.31	1.13	1.53	.0005	1.28	1.12	1.45	.0002
Mental health care	1.26	1.10	1.45	.0010	1.16	1.03	1.31	.0141
Body Mass Index	0.99	0.85	1.15	.8742	1.02	0.91	1.15	.7068
	Overvaluation of Weight/Shape				Fear of Gaining Weight
Thin ideal internalization	1.06	0.88	1.29	.5358	1.58	1.34	1.87	<.0001
Thinness expectancy	1.22	1.06	1.40	.0067	1.58	1.31	1.92	<.0001
Denial costs	1.25	0.99	1.59	.0649	1.43	1.17	1.74	.0005
Body dissatisfaction	1.52	1.23	1.88	<.0001	2.05	1.72	2.44	<.0001
Weight control behaviors	1.03	0.89	1.19	.6848	1.24	0.94	1.63	.1348
Dieting	1.26	1.11	1.44	.0003	2.03	1.72	2.39	<.0001
Negative affect	1.31	1.15	1.48	<.0001	1.82	1.60	2.08	<.0001
Overeating	1.56	1.28	1.92	<.0001	1.64	1.45	1.87	<.0001
Fasting	1.26	1.11	1.42	.0003	1.41	1.28	1.56	<.0001
Excessive exercise	1.38	1.24	1.53	<.0001	1.25	1.11	1.40	.0002
Functional impairment	1.26	1.12	1.43	.0002	1.72	1.48	2.00	<.0001
Mental health care	1.12	0.99	1.26	.0717	1.34	1.16	1.54	<.0001
Body Mass Index	1.03	0.92	1.15	.5737	1.06	0.92	1.23	.4030

Notes. Bolded entries significant at p<.05. CI = confidence interval, HR = hazard ratio, LB = lower bound, UB = upper bound.

Multivariate Predictors of Specific Eating Disorders

Multivariate models predicting onset of future subthreshold or threshold eating disorders, which included risk factors with significant univariate effects, revealed that elevated functional impairment (HR = 1.64, 95% CI [1.10, 2.44], p = .0142) and lower BMI (HR = 0.06, 95% CI [0.02, 0.14], p < .0001) remained unique predictors of sAN/AN onset. Elevated dieting (HR = 1.45, 95% CI [1.12, 1.88], p = .0048), negative affect (HR = 1.46, 95% CI [1.19, 1.79], p = .0003), and overeating (HR = 1.78, 95% CI [1.48, 2.13], p < .0001) showed unique predictive effects for sBN/BN onset. Elevated thin-ideal internalization (HR = 1.36, 95% CI [1.05, 1.75], p = .0184), body dissatisfaction (HR = 1.71, 95% CI [1.31, 2.23], p < .0001), and functional impairment (HR = 1.31, 95% CI [1.04, 1.65], p = .0209) showed unique predictive effects for sBED/BED onset. Elevated body dissatisfaction (HR = 1.63, 95% CI [1.21, 2.18], p = .0012), dieting (HR = 2.24, 95% CI [1.66, 3.03], p < .0001), and overeating (HR = 1.25, 95% CI [1.03, 1.52], p = .0224) showed unique predictive effects for sPD/PD onset.

Multivariate models predicting onset of future threshold eating disorders, which included risk factors that showed significant univariate effects, revealed that elevated negative affect (HR = 2.24, 95% CI [1.25, 3.99], p = .0064) and lower BMI (HR = 0.63, 95% CI [0.44, 0.91], p < .0135) showed unique predictive effects for AN onset. Elevated body dissatisfaction (HR = 1.50, 95% CI [1.08, 2.08], p = .0166), overeating (HR = 2.07, 95% CI [1.65, 2.61], p < .0001), and fasting (HR = 1.29, 95% CI [1.08, 1.53], p = .0046) showed unique predictive effects for BN onset. Elevated body dissatisfaction (HR = 1.49, 95% CI [1.07, 2.06], p = .0167), overeating (HR = 1.99, 95% CI [1.59, 2.49], p < .0001), and functional impairment (HR = 1.40, 95% CI [1.07, 1.83], p = .0142) showed unique predictive effects for BED onset. Elevated body dissatisfaction (HR = 1.88, 95% CI [1.29, 2.72], p = .0009) and dieting (HR = 1.94, 95% CI [1.35, 2.79], p = .0003) showed unique predictive effects for PD onset.

Multivariate models predicting onset of future eating disorder symptoms, which included risk factors that showed significant univariate effects, revealed that elevated dieting (HR = 1.26, 95% CI [1.08, 1.47], p = .0041), negative affect (HR = 1.34, 95% CI [1.17, 1.54], p < .0001), and overeating (HR = 1.50, 95% CI [1.29, 2.76], p < .0001) showed unique predictive effects for binge eating onset. Elevated thin-ideal internalization (HR = 1.16, 95% CI [1.01, 1.33], p = .0304), dieting (HR = 1.50, 95% CI [1.30, 1.73], p < .0001), negative affect (HR = 1.23, 95% CI [1.08, 1.39], p = .0019), overrating (HR = 1.31, 95% CI [1.15, 1.48], p < .0001), and excessive exercise (HR = 17.92, 95% CI [1.17, 275.55], p = .0385) showed unique predictive effects for compensatory behavior onset. Elevated body dissatisfaction (HR = 1.20, 95% CI [1.06, 1.37], p = .0048), negative affect (HR = 1.22, 95% CI [1.07, 1.40], p = .0037), fasting (HR = 1.17, 95% CI [1.01, 1.35], p = .0379), and excessive exercise (HR = 1.29, 95% CI [1.15, 1.44], p < .0001) showed unique predictive effects for onset of overvaluation of weight or shape. Elevated body dissatisfaction (HR = 1.44, 95% CI [1.19, 1.74], p = .0002), dieting (HR = 1.58, 95% CI [1.32, 1.89], p < .0001), negative affect (HR = 1.21, 95% CI [1.01, 1.45], p = .0343), overeating (HR = 1.24, 95% CI [1.08, 1.42], p = .0017), fasting (HR = 1.16, 95% CI [1.03, 1.30], p = .0019) and functional impairment (HR = 1.24, 95% CI [1.03, 1.49], p = .0220), showed unique predictive effects for onset of fear of weight gain/obesity. Lower baseline BMI (HR = 0.03, 95% CI [0.01, 0.11], p < .0001) showed a unique predictive effect for onset of low BMI.

Discussion

Thin-ideal internalization, positive expectancies for thinness, body dissatisfaction, dieting, overeating, and mental health care predicted onset of BN, BED, and PD, and denial of costs of pursuing the thin ideal and fasting predicted onset of two of these three disorders. Results for the first four risk factors largely converge with prior prospective studies that identified predictors of future onset of subthreshold/threshold BN, BED, and PD (Table 1) and provide support for the theory that pursuit of the culturally sanctioned thin ideal and the resulting body dissatisfaction, dieting, and maladaptive weight control behaviors increase risk for eating disorders involving binge eating and unhealthy compensatory behaviors. The overlap in predictors of BN, BED, and PD is striking, as each of these risk factors predicted onset of two or three of these eating disorders and no examined risk factor was specific to only one of these eating disorders classes.

Few studies have tested whether prodromal symptoms predict onset of syndromal eating disorders. The predictive relation between fasting and onset of binge-spectrum eating disorders converges with past findings (Stice et al., 2008a), but the effects for overeating and excessive exercise are novel. When interpreting the prodromal effects, it should be noted that there was partial criterion confounding, in that fasting and excessive exercise can be diagnostic features of AN, BN, and PD, and that three of the six overeating questions must be endorsed for a BED diagnosis. The former might have contributed to the relation of fasting and excessive exercise to BN onset and the relation of fasting to PD onset, but could not explain the relation of excessive exercise to BED onset. Moreover, if criterion confounding were solely driving the results, fasting and excessive exercise should have predicted onset of AN, but they did not. The latter criterion confounding could have contributed to the relation of overeating to BED onset, but could not explain why overeating predicted onset of BN and PD, as those overeating features are not required for these two diagnoses. The fact that it took 11.5 months on average to show onset of BED also implies that participants did not simply show a small increase in binge eating to fulfill BED criteria. Further, all of the participants at baseline who endorsed overeating did not met criteria for the other symptoms of each of the eating disorders; as such participants were excluded from survival models. Although effect sizes were generally small, overeating showed the largest predictive effects for BN and BED. Data imply that it might be useful for indicated prevention programs to focus on individuals with prodromal eating disorder symptoms.

Negative affect and impaired psychosocial functioning were the only risk factors that predicted onset of all four eating disorders. The former finding provides support for etiologic models that posit that negative affect increases risk for binge eating and compensatory weight control behaviors (Lavender et al., 2016; Stice, 2001). Theoretically, negative affect increases the reward value of high-calorie foods and binge eating improves mood (Lemmens, Rutters, Born, & Westerterp-Plantenga, 2011; Stice, 2001). Negative affect may also increase the likelihood of unhealthy weight control behaviors because they reduce anxiety about excess body fat. For others, negative affect may reduce appetite, contributing to unhealthy weight loss. The finding that impaired psychosocial functioning increased risk for all eating disorders is noteworthy because etiologic theory has not implicated this risk factor. Yet this finding converges with evidence that social withdrawal predicted onset of any eating disorder (except AN) over 3–20 year follow-up (Allen, Byrne, Oddy, Schmidt, & Crosby, 2014). Data suggest that impaired functioning precedes onset of the disorders examined herein and thus may play a role in the development of the full spectrum of eating disorders. Impaired functioning may contribute to negative affect, which could increase risk for binge eating, compensatory behaviors, and under-nutrition for the reasons enumerated above.

This is the first study to find that mental health care predicts onset of eating disorders. It is unlikely that mental health care prompts eating disorder onset. Rather, receiving mental health care may be a proxy risk factor for comorbid psychopathology (e.g., it is possible that mood and anxiety disorders contribute to mental health care and also increase risk for eating disorders).

One striking findings is that low BMI, low dieting, negative affect, and functional impairment predicted onset of AN and a low BMI, whereas the risk factors relating to cultural pressures for thinness and resulting body dissatisfaction and weight control behaviors did not. The evidence that a low BMI predicted onset of AN and a low BMI, which was the largest effect, and the evidence that low dieting predicted onset of a low BMI, converges with evidence that low BMI and low dieting predicted future onset of AN in the only other prospective study of risk factors for AN onset (Stice & Bohon, 2013). The earlier study used data from a community sample of adolescent girls followed over 5-years (early puberty, perceived pressure for thinness, thin-ideal internalization, body dissatisfaction, negative affect, and social support deficits did not predict AN onset). The evidence that low BMI predicted AN onset in two studies converges with the finding from retrospective studies that lifetime diagnoses of AN correlated with a low BMI at age 6 (Tyrka, Waldron, Graber, & Brooks-Gunn, 2002) and with low birth weight (Cnattingius, Hultman, Dahl, & Sparen, 1999), and low dieting, eating conflict, meal struggles, and unpleasant meals at age 6 (Kotler, Cohen, Davies, Pine, & Walsh, 2001). Contrary to accepted theorizing based on case-control studies, which implicate dieting, perfectionism, and low self-esteem (e.g., Fairburn, Cooper, Doll, & Welch, 1999), results imply that youth who are ambivalent to food or who undereat for other reasons, and therefore have lower BMIs and less need to diet, are at risk for AN. Although low body weight during early adolescence might be construed as a prodromal sign of incipit AN onset, fasting and excessive exercise did not predict AN onset. Negative affect and impaired psychosocial functioning also predicted AN onset, implying that youth who have high negative affect and difficulty getting along well with their family, friends, and peers are at risk for AN onset or that under-eating causes or exacerbates affective disturbances and interpersonal conflict. The fact that thin ideal internalization and body dissatisfaction did not predict AN implies that cultural pressure for thinness do not increase risk for AN.

Risk factors that predicted onset of subthreshold or threshold eating disorders generally predicted onset of threshold versions of the same eating disorder, implying that risk factors for subthreshold and threshold variants of these eating disorders are more similar than different. Further, the risk factors that predicted onset of binge eating and compensatory behaviors were similar to those that predicted onset of subthreshold and threshold BN, BED, and PD. There were only a few differences in risk factors identified in these analyses: whereas excessive exercise did not predict onset of subthreshold or threshold BN and BED, it did predict binge eating onset; and fasting showed a stronger relation to onset of subthreshold or threshold PD than compensatory behavior onset. Results suggest that these forms of eating pathology have similar risk processes. In contrast, perhaps the largest difference is that most of the examined risk factors predicted future onset of weight and shape overvaluation and of fear of weight gain, whereas only two of these risk factors predicted onset of subthreshold or threshold AN. These findings imply that sociocultural factors and prodromal symptoms play a more potent role in the attitudinal components of AN versus the low BMI component. These results also suggests that the etiologic processes that give rise to a low BMI and AN are distinct from those that predict onset of the other core behavioral and attitudinal symptoms.

Many of the risk factors that showed significant univariate effects did not account for unique variance in eating disorder onset in multivariate models, presumably because these factors are functionally correlated; subscription to the thin ideal theoretically contributes to body dissatisfaction, which motivates use of unhealthy weight control behaviors, and contributes to negative affect (Stice, 2001). Body dissatisfaction and dieting showed unique effects for onset of BN and BED, implying that these are cornerstone risk factors that should be targeted in prevention programs. Negative affect and thin-ideal internalization also showed unique predictive effects for onset of BN and BED respectively, implying that the effects of these factors are somewhat distinct from those of body dissatisfaction and dieting. It was noteworthy that overeating and functional impairment showed unique effects for onset of these two eating disorders and that overeating showed a unique effect for onset of PD, as is the first study to investigate these factors. Findings imply that these variables are tapping orthogonal risk processes that deserve greater attention in etiologic theories and prevention programs.

Diagnostic progression from subthreshold to threshold levels of eating disorders was higher in this high-risk sample (31%) than in studies of more representative samples (17%). Diagnostic progression was lower for AN (4%) than for BN, BED, and PD (21–46%), mirroring the pattern from community samples (AN = 4%, BN = 25%, BED = 21%). Individuals with full threshold versus subthreshold levels of eating disorders generally had similar deficits in functional impairment and negative affect, replicating past findings (Stice et al., 2009).

It is important to consider study limitations. First, most of the examined risk factors involved variables implicated in sociocultural etiologic models of eating disorders. Future risk factor studies should include biological variables. The evidence that a prevention program that reduces pursuit of the thin ideal decreases fMRI-assessed brain reward region response to thin models (Stice, Yokum, & Waters, 2015) suggests that this cornerstone sociocultural risk factor has a biological basis that maps onto reward valuation from the RDoC matrix. There may be value in examining biological processes that underlie other risk factors examined herein. Second, the focus on body-dissatisfied women may have obscured detection of other risk factors orthogonal to body dissatisfaction. However, the means and SDs for the risk factors were similar to those from a community sample of aged-matched adolescent girls, implying that this sample is representative and that there was no restriction in range that reduced sensitivity. Moreover, the risk factors identified in this high-risk study of adolescent females were similar to those identified in community studies of adolescent females (Table 1), suggesting that the high-risk design did not bias results. Third, participants in this study completed eating disorder prevention programs, though the effects of condition were statistically controlled and the predictive effects of the risk factors were not affected by the interventions. Fourth, we combined normalized responses to a negative affect measure and depressive symptom measures, but the two constructs are not isomorphic and combining them probably introduced error variance, though fortunately this did not prevent us from detecting predictive effects of this risk factor. Fifth, it is possible that some of the results could have emerged due to chance. We considered using a lower alpha, but decided against it for three reasons. First, it is important to balance Type I error risk (false positive findings) against Type II error risk (false negative findings); given that survival models that predict onset of low incident events have low sensitivity, we were more concerned about Type II errors. Second, the fact that 62% of the tested effects were significant communicates that it is unlikely that we are interpreting chance findings, which would have been closer to 5% of the tested effects. Third, the fact that findings from this report converge with those from prior studies also suggests it is unlikely that the present results are due to chance. Nonetheless, we report the exact p-values of each of the tests so that readers can focus on the effects that exceed more stringent alpha levels (e.g., p < .01) if they so desire.

With regard to implications, results suggest that examining high-risk samples is a useful way to identify risk factors for specific mental disorders within a diagnostic cluster. Second, the predictive role across eating disorders played by negative affect and functional impairment suggests that these should be examined as potential risk factors for the onset of other mental disorders, as they may represent transdiagnostic risk factors for psychopathology, which if targeted in prevention programs might reduce a range of mental health problems. Third, findings suggest a distinct risk profile for AN versus eating disorders in the binge eating/purging spectrum. Data imply it would be useful to target low BMI middle school students to advance knowledge of risk factors for AN. Fourth, the overall similarity in risk factors that predicted onset of threshold versus subthreshold or threshold eating disorders versus the core symptom dimensions that crosscut eating disorders implies that eating pathology exists along a continuum. Future research should focus on elucidating how best to model eating pathology in etiologic studies to maximize sensitivity for detecting risk factors. Fifth, results imply that the yield of eating disorder prevention programs, which typically focus on thin-ideal internalization and body dissatisfaction, might be improved by also targeting negative affect, overeating, and functional impairment. The evidence that negative affect and impaired interpersonal functioning predict a range of eating disorders implies that it might be fruitful to target these high-risk populations with selective prevention programs if the goal is to prevent the full spectrum of eating disorders. Also, existing prevention interventions may be more effective in reducing future binge/purge behaviors than the weight restriction present in AN, which should be investigated. Data suggest that interventions aimed at preventing AN may need to target qualitatively different high-risk populations (e.g., low weight middle school girls) and might need to focus on reducing distinct risk processes for optimal success.