SUBTYPING DIETARY RESTRAINT AND NEGATIVE AFFECT IN A LONGITUDINAL COMMUNITY SAMPLE OF GIRLS

Abstract

Objective

This study tests the validity of the ‘dietary-depressive’ subtype (typified by greater negative affect) and a ‘dietary’ subtype (typified by dietary restraint only) using a diverse longitudinal community sample.

Method

Girls at ages 10, 12 and 14 completed the Child Eating Attitudes Test, the Child Symptom Inventory-4, and Body Image Measure. Body Mass Index was assessed at each age.

Results

Unlike previous studies, cluster analysis revealed an at-risk ‘dietary-depressive’ (R+) subtype (18.7%,100/534) and a not at-risk (R−) subtype, distinguished by few depressive symptoms and little dietary restraint (81.3%,434/534), but no ‘dietary’ subtype. Compared to the R− subtype, the R+ subtype had significantly greater eating disordered behavior and attitudes. The R+ subtype at age 10 was a risk factor for binge-eating but not obesity at ages 12 and 14.

Conclusion

Dietary restraint and depressive symptoms combined predict binge-eating longitudinally in a diverse community sample of girls.

Binge eating is thought to be maintained by dietary restraint or affect regulation. Dietary restraint models describe binge eating as the body’s response to dieting (1) which in turn is prompted by over-concern with weight and shape(2). In contrast, affect regulation models describe binge eating as an attempt to influence, change, or control painful emotional states (3).

The dual-pathway model (4, 5) integrates dietary restraint, affect dysregulation and socio-cultural factors as triggers of binge eating. This model has been supported by subtyping studies involving treatment-seeking eating disorder samples, particularly with binge eating problems. Results from these studies (6, 7) (8–10) suggest that 30–40% of clinical binge eating samples fit a ‘dietary-depressive’ (or ‘dietary-negative affect’) subtype, whose binge eating is typified by greater negative affect, and 60–70% fit a ‘dietary’ subtype, whose binge eating is typified by dietary restraint and less by negative affect. Typically, negative affect has been assessed on the Beck Depression Inventory (BDI) and dietary restraint has usually been assessed on the Eating Disorder Examination (EDE) (11) Restraint subscale. The ‘dietary-depressive’ affect subtype typically scores moderate to severe depressed mood on the BDI and in the clinical range (compared to normative groups (11)) on EDE Restraint subscale. The ‘dietary’ subtype typically score within the normal to mild-moderate depressed mood on the BDI and in the clinical range on the EDE Restraint subscale..

Studies with clinical samples of adults find that individuals meeting the definition of the ‘dietary-depressive’ subtype compared to the ‘dietary’ subtype report more severe eating disorder behavior (e.g., binge eating) and weight and shape concerns, higher rates of other Axis I and II disorders, particularly affective disorders, greater social maladjustment, and poorer response to treatment, such as lower binge-purge abstinence (4, 6–9, 12). These results have been replicated in samples of treatment-seeking adults with bulimia nervosa (6, 9) and binge eating disorder (4, 7, 13). In addition, these results have also been found in adolescents with bulimia nervosa (12), adolescent inpatients with mixed eating disorders (8) and in children and adolescents reporting a loss of control with eating (14).

The present study adds to this literature by examining the validity of the subtyping approach in a non treatment-seeking sample of adolescent girls, and by testing the longitudinal predictive value of the subtyping strategy. This study utilizes data from the Pittsburgh Girls Study (PGS), a community-based sample of girls derived from a household enumeration of the City of Pittsburgh (15). Importantly, unlike previous subtyping studies, this study utilizes a non-referred sample. In addition the sample utilized was highly diverse economically and racially, making the results particularly ecologically valid. The primary aim of the present study was to examine whether or not the ‘dietary’ (dietary restrained only) and ‘dietary-depressive’ (reporting both dietary restraint and depressive symptoms) subtypes exist in a non-referred population of girls. If this subtyping schema is found, it is hypothesized that an ‘at-risk’ cluster would report more dysfunctional eating attitudes, poorer body image (discrepancy between ideal and current body size), higher body mass index and greater rates of obesity and binge eating behavior at age 10, and that these differences would continue through age 12 and age 14. It is further hypothesized that the ‘at–risk’ cluster would predict binge eating at ages 10, 12 and 14 independent of socio-economic influences, and differences in body size dissatisfaction and body mass indices.

The final exploratory hypothesis was to examine if the ‘at-risk’ cluster at age 10 predicts obesity at ages 10, 12 and 14 above and beyond binge eating, body size dissatisfaction and socio-economic status. While this analysis is not intended as comprehensive, given the multiple predictors of obesity not examined here (16, 17), it is intended as an exploration of whether the ‘at-risk’ subtype is a correlate and potential risk factor for obesity in girls. This analysis is important as there is evidence that dieting and depressed mood are independent contributors to binge eating or bulimic behaviors (10) (18, 19) and obesity (20, 21). However, to our knowledge, this subtyping strategy has not been utilized to examine correlates and risk factors for obesity.

METHOD

Participants

The PGS used a stratified, random household sampling strategy, with over-sampling of households in low-income neighborhoods. This strategy identified 2,992 of girls aged 5 to 8 years, of whom 85% (2,451) agreed to participate. Details of the PGS, including the sampling methodology used have been described in detail (15, 22). The University of Pittsburgh Institutional Review Board approved all study procedures. Written informed consent was obtained from the primary caregiver and verbal assent from the child.

The initial sample of 2,451 ten-year old girls was predominately African-American (52.9%), with 41.2% Caucasian, 4.9% Multi-racial and 0.9% Asian. About a third of girls (39.8%) came from single parent households and about one-third (34.8%) of the girls were in families receiving public assistance. Of this sample, 15.8% of parents reported less than 12 years of education.

The accelerated longitudinal design allows examination of developmental patterns over time by linking cohorts of different ages and by including overlapping ages at assessment. As not every individual was administered the same set of measures at each longitudinal assessment point, in the present study we only included girls who had data pertinent to the study at ages 10, 12 and 14, i.e., N=543. Of this sample of 543 girls, the majority were from a minority group (59.3%, with 54.1% African American, 4.8% Multi-racial and 0.4% Asian) and 40.5% were Caucasian. Almost half of the girls (43.8%) came from single parent households and about one-third (29.5%) of the girls were in families receiving public assistance. Of this sample, 19.5% of parents reported less than 12 years of education. When compared to the initial sample of 2,451, the final sample of 543 differed in having a greater percentage with parents with less than 12-years of education but a higher percentage of families not on public assistance, although there was no difference in rates of minority status or single parent households (p<.05).

Measures

Child Eating Attitudes Test (ChEAT) (23)

The ChEAT is a 26-item questionnaire assessing attitudes and behaviors associated with anorexia nervosa and bulimia nervosa. This measure is a simplified version of the adult Eating Attitudes Test. Each item is rated on a likert scale from 1 (always) to 6 (never). In coding, the most symptomatic score is recoded to 3, the next, a 2, and the least symptomatic, a 1, with the three remaining choices each coded 0. Total raw scores can range from 0 to 78, with studies finding that a score of 20 or more identifying disturbed versus normal eaters in 6^th grade (22) and high school girls (23). The internal consistency (Cronbach’s alpha= .76), test-retest reliability (.81) and concurrent validity for the ChEAT are good (22). Examination of the factor structure of the ChEAT in adolescent girls yielded four factors (Dieting [DIET], Restricting and Purging, Food Preoccupation and Oral Control) that accounted for about 48% of the variance in total ChEAT scores (23). Questions that loaded on the DIET factor included those about thinking about being thinner and being scared about being overweight. Questions about vomiting and the urge to vomit loaded on the Restricting and Purging factor. Questions such as ‘thinking about food a lot’ loaded on the Food Preoccupation factor, and questions about ‘feeling as if others pressure me to eat’ loaded on the Oral Control factor. In this study, calculation of these factors involved the summing of raw scores.

Body Image Measure (BIM) (24))

Silhouettes of seven girls were presented to illustrate a range in body weight from very thin to obese. The girls were asked to rate: 1) which picture looks the most like you look, and 2) which picture shows the way you want to look. A discrepancy score was calculated where the current was subtracted from the ideal score. Negative scores denoted body size dissatisfaction, with the larger the negative score, the greater the discrepancy between current and ideal. Test-retest reliability of this measure among children found this to be .71 for current figure and .59 for the ideal (24). Criterion-related validity derived by comparing the current figure with actual body mass index yielded significant correlation coefficients (.37) (24).

Body Mass Index (BMI) and obesity

Height in inches and weight in pounds were measured using a calibrated stadiometer and digital scale. In adolescents, BMI (kg/m2) has been found to correlate with measures of total body fat and with health measures such as diabetes, insulin and blood pressure measures (25) and is the recommended proxy measure for adiposity. Obesity was defined using the age and sex-adjusted 95^th percentiles of BMI for girls based upon United States normative data (26).

Child Symptom Inventory-4 (CSI-4) (27)

Depressive symptoms were measured using the CSI Symptom Inventory-4, which is based upon the child’s report. The CSI-4 is a DSM-IV based checklist that has been normed and validated for children 5 to 13 years of age (28). The CSI-4 has also been used for adolescents age 14 (29, 30). For generating symptom counts, we summed the nine items reflecting DSM-IV symptoms of depression that were endorsed as ‘a lot’, ‘all the time’ or ‘yes’ as in a previous paper (29). The CSI-4 also has a symptom severity score that is the sum of scores for a particular disorder. The CSI-4 has adequate reliability (27) (30), and validity (30). Internal consistency for symptom counts for major depression was .63 and one-month test-retest reliability was .68 (31).When compared to a structured interview for child and adolescent disorders, sensitivity for depressive disorders was 0.90 (28).

Data analysis

The analyses were conducted using a weight variable to correct for the oversampling of low-income neighborhoods, enabling the results to be representative of girls living in the city of Pittsburgh. This weight variable was generated by comparing the proportions of neighborhoods represented in the study to the proportions of neighborhoods in the city of Pittsburgh in which girls in the same age range were living (using data from the 2000 U.S. Census). The weighted analysis accounts for the oversampling of minority and low income groups.

Cluster analysis was performed to identify a subtype of girls ‘at risk’ for eating disordered behavior. To do this, two clusters (k) were specified using scores on Smolak’s DIET factor (on the ChEAT) and Symptom Count for the CSI-4 depressive symptoms. Two clusters were specified as we wished to examine the same subtyping strategy used in previous studies with clinical samples (12) (8) (7) (7) (9) (4). The ChEAT DIET factor and the CSI-4 symptom counts used for subtyping the sample were reliable (Cronbach’s alpha respectively, 0.75 and 0.61 respectively). As in previous studies (6) (9), squared Euclidean distances were used as the distance measure and a non-hierarchical or iterative clustering algorithm, rather than a hierarchical algorithm was used as a non-hierarchical clustering is less sensitive to outliers and more prone to detect and correct early misclassifications.

To assess simple differences between clusters on measures of disordered thoughts and behaviors associated with eating (ChEAT factors, BMI, obesity, binge eating, vomiting) and discrepancy score on BIM a series of t-tests and χ² analyses were performed on the data for subjects at ages 10, 12 and 14. Raw scores on the two ChEAT questions regarding 1) vomiting after eating and 2) having uncontrollable eating binges were coded as ‘yes’=1 or ‘no’ =0 where ‘yes’ referred to a subject validating that they did this ‘always’, ‘very often’ or ‘often’ while ‘no’ was when a subject validated ‘sometimes’, ‘rarely’ and ‘never’. These items were collapsed in this way as this data was skewed and transformations did not rectify this. To control for the number of tests at each time point a bonferroni corrected alpha value of p < .006 (.05/9) was used. To assess effect size, Cohen’s d (32)was used for t-test comparisons and Φ was used for χ² analyses. Tests were 2-sided.

To assess extent to which the membership in the at-risk group is a correlate of (age 10) or predicts likelihood (at ages 12 and 14) of binge eating independent of socioeconomic differences, body size dissatisfaction (BIM) and weight (BMI), three hierarchical logistic regressions were conducted for data at ages 10, 12 and 14. All predictor variables were assessed at age 10 years. For each logistic regression, presence of binge eating was the criterion variable. Minority status, membership of a two-parent family or not, whether or not the primary caregiver was a high school graduate, family reliance on public assistance or not, BIM discrepancy score and BMI were entered in the first step, and cluster membership was entered in the second step. For minority status, the sample was divided into two groups: minority (African-American, Asian and Multi-racial) or non-minority (Caucasian). In order to examine predictors of binge eating at ages 12 and 14, individuals who reported binge eating at age 10 were excluded from these analyses in order to provide a prospective test of our hypothesis.

Finally, in order to examine whether membership of the at-risk group predicts obesity status at ages 10, 12 and 14 years, three hierarchical logistic regressions were conducted. In these regressions, predictors entered in the first step were age 10 predictors for minority status, public assistance status, intact family status, High School status of the primary caregiver, BIM score and presence of binge eating. Cluster membership was entered in the second step. So as to provide a prospective test of whether the independent variables at age 10 predicted obesity at ages 12 and 14 years, individuals who were obese at age 10 were excluded from these analyses.

RESULTS

Cluster Analysis

A Cluster analysis revealed an at-risk ‘dietary-depressive’ (R+) subtype (100/534, 18.7%) and a not at-risk (R−) subtype that reported few depressive symptoms and little dietary restraint (434/534,,81.3%), but no ‘dietary’-only subtype. The ‘dietary-depressive’ subtype is described as ‘at-risk’ because the levels of dietary restraint and depressive symptoms either are or are almost clinically impairing, and are higher than that observed in the non-referred child. For instance, mean total ChEAT scores in the R+ subtype (M=17.18, SD=8.47) were close to the standard clinical cut-off of 20 (23, 33) unlike that for the R− subtype (M=5.54, SD=3.75). Compared to the mean total CSI-4 symptom severity score for a referred sample (M =10.2, SD=4.1, (31), the mean total CSI-4 symptom severity score for the R+ subtype (M=13.03, SD=5.22) was higher while the mean total CSI-4 symptom severity score for the R− subtype (M=8.59, SD=4.47) was lower.

Compared to the R−, the R+ subtype reported significantly more symptoms of depression on the CSI-4 (M=2.87, SD=1.72 versus M=1.60, SD=1.54, t=−6.80, d=.78, p<.006) and dieting on the ChEAT DIET factor (M=9.90, SD=4.44 versus M=.92, SD=1.45, t=−19.98, d=2.72, p<.006).

Demographic and Socio-economic Analyses

R+ and R− groups did not differ with respect to age, single family, parent education, and public assistance status. R+ girls were significantly more likely to be from a minority group (see Table 1).

Table 1.

Demographic and Socio-economic Characteristics of Participants (N = 534)^a

	R− (n= 434, 81.3%)	R+ (n= 100, 18.7%)	t/χ2	p
Age in years: M (SD)	10.69(0.34)	10.74 (0.35)	−1.26	.21
Minority status: n (%minority)	247 (57.0)	71 (71.0)	6.58	.01*
Public Assistance n (% receiving)	129 (29.9)	31 (31.0)	.05	.82
Intact Family n (%intact)	246 (56.9)	49 (49.0)	2.08	.15
Parent - High School Graduate (%graduate)	353 (81.3)	83 (83.0)	0.15	.70

Note: R− = not at-risk group; R+ = at-risk ‘Dietary-Depressive subtype

^aNine participants had missing data in these analyses

t Equal variances not assumed

χ² Pearsons chi square test

p 2-sided tests.

^*p < .05

^**p < .01

^***p<.001

Primary Analyses

Comparison of At-Risk (R+) compared to not at-risk (R−) subtypes

With the exception of vomiting at age 14 and scores on the Oral Control subscale of the ChEAT at age 12, the R+ subtype was significantly more impaired relative to the R− subtype across all time points and all measures (see Table 2). These results are similar when obesity at each age was controlled for, with the only exceptions that the score on the Oral Control subscale of the ChEAT at age 12 was significant (F(1,495)=11.80, p=.001) and BIM scores at 12 and 14 were non-significant (respectively, (F(1,495)=0.10, p=.76; F(1,513)=0.18, p=.68).

Table 2.

Weight and Eating Variables as a function of subtype

	R−	R+	t/χ2	p	d/Φ
Age 10 (N = 534)a	(n = 434, 81.3%)	(n = 100, 18.7%)
ChEAT –RP: M (SD)	0.21 (0.70)	2.99 (4.27)	−6.49	0.000**	−.91
ChEAT -FP: M (SD)	0.52 (1.25)	1.93 (2.68)	−5.13	0.000**	−.67
ChEAT -OC: M (SD)	0.92 (2.10)	2.07 (2.97)	−3.67	0.000**	−.45
BIM: M (SD)	−0.15 (0.87)	−.91 (1.19)	6.02	0.000**	.73
BMI: M (SD)	19.70 (4.38)	23.37 (6.03)	−5.75	0.001*	−.70
Obesity n (% obese)	77 (77.7)	50 (50.00)	46.66	0.000**	.30
Binge eating n (%)	19 (2.3)	19 (19.0)	43.99	0.000**	.29
Vomiting n (%)	3 (0.7)	8 (8.0)	21.52	0.000**	.20
Age 12 (N = 527)b	(n = 427, 81.0%)	(n = 100, 19.0%)
ChEAT –RP: M (SD)	0.27 (0.95)	1.59 (3.65)	−3.58	.001*	−.49
ChEAT -FP: M (SD)	0.36 (1.04)	1.12 (1.95)	−3.79	.000**	−.49
ChEAT -OC: M (SD)	0.68 (1.64)	1.11 (2.35)	−1.75	.083	−.21
ChEAT -DIET: M (SD)	1.37 (2.46)	5.88 (6.12)	−7.24	.000**	−.97
BIM: M (SD)	−0.21 (0.77)	−0.61 (1.06)	3.55	.001*	.43
BMI: M (SD)	21.85 (4.83)	26.19 (6.74)	−7.27	.000**	−.74
Obesity n (% obese)	82 (18.9)	50 (50.0)	42.26	.000**	.28
Binge eating n (%)	8 (1.9)	9 (9.1)	13.34	.000**	.16
Vomiting n (%)	1 (0.2)	3 (3.0)	8.23	.004*	.13
Age 14 (N = 534)a	(n = 434, 81.3%)	(n = 100, 18.7%)
ChEAT –RP: M (SD)	0.53 (2.21)	2.08 (4.20)	−3.59	.000**	−.46
ChEAT -FP: M (SD)	0.56 (1.45)	1.32 (2.27)	−3.21	.002*	−.40
ChEAT -OC: M (SD)	.62 (1.57)	1.37 (2.51)	−2.87	.005*	−.36
ChEAT -DIET: M (SD)	2.22 (4.26)	5.63 (6.91)	−4.72	.000**	−.59
BIM: M (SD)	−0.21 (0.83)	−0.59 (1.08)	3.32	.001*	.39
BMI: M (SD)	23.50 (5.24)	28.38 (7.91)	−5.81	.000**	−.73
Obesity n (% obese)	82 (18.9)	47 (47.0)	35.04	.000**	.26
Binge eating n (%)	13 (3.0)	10 (10.0)	9.68	.002*	.14
Vomiting n (%)	2 (0.5)	3 (3.0)	5.645	.017	.10

Note: R− = not at-risk group; R+ = at-risk ‘Dietary-Depressive subtype; ChEAT-DIET= Child Eating Attitudes Test–Dieting factor; ChEAT-RP=Child Eating Attitudes Test–Restricting and Purging factor; ChEAT-FP= Child Eating Attitudes Test–Food Preoccupation factor; ChEAT-OC= Child Eating Attitudes Test–Oral Control factor; BIM= Discrepancy between ideal and current scores on the Body Image Measure; BMI= Body Mass Index.

^aNine participants had missing data in these analyses

^b16 participants had missing data in these analyses

^*p < .006 (.05/9)

^**p < .001

Predictive value of R+ subtype on binge eating

At age 10, after controlling for socio-economic variables and BMI and BIM score, R+ subtype status was a significant correlate of binge eating (B=2.60, Wald=28.62) as were lower BMI at age 10 (B=−.14, Wald=5.34) and high school graduate status of the primary caregiver (B=−2.28, Wald=4.46). At age 12, after controlling for variables assessed at age 10: socio-economic variables, BIM and BMI, R+ subtype status was a significant risk factor for binge eating (B =1.52, Wald=6.86) as well as a higher BIM score, denoting less dysfunction (B =.68 Wald=1.21). At age 14 the R+ subtype was a significant risk factor for binge eating (B=1.29, Wald=5.98) as was belonging to a family on public assistance (B =1.17, Wald=4.58), with predictors assessed at age 10. See Table 3 for details of these results.

Table 3.

Hierarchical logistic regression results for binge-eating with Minority Status, Public Assistance Status, Intact Family Status, High School Graduate status, Body Image Measure Discrepancy Score, Body Mass Index and Subtype as predictors.

	Binge-Eating at 10 years (N=543)		Binge-Eating at 12 years (N=504)		Binge-Eating at 14 years (N=504)
Variable	OR (95%CI)a,b	pc	OR (95%CI)a,b	pc	OR (95%CI)a,b	pc
Block1:
Minority Status	0.77 (.30–1.94)	.58	4.03 (.82–19.74)	.09	1.58 (.49–5.14)	.45
Public Assistance	2.26 (.89–5.72)	.09	.73 (.22–2.46)	.61	3.22 (1.10–9.39)	.03*
Intact Family	.62 (.26–1.52)	.30	.71 (.23–2.14)	.54	.75 (.28–2.03)	.58
High School Status	.10 (.01–.85)	.04	1.98 (.59–6.68)	.27	1.25 (.41–3.80)	.69
BIM	.72 (.44–1.17)	.18	1.97 (1.07–3.64)	.03*	.66 (.33–1.30)	.23
BMI	.87 (.77–.98)	.02*	1.03 (.93–.1.14)	.62	.91 (.81–1.02)	.11
Block 2:
R+/R− subtype	13.48 (5.20–34.95)	.000***	4.56 (1.46–14.19)	.009**	3.64 (1.29–10.23)	.01*

^aOR=Odds Ratio which is an estimate of effect size for each predictor from the final hierarchical logistic regression model generated.

^b95^th CI = 95^th Confidence Interval for each predictor from the final hierarchical logistic regression model generated.

^cp=probability

Note: BIM= Body Image Measure Discrepancy score, BMI= Body Mass Index.

^*p < .05

^**p < .01

^***p<.001

Predictive value of R+ subtype on obesity

At age 10, after controlling for variables assessed at age 10: socio-economic, BIM score and binge eating, subtype status was a significant correlate for obesity status (B=.99, Wald=11.14), as was minority membership (B=.54, Wald=3.91), and more dysfunctional BIM discrepancy score (B=−1.23, Wald=62.73). At age 12, after controlling for variables assessed at age 10: socio-economic variables, BIM and binge eating status, only more dysfunctional BIM discrepancy score (B =−1.29, Wald=21.37) was significant. At age 14, minority race (B =1.18, Wald=5.55) and more dysfunctional BIM score (B=−1.83, Wald=33.73) were significant risk factors for obesity status, with predictors assessed at age 10. See Table 4 for details of these results.

Table 4.

Hierarchical logistic regression results for Obesity with Minority Status, Public Assistance Status, Intact Family Status, High School Graduate status, Body Image Measure Discrepancy Score, Binge-Eating status and Subtype as predictors.

	Obesity at 10 years (N=416)		Obesity at 12 years (N=416)		Obesity at 14 years (N=416)
Variable	OR (95%CI)a,b	pc	OR (95%CI)a,b	pc	OR (95%CI)a,b	pc
Block1:
Minority Status	1.71 (1.01–2.91)	.048*	1.45 (.60–3.54)	.41	3.24 (1.22–8.61)	.02*
Public Assistance	.80 (.45–1.42)	.45	1.10 (.42–2.86)	.85	1.80(.68–4.75)	.24
Intact Family	.88 (.53–1.45)	.61	1.85 (.79–4.36)	.16	1.09 (.46–2.59)	.84
High School Status	.55 (.26–1.10)	.09	.77(.26–2.28)	.63	.46 (.14–1.56)	.32
BIM	.29 (.22–.40)	.000***	.27 (.16–.47)	.000***	.16 (.09–.28)	.000***
Binge-eating	.38 (.13–1.11)	.08	2.05 (.31–13.48)	.46	.53 (.06–4.84)	.58
Block 2:
R+/R− subtype	2.68 (1.50–4.79)	.001**	1.01 (.35–2.87)	.99	1.59 (.55–4.63)	.39

^aOR=Odds Ratio which is an estimate of effect size for each predictor from the final hierarchical logistic regression model generated.

^b95^th CI = 95^th Confidence Interval for each predictor from the final hierarchical logistic regression model generated.

^cp=probability

Note: BIM= Body Image Measure Discrepancy score; BMI= Body Mass Index.

^*p < .05

^**p < .01

^***p<.001

CONCLUSION

Previous studies have found that eating disorder samples can be subtyped into ‘dietary’ (reporting dietary restraint only) and ‘dietary-depressive’ (reporting dietary restraint and depressive symptoms) subtypes. In the present study, only the ‘dietary-depressive’ subtype was represented. The at-risk ‘dietary-depressive’ subtype compared to the not at-risk subtype reported significantly greater dysfunction in most eating disorders attitudes (ChEAT and BIM) and greater likelihood of binge eating and vomiting at ages 10, 12 and 14. Only vomiting at age 14 and Oral Control at age 12 became non-significant trends after correcting for family-wise error. When obesity at these ages were controlled for in these same analyses, this yielded similar results, with the exceptions that greater dysfunction in the at-risk subtype on the Oral Control ChEAT subscale at age 12 became significant while those on the BIM at ages 12 and 14 became non-significant. In addition, individuals in the at-risk ‘dietary-depressive’ subtype reported higher body mass indices and were significantly more likely to be obese at ages 10, 12 and 14, when other variables such as socio-economic variables and binge eating and body dissatisfaction were not controlled for.

Belonging to the at-risk ‘dietary-depressive’ subtype at age 10 was a significant correlate of binge eating at age 10 and a significant risk factor (34) for binge eating at ages 12 and 14 when socio-economic variables, body size dissatisfaction and body mass index at age 10 were controlled for. Binge eating at age 10, unlike other years, was also predicted by lower age 10 BMI and high school graduate education status of the primary caregiver. At age 12, binge eating was additionally predicted by less age 10 body size dissatisfaction and at age 14 by public assistance status. The odds of binge eating at age 10 were 13.5 times more likely in the at-risk ‘dietary-depressive’ than not at-risk group and at ages 12 and 14, these odds were approximately four-fold.

Membership of the ‘dietary-depressive’ subtype was a significant correlate with obesity at age 10, with the odds of being obese at age 10 being about two and a half times more likely in the at-risk ‘dietary-depressive’ than in the not at-risk group. However membership of the ‘dietary-depressive’ subtype was not a significant risk factor for obesity at ages 12 and 14 when socio-economic factors (minority status, public assistance status, intact family status, parent high school status), binge eating and BIM scores at age 10 were controlled for. Notably when these other possible predictors for later obesity were not controlled for, membership of the ‘dietary-depressive’ subtype predicted obesity at ages 12 and 14 years. This discrepancy between the t-test results and the regression results suggest that factors other than subtype membership are important predictors of obesity. Results from the regressions indicate that at age 10 in addition to membership of the ‘dietary-depressive’ subtype, obesity was predicted by minority status, and greater age 10 body size dissatisfaction. At age 12, the risk factors for obesity were greater age 10 body size dissatisfaction while at age 14, minority status and greater age 10 body size dissatisfaction were risk factors for obesity. So while the at-risk subtype is a risk for binge eating in early adolescence, it is not a risk factor for obesity in these age groups.

It was counterintuitive that the at-risk ‘dietary-depressive’ subtype reported high levels of binge eating but not later obesity as some studies suggest that binge eating is related to later obesity (35). It may have been that girls who reported binge eating actually had subjective rather than objective binges. Alternatively, it may be that girls who binge-ate became overweight rather than obese. Or another explanation for the results is that binge eating girls may have used compensatory behaviors such as exercise to counteract binge eating. These questions suggest the need for future studies with more extensive eating disorder assessments to untangle this relationship in girls.

This work extends the subtyping literature for BED and BN (6, 8–10) to a longitudinal diverse community sample of adolescent girls. The findings of this study with a non-clinical sample broadly fit with the dual-pathway model of binge eating. This model predicts that binge eating is a result of both dietary restraint and affect dysregulation. As in previous studies with clinical samples, the at-risk ‘dietary-depressive’ subtype in this non-clinical sample was associated with more eating disordered thoughts and behaviors. These results support the finding that later binge eating in non-clinical samples of adolescent girls may develop from a combination of depressive symptoms and dieting (5, 10, 19) and that precursors to binge eating in the preadolescent period may be more undifferentiated symptoms rather than distinct disorders. Alternatively, it may be argued that the ‘at-risk’ subtype is more an indicator of general psychological severity rather than a qualitatively distinct depressed and dietary restrained subtype.

There are various potential implications of not replicating the previously found dietary subtype but instead finding an ‘at-risk’ ‘dietary-depressive subtype’ and a not ‘at-risk’ subtype. It may be that dietary restraint and depressed mood are overlapping rather than distinct constructs in a racially and economically diverse sample of girls. This may be evidenced by the lower rates of dieting in black compared to white adolescent girls (36). Alternatively, one might argue that perhaps in younger samples such as this, dietary restraint and depressed mood are overlapping constructs that may become distinct in older ages.

While the ‘at-risk’ dietary-depressive subtype was a correlate of obesity at age 10 years, it was not a risk factor for obesity at ages 12 and 14 years. This subtype did not appear to have the same relationship with obesity as it did with binge eating despite previous findings suggesting that depressive symptoms and dieting (21) or dieting alone (37) (20) or depression alone (38) predict later obesity It may be that the baseline age of this sample (10 years old predicting 12 and 14 years) was younger than those utilized in previous studies (typically 13–14 years predicting 17–20 years) (20) (38) (37) (21). Given this, future studies may need to examine if the at-risk dietary-depressive subtype predicts obesity in later adolescence and young adulthood. This study points to the importance of examining younger aged samples and the need for following up such samples into young adulthood to examine what factors predict obesity.

Body size dissatisfaction was the only consistent risk factor for obesity at ages 12 and 14 years. This finding has been found in other adolescent studies (37) (39) (40). However, body size dissatisfaction was not a risk factor for binge eating and in fact the converse was true at age 12, i.e., less body size dissatisfaction, predicted binge eating. It may be that the relationship between body dissatisfaction and binge eating symptoms is mediated by dieting and negative affect as found in (10), resulting in a weaker relationship between body dissatisfaction and binge eating. In addition, binge eating was not a risk factor for obesity despite previous findings, although this relationship appears somewhat controversial with some longitudinal data with older samples providing evidence of this relationship e.g., (19) while others do not e.g., (21).

Other risk factors of binge eating and obesity were at age 14, minority status predicting obesity and in the same age, public assistance status predicting binge eating. These results were statistically weighted so that these findings were not due to the oversampling of minority or economically diverse groups.These findings support those of other studies showing that minority status (for instance, (41)) is associated with obesity in adolescents and that lower socio-economic status is associated with binge eating in women, e.g., (42) (43). The relationships between minority status and obesity may result from genetic differences in adiposity, co-occurring poverty, different cultural views with regards to obesity or to cultural differences in diet or a combination of these factors (44). The relationship between public assistance and binge eating may have resulted because of the direct relationship between lack of access to food, with over-eating resulting when food is available. Alternatively, impoverished individuals may only have access to cheap foods, such as fast food, that are typically presented in large quantities. These speculations are hypothetical with further research needed to examine the relationship between minority status and/or public assistance and binge eating or obesity. While there is evidence that these factors are important in obesity and binge eating there is a need to assess these factors in longitudinal community samples to examine whether these factors particularly emerge during adolescence.

These findings extend those of previous studies by subtyping a diverse community sample of girls and investigating whether this predicts binge eating and obesity longitudinally. Strengths of this longitudinal study included the use of a non treatment-seeking sample, which had a large proportion of minority girls from low-income families.

However, there are a number of limitations with the current findings that effect their interpretation. First, there were limitations in the measures used. The measures utilized were different to those used in other studies. Another limitation is that of the reliability and validity of the single item of the ChEAT used to assess binge eating. Replication of these results is needed using the Eating Disorders Examination (11), which has been used in previous studies, that makes eating disorders diagnoses and includes multiple questions for eating disorder behaviors. Second, the cluster analysis technique utilized is limited as there is no objective criteria for determining the optimal number of clusters. Given this, it is important that this finding is replicated using other data analytic strategies such as latent profile analyses. Third, the design of the study could have been improved by using a non-accelerated longitudinal design where each individual is assessed at each time point, which provides a better estimate for the onset and longer term course of problems in an individual. Fourth, examination of the ‘dietary-depressive’ subtype is needed in younger and in older samples. For instance, it is not clear from these results whether membership of the ‘dietary-depressive subtype’ at age 10 results from the experiences of earlier childhood obesity or earlier binge eating. Fifth, the examination of the contribution of the ‘dietary-depressive subtype’ to obesity was limited by not including other factors known to be associated with obesity in children (e.g., amount of television watched, amount of soda drank, amount and access to opportunities for physical activity, experiences of weight-related teasing, eating breakfast etc) (16) as potential predictors. Future studies will have to also examine the contribution of these factors.

Finally, the implications of the main results of this study are that prevention programs developed to prevent binge eating must specifically target children who have some depressive symptoms who are also engaged in dieting and who are on public assistance. With regards to obesity, the implications of this study are that prevention programs for obesity must particularly target minority girls with poor body image. Additionally further studies are needed to explore the mechanisms by which minority status and/or public assistance status predicts binge eating or obesity.