Prevalence of food addiction in children and adolescents: A systematic review and meta‐analysis

Summary Food addiction (FA) has been as a construct that is associated with childhood obesity. However, relatively little is known regarding the prevalence of FA among children and adolescents. An instrument designed to assess FA among youth, the Yale Food Addiction Scale for Children and Adolescents (YFAS‐C), has been developed and used to estimate FA prevalence among pediatric populations. The present systematic review and meta‐analysis aimed to synthesize the results of FA prevalence among youth. Using keywords related to FA and children to search PubMed, Embase, Scopus, and Web of Science, we identified and analyzed 22 cross‐sectional studies. No longitudinal studies were identified in the search. Meta‐analysis with Freeman‐Tukey Double Arcsine transformation was conducted to estimate FA prevalence. Meta‐regression was applied to understand whether weight status (i.e., data from community samples vs. overweight/obese samples) is associated with FA. Eligible studies (N = 22) were analyzed using 6,996 participants. The estimated FA prevalence was 15% (95% CI 11–19%) for all samples, 12% (95% CI 8–17%) for community samples, and 19% (95% CI 14–26%) for overweight/obese samples. Meta‐regression indicated that weight status was associated with FA severity (p = 0.002) and marginally with FA prevalence (p = 0.056). Healthcare providers should consider and address the high FA prevalence among pediatric population.


| INTRODUCTION
Childhood obesity is health challenge globally with increasing prevalence 1 and associated morbidity. 2,3 Poor health associated with obesity include physical and psychological factors among individuals [4][5][6][7][8] and economic burdens for societies. 9,10 Therefore, finding ways to decrease the prevalence of childhood obesity is an important topic. In order to address childhood obesity, several prevention and treatment approaches have been discussed, particularly as obesity is difficult to treat perhaps given heterogeneous etiologies. 11 One potential factor, food addiction (FA), has attracted attention. [11][12][13][14][15] FA has been a debated concept, 16 with some people questioning its validity. 15 However, the development and refinement of the Yale Food Addiction Scale (YFAS) 17,18 have supported the clinical relevance of defining FA. Although FA has not been recognized or defined by the Diagnostic and Statistical Manual of Mental Disorders fifth edition (DSM-5), 19 the YFAS follows criteria of substance use disorders proposed by the DSM to make provisional FA diagnoses.
Specifically, the YFAS characterizes food addiction via the following items and criteria: Items 1 to 3 assess food is taken in larger amount and for longer period than intended (Criterion 1); Items 4, 17, 18, and 25 assess persistent desire or repeated unsuccessful attempts to quit (Criterion 2); items 5 to 7 assess much time is spent to obtain and to eat food or to recover from its effects (Criterion 3); Items 8 to 11 assess important social, occupational, or recreational activities are given up or reduced because of food addiction (Criterion 4); Item 21 assesses food is continued to be used despite knowledge of adverse consequences (Criterion 5); Items 22 and 23 assess tolerance (Criterion 6); Items 12 to 14 assess characteristic withdrawal symptoms and food is taken to relieve withdrawal (Criterion 7); and Items 15 and 16 assess clinically significant impairment or distress that is not listed in the the inclusionary criteria for substance-use disorders but are rather mentioned prior to the specific inclusionary criteria. A 5-point Likert scale (0 = never; 4 = always) is applied to all 18 YFAS items, and a dichotomous (yes/no) scale is first used for the seven items. Following this, all the items are converted dichotomously (0 = no; 1 = yes) according to specific scoring thresholds for each item. Using the converted dichotomous scores, a symptom count scoring version (ranging between 0 and 7) and a diagnostic scoring version (having 3 or more criteria met plus having clinically significant impairment or distress) can be generated.
Recent neurobiological studies provide insight into potential mechanisms underlying FA. 12,14 Given the health relevance, more research on FA is needed. 13,15 For example, literature on adults found that FA is associated with weight gain, obesity, psychological distress such as anxiety and depression, and eating disorders. 13,16,20,21 With respect to understanding the public health impact of FA, an assessment of its prevalence is important.
Currently, several reviews have evaluated the prevalence of FA. 20 23 additionally stated that higher prevalence was found among individuals aged over 35 years, females, and those with obesity. Imperatori et al., 20 though not using systematic review or meta-analysis, gathered findings from FA prevalence studies and concluded that FA is more prevalent among individuals having eating disorders (compared with those not having eating disorders) and those with obesity (compared with those of lean/normal weight). In addition, Imperatori et al. 20 reported on limited research studying FA among children and adolescents; the prevalence was between 7.2% and 29% in these studies. More recently, Penzenstadler et al. 21 systematically reviewed studies using the YFAS and reached a similar conclusion to other reviews 20,22,23 : namely, FA prevalence is relatively high. In addition, Penzenstadler et al. 21 noted that FA was associated with higher body mass index (BMI) and having eating disorders.
Although FA studies are increasing and initial information regarding FA prevalence has been disseminated, little is known about FA prevalence among children and adolescents. Indeed, among the reviews mentioned above, only Imperatori et al. 20 have reported information on children and adolescents. However, the information of FA prevalence in children and adolescents reported by Imperatori et al. 20 is limited given the following two reasons. First, Imperatori et al. 20 did not use a systematic review or meta-analysis to synthesize the findings. Second, few studies had investigated FA among children and adolescents when Imperatori et al. 20 conducted the review.
In order to understand FA prevalence in children and adolescents (individuals aged below 21 years), we conducted a systematic review and meta-analysis. Specifically, the present systematic review and meta-analysis is now feasible because (1) a children and adolescents version of the YFAS has been developed (i.e., YFAS-C) with promising psychometric properties (the YFAS-C was developed via [1] rewording the sentence descriptions relevant to children and adolescents' daily livings; for example, school instead of employment is mentioned in the YFAS-C and [2] editing the sentence descriptions for a reading level equivalent to grade 2.7) 24 ; and, (2) more studies of FA among children and adolescents have been conducted. Thus, the present systematic review and meta-analysis is both feasible and timely.

| METHODS
The study was registered at PROSPERO (CRD42020142198).

| Literature search strategy
This systematic review and meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) checklist. 25 A search was conducted in PubMed, Embase, Scopus, and Web of Science through August 20, 2020. The search strategy included the key terms "food addiction" or "compulsive eating" and "children" or "adolescent" or "child" or "child, preschool." After removing the duplicate studies, 402 studies remained, and the titles and abstracts of those studies were evaluated by two independent reviewers

| Type of study
The search included observational studies including cohort, case-control, and cross-sectional studies.

| Type of participants
All studies that reported prevalence of FA in children as an outcome were reviewed. Peer-reviewed publications were included if they reported the YFAS-C score (in children aged under 21 years). We defined 21 years as the upper limit of the children's age because the statement made by the American Academy of Pediatrics identified the upper age limit of pediatrics as 21 years. 26 Only publication in English was included. Other systematic reviews and letters to editors were excluded. Single case reports or case series and findings of clinical trials were excluded.

| Clinical appraisal
After finalizing the included papers, each was assessed for quality by calculating the Newcastle-Ottawa score (NOS) via the Newcastle-Ottawa checklist. The Newcastle-Ottawa checklist assesses quality of a study by considering eight domains. It considers the use of appropriate statistical tests, assessment of outcomes, controlling for potentially confounding factors, ascertainment of exposure, approaching non-respondents, sample size, representativeness of the sample, and clear goals as main determinants of a study's quality. Highest quality papers get the maximum score (16), and those with a NOS of 5 or less were excluded. 27

| Outcome measures
The prevalence of FA among children as measured by the YFAS-C score was calculated, along with the mean YFAS-C score and the correlations between BMI z-scores and FA.

| Data synthesis and statistical analysis
After extensive review, a meta-analysis was conducted using the STATA software version 15.0 (StataCorp LLC). The STATA metaprop and metan commands were used. Forest plots are provided to illustrate the point and interval estimations. In meta-analysis for prevalence, Freeman-Tukey Double Arcsine transformation was used to stabilize the variances. The metaninf command was used for sensitivity analysis via evaluating the effect of each study on overall estimates. The I 2 index and Egger's and Begg tests were performed to determine heterogeneity and publication bias. Funnel plots are also provided in this regard.
Studies were divided into community samples and overweight/ obese samples among studies reporting this information. Stratified analysis for prevalence, mean score of YFAS-C and linear correlation of FA and BMI z-score were performed according to this categorization.
In all analyses, BMI z-scores were used if available. BMI z-scores were available or calculated from percentile values for 15 studies (please see Table 1 for detailed information). BMI z-scores were based on objective height and weight measures in 11 studies, and in four studies, height and weight values were based on self-reports (please see Table 1 for detailed information). The mean percentiles of BMI values of three studies (please see Table 1 for detailed information) were transformed to BMI z-scores using the inverse cumulative distribution function of normal distribution.
Meta-regression analysis on the effects of age, gender ratio, target population (i.e., overweight/obese samples vs. community samples), NOS score, sample size, and continent on prevalence of FA and YFAS-C scores was performed in separate models for each covariate using metareg command in STATA.

| Included studies
Extensive systematic review of the literature initially identified 400 studies. Among these, 320 were excluded after screening the titles and abstracts. Full texts of the remaining 80 papers were further evaluated. After removing adult studies, animal studies, other review articles, and those without reported YFAS-C scores, 22 studies were included (Table 1). Sample size across studies was 6,996 participants.
Kappa for the abstract review was 0.88 (95% CI; 0.77-0.98), and Kappa for the full text review was 0.92 (95% CI; 0.84-0.99). All 22 studies were cross-sectional. 24, No longitudinal studies were identified in the search.

| Quality assessment
The NOSs for the 22 included publications 24,28-48 were calculated.
No article had a NOS of 5 or less (Table 1). A graphical representation of each NOS domain is provided (Figure 1).

| Heterogeneity and publication bias
Egger and Begg tests both showed no significant publication bias for the estimated prevalence of FA (p = 0.084 and 0.82, respectively) ( Figure 2). These tests also showed no significant publication bias for the estimated mean YFAS-C score (p = 0.295 and p > 0.90, respectively). Heterogeneity assessed by I 2 values is depicted in each of the corresponding forest plots (Figures 3, 4, and 5). (95% CI: 14-26%) in the overweight/obese samples ( Figure 6).

| YFAS-C scores
The mean YFAS-C score was calculated to be 2.06 (95% CI:

| Meta-regression analyses
In the meta-regression analysis, the target sample had a significant effect on YFAS-C score (p = 0.002) and borderline significant effect on prevalence (p = 0.056) ( Table 4). Specifically, studies on overweight/obese samples had a higher YFAS-score and FA prevalence than did the community samples. Age

| Sensitivity analysis
Sensitivity analysis showed that no considerable impacts were found in the estimates of FA prevalence and YFAS-C scores after omitting any of the included studies (Tables 2 and 3). The relatively high FA prevalence found in children and adolescents is comparable with that in adults. [21][22][23] Therefore, FA appears relevant to both pediatric and adult populations. Indeed, Pursey et al., 23 Imperatori et al., 20 and Penzenstadler et al. 21 all reported high FA prevalence in adult populations. Moreover, similar to adults, a trend of higher FA prevalence was found among children with overweight/obesity when compared with those without overweight/ obesity in the community. Specifically, higher FA prevalence was found in adults with obesity than those without obesity in a metaanalysis 23 and two narrative reviews. 20,21 Our meta-analysis on children and adolescents resonates with these findings and extends them findings to a younger age group. In addition to FA prevalence, higher YFAS-C score found in our meta-analysis supports the importance to considering FA in youth. Therefore, healthcare providers, policy makers, parents, and other stakeholders should attend to FA in youth.
However, unlike weight status, other factors proposed to be relevant to FA prevalence by Pursey et al. 23 in adults were not significantly associated with FA prevalence in our pediatric population.
Pursey et al. 23 [49][50][51] Given that FA has been found to be associated with eating disorders, 13 future studies are warranted to examine the mechanisms why gender-related differences have been found in eating disorders but not FA among pediatric population. As age did not influence results, comparable risks for FA may exist throughout much of childhood and adolescence. Thus, these findings suggest that resource allocation may be best evenly distributed across genders and child/adolescent age groups, although this possibility warrants more study.
All analyzed studies used the same instrument, the YFAS-C, to assess FA. The YFAS-C is a "gold standard" in assessing FA among children and adolescents. Specifically, it has been developed with rigorous methodologies, including the adoption of addiction criteria proposed in the DSM-5 19 and the descriptions modified for use among youth. 24 Moreover, the strong psychometric properties of the YFAS-C have been reported in many studies using multiple testing methods, including classical and modern test theories. 24,[38][39][40]48 Moreover, using the same instrument to assess FA across the synthesized studies included in the present meta-analysis helps to ensure measurement quality. Therefore, together with the nonsignificant publication bias suggested by the Egger and Begg tests, we have confidence that the estimated FA prevalence and YFAS-C score are very reliable.
The present systematic review and meta-analysis has the following strengths. First, the comprehensive search uses multiple major databases (PubMed, Embase, Scopus, and Web of Science) and recommended keywords (identified according to the PECO framework).
Therefore, the included studies are relevant to the study aims. Second, rigorous methodology was adopted in this systematic review and meta-analysis. We used NOS to verify quality assurance; that is, only publications passing a specific study quality threshold were included.
Moreover, publication bias, meta-regression, and sensitivity testing were performed. Therefore, there is considerable confidence regarding the accuracy of the estimates in our meta-analysis. Third, as mentioned earlier, all included studies used the same instrument with promising psychometric properties (i.e., the YFAS-C). Therefore, the present systematic review and meta-analysis has little instrument measurement bias. be beneficial for healthcare providers to design appropriate prevention programs targeting weight gain. Third, studies investigating differences in YFAS-C prevalence across countries would be an important future direction as well.

| CONCLUSION
In conclusion, FA is an important topic among youth. With the relatively high prevalence of FA among children and adolescents found in the present systematic review and meta-analysis, healthcare providers, policymakers, and other stakeholders should design appropriate interventions to address FA in this age group. Moreover, higher estimates of FA were observed among children and adolescents with overweight/obesity as compared with lean/normal-weight individuals.
Thus, targeted interventions may be particularly relevant to children and adolescents with overweight/obesity.