Associations between severe food insecurity and disordered eating behaviors from adolescence to young adulthood: Findings from a 10-year longitudinal study

Abstract

Emerging evidence suggests a cross-sectional association between food insecurity (FI) and disordered eating among adults, while evidence among adolescents is limited. Longitudinal research is needed to elucidate the temporality of this relationship and clarify whether the association differs by age. Three waves of prospective data came from 1,813 participants in the Project EAT (Eating and Activity in Teens and Young Adults) cohort study. Data were collected at five-year intervals, with the baseline survey in 1998–1999 (EAT-I; M_age=14.9 years) and follow-up surveys in 2003–2004 (EAT-II; M_age=19.5 years) and 2008–2009 (EAT-III; M_age=24.9 years). Severe FI was assessed as any past-year hunger with one item from the U.S. Household Food Security Survey Module, and a range of disordered eating behaviors were self-reported. Associations adjusted for sociodemographic characteristics were examined with generalized estimating equations. Effect modification by age was also tested. Cross-sectionally, severe FI was significantly associated with greater prevalence of all disordered eating behaviors examined, with the strongest associations observed for extreme weight-control behaviors (prevalence ratio [PR]=1.49, 95% confidence interval [CI]: 1.13–1.95) and binge eating (PR=1.49, 95% CI: 1.04–2.12). Longitudinally, severe FI significantly predicted 1.41 (95% CI: 1.05–1.90) times greater prevalence of binge eating five years later after accounting for prior binge eating. Effect modification by age indicated a stronger cross-sectional association between severe FI and unhealthy weight-control behaviors among younger participants. Results support a cross-sectional link between severe FI and disordered eating and provide longitudinal evidence suggesting severe FI is a risk factor for binge eating.

Introduction

Food insecurity (FI)—defined as inconsistent or lack of access to adequate food for active, healthy living¹—has affected over one in ten U.S. households in recent years.^2,3 Food-insecure households may experience fluctuations in food availability throughout the month, such that periods of relative food scarcity may alternate with periods of relative food abundance (e.g., after receiving a paycheck or food assistance benefits). Fluctuations in food availability may result in corresponding fluctuations between restricted food intake and opportunities to increase food intake.⁴ Importantly, restriction of food intake—regardless of whether the restriction is voluntary or involuntary—has been shown to lead to preoccupation with food, as well as a propensity toward binge eating once imposed restrictions are eliminated.⁵ In addition, limited access to nutritious food has been reported as a barrier to weight management among individuals experiencing FI,⁶ which may lead food-insecure individuals with a desire to lose or maintain their weight to turn to alternative, and potentially harmful, weight-control strategies (e.g., fasting, self-induced vomiting). FI may therefore serve as a risk factor for a range of disordered eating behaviors.

Disordered eating behaviors represent a significant public health concern, considering that they are associated with increases in depressive symptoms over time,^7–9 future suicidality,^10,11 and increased risk for the development of full-threshold eating disorders,^12,13 as well as excess weight gain over time,^9,14,15 which may lead to metabolic consequences.¹⁶ Thus, if FI serves as a risk factor for disordered eating, disordered eating could be an important target alongside FI for multilevel interventions aiming to mitigate FI and its sequelae.

There is growing evidence that FI may be associated with disordered eating, based predominantly on cross-sectional studies among adults, with the strongest associations observed for severe FI (i.e., involving reduced food intake/hunger).^17–20 However, longitudinal studies are needed to elucidate the temporality of the relationship between FI and disordered eating behaviors and guide the development of intervention efforts. Additionally, while substantially fewer studies have examined the association between FI and disordered eating among adolescents as compared to adults, a less consistent pattern has been observed among adolescents,¹⁷ raising the possibility that FI may be more strongly associated with disordered eating in adulthood than in adolescence. Such a possibility is plausible, and young adulthood may be a particularly high-risk developmental stage, as the transition into adulthood is a critical period involving not only increased independence and responsibility,²¹ but also the loss of access to some of the resources devoted to young people experiencing FI or at risk for FI (e.g., school meal programs).²² Therefore, the present study aimed to examine cross-sectional and longitudinal associations between FI and disordered eating behaviors in a population-based cohort followed from adolescence into young adulthood, as well as to examine potential differences in these associations by age. This study focused on severe FI, the form of FI found to be most robustly associated with disordered eating in prior cross-sectional research.

Methods

Study Design and Population

Project EAT (Eating and Activity in Teens and Young Adults) is a population-based, longitudinal investigation of eating- and weight-related factors among young people. At baseline in 1998–1999 (EAT-I), classroom surveys were completed by 4,746 students ages 11–18 years from 31 primarily urban (27 inner-city and 4 inner-ring suburban) public middle schools and high schools in the Minneapolis-St. Paul metropolitan area.^23,24 EAT-I was originally designed as a cross-sectional study; however, given growing interest in the eating and weight-related health of young people, follow-up data were collected at five-year intervals with participants who provided sufficient contact information (n=3,672). Follow-up assessments were conducted in 2003–2004 (EAT-II; ages 16–23 years), 2008–2009 (EAT-III; ages 21–31 years), and 2015–2016 (EAT-IV).^25–27 Data from the fourth wave were not used in the present study because FI was not assessed at EAT-III, precluding examination of the longitudinal associations of interest from EAT-III to EAT-IV. The University of Minnesota’s Institutional Review Board Human Subjects Committee approved all protocols, and informed consent was obtained from all study participants.

From the original EAT-I sample, 2,516 responded at EAT-II (68.4% of those with sufficient contact information) and 2,287 responded at EAT-III (66.4% of those with sufficient contact information). Because attrition from the original EAT-I sample did not occur at random (e.g., continued participation at EAT-III was more likely among participants who were female, white, and of higher socioeconomic status), inverse probability weighting (IPW)²⁸ was used to account for missing data. IPW is the recommended method for handling missing data in longitudinal studies, where individuals who do not respond to surveys at various assessment time points have missing values on all variables at those time points.²⁹ IPW minimizes potential response bias due to missing data and allows for extrapolation back to the original EAT-I school-based sample. Weights for IPW were derived as the inverse of the estimated probability that an individual responded to both the EAT-II and EAT-III surveys based on several EAT-I covariates, including demographics, weight status, parental living situation, and grades in school. The weighting method resulted in estimates representative of the demographic make-up of the original, school-based EAT-I sample.

Of the 1,902 participants who responded across all three waves, those with missing data on time-invariant demographic characteristics (n=44) were excluded. Participants who did not provide data on time-varying variables of interest at two or more consecutive survey waves were also excluded (an additional n=45). Thus, a total of 89 participants (4.7% of those who participated across all three survey waves) were excluded from the present analyses due to missing variable-level data. The final analytic sample for the present study included 1,813 participants (see Table 1 for sociodemographic characteristics).

Table 1.

Sample Characteristics (N = 1,813)

	Time-Invariant Characteristics
Sex, % (n)
Male	42.9 (778)
Female	57.1 (1,035)
Ethnicity/race, % (n)
Non-Hispanic White	50.7 (1,247)
Non-Hispanic Black/African American	17.8 (155)
Hispanic/Latinx	5.7 (64)
Asian/Asian American	17.9 (259)
Mixed/other	7.9 (88)
Socioeconomic status, % (n)
Low	17.4 (197)
Low-middle	18.6 (284)
Middle	26.6 (458)
Upper-middle	23.6 (541)
Upper	13.9 (333)

	Time-Variant Characteristics, by Survey Wave
	EAT-I	EAT-II	EAT-III
Age in years, M (SD)	14.9 (1.6)	19.5 (1.6)	24.9 (1.6)
Severe food insecurity, % (n)	8.9 (118)	20.9 (321)	not assessed
Unhealthy weight-control behaviors, % (n)	45.7 (756)	48.9 (850)	43.3 (745)
Extreme weight-control behaviors, % (n)	7.9 (131)	15.2 (270)	14.6 (249)
Chronic dieting, % (n)	13.1 (207)	13.8 (227)	8.0 (126)
Overeating, % (n)	13.1 (208)	10.3 (194)	18.2 (338)
Binge eating, % (n)	8.6 (128)	6.9 (129)	12.0 (221)

Note. M = mean; SD = standard deviation. All statistics except n’s, which represent observed counts, are weighted to account for attrition over time and allow for extrapolation to the original population-based sample.

Measures

The Project EAT survey assessed a broad range of weight-related variables at each wave. Test-retest reliability of survey items was examined in a sample of 161 adolescents at EAT-I and a separate sample of 66 young adults at EAT-III. We report the test-retest reliability of survey items at EAT-III or at EAT-I if the item was only assessed at the earlier wave.

Severe FI.

Past-year severe FI was assessed at EAT-I and -II via self-report with an item adapted from the U.S. Household Food Security Survey Module.³⁰ FI was not assessed at EAT-III. At EAT-I, participants were asked, “How often during the last 12 months have you been hungry because your family couldn’t afford food?” (EAT-I test-retest Spearman’s r=.37). At EAT-II, participants in high school were asked the same question as was asked at EAT-I, and young adults were asked, “How often during the last 12 months have you been hungry because you couldn’t afford more food?”. For primary analyses, responses of almost every month, some months but not every month, and only one or two months were collapsed to represent severe FI, and participants who responded I have not been hungry for this reason were not considered to have experienced past-year severe FI. To conduct sensitivity analyses investigating the relevance of the frequency with which severe FI occurred, responses of almost every month and some months but not every month were considered to represent frequent past-year severe FI (these responses were combined due to few participants endorsing the almost every month response), and responses of only one or two months were considered to represent infrequent past-year severe FI.

Unhealthy and extreme weight-control behaviors.

Past-year weight-control behaviors were assessed across all survey waves with the question, “Have you done any of the following things in order to lose weight or keep from gaining weight during the past year?”.²³ To distinguish extreme behaviors that may have severe immediate consequences from inherently unhealthy but less extreme behaviors, weight-control behaviors were categorized as unhealthy (fasting, skipping meals, eating very little food, using food substitutes, smoking more cigarettes) or extreme (self-induced vomiting, diet pill use, laxative use, diuretic use). Response options were yes and no for each behavior. Responses were used to create dichotomous variables for endorsing one or more of the unhealthy weight-control behaviors (EAT-III test-retest agreement=83%) and one or more of the extreme weight-control behaviors (EAT-III test-retest agreement=97%). To conduct sensitivity analyses investigating associations between severe FI and the number of weight-control behaviors participants endorsed, two count variables were also created, representing the number of unhealthy weight-control behaviors (possible range: 0–5) and the number of extreme weight-control behaviors (possible range: 0–4) endorsed.

Chronic dieting.

Past-year chronic dieting was assessed across all survey waves with the question, “How often have you gone on a diet during the last year? By ‘diet’ we mean changing the way you eat so you can lose weight.” Participants who reported dieting five or more times in the past year were considered to be chronic dieters, as has been done in previous Project EAT analyses²³ (EAT-III test-retest agreement=97%).

Overeating and binge eating.

Past-year overeating and binge eating were assessed across all survey waves with the following two questions from the Questionnaire on Eating and Weight Patterns-Revised³¹: “In the past year, have you ever eaten so much food in a short period of time that you would be embarrassed if others saw you (binge-eating)?” and, if yes, “During the times when you ate this way, did you feel you couldn’t stop eating or control what or how much you were eating?”. Response options were yes and no for each question. Participants responding yes to the first question and no to the second question were considered to have endorsed overeating (EAT-III test-retest agreement=92%). Participants responding yes to both questions were considered to have endorsed binge eating (EAT-III test-retest agreement=84%).

Sociodemographic characteristics.

Age, sex, and structurally racialized categories labelled as ethnicity/race were self-reported. Based primarily on participant report of the highest level of educational attainment by either parent at EAT-I, household socioeconomic status (SES) was coded in 5 groupings: low, low-middle, middle, upper-middle, and high. Other variables used to assess SES included family receipt of public assistance, eligibility for free or reduced-cost school meals, and parental employment. An algorithm was developed using Classification and Regression Trees to avoid classifying participants as high SES based on parental education levels if they were receiving public assistance, were eligible for free or reduced-cost school meals, or had two unemployed parents (or one unemployed parent if from a single-parent household).^24,32 These variables were also used to categorize SES in instances of missing parental education data.

Statistical Analysis

Descriptive statistics.

Prevalences of severe FI and disordered eating behaviors at each survey wave were summarized.

Examining associations between severe FI and disordered eating behaviors (main effect models).

Generalized estimating equations (GEEs) were used to examine associations between severe FI and disordered eating behaviors. By accounting for correlated data within individuals across study waves, a GEE can produce a single estimate representing multiple waves of data. In models we refer to as repeated cross-sectional, concurrent severe FI was examined as a predictor of each disordered eating variable across EAT-I and -II (data from EAT-III were not included in this model given that FI was not assessed at EAT-III). In models we refer to as longitudinal, severe FI was examined as a predictor of each disordered eating variable at the subsequent survey wave (i.e., examining severe FI at EAT-I and -II predicting disordered eating behaviors at EAT-II and -III, respectively), controlling for the respective disordered eating variable at the time of FI assessment. Hence, the longitudinal models estimate the association between severe FI and future disordered eating behaviors while taking into account current disordered eating behaviors. Modified Poisson regression models (i.e., with robust standard errors)³³ were used to obtain prevalence ratios. Models were adjusted for the following potential confounding variables: age (continuous, time-varying), sex (dichotomous, time-invariant), ethnicity/race (five-level categorical, time-invariant), and SES (five-level categorical, time-invariant). Body mass index was considered for inclusion as a covariate but was ultimately excluded from the models because literature suggests it may act as a mediator (i.e., on the causal pathway from FI to disordered eating behaviors) or collider (i.e., common effect of both FI and disordered eating behaviors),³⁴ in which case adjustment for body mass index could introduce overadjustment bias or selection bias, respectively.

Examining potential differences in associations by age (interaction models).

To test whether associations between severe FI and disordered eating differ by age, interaction terms between FI and age were added to the models. Interactions were tested on both the multiplicative (differences in relative risk) and additive (differences in absolute risk) scales, and those demonstrating significant interaction were probed using the Johnson-Neyman technique to identify regions of significance.³⁵

Sensitivity analyses.

Three sets of sensitivity analyses were conducted for the main effect models. First, sensitivity analyses not adjusting for SES were conducted, as although SES was included as a covariate in primary analyses in order to examine the unique role of severe FI above and beyond that of SES broadly, the SES variable was in part derived from possible indicators of FI (i.e., family receipt of public assistance, eligibility for free or reduced-cost school meals). Second, sensitivity analyses using the three-level FI variable (frequent versus infrequent versus no past-year severe FI) were conducted to investigate the relevance of the frequency with which severe FI occurred (results reported in Supplemental Table 1). Third, sensitivity analyses using the unhealthy and extreme weight-control behavior count variables and standard Poisson regression models were conducted to investigate associations between severe FI and the number of weight-control behaviors participants endorsed (results reported in Supplemental Table 2).

Missing data.

All analyses incorporated IPWs to account for missing data due to attrition across survey waves; missing data at the variable level were not imputed given that the variable-level missing rate was less than 5% and thus considered nominal.³⁶

Statistical software.

Johnson-Neyman probing was conducted with the interactions package³⁷ in R version 4.0.3; all other analyses were conducted with Stata 16.1. Results were considered statistically significant at p<.05.

Results

Past-year severe FI was reported by 8.9% of the sample at EAT-I and 20.9% of the sample at EAT-II (Table 1). Disordered eating behaviors were common across survey waves, with unhealthy weight-control behaviors (e.g., fasting, skipping meals) being the most commonly reported type of disordered eating behavior at each time point.

Cross-Sectional Associations Between Severe FI and Disordered Eating Behaviors

Across EAT-I and -II, severe FI was cross-sectionally associated with greater prevalence of every type of disordered eating behavior examined (Table 2). The strongest associations were observed for extreme weight-control behaviors and binge eating, with 49% greater prevalence of each of these behaviors among participants who had experienced past-year severe FI compared to those who had not after adjusting for sociodemographic covariates. Results based on sensitivity analyses not adjusting for SES were not substantively different.

Table 2.

Associations Between Severe Food Insecurity and Disordered Eating Behaviors and Tests of Effect Modification by Age

	Unhealthy Weight-Control Behaviors	Extreme Weight-Control Behaviors	Chronic Dieting	Overeating	Binge Eating
	PR (95% CI)
Repeated cross-sectional models a
Severe FI main effect	1.27 (1.15, 1.39)***	1.49 (1.13, 1.95)**	1.43 (1.07, 1.93)*	1.47 (1.11, 1.96)**	1.49 (1.04, 2.12)*
Severe FI × age interaction, multiplicative scale	p = .004	p = .32	p = .16	p = .85	p = .74
Severe FI × age interaction, additive scale	p = .009	p = .99	p = .33	p = .84	p = .87
Longitudinal models b
Severe FI main effect	1.07 (0.97, 1.19)	0.99 (0.74, 1.32)	1.27 (0.89, 1.81)	1.20 (0.95, 1.53)	1.41 (1.05, 1.90)*
Severe FI × age interaction, multiplicative scale	p = .05	p = .20	p = .27	p = .36	p = .24
Severe FI × age interaction, additive scale	p = .35	p = .59	p = .18	p = .35	p = .18

Note. PR = prevalence ratio; CI = confidence interval; FI = food insecurity.

^aModels examining severe food insecurity predicting concurrent disordered eating behavior across EAT-I and -II, adjusted for age, sex, ethnicity/race, and socioeconomic status.

^bModels examining severe food insecurity at EAT-I and -II predicting disordered eating behavior five years later at EAT-II and -III, respectively, adjusted for age, sex, ethnicity/race, socioeconomic status, and disordered eating behavior at the time of food insecurity assessment.

^*p < .05,

^**p < .01,

^***p < .001.

Longitudinal Associations Between Severe FI and Disordered Eating Behaviors

Severe FI was significantly associated with 41% (95% confidence interval: 5%–90%) greater prevalence of binge eating five years later after accounting for binge eating at the time of FI assessment and sociodemographic covariates (Table 2). No other significant associations were observed longitudinally. Results were essentially identical in sensitivity analyses not adjusting for SES.

Effect Modification by Age

Cross-sectionally, a significant interaction between severe FI and age was observed in relation to use of unhealthy weight-control behaviors (illustrated in Figure 1) on both the multiplicative (p=.004) and additive scales (p=.009). Probing of the interaction revealed a significant association between severe FI and use of unhealthy weight-control behaviors only up until 20 years of age. Specifically, this interaction indicated that unhealthy weight-control behaviors were significantly more prevalent among participants who had experienced past-year severe FI compared to those who had not across the ages of 11–20 years, but that severe FI was not significantly associated with unhealthy weight-control behaviors in participants older than 20 years of age. No other significant interactions between severe FI and age were observed cross-sectionally, and none were observed longitudinally (Table 2).

Figure 1.

Predicted prevalence of unhealthy weight-control behaviors by severe food insecurity and age, from repeated cross-sectional model adjusting for sex, ethnicity/race, and socioeconomic status (error bars represent standard errors)

Discussion

This study examined associations between severe FI and disordered eating behaviors in a population-based cohort followed from adolescence into young adulthood. Cross-sectionally, severe FI was associated with greater prevalence of each type of disordered eating behavior examined, with the strongest associations observed for extreme weight-control behaviors (e.g., self-induced vomiting, diet pill use) and binge eating, arguably the behaviors of greatest concern. Longitudinally, severe FI also significantly predicted greater prevalence of binge eating five years later. Overall, these findings are consistent with a growing body of literature indicating a cross-sectional link between FI and disordered eating^17–20 and also provide novel evidence for a longitudinal relationship between severe FI and binge eating. Notably, the five-year longitudinal association between severe FI and binge eating was nearly as strong as the corresponding cross-sectional association, indicating robustness of this link over time.

The prevalence of past-year severe FI was quite high at both assessment time points in the present study relative to national U.S. prevalences in the same years.³⁸ The high prevalences observed in this study are likely related to the urban setting from which the sample was recruited, as FI rates are generally highest in urban areas.³ In addition to the relatively high prevalences observed in this study overall, past-year severe FI was reported by over twice as many participants at the second wave of the study than at the first wave. This increase may be explained, in part, by the fact that parents generally try to shield their children from the effects of household-level FI.³⁹ Thus, at the first study wave—when participants were younger—more participants living in food-insecure households may have had enough food to eat as a result of parents forgoing food for themselves to ensure that their children were not hungry.

In the limited number of studies on FI and disordered eating to date, some inconsistencies in findings across developmental stages have suggested the possibility that FI may be more strongly associated with disordered eating in adulthood than in adolescence.¹⁷ To investigate this possibility, the present study examined differences in associations by age. In contrast with the idea that the association between FI and disordered eating might be stronger in adulthood than adolescence, our results instead indicated that cross-sectionally, severe FI was associated with greater prevalence of unhealthy weight-control behaviors (e.g., fasting, skipping meals) only up until 20 years of age. This finding suggests that FI in adolescence may in fact be more, rather than less, strongly associated with restrictive disordered eating behaviors such as fasting and skipping meals than FI in young adulthood. Less control over one’s food-purchasing choices during adolescence as compared to adulthood may help to explain this finding. Even with difficulties affording nutritious food in the context of FI, adults with weight-management goals may have greater capacity to prioritize certain foods for weight management in their budget, whereas adolescents with weight management goals may be more likely to turn to unhealthy strategies. It is worth noting, however, that the absence of a significant association between severe FI and unhealthy weight-control behaviors beyond the age of 20 years in this study does not necessarily imply that young adults experiencing severe FI are engaging in healthful weight-management behaviors, nor does it imply that they are not engaging in disordered eating behaviors. Rather, it only signifies that their likelihood of engaging in the specific set of unhealthy weight-control behaviors assessed in this study may be similar to that of young adults not experiencing severe FI. The observed difference by age could also potentially be related to greater parental use of controlling feeding practices (e.g., restrictive feeding) in the food-insecure context,⁴⁰ which may contribute to disordered eating.⁴¹ Another possible explanation for the difference observed by age is more methodological in nature. Prior qualitative work indicates that adolescents’ perceptions of FI are distinct from adults’ perceptions, such that adolescents recognize if there is less food available in the home but may not attribute this to the economic context.^42,43 Thus, it is possible that the stronger association observed during adolescence between FI and disordered eating behaviors such as fasting and skipping meals to control weight may, to some extent, reflect conflation of financial and weight-related reasons underlying reduced food availability if the rationale for the reduction in food availability is not clearly understood. Further research, particularly research incorporating qualitative methods, is needed to explore this possibility and better understand the relationship between FI and unhealthy weight-control behaviors during adolescence.

The large, population-based sample and longitudinal design, which allowed us to examine the temporality of the relationship between severe FI and disordered eating behaviors, represent important strengths of the present study. However, study limitations must also be considered when interpreting our results. As noted above, the extent to which adolescents may or may not understand the reasons underlying reductions in food, the self-report nature of both FI and disordered eating behaviors are limitations of this study, as is the fact that FI was assessed with a single item. However, assuming a true prevalence ratio greater than one and nondifferential misclassification, misclassification of severe FI due to the use of this single-item measure would be expected to bias our results toward the null (i.e., demonstrating weaker associations than they would otherwise).⁴⁴ Additionally, the lack of FI assessment at the third wave of the study precluded longer-term examination of the association between FI and disordered eating. With regard to the measurement of disordered eating behaviors, another limitation beyond the self-report nature of these behaviors is that data were only collected on whether or not participants had engaged in behaviors. Thus, we were unable to examine associations of FI with the severity or frequency of these behaviors, but such associations should be investigated in future research. Importantly, another limitation of this study is the observational nature of the data, as although we were able to clarify temporality in longitudinal analyses, causality cannot be determined. Finally, all participants were drawn from the St. Paul-Minneapolis metropolitan area, limiting the generalizability of findings from this study to other geographic areas.

Conclusions

Results from the present study bolster prior evidence for a cross-sectional link between FI and a range of disordered eating behaviors, as well as provide longitudinal evidence to suggest that severe FI serves as a risk factor for binge eating. These findings have important clinical and public health implications. Clinically, providers should screen for both FI and disordered eating, recognize that patients experiencing FI may be at risk for disordered eating and the accompanying adverse health consequences, address disordered eating behaviors among food-insecure patients presenting with these behaviors, and provide food-insecure patients with referrals for food resources. From a public health perspective, multilevel community interventions are needed to reduce FI itself, as well as its consequences. Importantly, because FI disproportionately affects people who identify as Black, Indigenous, or a person of color,^2,45 interventions must also address the systemic processes underlying these inequities.

Highlights.

• Severe food insecurity (FI) was linked with higher prevalence of disordered eating

• Associations were strongest for extreme weight-control behaviors and binge eating

• Longitudinal results suggest severe FI is a risk factor for binge eating

Abbreviations:

EAT

Eating and Activity in Teens and Young Adults

SES

socioeconomic status