Prevalence estimates of binge eating, vomiting, laxative use, fasting, and thinness using complete case analysis are presented in Table 2. Using only adolescent report, estimates are comparable to what has been found in other population-based Western studies of eating disorder symptoms,or would be expected from such studies of full eating disorders, in youth^{22, 23}. Over 5% of adolescents reported binge eating, while reports of purging behaviors were slightly more rare; behaviors were more common at age 16 than at age 14. Relying only on parent report, prevalence estimates for each symptom were lower, with the largest discrepancies seen for purging behaviors (e.g., vomiting at age 16: 4.9% according to adolescents vs. 0.3% according to parents). The estimated prevalence for each symptom was highest when taking an “OR rule” approach, with estimates often greater than the sum of parent-only and adolescent-only estimates due to missing data patterns.

For study participants for whom we had both the adolescent and parent report, we estimated the concordance between the two informants with kappa coefficients, overall and stratified by sex. Kappa estimates were all below 0.3, indicating at most slight to fair agreement. Given the universally low concordance, we have included these kappa estimates as supplemental material only, along with two specific examples to illustrate the patterns of discordance (eTable 1-3).For vomiting and laxative use, adolescents tended to endorse the behavior more often than parents, and parents only rarely indicated the presence of the behavior when the adolescent denied it. In contrast, for binge eating, fasting, and thinness, parents and adolescents reported more similar magnitudes of prevalence, but were in disagreement about their presence.

We modeled prevalence of each symptom at each age by covariates and informant in Table 3. For binge eating, vomiting, laxative use, and fasting, these models indicate that prevalence estimates for each symptom are higher when reported by the adolescent relative to their parent; for thinness, the inverse was observed. An informant-sex product term was significant for vomiting, fasting, and binge eating measured at both ages, indicating the magnitude of any differences in prevalence of these symptoms between females and males depends on the source of information. An example of model-based prevalence estimates of each symptom by informant, age group, and sex are presented in eFigure 1.

In Tables 4, we present models predicting BMI at age 17.5 years, with sex, parity, maternal education, each eating disorder symptom, and informant as predictors. Models indicate that within an informant, symptom reports were predictive of BMI (e.g., using these models, we would expect the child of a parent reporting binge eating at 14 years would have a BMI 2.5 kg/m² higher at 17.5 years than the child of a parent reporting no binge eating at age 14). Generally, BMI at 17.5 years were nearly identical whether the parent reported a lack of symptom or the adolescent reported a lack of symptom. The symptom-informant product terms indicate parents’ report of binge eating was more predictive of BMI than the adolescents’ report of binge eating; this was less consistently seen for other symptoms. In eTables 4 and 5, we present similar findings for predicting absolute fat and lean mass at 17.5 years using parent and adolescent eating disorder symptom reports; we further provide example model-based estimates of expected BMI, fat mass, and lean mass by informant and symptom endorsement in eFigures 2 and 3. We ran similar models for predicting eating disorder treatment, but the rarity of this outcome limited interpretability (results not shown). Of note, even though treatment status was reported by the adolescent, the relationship between symptom reports and the adolescents’ reported treatment status was similar regardless of the informant reporting the symptom status.

Epidemiologic studies of eating disorders in youth have traditionally relied on only self-report, while the potential usefulness of parent report was unknown. In the present study, parents and adolescents generally were discordant when reporting on eating disorder symptoms, with purging behaviors generally reported more frequently by the adolescents, while discordance for binge eating, fasting, and thinness was perhaps due more to disagreement than possibly over- or under-reporting by one informant over another. Thus, prevalence estimates based adolescent only, parent only, or “OR rule” measures varied considerably. Our prevalence models highlight the importance of the source of information, particularly by sex: adolescent reports aligned with other epidemiologic studies finding an increased risk in females^22,25, while parent reports had a more muted sex difference. Finally, eating disorder symptom reports by either the parent or adolescent were generally predictive of subsequent body mass and composition measures, although parent report of binge eating may be considered more predictive.

The patterns of discordance between these reports may be unsurprising given clinical knowledge of how these symptoms manifest in eating disorder patients. Adolescents reported purging behaviors (vomiting and laxative use) more frequently than their parents, a pattern that aligns with the often hidden nature of bulimic behaviors: similar to other psychiatric constructs that often involve secrecy (e.g., deliberate self-harm, substance use), we expected some adolescents to report these symptoms on an anonymous questionnaire even though the behavior may be hidden from their parents. Indeed, when adolescents were asked to explain why they thought their parents would fail to report the adolescent used drugs or alcohol when the adolescent reported they did use one of these substances, most expected that the parents were simply unaware of the behavior⁵. Information on purging behaviors provided by parents, then, may aid investigators in interpreting frequency of, severity of, or treatment-seeking for the behaviors rather than affirming presence or absence. For binge eating and thinness, parents and adolescents may be offering different perspectives or interpretations of these constructs. In particular, parents and adolescents may simply disagree on what eating “objectively large” amounts of food looks like (part of the definition of binge eating), or use different anchors to assess bodyweight. Finally, fasting exhibited both styles of discordance: at age 14 years, prevalence of fasting as reported by parent and adolescent was not greatly dissimilar (yet still discordant), while adolescents endorsed fasting at 16 years at a much greater frequency than parents. Like purging behaviors, perhaps fasting is a behavior the adolescent would largely try to hide. However, at younger ages, parents may be more involved in adolescents’ eating habits (e.g., family meals) and thus the behavior is not as easily hidden. Further research may explore the importance of family meals for the early identification of disordered eating.

Our models of symptom prevalence indicate that measuring sex differences may be complicated by the source of information. Largely based on results from treatment-seeking samples, it has been argued that eating disorders are much more common in females than males, with gender ratios as high as 9:1²⁴. Prior population-based studies indicate that this ratio is perhaps not as extreme, with estimated gender ratios closer to 3:1^{22, 25}. However, the prior studies in youth rely on self-report. Indeed, when focusing on adolescent reports, our results align with these prior epidemiologic studies (e.g., binge eating as reported by the adolescent was 2-4 times more common in females than males). When comparing parent reports for boys and girls, the gender ratio appeared to be nearly 1:1. At minimum, these results underscore the importance of the source of information when assessing the prevalence and distribution of disorders in a population. Long-held beliefs that eating disorders are much more common in females than males are rooted in research focused on treatment-seeking samples or epidemiologic studies using only self-report, and thus the presentation and course of illness in males has been a neglected area of research. Indeed, several recent studies have suggested current measurements of eating disorders are not well suited to boys. For example, cognitive features may manifest differently due to different ideal body types (valuing muscularity over thinness), and males may not accurately or consistently report objective overeating and loss of control (i.e., components of binge eating)^26-28. As eating disorder assessments continue to evolve, incorporating parental reports could aid our understanding of these constructs as they present in males.

As stated above, this study does not address whether the information provided by the parent or adolescent is more or less valid across all possible studies or research questions. Nonetheless, we can describe their relative relevance in predicting later outcomes expected to be strongly correlated with bulimic behaviors and prior bodyweight²³. Indeed, both informants’ reporting of symptoms was predictive of future BMI and body composition measures. The magnitude of difference between the expected outcomes with binge eating compared to without was slightly larger when considering parent reports. One possible interpretation is parental reports better recognize the facets of the binge eating construct that are predictive of later BMI and body composition; alternatively, parents may only be recognizing more severe behaviors, and thus when they do recognize the behavior it is strongly predictive.

Some limitations warrant consideration. Because different questionnaires were given to the parents and adolescents, question wording for each symptom did not perfectly align. Different answer choices forced us to dichotomize behaviors as any vs. none; future research is needed to assess how frequency of behaviors may modify concordance. Parents were asked whether the behaviors occurred in the prior three months while adolescents were asked whether they occurred in the prior year; thus, it is possible that some (although likely not all) of the discordance when an adolescent endorsed a behavior that the parent did not is due to symptoms that occurred between three and twelve months ago. In particular, it is unlikely that the discordance is completely explained by this temporal discrepancy because parents more often reported thinness. Moreover, given the relationship between lifetime and period prevalences seen in prior studies, it is unlikely the temporal discrepancy would influence these results^{22, 25}. The strengths of this study likely outweigh these limitations. It is a large cohort with repeated measurements of many important eating disorder symptoms, as well as important body mass and composition measures. Our modeling approaches allowed us to present interpretable results while efficiently using all collected data and advanced techniques to address missingness^{12, 18}.

Eating disorders are of great public health concern, particularly during adolescence, because these disorders are prevalent and are risk factors for severe medical complications and suicide^{22, 25, 29, 30}. Thus, it is critical that epidemiologic studies of eating disorders obtain all relevant information so we can study prevention, intervention, and consequences of these disorders using valid measurements. Future studies of eating disorders may consider incorporating both parent and adolescent report, particularly if studying sex differences, bearing in mind that each informant may be providing different yet potentially valuable information.