Disordered eating includes behaviors such as binge eating or using laxatives, diet pills or vomiting for weight control, and is common in adolescents and young adults [1]. Disordered eating behaviors confer serious health risks [2] and are on the rise [3, 4]. Health-related quality of life (HRQL) has been widely measured to assess the degree of impairment in psychosocial and physical functioning associated with disease states, such as clinical eating disorders [ED, 5]. Given that approximately 30 million Americans are thought to experience an ED across their lifetime [6], cost-effectiveness analyses using HRQL instruments are needed to assess the relative value of prevention and treatment interventions for people affected by EDs and to guide policy and funding decisions [7].

Although many studies have implemented either generalized or ED-specific HRQL instruments with clinical samples [see: 5], very few studies have used a preference-based HRQL instrument. Estimating preference-based HRQL decrements associated with EDs is the first step in conducting cost-effectiveness evaluations, which can be used by policy makers to understand the value and appropriately fund interventions for these serious mental illnesses [8]. One preference-based HRQL measure that has been very widely used for cost-effectiveness analyses is the EuroQol five-dimension, five-level (EQ-5D-5L), a standardized instrument for describing and valuing health [9]. Its strength is that it is a comprehensive, validated instrument that captures public preferences for health states and is easily implemented through surveys of community-based samples. However, to date, the EQ-5D-5L has not been examined among individuals with EDs or those reporting disordered eating behaviors in a community sample.

The literature on decrements in HRQL associated with various diagnoses of EDs among women is now well developed [for a review see: 10]. However, a much smaller amount of work has focused on the effects of disordered eating behaviors—such as binge eating, dietary restriction, or use of diet pills, vomiting or laxatives for weight control—and only seven studies have reported on the relationship between disordered eating behaviors and HRQL in community samples including both men and women [11,12,13,14,15,16,17]. Consistent in this research are findings that include: scores on measures of HRQL appear lower in women than in men [11, 13,14,15,16], disordered eating behaviors are more frequent in women than men [11,12,13, 15,16,17], and the presence of disordered eating behaviors confer significant decrements in HRQL [12,13,14,15,16,17]. Three of these seven studies formally analyzed effect modification by gender on the relationship between disordered eating behaviors and HRQL [12, 15, 16], and these present a much more nuanced picture.

The first study [12] was conducted in Australia and implemented the Medical Outcomes Study Short Form-36 (SF-36) and the Eating Disorder Examination to assess whether gender modified the association between the presence versus absence of disordered eating behaviors and scores on the mental and physical dimensions of the SF-36. Objective binge eating (OBE) was found to be associated with significantly greater decrements on the mental health dimension for men than for women, but significantly greater decrements on the physical health dimension for women than men. This latter result is largely consistent with the second study [16] which was conducted in Hawai’i using the Eating Disorder Examination Questionnaire (EDE-Q) and SF-36 with college students. Here, OBE was significantly associated with poorer scores on the physical health dimension for women, but not for men. However, when a moderation analysis was conducted to take into account the base rate differences of HRQL across men and women who did not report disordered eating, there was no significant gender moderation effect [16]. Despite this, a mediation analysis found that the global EDE-Q mediated the relationship between gender and mental HRQL; women were found to have a significantly lower mental HRQL as their eating pathology increased, whereas this was not true for men [16]. The third study [15]—also from Australia—used the EDE-Q and the K6 as a measure of general psychological distress. In contrast to previous results, the study found no effect modification by gender on the relationship between OBE and K6 scores. Despite testing many disordered eating behaviors, just one significant interaction was found; dietary restriction was associated with higher levels of distress in women than in men.

The contribution of body mass index (BMI) to associations between disordered eating behaviors and HRQL has received little attention. Although all three of the studies described above controlled for participant weight [12, 15, 16], none examined the effect of including weight in their estimation models. Previous research has shown that BMI is differentially associated with the type and severity of eating pathology [18], but to what extent BMI might contribute to or confound associations between ED symptoms and HRQL remains unknown.

Despite the important findings of previous studies, the current literature presents important gaps in our knowledge. In light of these gaps, the aims of the current study were the following: (1) Examine associations between disordered eating behaviors, a lifetime history of ED diagnosis, and HRQL in a young adult community sample from the USA using a preference-based measure—the EQ-5D-5L; (2) Evaluate whether gender moderates the relationship between disordered eating behaviors or ED diagnosis and HRQL by examining effect modification by gender; (3) evaluate whether BMI might contribute to or confound the relationship between eating pathology and decrements in HRQL.

Methods

Study participants

Participants were from the US Growing Up Today Study (GUTS), a prospective cohort of children of women in the Nurses Health Study 2 (NHS2). The GUTS cohort was initiated in 1996 with 16,882 girls and boys aged 9–14 years (GUTS1) and expanded in 2004 with the addition of 10,923 children, aged 9–15 years (GUTS2). Questionnaires have been sent to all participants annually or biennially. The sample is predominantly white (93%) and has a restricted socioeconomic range as all participants’ mothers have 4-year nursing degrees. In 2013, the year in which the data for the present study were collected, GUTS participants were aged 18–31 years. The study protocol was approved by the institutional review board of Brigham and Women’s Hospital.

Data were available for n = 12,748 participants. Those who did not provide information on disordered eating behaviors or eating disorders diagnosis (n = 186), HRQL (n = 2118), weight (n = 997), or gender (n = 7) were excluded from analyses (3308 total exclusions); the final analytic sample therefore included 9440 participants.

Measures

Outcome: health-related quality of life

The validated and generic EQ-5D-5L measure was used to assess HRQL [19, 20]. The EQ-5D-5L comprises five dimensions (mobility, self-care, usual activities, pain/discomfort, and anxiety/depression) each scored on a five-level scale of severity (1. no problems, 2. slight problems, 3. moderate problems, 4. severe problems, and 5. extreme problems). Participants indicate their current perceived health state by selecting the most appropriate level in each dimension. This results in a 1-digit dimension number, and a 5-digit health state number describing the patient across all dimensions.

The EQ-5D defines 3125 potential health states. Although the previous 3-level version (EQ-5D-3L) [19, 20] has been valuated in the USA, at the time of this study, the 5-level version (EQ-5D-5L) value sets were not yet available for the USA, so we used crosswalk between 3L and 5L versions estimated by the EuroQol Group [21] to provide U.S-based population values for the EQ-5D-5L.

Primary predictors: disordered eating behaviors and diagnosis

Items assessing weight control behaviors were adapted from the Youth Risk Behavior Surveillance System (YRBSS) questionnaire [22]. To assess weight control behaviors, participants were asked, ‘‘During the past year, did you do any of the following to lose weight or keep from gaining weight?’’ with “go on a diet” (never, a couple times, several times, often, always on a diet), “use diet pills”, ‘‘make yourself throw up,’’ and ‘‘take laxatives’’ (diet pills, throw up, laxatives: never, less than a month, 1–3 times per month, once a week, more than once per week) as the methods listed. Respondents reporting any weight control behaviors were coded as “yes” to the behavior present, except for dieting where only the two most frequent response options (often, always) were coded as “yes” to capture the more severe behavior likely to represent eating pathology.

To assess OBE, participants were first asked about frequency of binge eating in the past year: ‘‘Sometimes people will go on an ‘eating binge,’ where they eat an amount of food that most people, like their friends, would consider to be very large, in a short period of time. During the past year, how often did you go on an eating binge?’’ Five frequency response options were then provided, ranging from ‘‘never’’ to ‘‘more than once a week.’’ Those who reported binge eating were then asked, ‘‘Did you feel out of control, like you couldn’t stop eating even if you wanted to stop?’’ with the response options ‘‘no’’ or ‘‘yes.’’ Those who reported binge eating at least monthly in the past year while feeling out of control were coded as “yes” to the behavior present. In addition, because previous research has focused on weekly binge eating [12, 15, 16], a second variable was created to assess higher frequency binge eating and compare results across the present study and earlier findings.

A composite “any disordered eating” binary variable was created with participants indicating any binge eating at least monthly, or any dieting (often, always), laxatives, diet pills, or vomiting coded as “yes” and those indicating none coded as “no.” In adolescents, self-report items assessing vomiting and laxative use for weight control have been found to have high sensitivity (0.93) and specificity (0.86) and those assessing binge eating have been found to have moderate sensitivity (0.53) and specificity (0.78) [23].

To assess lifetime history of diagnosis, a further question asked “Have you ever been told by a health care provider that you have any of the following illnesses?” with response options: “anorexia nervosa,” “bulimia nervosa,” “binge eating disorder,” or “other eating disorder.” A composite binary variable was created (ED diagnosis), with participants indicating any ED coded as “yes” and those with no diagnosis coded as “no.”

As an omnibus survey, the Growing Up Today Study measures a range of health, occupational, and education outcomes among young Americans. Given the significant time burden omnibus surveys can place on participants, the current study was limited to previously validated brief measures and was not able to implement broader measures such as body image instruments.

Covariates

Mother’s report of their child as either girl or boy at the time of enrollment was used for assigning participant gender; however, in 2011 when gender identity was measured via self-report [24], seven participants reported identifying as neither male nor female. Those participants were excluded from these analyses.

BMI was calculated from self-reported height and weight. For the current study, final adult height for a participant was taken to be the first non-missing height report at or after 19 years of age. For GUTS1 participants this was in 2007 and for GUTS2 participants this was in 2011. We used CDC growth charts with percentiles to categorize participants into weight status categories. Young adults self-reporting weight and height generally provide valid information [25, 26]. We analyzed BMI as categorical rather than continuous data for two reasons. First, the relationship between outcomes and BMI is nonlinear, hence presenting BMI categories provides a more easily interpreted and intuitive picture of the nonlinear relationship than if we had modeled a nonlinear curve. Second, BMI as categories has more clinical relevance because health professionals tend to weigh clients presenting with disordered eating or eating disorders, and use BMI categories diagnostically, especially because low body weight is a criterion for anorexia nervosa and because binge eating disorder is associated with high body weight.

Potential confounders included in estimation models were age (in years), partner status (married/living with partner versus never married/separated/divorced/widowed) and ethnicity (white versus all other race/ethnicities), as these have been found to be associated with HRQL [27,28,29] and severity and prevalence of disordered eating behaviors among young adults [30].

Statistical analyses

In analyses conducted in 2018, cross-sectional multivariable linear regression models were used to examine whether disordered eating predicted HRQL, after controlling for age, ethnicity, partner status, and gender in Model 1 with Model 2 adding BMI category to assess how the effect estimates for the ED variables were impacted compared to Model 1. Separate models were examined for each disordered eating variable and for the ED diagnosis variable. Analyses then followed a two-step approach recommended for analyzing health utility scores with bimodal distributions [31]. First, the health utility score was dichotomized (1 versus < 1) and examined as a binary outcome in models predicted by disordered eating behavior/ED diagnosis (referent = no behavior/diagnosis present) using generalized estimating equations (GEE) with log link to estimate risk ratios (RR) and 95% confidence intervals (CI). Second, the health utility score was examined as a continuous outcome in models including all participants and then again when restricted to those with health utility scores < 1; the linear association between disordered eating behavior or ED diagnosis and HRQL was estimated again using GEE and the robust sandwich estimator to account for sibling clustering in the cohort [32].

Possible effect modification by gender of the association between disordered eating behavior or ED diagnosis and health utility was tested using gender by primary predictor interaction terms. Effect estimates for the interaction term are shown only for models with a p < 0.10.

Additional multivariate analyses were conducted with the total sample to estimate the relative risk (RR and 95%CI) for any impairment (dimension score > 1 versus 1) on each of the five EQ-5D-5L dimensions (mobility, self-care, usual activities, pain/discomfort, anxiety/depression). However, these were conducted on only two variables (any disordered eating and eating disorder diagnosis) and using only one model for confounders (age, ethnicity, partner status, gender, and BMI category) because the number of participants reporting each disordered eating behavior within each impairment dimension did not allow sufficient power for further analyses.

Results

As shown in Table 1, across the whole sample, men reported significantly better HRQL than women on both the health utility score and on most of the five dimensions, except for self-care where scores were equal across men and women, and on mobility, where scores were different only when analyzing across the whole sample and not the subset who reported less than full health. Women were significantly more likely to report any disordered eating behavior (20% vs 10%) and ever receiving any ED diagnosis (4% vs 0.2%).

Table 1 Sample characteristics by gender in the US cohort of young adults aged 18–31 years (n = 9440)
Full size table

Table 2 presents the multivariable relative risk of less-than-full health, defined as an EQ-5D-5L health utility score < 1, across the full sample. Here, participants reporting disordered eating behaviors, and those reporting an ED diagnosis, had a significantly greater risk of less-than-full health than their counterparts who did not report disordered eating or diagnosis. The effect modification by gender analysis found that for diet pills there was a significance level of p = .06 for Model 1, such that men who reported diet pill use had a greater risk of reporting less-than-full health compared to their same-gender counterparts with no diet pill use than women who reported diet pill use compared to their same-gender counterparts with no diet pill use. The only significant effect modification by gender was found for ED diagnosis; men reporting having ever received a diagnosis had a greater risk of reporting less-than-full health as compared to their no-diagnosis same-gender counterparts than women who reported ever having received a diagnosis compared to their no-diagnosis same-gender counterparts. Model 2 results, which adjusted for the effects of BMI category, revealed small attenuation in the associations between the disordered eating variables and HRQL. In addition, BMI categories showed a U-shaped association with HRQL, where both the lowest and highest BMI categories experienced elevated risk of less-than-full health.

Table 2 Multivariable relative risk (RR) and 95% confidence interval (CI) of less-than–full health in the US cohort of young adults aged 18–35 years, reporting disordered eating behaviors or eating disorder diagnosis (n = 9440)
Full size table

When the significant effect modification by gender for ED diagnoses was further examined, it was found that within males, those with a lifetime history of an ED diagnosis had 1.85 times (95% CI 1.62, 2.11) the risk of less than full health, as estimated by Model 1 in Table 2, and estimates changed only slightly in Model 2 (RR 1.86 [95% CI 1.64, 2.12]). Within females, those with a lifetime history of an ED diagnosis had 1.34 times (95% CI 1.23, 1.41) the risk of less than full health, as estimated by Model 1 in Table 2, and estimates changed only slightly in Model 2 (RR 1.33 [95% CI 1.24, 1.42]).

Table 3 shows the outcome of multivariable linear associations in the full sample of participants. These estimated the linear relationship between disordered eating behaviors, or ED diagnosis, and EQ-5D-5L score. Similar to Table 3, Table 4 shows the outcome of multivariable linear associations but is restricted to the subset of participants who reported less than full health (EQ-5D-5L health utility score < 1). As shown in both Tables 3 and 4, participants reporting disordered eating behaviors or ED diagnosis reported significantly lower scores than their non-disordered counterparts. No significant effect modifications by gender were found. Again, the addition of BMI category as a covariate in Model 2 appeared to account for only small attenuations in effect estimates and BMI categories showed an inverted U-shaped association with HRQL.

Table 3 Multivariable linear associations between HRQL and disordered eating behaviors or eating disorder diagnosis in the US cohort of young adults aged 18–35 years (n = 9440)
Full size table
Table 4 Multivariable linear associations between HRQL and disordered eating behaviors or eating disorder diagnosis, in subsample with less-than-full health (n = 5263)
Full size table

Table 5 shows the multivariable relative risk of any impairment, versus no impairment, within the five dimensions of the EQ-5D-5L reported by the full sample. Here participants reporting any disordered eating behavior had a significantly greater risk of impairment on the usual activities and anxiety/depression dimensions than their non-disordered counterparts. No other dimensions showed significantly greater risk of impairment, though the proportions of participants reporting impairments across these dimensions were generally low. Having received an ED diagnosis was associated with a greater risk of impairment across all dimensions except for self-care. No significant effect modification by gender was found.

Table 5 Multivariable relative risk (RR) and 95% confidence interval (CI) of any impairment in the US cohort of young adults aged 18–31 years reporting disordered eating behaviors or eating disorder diagnosis (n = 9440)
Full size table

Discussion

This study was the first to examine the associations between disordered eating behaviors in the last year, ever having received an ED diagnosis, and HRQL in a young adult community sample using a preference-based measure—the EQ-5D-5L. It found that the presence of disordered eating or ED diagnosis were associated with strong and significant decrements to HRQL as compared to peers with no disordered eating behaviors or ED diagnosis.

Individuals reporting binge eating or ED diagnosis reported the highest relative risk of less-than-full health. In addition, among those with less-than-full health, weekly binge eating conferred a HRQL decrement of − .03, and ED diagnosis − .02, decrements of a magnitude that have been previously found to be clinically meaningful [33, 34]. The relative risk of less-than-full health was similar for weekly binge eating and monthly binge eating, consistent with other research that suggests lower frequency behavior still appreciably impacts on quality of life and warrants diagnosis and treatment in both men and women [35].

An aim of this research was to evaluate whether gender moderates the relationship between disordered eating behaviors and HRQL. The proportion of women reporting any level of impairment was significantly higher than in men for the health utility index score and dimensions of usual activities, pain/discomfort, and anxiety/depression; however, in the remaining dimensions (mobility, self-care), few gender differences were observed. Examination for possible effect modification by gender conducted on the full sample (Table 2) revealed that ED diagnosis (significant) and diet-pill use (marginally significant) were associated with a greater relative risk of less-than-full health in men. The finding relating to diagnosis is consistent with past research suggesting that men typically experience a longer treatment delay and may be subject to more stigmatizing beliefs about EDs, creating barriers to care that entrench pathology before diagnosis is received [12, 36]. The finding that diet-pill use seems to confer greater decrements among men than women is novel. Because the number of men reporting diet-pill use was small (n = 70, 2.3%), data were underpowered to assess gender differences across all five dimensions, which made it more difficult to identify subtle trends emerging for men (i.e., whether diet-pill use had a stronger association with anxiety/depression perhaps due to stigmatizing beliefs that diet-pill use is normative for women but not for men). With dietary supplement and diet-pill use on the rise [37], this is an area that requires further research.

The final aim of this research was to evaluate the contribution of BMI to associations between disordered eating or ED diagnosis and decrements in HRQL. In multivariable models, both the lowest and highest BMI categories experienced pronounced decrements in HRQL. Interestingly though, the addition of BMI categories to models (see Model 2 in Tables 2 and 3) resulted in only small or in some cases no attenuation of effect estimates for the disordered eating and the ED diagnosis primary predictors. This finding suggests that perhaps BMI category mediates associations between disordered eating or ED diagnosis and HRQL, but the impact is minor. Alternatively, the small attenuations observed may indicate that BMI category, at the lowest and highest ends of the spectrum, are reflecting residual confounding by disordered eating severity. It may be that individuals with very low or higher body weight have more severe eating pathology driving greater decrements in HRQL. Our primary measures of disordered eating and diagnosis did not well capture severity. Either way, this finding reiterates the need for ED screening across the weight spectrum and a shift away from the erroneous belief that individuals need to be underweight to experience clinically significant decrements in HRQL.

A strength of this study was its large community cohort and implementation of a well validated, preference-based measure of HRQL. Despite the large sample, some disordered eating behaviors were not commonly reported, especially among men, and analysis across the individual EQ-5D-5L dimensions could not be conducted for each behavior. Further, symptoms were dichotomized (presence vs absence) to allow for a well-powered sample, though further investigation of how severity impacts on dimensions and possible effect modification by gender would require larger sampling of those reporting behaviors not commonly found in this study. Future research should include targeted recruitment of men with ED and examine how ED severity across the weight spectrum affects HRQL; this would improve understanding of the mechanisms of impact that might underlie the reported decrements in quality of life in individuals with disordered eating. Limitations of the current study include that relevant data were available only for 9440 participants (34%) of the original 27,805 total participant sample. In a sensitivity analysis (data not shown) comparing those excluded in the analyses to those included, we found those excluded reported slightly higher prevalence of binge eating weekly, dieting, vomiting for weight control, and any disordered eating, yet a slightly lower prevalence of eating disorder diagnosis; therefore, prevalence of these indicators in our study are likely underestimates. Further limitations include the cross-sectional nature of the data analyzed, the self-report of eating disorder diagnosis, and the limited generalizability of results to groups other than predominantly white, and young adult cisgender Americans from middle-income families. In addition, all participants’ mothers had nursing degrees. Despite these limitations, the current study provides valuable findings on HRQL that may be used in future health economic analyses and screening initiatives.

Conclusions

This study provides new data on preference-based HRQL that can be used to inform future cost-effectiveness analyses of ED interventions. Findings suggest that disordered eating behaviors and ED diagnoses are associated with significant reductions in HRQL during young adulthood. The presence of binge eating, and ED diagnosis, appear particularly detrimental to health, especially for males. High or very low body weight in individuals experiencing eating pathology leads to further decrements in HRQL, but the mechanism of this effect remains unclear. Further longitudinal research is required to understand the role of eating disorder severity, body weight, gender, and decrements in HRQL.