Range of environmental exposures/constructs examined

Table 1 presents common environmental constructs and corresponding examples that emerged from the articles reviewed organized by level of influence (interpersonal, behavioural settings, sectors of influence and sociocultural context). As several studies examined more than one type of environmental exposure, Fig. 2 presents the frequency and percentage of major categories of environmental exposures that were examined in the literature. Studies that analysed multiple environmental constructs were counted multiple times, with ninety-three studies and a total of 297 constructs counted. Family (51·5 %), peer (22·9 %) and media influences (9·4 %) represented the most commonly studied environmental exposures. Family- and peer-related influences had the greatest number and range regarding type of environmental construct examined in association with DWCB. Behavioural settings, including the school and home environment, were much less frequently examined as an exposure (5·4 %). Of the many sectors that may influence weight-related behaviours⁽ Reference Swinburn, Sacks and Hall ^{20 )}, only media was examined in association with DWCB. General sociocultural influences represented 4·4 % of environmental exposures, and no studies were conducted on the built environment or policy influences (i.e. legislation regarding sale of diet pills and laxatives to minors) as exposures related to DWCB.

Table 1

Common environmental constructs examined by level of influence

DWCB, disordered weight-control behaviours; SES, socio-economic status; TV, television.

Fig. 2

Frequency of environmental constructs* (, family, n 153; , peers, n 68; , other interpersonal influences, n 9; , behavioural settings, n 16; , media, n 28; , sociocultural influences, n 13; , environmental influences (from twin studies), n 10; , built environment (BM), n 0; , policy, n 0) examined as exposure variables to disordered weight-control behaviours among youth. *Studies that examined multiple environmental constructs were counted multiple times, with ninety-three studies and a total of 297 constructs counted (grouped by the categories shown); n refers to the number of times a particular construct was examined, not the number of studies

Theoretical basis

Most studies (69 %; n 64) did not explicitly identify a theoretical framework guiding the research. Over half of longitudinal studies (52 %; n 12) identified a theory or conceptual framework compared with 24 % (n 17) of cross-sectional studies. The following theories or constructs were cited as explicitly guiding the research: Social Cognitive Theory, gene–environment theory, socialization theory, achievement goal theory, social support, self-discrepancy theory, cumulative risk model, ecodevelopmental perspective, biosocial model, biopsychosocial model, the Catalyst Theory, feminist theory of conflicting gender roles, and thin-ideal internalization. Four studies developed and tested a specific conceptual model of the development of DWCB⁽ Reference Franko, Thompson and Affenito ^{48 –} Reference Neumark-Sztainer, Hannan and Story ^{51 )}.

Study populations

About half of the studies (n 45) were conducted with all-female samples, with no studies conducted on all-male samples. Less than half of the studies (46 %; n 43) included diverse or representative populations with respect to race/ethnicity and socio-economic status (SES). Most studies (61 %; n 57) had sample sizes of youth under 1000; twenty-six studies had 1000–9999 participants and ten studies had 10 000 or more participants. Two studies specifically recruited youth who participated in an aesthetic physical activity (i.e. gymnastics or ballet)⁽ Reference Thomas, Keel and Heatherton ^{52 ,} Reference de Bruin AP Bakker and Oudejans ^{53 )}. Five studies examined genetic and environmental influences on DWCB using twin samples⁽ Reference Klump, McGue and Iacono ^{54 –} Reference Ferguson CJ, Winegard and Winegard ^{58 )}, and nine studies included parent/caregivers of adolescents in the study⁽ Reference Keel, Heatherton and Harnden ^{13 ,} Reference Fulkerson, McGuire and Neumark-Sztainer ^{15 ,} Reference Field, Javaras and Aneja ^{37 ,} Reference Maharaj, Rodin and Olmsted ^{59 –} Reference Keery, Eisenberg and Boutelle ^{64 )} to obtain multiple methods of data collection on measures of interest.

Study measures

The majority of environmental exposures and DWCB were assessed by various self-report surveys completed by youth with diverse measures used to assess the various environmental exposures of interest (see Supplemental Tables 2 and 3, online supplementary material). Commonly validated measures for DWCB assessment included scales or subscales such as the Eating Attitudes Test (EAT-26, EAT-40)⁽ Reference Lee and Lee ^{65 –} Reference Espinoza, Penelo and Raich ^{68 )}, the Bulimia Test and Bulimia Test Revised (BUILT, BUILT-R)⁽ Reference de Bruin AP Bakker and Oudejans ^{53 ,} Reference Pike ^{69 ,} Reference Stice ^{70 )} and various subscales of the Eating Disorders Inventory (EDI)⁽ Reference Baker, Maes and Lissner ^{12 ,} Reference Wertheim, Koerner and Paxton ^{50 ,} Reference Klump, McGue and Iacono ^{54 ,} Reference Klump, McGue and Iacono ^{55 ,} Reference Borjorquez and Unikel ^{71 –} Reference Vincent and McCabe ^{82 )}, as well as validated items on DWCB used in study questionnaires, such as the Project Eating Among Teens (Project EAT) survey⁽ Reference Fulkerson, McGuire and Neumark-Sztainer ^{15 ,} Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Neumark-Sztainer, Hannan and Story ^{51 ,} Reference Keery, Eisenberg and Boutelle ^{64 ,} Reference Haines and Neumark-Sztainer ^{83 –} Reference Utter, Neumark-Sztainer and Wall ^{91 )}, and items adapted from the Youth Risk Behavior Surveillance (YRBS) survey⁽ Reference Field, Camargo and Taylor ^{36 ,} Reference Field, Javaras and Aneja ^{37 ,} Reference Ackard and Neumark-Sztainer ^{92 –} Reference Field, Camargo and Taylor ^{94 )}. A number of studies (n 19) used items or surveys developed specifically for the study to assess DWCB.

Study design

The majority of studies (75 %; n 70) used a cross-sectional design. Figure 3 presents the length of follow-up and MQRS scores (mean = 10·8, range 5·7–20·0) of the twenty-three longitudinal studies. Average length of follow-up time was 3·6 years, with eleven studies having follow-up periods of 8 months to 2 years⁽ Reference Field, Camargo and Taylor ^{36 ,} Reference Wertheim, Koerner and Paxton ^{50 ,} Reference Shomaker and Furman ^{67 ,} Reference Espinoza, Penelo and Raich ^{68 ,} Reference Stice ^{70 ,} Reference McCabe and Ricciardelli ^{77 ,} Reference Field, Camargo and Taylor ^{94 –} Reference Haines, Gillman and Rifas-Shiman ^{98 )} and the remaining twelve studies with follow-up periods ranging from 4 to 10 years⁽ Reference Eisenberg and Neumark-Sztainer ^{17 ,} Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Field, Javaras and Aneja ^{37 ,} Reference Franko, Thompson and Affenito ^{48 ,} Reference Klump, Burt and McGue ^{56 ,} Reference Spanos, Klump and Burt ^{57 ,} Reference Ferguson CJ, Winegard and Winegard ^{58 ,} Reference Haines and Neumark-Sztainer ^{83 ,} Reference Neumark-Sztainer, Eisenberg and Fulkerson ^{88 ,} Reference Neumark-Sztainer, Wall and Story ^{89 ,} Reference van den Berg, Neumark-Sztainer and Hannan ^{99 ,} Reference Ferreiro, Seoane and Senra ^{100 )}. Ten longitudinal studies had attrition rates under 20 %.

Fig. 3

Length of study () and methodological rigour (Methodological Quality Rating Scale (MQRS) score, ) of the longitudinal studies reviewed (n 23)

Overall results across studies

The wide range of definitions, operationalizations, measurement and analysis of environmental exposures in association with DWCB across studies limited our ability to standardize comparisons of studies’ effect size estimates; thus, we first describe general patterns in associations between environmental exposures and DWCB across all ninety-three studies, followed by a methodological critique of the longitudinal studies and synthesis of their findings.

Overall, the following family, peer and media influences were found to be positively associated with DWCB: negative weight-related comments and/or teasing from family members and peers⁽ Reference Keel, Heatherton and Harnden ^{13 ,} Reference Haines, Kleinman and Rifas-Shiman ^{34 ,} Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Field, Javaras and Aneja ^{37 ,} Reference Shomaker and Furman ^{67 ,} Reference Rodgers, Faure and Chabrol ^{79 ,} Reference Haines and Neumark-Sztainer ^{83 ,} Reference Neumark-Sztainer, Bauer and Friend ^{90 ,} Reference Tremblay and Lariviere ^{101 )}; parental and peer modelling and social reinforcement of dieting behaviours⁽ Reference Keel, Heatherton and Harnden ^{13 ,} Reference Shomaker and Furman ^{67 ,} Reference Stice ^{70 ,} Reference Stice and Agras ^{95 ,} Reference Eisenberg, Wall and Shim ^{102 )}; peer acceptance concerns⁽ Reference French, Story and Neumark-Sztainer ^{103 ,} Reference Story, French and Neumark-Sztainer ^{104 )}; parent and peer attitudes on dieting and weight concerns⁽ Reference Maharaj, Rodin and Olmsted ^{60 ,} Reference Levine, Smolak and Hayden ^{105 ,} Reference Paxton, Schutz and Wertheim ^{106 )}; poor parental–child communication and/or relationships⁽ Reference Maharaj, Rodin and Olmsted ^{59 ,} Reference Maharaj, Rodin and Olmsted ^{60 ,} Reference Hautala, Junnila and Helenius ^{107 )}; high parental supervision and monitoring⁽ Reference Fonseca, Ireland and Resnick ^{108 )}; family conflict⁽ Reference Maharaj, Rodin and Olmsted ^{59 ,} Reference Lee and Lee ^{65 )}; lack of trust in parents⁽ Reference Maharaj, Rodin and Olmsted ^{59 )}; fewer family meals⁽ Reference Croll, Neumark-Sztainer and Story ^{1 ,} Reference Fulkerson, Story and Mellin ^{33 ,} Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Neumark-Sztainer, Wall and Story ^{87 ,} Reference Neumark-Sztainer, Eisenberg and Fulkerson ^{88 ,} Reference Neumark-Sztainer, Wall and Story ^{89 ,} Reference Hautala, Junnila and Helenius ^{107 )}; high use of media (particularly content that portrays thin ideals)⁽ Reference Grigg, Bowman and Redman ^{14 ,} Reference Neumark-Sztainer, Wall and Haines ^{35 –} Reference McCabe, Ricciardelli and Mellor ^{38 ,} Reference Harrison ^{74 ,} Reference Harrison ^{75 ,} Reference Peterson, Paulson and Williams ^{78 ,} Reference Haines and Neumark-Sztainer ^{83 ,} Reference Neumark-Sztainer, Wall and Story ^{89 ,} Reference Field, Camargo and Taylor ^{94 ,} Reference Harrison and Hefner ^{97 ,} Reference Levine, Smolak and Hayden ^{105 )}; perceived pressure to be thin from various sources⁽ Reference Grigg, Bowman and Redman ^{14 ,} Reference Shomaker and Furman ^{67 ,} Reference Stice ^{70 ,} Reference Peterson, Paulson and Williams ^{78 ,} Reference Stice and Agras ^{95 ,} Reference Tremblay and Lariviere ^{101 ,} Reference Paxton, Schutz and Wertheim ^{106 ,} Reference Goodwin, Haycraft and Meyer ^{109 )}; thin-ideal internalization⁽ Reference Stice ^{70 ,} Reference Harrison ^{74 ,} Reference Harrison ^{75 ,} Reference Stice and Agras ^{95 )}; and traumatic experiences, such as history of physical and/or sexual abuse or being bullied by peers⁽ Reference Neumark-Sztainer, Story and Toporoff ^{49 ,} Reference Ackard and Neumark-Sztainer ^{92 ,} Reference French, Story and Neumark-Sztainer ^{103 ,} Reference Story, French and Neumark-Sztainer ^{104 ,} Reference Ackard, Neumark-Sztainer and Hannan ^{110 –} Reference Kaltiala-Heino, Rissanen and Rimpela ^{117 )}. One study found that negative comments from parents regarding body shape, eating, importance of physical appearance, or comparisons of shape and weight of other people explained a larger amount of the variance in girls’ disordered eating compared with boys⁽ Reference Rodgers, Faure and Chabrol ^{79 )}.

Environmental exposures that were protective against DWCB included: family connectedness and support⁽ Reference Croll, Neumark-Sztainer and Story ^{1 ,} Reference Neumark-Sztainer, Wall and Story ^{89 ,} Reference Fonseca, Ireland and Resnick ^{108 ,} Reference Perkins and Luster ^{116 )}; positive family communication⁽ Reference Fulkerson, Strauss and Neumark-Sztainer ^{84 ,} Reference Fonseca, Ireland and Resnick ^{108 ,} Reference Perkins and Luster ^{116 )}; frequent family meals or dinners⁽ Reference Fulkerson, Story and Mellin ^{33 ,} Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Franko, Thompson and Affenito ^{48 ,} Reference Neumark-Sztainer, Hannan and Story ^{51 ,} Reference Fulkerson, Strauss and Neumark-Sztainer ^{84 ,} Reference Neumark-Sztainer, Wall and Story ^{87 ,} Reference Neumark-Sztainer, Eisenberg and Fulkerson ^{88 )}; positive family meal environment (i.e. high priority for family meals, a positive atmosphere at family meals and a structured family meal environment)⁽ Reference Fulkerson, Strauss and Neumark-Sztainer ^{84 ,} Reference Neumark-Sztainer, Wall and Story ^{87 )}; and connectedness to adults (non-family members from school or church)⁽ Reference Story, French and Neumark-Sztainer ^{104 ,} Reference Perkins and Luster ^{116 )}. Family connectedness and positive family communication were strong protective factors for girls, and high perceived parental expectations were a strong protective factor for boys⁽ Reference Fonseca, Ireland and Resnick ^{108 )}. One study of a large co-gender sample⁽ Reference Field, Camargo and Taylor ^{94 )} found that parental, peer and media influences independently predicted constant dieting, with no one factor emerging as the primary source of influence.

In terms of behavioural settings, school-wide prevalence of trying to lose weight was associated with unhealthy weight-control behaviours among healthy-weight females and marginally associated among overweight females, adjusting for friends’ dieting behaviours⁽ Reference Eisenberg and Neumark-Sztainer ^{17 )}. Studies that examined the broader sociocultural environment or overall sociocultural influences found that perceived sociocultural or societal pressures related to appearance⁽ Reference Griffiths and McCabe ^{72 ,} Reference White and Halliwell ^{118 )}, stronger identification with the Anglo culture among Latino youth living in the USA⁽ Reference Ayala, Mickens and Galindo ^{119 )} and higher agreement with socially sanctioned standards of beauty and thinness⁽ Reference Ayala, Mickens and Galindo ^{119 )} were associated with DWCB. One study that explored percentage of students eligible for free or reduced-priced school lunch and percentage of residents living at or below poverty in the student's neighbourhood found that the latter was positively associated with DWCB among boys⁽ Reference Austin, Spadano-Gasbarro and Greaney ^{9 )}. School SES was not found to be associated with DWCB among adolescent females⁽ Reference Kaluski, Natamba and Goldsmith ^{120 )}. No studies that examined the built environment, neighbourhood/community influences, industries such as those promoting fashion, weight loss, diet pills, laxatives or cosmetic surgery, or policy influences on DWCB were found.

Gene–environment studies using twin samples examined shared and non-shared environmental influences on DWCB. According to Klump et al.⁽ Reference Klump, McGue and Iacono ^{54 –} Reference Klump, Burt and McGue ^{56 )}, non-shared environmental influences are those that are ‘unique to individuals reared in the same family and contribute to differences among family members’, whereas shared environmental influences are ‘those that are common to individuals reared in the same family and that contribute to familial similarity’. However, most of these studies did not specify or measure shared and non-shared environment factors; shared environmental influence was inferred if both monozygotic and dizygotic twin correlations were similar in magnitude and statistically significant, whereas non-shared environmental influence was suggested if these correlations were small. Thus, the specific type of environmental influence that contributed significantly to DWCB was often unclear from these studies.

Longitudinal studies’ findings

Finally, we examined longitudinal studies’ findings within the context of their MQRS scores, which were heavily correlated with study length of follow-up (see Fig. 3). Similar to cross-sectional studies, parental, peer and media influences emerged as the most frequent exposures examined in relation to DWCB (see Supplemental Table 3, online supplementary material). Studies varied in their presentation of outcome effects, including odds ratios, mean scores on various DWCB instruments, percentage of participants who engaged in DWCB at various time points, correlations and percentage of variance explained. The following findings across studies scoring in the top 50th percentile (MQRS ≥ 12) were significant in predicting DWCB for males and females: friend/peer dieting behaviours as a risk factor (1·4–3·7 times increase in the percentage participating in DWCB⁽ Reference Eisenberg and Neumark-Sztainer ^{17 )}; OR = 1·2–1·6)⁽ Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Neumark-Sztainer, Wall and Story ^{89 )}; weight-related teasing as a risk factor (OR = 1·4–1·8)⁽ Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Haines and Neumark-Sztainer ^{83 )}; exposure to weight-loss magazines (OR = 1·4–1·6)⁽ Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Neumark-Sztainer, Wall and Story ^{89 )}; and family connectedness as protective factors (OR = 0·8–0·9)⁽ Reference Neumark-Sztainer, Wall and Story ^{89 )}. Several gender differences emerged in studies’ findings: parental weight concerns and behaviours were a risk factor for females (OR = 1·2–1·3)⁽ Reference Neumark-Sztainer, Wall and Haines ^{35 )}; frequent family meals were protective against DWCB among females (OR = 0·7–0·9) but increased risk for skipping meals and eating very little food among males (OR = 1·8)⁽ Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Neumark-Sztainer, Eisenberg and Fulkerson ^{88 )}; positive family atmosphere was protective for females (OR = 0·6–0·8)⁽ Reference Neumark-Sztainer, Wall and Haines ^{35 ,} Reference Neumark-Sztainer, Wall and Story ^{89 )}; maternal history of an eating disorder (OR = 2·8) and trying to look like females in the media were risk factors for females (OR = 1·5)⁽ Reference Field, Javaras and Aneja ^{37 )}; and importance of weight to peers (OR = 3·4)⁽ Reference Field, Javaras and Aneja ^{37 )}, friend connectedness (OR = 1·7)⁽ Reference Neumark-Sztainer, Wall and Haines ^{35 )} and low social support from family and friends (β = 0·2) were risk factors among males⁽ Reference Ferreiro, Seoane and Senra ^{100 )}. A dose–response association was observed between frequency of reading diet/weight-loss magazine articles and DWCB among females (OR = 2·0–3·2 for those who often read these types of articles v. those who never read these articles)⁽ Reference van den Berg, Neumark-Sztainer and Hannan ^{99 )}.

Two longitudinal studies conducted with twins explicitly measured specific types of environmental influence (parent–child conflict⁽ Reference Klump, Burt and McGue ^{56 )}; maternal warmth, school problems and media use⁽ Reference Ferguson CJ, Winegard and Winegard ^{58 )}) in the gene–environment analysis. Results indicated that differences in parent–child conflict between identical adolescent twins were a consequence, rather than a predictor, of disordered eating (r _s = 0·1–0·3)⁽ Reference Spanos, Klump and Burt ^{57 )} and that media use was associated with reduced DWCB (β = −0·3) among dizygotic twins⁽ Reference Ferguson CJ, Winegard and Winegard ^{58 )}. Franko and colleagues’ 10-year longitudinal study, consisting of annual assessments of family meal frequency, family cohesion, extreme weight-control behaviours and bulimia among 2379 white and black females (age 9·5 years at baseline), yielded the highest MQRS score (20). This particular study demonstrated that the protective association between family meal frequency during years 1 and 3 and scores on a bulimia scale at year 10 (score = 2·1 for never/almost have family meals v. score = 1·4 always/usually have family meals) was mediated by greater family cohesion (β = 0·05). Guided by a clear conceptual model, Franko et al.'s longitudinal study provided the strongest evidence for family cohesion, promoted through eating together as a family, as protective against DWCB 10 years later among female youth⁽ Reference Franko, Thompson and Affenito ^{48 )}.