**Food Addiction in Eating Disorders and Obesity: Analysis of Clusters and Implications for Treatment**

**Susana Jiménez-Murcia 1,2,3,\*, Zaida Agüera 1,2,4, Georgios Paslakis 5,6, Lucero Munguia 3, Roser Granero 2,7, Jéssica Sánchez-González 1, Isabel Sánchez 1,2, Nadine Riesco 1,2, Ashley N Gearhardt 8, Carlos Dieguez 2,9, Gilda Fazia 10,11, Cristina Segura-García 11,12, Isabel Baenas 1, José M Menchón 1,3,13 and Fernando Fernández-Aranda 1,2,3,\***


Received: 31 July 2019; Accepted: 14 October 2019; Published: 3 November 2019

**Abstract:** Food addiction (FA) has been associated with greater psychopathology in individuals with eating disorders (ED) and obesity (OBE). The current study aims to provide a better phenotypic characterization of the FA construct by conducting a clustering analysis of FA in both conditions (ED and OBE). The total sample was comprised of 234 participants that scored positive on the Yale Food Addiction Scale 2.0. (YFAS-2) (119 bulimia nervosa (BN), 50 binge eating disorder (BED), 49 other specified feeding or eating disorder (OSFED) and 16 OBE). All participants completed a comprehensive battery of questionnaires. Three clusters of FA participants were identified. Cluster 1 (dysfunctional) was characterized by the highest prevalence of OSFED and BN, the highest ED severity and psychopathology, and more dysfunctional personality traits. Cluster 2 (moderate) showed a high prevalence of BN and BED and moderate levels of ED psychopathology. Finally, cluster 3 (adaptive) was characterized by a high prevalence of OBE and BED, low levels of ED psychopathology, and more functional personality traits. In conclusion, this study identified three distinct clusters of ED-OBE patients with FA and provides some insight into a better phenotypic characterization of the FA construct when considering psychopathology, personality and ED pathology. Future studies should address whether these three food addiction categories are indicative of therapy outcome.

**Keywords:** food addiction; eating disorders; bulimia nervosa; binge eating disorders; obesity; other specified feeding or eating disorders; cluster analysis

### **1. Introduction**

Food addiction (FA) is a concept that has been of increasing scientific interest and debate. An immense body of literature within the field of eating disorder (ED) research has emerged, with whole special issues of scientific journals being dedicated to its characterization [1,2]. In addition to obvious phenomenological similarities between addiction and ED (e.g., loss of control, continued use despite negative consequences, cravings), a great number of neurobiological findings have emerged to additionally support the new concept, not only in preclinical studies, but also in humans [3–6].

Still, not all controversies have yet been resolved. Starting off as a concept to explain a potential subtype of obesity (OBE) [7–10], FA has also been associated with ED, such as bulimia nervosa (BN) [11–13], binge eating disorder (BED) [14–16], and even anorexia nervosa (AN) [17]. In a systematic review of studies on FA in non-clinical and clinical cohorts, it was especially BED that was associated with the most severe FA symptoms [15]. FA also seems to be prevalent in individuals with OBE waiting for bariatric surgery [18,19] and predicts less effective weight reduction throughout dietary intervention before surgery [20]. Interestingly, surgery-induced weight loss may lead to remission of FA [21,22].

The Yale Food Addiction Scale (YFAS) is the main instrument for the assessment of FA and has been developed to assess FA based on the known criteria used for the assessment of substance dependence, but applied for high palatable foods [23,24]. Higher scores on the YFAS have been associated with higher body mass index (BMI), binge eating episodes, impulsivity, and cravings for highly palatable food [25], as well as with neural responses similar to those found in substance use disorders [26–29].

There is scarce evidence with regard to the identification of key determinants of FA based on personality traits or ED-related symptoms and most of what is known is derived from non-clinical cohorts. In a non-clinical sample, negative urgency (irrational acting in aversive affective states) and low levels of task persistence (lack of perseverance) were shown to be significantly and directly associated with FA and FA mediated their association to BMI [30]. In another study in undergraduates, negative urgency, impulsivity when under distress, and emotion dysregulation positively predicted symptom count on the YFAS [31]. Similar findings were shown in a clinical ED cohort, although negative urgency appeared to be the only independent predictor for FA, while self-directedness and emotion dysregulation predicted negative urgency and were highly related to ED-related symptomatology, but not to food addiction itself [32,33]. In individuals with OBE awaiting bariatric surgery, FA was associated with personality traits such as neuroticism, impulsivity, and alexithymia [34], but also more emotion dysregulation, more harm avoidance, and less self-directedness [35].

FA has also been associated with a positive screen for more severe variants of ED-related psychopathology [7] as well as with a positive screen for major mental health symptoms, major depressive episode, anxiety, early life adversities, such as psychological and sexual abuse, and an overall reduced quality of life [36–38]. Finally, female gender was a predictor of severe food addiction [36] and high reward sensitivity was significantly associated with more severe FA symptoms in females [39].

Due to these meaningful interrelationships in the literature between FA and personality traits, as well as psychopathology, it is important to consider the association of FA with these factors in an integrated way (rather than looking at each construct in isolation). Evaluating how personality and

psychopathology cluster within the FA construct could lead to a better understanding of potential distinct phenotypes within FA that could have different clinical profiles.

Based on this premise, the aim of the present paper was to explore empirically the severity of clusters of FA-positive (FA+) participants based on psychopathological symptomatology (namely ED-related psychopathology and general psychopathology) and personality traits, and to investigate how the clinical features and diagnosis of ED and OBE were distributed among them. This is the first study that attempts this type of analysis in order to identify subgroups among participants with positive FA in order to provide a better phenotypic characterization of the FA construct. We hypothesized that, despite overlaps, patients with ED would predominantly fit into a cluster characterized by a more severe ED-related psychopathology that would be different than the cluster predominantly found among individuals with OBE. We then evaluated whether there were differences by ED subtypes, with a subgroup of patients (basically BED, but also BN) who are more similar to participants with OBE. We also hypothesized that, among others, ED-related severity, general psychopathology, and personality traits would be important determinants of the FA phenotypes. The rationale for performing this kind of analysis was to gain insight into the possibility of different FA clusters that would ideally translate into future symptom-targeted treatments for FA phenotypes in individuals with OBE and for those suffering from ED.

### **2. Materials and Methods**

### *2.1. Participants*

From an initial sample of 395, the final participants of the current study were 234 females who scored positive for the FA (based on YFAS 2.) and who also had a diagnosis of ED or OBE (119 with BN, 50 with BED, 49 with other specified feeding or eating disorder (OSFED), and 16 with OBE). All participants in the OBE group (*n* = 16) as well as *n* = 42 participants in the BED group (84.0%), *n* = 27 in the BN group (22.7%), and *n* = 2 in the OSFED group (4.1%) had a BMI higher than 30. All participants included in the study were consecutively referred for assessment and treatment at the Unit of Eating Disorders of the Department of Psychiatry of the University Hospital of Bellvitge in Barcelona, between May of 2016 and November 2018, diagnosed according to the DSM-5 [40] criteria, and were between 18 and 60 years old.

Male participants referred for the Unit in that time period were excluded from the study, as the number of them in our sample was too small for meaningful statistical comparisons (*n* = 21; four with BN, eight with BED, two with OSFED, and seven with OBE). Considering the controversial results of the presence of FA in AN [17], as well as the characteristic fears and cognitive distortions about food and weight involved in AN that may influence patients understanding of what is considered excessive food intake or abnormal eating behavior [41], patients diagnosed with AN were excluded from the study as well. Likewise, due to the different reported prevalence of FA in OBE patients [20] compared to OBE with a comorbid ED [14,15,17], the homogeneity of the sample was preserved by only inducing patients with OBE without ED (*n* = 53), and only those with positive FA were included (*n* = 16).

According to the Declaration of Helsinki, the present study was approved by the Clinical Research Ethics Committee (CEIC) of Bellvitge University Hospital (ethic approval code: PR205/17), and written informed consent was obtained from all participants. All the assessments were conducted by experienced psychologist and psychiatrists.

### *2.2. Assessment*

For the assessment, anthropometric measures such as weight and height (without the participants wearing clothes or shoes) were taken to calculate the BMI (i.e., weight (kg)/height (m2)). In addition to clinically relevant variables (like age of onset or duration of the disorder) and sociodemographical characteristics, a battery of the Spanish-validated versions of the following instruments was used.

### 2.2.1. Eating Disorders Inventory 2 (EDI-2)

The Eating Disorders Inventory 2 (EDI-2) [42] is a 91-item multidimensional self-report questionnaire answered on a 6-point Likert scale that assesses different cognitive and behavioral characteristics typical for ED: Drive for Thinness, Bulimia, Body Dissatisfaction, Ineffectiveness, Perfectionism, Interpersonal Distrust, Interoceptive Awareness, Maturity Fears, Asceticism, Impulse Regulation, and Social Insecurity. The validated version for the Spanish population was developed by Garner, 1998 [43], with a mean internal consistency of 0.63 (coefficient alpha).

### 2.2.2. Symptom Checklist-90-Revised (SCL-90-R)

The Symptom Checklist-90-Revised (SCL-90-R) [44] is used to evaluate a broad range of psychological problems and symptoms of psychopathology. It consist of a 90-item questionnaire that measures nine primary symptom dimensions: Somatization, Obsession-Compulsion, Interpersonal Sensitivity, Depression, Anxiety, Hostility, Phobic Anxiety, Paranoid Ideation, and Psychoticism; and includes three global indices: global severity index (overall psychological distress), positive symptom distress index (the intensity of symptoms), and a positive symptom total (self-reported symptoms). The global severity index can be used as a summary of the test. The validation of the scale in a Spanish population [45], obtained a mean internal consistency of 0.75 (coefficient alpha).

### 2.2.3. Yale Food Addiction Scale 2.0 (YFAS-2)

The Yale Food Addiction Scale 2.0 (YFAS-2) [24] is a 35-item self-report questionnaire for measuring FA during the previous 12 months. This original instrument (YFAS) was based on the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV-TR) criteria for substance dependence and was adapted to the context of food consumption. The newer version of the instrument, YFAS-2, is based on DSM-5 Criteria and evaluates 11 symptoms. The scale produces two measurements: (a) a continuous symptom count score that reflects the number of fulfilled diagnostic criteria (ranging from 0 to 11), and (b) a food addiction threshold based on the number of symptoms (at least 2) and self-reported clinically significant impairment or distress. This final measurement allows for the binary classification of food addiction (present versus absent). Additionally, based on the revised DSM-5 taxonomy, is possible to establish severity cut-offs: mild (2–3 symptoms), moderate (4–5 symptoms), and severe (6–11 symptoms). The translation and validation of the YFAS-2 for Spanish speaking samples with ED was carried out by Granero et al. (2018) [13], showing excellent internal reliability coefficient (α = 0.94), as well as an excellent accuracy in discriminating between healthy controls and ED subsamples (κ = 0.75).

### 2.2.4. Temperament and Character Inventory-Revised (TCI-R)

The Temperament and Character Inventory-Revised (TCI-R) [46] is a 240-item, 5-point Likert scale, questionnaire that measures four temperament dimensions (Harm Avoidance, Novelty Seeking, Reward Dependence, and Persistence) and three character dimensions (Self-Directedness, Cooperativeness, and Self-Transcendence) of personality. The Spanish validation in an adult population was carried out by Gutiérrez-Zotes et al. [47]; the internal consistency (coefficient alpha) of the scales was 0.87.

### *2.3. Statistical Analysis*

Statistical analysis was carried out with SPSS23 for Windows. Empirical clusters were explored with the two-step-cluster procedure, using as indicators the ED severity level (EDI-2 scores), the global psychopathological state (SCL-90-R scores), the personality profile (TCI-R scores), and the DSM-5 diagnostic subtype. Two-step-cluster used the log-likelihood distance measure through a multinomial probability mass function for categorical variables and a normal density function for continuous variables. This clustering technique has desirable features which makes it different from traditional grouping and latent class techniques [48,49]: (a) scalability, which allows analysis of large data files by constructing a cluster-features-tree which is used as a summary of the records; (b) automatic selection of the number of clusters-classes; and (c) handling of categorical and quantitative variables. Criteria for the final model selected in this study were adequate goodness-of-fit (based on a cohesion and separation index) and adequate clinical interpretability [50]. In this study, the Silhouettes coefficient was used as a measure of the goodness of the final cluster solution. This coefficient estimates the cohesion of the elements within a cluster and the separation between the clusters, and it ranges from –1 to +1 (being the values lower than 0.30 interpreted as poor fitting, between 0.30 and 0.50 as fair fitting, and higher than 0.50 as good fitting [51]).

Comparison between clusters was based on chi-square tests (χ2) for categorical variables and analysis of variance (ANOVA) for quantitative criteria. Effect size was measured through Cohen's-d coefficient for mean and proportion comparisons (low effect size was considered for |*d*| > 0.2, moderate for |*d*| > 0.5, and large for |*d*| > 0.8; [52]). The Finner's method was used to control Type-I error due to multiple statistical comparisons (this procedure is included into the Familywise error rate stepwise procedures and offers a more powerful test than the classical Bonferroni correction) [53].

### **3. Results**

### *3.1. Characteristics of the Sample*

All the participants in the analyses were women who met criteria for FA positive screening score based on the YFAS-2 questionnaire *n* = 234. The distribution of the whole sample according to diagnostic group was 16 OBE (without ED) (6.8%), 50 BED (21.4%), 119 BN (50.9%), and 49 OSFED (20.9%). Most of the participants were single (62.0%, compared to 24.8% married or living with a stable partner and 13.2% separated or divorced) and achieved primary (40.6%) or secondary (44.9%) education levels. Table A1 (Appendix A) includes the distribution of the clinical profiles stratified and compared by the diagnostic subtype.

### *3.2. Cluster Composition*

The best grouping structure selected for the sample of the study was the three-cluster solution, which coincided with the most optimal solution chosen by the two-step-cluster procedure. This solution obtained a Silhouettes index (0.30) into the fair range, suggesting mild-moderate evidence of the cluster structure in the data. The comparison between the largest cluster size (*n* = 90, 38.5%) and the smallest (*n* = 60, 25.6%) yielded a ratio of 1.50.

Figure 1 summarizes the results of the clustering procedure: the bar-graph with the indicator relevance for each variable (which reports how good each variable was for the grouping and can differentiate between the derived clusters: the higher the importance of the measure, the less likely it is that the variation for the variable between clusters is due to chance and the more likely it is due to underlying differences) and the centroids (which summarizes the cluster patterns for the set of variables and allows clinical interpretation of the empirical clusters). The indicator variables with the highest contribution into the clustering were SCL-90-R scales measuring the symptom levels (concretely psychotic, depressive, interpersonal sensitivity, anxiety, paranoia), followed by the EDI-2 scales measuring ED severity (impulse regulation, social insecurity, and ineffectiveness). The personality traits measured with the TCI-R obtained low relevance for the clustering, except for harm avoidance (which achieved moderate-low capacity), as well as the diagnostic subtype (which achieved poor capacity for the differentiation between the groups).

**Figure 1.** Results of the clustering procedure. Note. EDI-2: Eating disorders Inventory 2. SCL-90-R: Symptom Checklist-90-Revised. SCL-90-R Obsess-comp.: Obsession-Compulsion SCL-90-R subscale. TCI-R: Temperament and Character Inventory-Revised.

### *3.3. Comparison between the Empirical Clusters*

The first part of Table 1 contains the comparison between the three empirical clusters for the civil status and the education levels. Differences between the groups only were found for the civil status: compared with the other two groups, cluster 1 had the highest proportion of single participants.

The second block of Table 1 contains the comparison between the clusters for the clinical profile. As a whole, clusters were ordered by the psychopathological state and personality profiles. Cluster 1 (*n* = 60, 25.6%) included the youngest participants (mean 28.8 years-old), with the lowest age of onset (mean 18.5 years-old), the shortest duration of the eating problems (mean 10.5 years), and the lowest BMI (mean 27.6 kg/m2), but with the highest severity in eating problems (the highest means in the EDI-2), the worst psychopathological state (the highest means in the SCL-90R), the highest levels in the personality traits of harm avoidance and self-transcendence, and the lowest levels in self-directedness and cooperativeness. Cluster 1 was labeled in this study as the "dysfunctional cluster".


**Table 1.** Comparison between clusters for variables of the study.

Note. SD: standard deviation. \* Bold: significant comparison (0.05 level). † Bold: effect size into the moderate (|*d*| > 0.50) to high range (|*d*| > 0.80). YFAS-2: Yale food addiction scale 2.0. BMI: Body mass index. EDI-2: Eating disorders Inventory 2. SCL-90-R: Symptom Checklist-90-Revised. TCI-R: Temperament and Character Inventory-Revised.

Cluster 2 (*n* = 90, 38.5%) included the participants with the longest duration of the eating problems (mean 14.3 years) and the highest level of food addiction (mean 9.6). The mean for BMI (30.3 kg/m2) was higher than in cluster 1, but similar to that obtained in cluster 3. The mean scores in the EDI-2 and the SCL-90R were high for participants into cluster 2, but clearly lower than values registered for cluster 1. Cluster 2 was labeled in this study as the "moderate cluster".

Cluster 3 (*n* = 84, 35.9%) was characterized by similar mean scores in age (34.5 years-old), onset of eating problems (21.7 years) and duration of the disorder (12.6 years) than cluster 2. The most adaptive scores in the clinical profile was registered for cluster 3 (the lowest scores in the EDI-2 and the SCL-90R), as well as the highest means in the personality traits of persistence, self-directedness and cooperativeness. Cluster 3 was labeled in this study as the "functional cluster".

Figure 2 includes the 100%-stacked bar chart with the distribution of the DSM-5 ED diagnostic subtype into each empirical cluster. OBE patients were primary included into cluster 3 (68.85), while BED patients were mostly distributed into clusters 3 and 2 (48.0% and 44.0%, respectively). Approximately half of the BN patients were into cluster 2 (52.1%), and the remaining participants into this group were distributed between cluster 1 (27.7%) and cluster 3 (20.2%). A little more than half of the OSFED patients were in cluster 3 (51%), and almost 41% in cluster 1.

**Figure 2.** Distribution of the diagnostic subtype and the food addiction (FA) severity group within the empirical clusters. Note. OBE: obesity. BED: binge eating disorder. BN: bulimia nervosa. OSFED: other specified feeding eating disorder.

As a synthesis of the results, Figure 3 contains the radar-chart comparing the empirical clusters for the main clinical variables of the study. To allow adequate interpretability, z-standardized scores were plotted.

**Figure 3.** Radar-chart comparing the z-standardized mean scores between the empirical clusters Note. FA: Food addiction. BMI: Body mass index. EDI-2: Eating disorders Inventory 2. SCL-90-R GSI: Global Severity Index of the Symptom Checklist-90-Revised.

### **4. Discussion**

The aim of the present study was to explore empirical severity clusters with FA-positive (FA+) females, and to investigate whether these FA clusters differed by ED and OBE. A three-cluster structure

was detected based on general psychopathology, ED severity, and personality traits. These clusters ranged from a more functional cluster, to moderate and highly dysfunctional group.

Although high FA has previously been associated with high eating psychopathology and more dysfunctional personality traits, [32], this is the first study that analyzes a potential heterogeneity within patients with FA. Although all the participants in the current study were FA+, we found that the identified clusters followed a linearity with respect to FA severity with the most dysfunctional clusters (1 and 2) having the highest FA symptoms level, and the most functional one, the lowest. For the "dysfunctional cluster", we found a higher prevalence of OSFED and BN, both ED conditions characterized by more dysfunctional personality traits, greater impulsivity, and more general psychopathology, [54–57], as well by their worse prognosis [56]. Consistent with this literature, the cluster with more OFED and BN had the worst psychopathological state and highest severity in ED symptomatology.

Personality traits that were elevated in the most dysfunction cluster were greater difficulties in establishing goals and objectives to guide their lives (self-directedness), the highest levels of anxiety, worry, fear (harm avoidance), and being a more self-centered person (lower cooperativeness). All of these characteristics have not only been associated with FA [32,35], but also with BN, substance use disorders (SUD), and comorbidity between BN and SUD [58,59]. This cluster was also associated with higher levels of novelty seeking, which is a predictor of risky behaviors and associated with SUD and ED pathologies [59–64]. Therefore, the commonality of these extreme personality features in addictive profiles, suggests it could be important in the etiology and maintenance of FA and could be the focus of intervention efforts. Additionally, the aforementioned characteristics have been related to higher risk of dropout and poor outcome in OSFED patients with purging behaviors [56], suggesting important clinical relevance. Although the association of FA and treatment outcome was not the aim of this study, prior research suggests that higher initial FA scores have been associated with a worse prognosis in BN [65].

In sum, the severity of this dysfunctional cluster may be driven by the comorbid ED provided and characteristics associated with these disorders. Thus, in order to be successful, the treatment aimed at patients who are in this cluster should target its principal psychological and psychiatric characteristics, in addition to factors that may maintain FA. As previous literature has reported, changes in personality traits are related to an overall improvement in eating pathology [66], and interventions that could target the personality factors elevated in this cluster would likely be of benefit. Additionally, due to the severity of the psychopathological state in this cluster, pharmacological approaches to address comorbid psychiatric conditions could be also considered if it is needed.

The second cluster, the moderate cluster, was the one that presented the highest level of FA, although functioning was more adaptive than cluster 1 (i.e., lower ED severity, intermediate psychopathology levels). Therefore, we hypothesize that it is FA which mostly determines the characteristics of this cluster. Regarding the personality traits, high levels of harm avoidance and low self-directedness were present in this cluster, but at moderate levels compared to cluster 1 (the dysfunctional). Related to the ED pathology, there was a high presence of patients with BN followed by patients with BED within this cluster. Higher FA scores have been already associated with bingeing ED-subtype patients [13,14,17,67], given the tendency to consume more high-fat/high-sugar caloric food during binges episodes, which may result in a higher number of FA symptoms [17] given the similar neural responses in reward pathways modulated by dopamine by those types of food and addictive drugs [6,16,29,68–70]. In fact, new maintenance models of BN and BED have emerged, that take into consideration the addictive response to palatable foods [71]. Related studies have also found that similar patterns of neural underpinning of tolerance and dependence observed in SUD appear to be related to binge-type eating disorders as well [72,73]. This is consistent with the high comorbidity between SUD and binge-type ED [74,75]. Finally, another important aspect to mention is that it has been shown that individuals with BN or BED experience more frequent and more intense food cravings than persons without binge eating [76–79] and that both conditions show significantly larger food cue reactivity (self-reported craving) [80]. This intense desire or urge to eat a particular type of food also, at a neural level, resembles

responses to drug cues in SUD [26]. Therefore, it could be suggested that the treatment directed to patients in this cluster should target the reward related neural processes that maintain addictive disorders. In this regard, it has been suggested that previously developed interventions for addictions could be applied to binge eating behaviors [71], such as training in the reduction of food cue reactivity in order to reduce craving [81,82] and a reduction in the intake of high-fat/high-sugar caloric food which hyper-activate reward systems.

The "adaptative" cluster presents with more functional personality traits and low levels of general psychopathology, as well as the lowest levels of FA. Thus, the FA in this cluster may be the result of different factors than patients in clusters 1 and 2, which could have important implications in the treatment. First, it is important to mention that within this cluster, there was the highest presence of patients with OBE without ED. This is consistent with prior findings that patients with OBE with a comorbid ED have a higher level of psychopathology than OBE patients without ED [83–86]. Second, this cluster presents the most functional personality traits, the lowest levels of harm avoidance and self-transcendence, and the highest in cooperativeness and self-directedness. Similar results have been found in healthy control groups when comparing them with ED and behavioral addiction patients [87,88]. Thus, BMI could be playing an important role in this cluster, considering that in our sample there are statistically significant differences between cluster 1 (the dysfunctional) and 3 (the adaptative) in BMI. It has been suggested that visceral adiposity levels could be a mediator of the relationship between middle-dorsal insula network connectivity (insula region relevant for eating behavior) and food craving [28], being that visceral fat disrupts insula coding of bodily homeostatic signals, which may boost externally driven food cravings, and also, there are positive associations between food craving and excessive overeating [89,90]. Given the characteristics of this cluster, for which no dysfunctional personality traits of severe ED symptomology or psychopathology must be addressed, it could be hypothesized that nutritional changes that would have a positive impact on a reduction of BMI could be also be beneficial for the reducing FA, through reducing craving episodes. This is consistent with previous studies that find that FA decreases significantly after bariatric surgery [91], and that the induced weight loss by the surgery resulted in the remission of FA in 93% of patients [22]. Finally, there is a moderate representation of patients with BED in this cluster, which may be due to the common co-occurrence between both conditions (OBE-BED). This may be due to the association of features such as grazing [92], craving [93–95], or hedonic [96] and emotional eating [97,98] with OBE and BED. Addressing these factors (and their negative consequences) could be an important focus of treatment in this FA cluster, as grazing is associated with poorer weight loss treatment outcomes in OBE [92] and craving and the use of food to regulate mood are potential triggers for overeating [41,99,100].

The results of this study should take into account the following limitations. First, the sample only included women, so the results cannot be generalized to males; for future studies, it will be important to explore if the FA cluster structure found in this study is replicated in males. Another limitation is that the size of some of the participant groups, divided by ED subtype and OBE, is small. Finally that, due to the fact that the participants were recruited in the same geographical area, the results may not be generalizable to other samples

### **5. Conclusions**

The findings in the present study describe a three-cluster structure of FA-positive (FA+) participants that differ by ED and OBE profile. The clusters range from more to less functional, depending on psychopathology and personality traits. The identification of phenotypes in FA will not only increase knowledge of each cluster's characteristics, but may allow for better individualization of treatment by identifying novel intervention targets and improving treatment outcomes. Likewise, the present study identifies future lines of research, as longitudinal studies that could analyze the predictive validity of this cluster structure on treatment outcome could be of importance.

**Author Contributions:** The authors had the following roles in this paper: conceptualization, S.J.-M., R.G., A.N.G., C.D., F.F.-A. and R.G.; methodology, F.F.-A, S.J.-M., Z.A. and R.G.; formal statistical analysis, R.G.; data collection and assessment: Z.A., G.P., L.M., J.S.-G., I.S. and N.R.; writing—original draft preparation, Z.A., G.P., L.M., J.S.-G., R.G., A.N.G., F.F.-A. and S.J.-M.; writing—review and editing, Z.A., G.P., L.M., J.S.-G., I.B., J.M.M., A.N.G., C.D., G.F., C.S.-G., F.F.-A., S.J.-M. and R.G.; supervision, F.F.-A. and S.J.-M.; funding acquisition, F.F.-A., J.M.M., C.D. and S.J.-M.; Investigation, S.J.-M., Z.A., G.P., L.M., J.S.-G., G.F. and F.F.-A.; Project administration, S.J.-M. and F.F.-A.; Resources, S.J.-M. and F.F.-A.

**Funding:** We thank CERCA Programme/Generalitat de Catalunya for institutional support. This manuscript and research was supported by grants from the Ministerio de Economía y Competitividad (PSI2015-68701-R), by the Delegación del Gobierno para el Plan Nacional sobre Drogas (2017I067), by the Instituto de Salud Carlos III (ISCIII) (FIS PI14/00290 and PI17/01167), by the SLT006/17/00246 grant, funded by the Department of Health of the Generalitat de Catalunya by the call "Acció instrumental de programes de recerca orientats en l'àmbit de la recerca i la innovació en salut", and co-funded by FEDER funds/European Regional Development Fund (ERDF), a way to build Europe. CIBERObn and CIBERSAM are both initiatives of ISCIII.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

### **Appendix A**



Note. α: Cronbach's alpha in the sample. BED: binge eating disorder; BN: bulimia nervosa; OSFED: other specified feeding eating disorder. SD: standard deviation. \* Bold: significant comparison (.05 level). YFAS: Yale food addiction scale 2.0. BMI: Body mass index. EDI-2: Eating disorders Inventory 2. SCL-90-R: Symptom Checklist-90-Revised. TCI-R: Temperament and Character Inventory-Revised.

### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Food Addiction Is Associated with Irrational Beliefs via Trait Anxiety and Emotional Eating**

### **Laurence J. Nolan \* and Steve M. Jenkins**

Department of Psychology, Wagner College, 1 Campus Rd., Staten Island, NY 10301, USA **\*** Correspondence: LNolan@wagner.edu; Tel.: +1-718-390-3358

Received: 31 May 2019; Accepted: 22 July 2019; Published: 25 July 2019

**Abstract:** Irrational beliefs (IB) are believed, in cognitive behavioral therapies, to be a prime cause of psychopathologies including anxiety, depression, problem eating, and alcohol misuse. "Food addiction" (FA), which has been modeled on diagnostic criteria for substance use disorder, and emotional eating (EE) have both been implicated in the rise in overweight and obesity. Both FA and EE are associated with anxiety. Thus, in the present study, the hypothesis that IB is associated with FA and with EE was tested. Furthermore, possible mediation of these relationships by trait anxiety and depression (and EE for IB and FA) was examined. The responses of 239 adult participants to questionnaires measuring FA, IB, EE, depression, trait anxiety, and anthropometrics were recorded. The results revealed that IB was significantly positively correlated with FA and EE (and depression and trait anxiety). Furthermore, only EE mediated the effect of IB on FA and this was not moderated by BMI. Finally, trait anxiety (but not depression) mediated the effect of IB on EE. Exploratory analysis revealed a significant serial mediation such that IB predicted higher FA via elevated trait anxiety and emotional eating in that order. The results of this study suggest that IB may be a source of the anxiety that is associated with EE and FA and suggest that clinicians may find IB a target for treatment of those persons who report experiences of EE and FA. IB may play a role in food misuse that leads to elevated BMI.

**Keywords:** food addiction; irrational beliefs; emotional eating; anxiety; food misuse

### **1. Introduction**

"Food addiction" (FA) has been suggested as a factor in the increased prevalence of overweight and obesity. Proponents of FA suggest that some energy-dense highly palatable foods (or specific additives to foods such as salt or refined sugar) generate addiction-like behaviors in those who consume them [1]. FA, as measured by the Yale Food Addiction Scale (YFAS), is associated with binge eating behavior (BED) [2–4], bulimia nervosa [5], night eating syndrome [6], and with elevated BMI even in the absence of BED [4,7,8]. The FA concept is not without controversy. Some critics prefer an alternate description that focuses on the behavior (i.e., "eating addiction") and suggest there is little evidence for an addicting substance in food. They instead suggest that overeating may be a form of habitual food "abuse" [9] or represent a possible food use disorder [10]. Others have suggested that there is not enough evidence yet to conclude that FA is a distinct entity that explains overeating [11]. Nonetheless, there has been significant interest in FA in the scientific community in recent years [12]. The YFAS, which was based on DSM IV substance dependence criteria, has now been updated to reflect the substance use disorder criteria described in the DSM-5 and dubbed the YFAS2 (which has since been provided in a shortened form [13]).

In cognitive behavioral therapies (CBT), irrational beliefs are believed to be a prime cause of psychopathologies including problem eating and addictive behavior. Ellis [14] and Beck [15] proposed that individuals often have habitual affect-eliciting thought patterns (referred to as irrational beliefs by Ellis) that can lead to dysfunctional emotional and/or behavioral responses. These irrational beliefs originate from a core process of perfectionism [14] or absolutist thinking [16] and the idea that one should be extremely upset when things go wrong and that it is crucial to be successful and approved of by everyone [15]. This absolutist thinking inevitably leads to anxiety and, in turn, may lead to irrational coping strategies such as substance use and uncontrolled eating typified by emotional eating and FA.

In a meta-analysis of 100 independent samples, irrational beliefs were found to be moderately correlated with psychological distress [17]. More specifically, irrational beliefs were associated with trait anxiety [18,19]. While Rohsenow and Smith [18] did not find a connection between irrational beliefs and depression (as measured by Minnesota Multiphasic Personality Inventory), in daily reports of mood over several months, there was an association of irrational beliefs and reports of depression. Others reported that irrational beliefs were related to depression as measured by the Beck Depression Inventory in a sample of women [20]. Mayhew and Edeleman [21] found that irrational beliefs were predictive of poor coping strategies and low self-esteem. Irrational beliefs have been associated with addictive behaviors such as drug misuse [22–25] and problem gambling (see [26] for review) although, in the gambling studies, irrational beliefs are often assessed using different measures than they are in studies of depression and anxiety.

Several studies (mostly involving samples of undergraduate women without eating disorder diagnoses) have reported a link between irrational beliefs and problem eating. Ruderman [27] reported that irrational beliefs were associated with dietary restraint (the cognitive control of food consumption as measured by the Revised Restraint Scale or RRS), particularly the concern with dieting subscale. Studies examining the relationship between irrational beliefs and subclinical eating disorder symptoms are more common. Irrational beliefs predicted a number of bulimia symptoms in undergraduate women [20,28,29]. In addition, irrational beliefs were found to be predictive of drive for thinness, body dissatisfaction, ineffectiveness, and poor interceptive awareness as measured by the Eating Disorders Inventory [21]. More recently, Tomotaki et al. [30] reported that obsession with eating, dieting, and obese-phobia were predicted by irrational beliefs. Irrational beliefs were found to be higher in women with high body dissatisfaction when compared to a group diagnosed with eating disorders and a group with low body dissatisfaction [31].

While irrational beliefs have been associated with dietary restraint, it has not been examined in relation to emotional eating. Emotional eating is generally viewed as a response to negative emotion or distress [32,33] or ego-threat [34], and has been associated with overeating, binge eating, bulimia nervosa, and obesity (see [32]). There is a positive association between emotional eating and anxiety in persons with obesity (but not in persons with a BMI between 18 and 25) [35] and in a sample of children and adolescents [36]. Irrational beliefs and depression were positively correlated in a sample of women [37]. Thus, irrational beliefs may be associated with emotional eating, possibly as the source of anxiety and/or depression.

The research findings described above suggest that irrational beliefs could predict FA and emotional eating. If they do, it is likely that there would be mediating variables. FA is positively correlated to depression in persons with obesity. Furthermore, FA has been associated with depression in persons with obesity [2,3] and in students and the general population [6,38]. FA has also previously been associated with emotional eating [1,2] and with anxiety [39]. The present study was conducted to determine whether the presence of irrational beliefs predicts higher FA symptoms. Furthermore, if such a relationship exists, it may be mediated by depression, trait anxiety, and/or emotional eating, and may depend on BMI. Absolutist irrational beliefs are predicted to produce psychological distress via activation of anxiety. Maladaptive responses such as emotional or uncontrolled eating (e.g., FA) may occur in response to this anxiety. Thus, the following hypotheses (H) were tested. Irrational beliefs and FA are positively correlated (H1). Furthermore, the effect of irrational beliefs on FA is mediated by trait anxiety, depression, and/or emotional eating (H1a). It was also hypothesized that there would be a positive relationship between irrational beliefs and emotional eating (H2) and that

the effect of irrational beliefs on emotional eating is mediated by trait anxiety and/or depression (H2a). Finally, it was hypothesized that there would be a moderation of these relationships by BMI; that the effect of irrational beliefs would depend on the value of BMI (H3). Moderation by gender was also hypothesized but not examined in the present study due to the relatively low number of men in the sample.

### **2. Materials and Methods**

### *2.1. Participants*

The participants were 239 adults; the sample included individuals who identified as women (*n* = 176), men (*n* = 60), or as having non-binary gender (*n* = 3). The sample was mostly composed of undergraduate students. The mean age of participants was 20.72 years (SEM = 0.42; range = 18–57) and their mean BMI was 24.07 kg/m2 (SEM = 0.32; 5.4% had a BMI less than 18.5; 62.3% between 18.5 and 24.9; 20.9% were between 25 and 29.9; 11.3% ≥ 30). When asked in an open-ended question to describe their ethnic or racial background, participants described themselves as having Arab (1.3%), African (6.3%), Asian (4.2%), European (73.2%), South Asian (3.8%), or more than one primary (4.6%) ancestry. In addition, some of the participants (6.7%) also indicated Hispanic ancestry. In total, 17.8% of the participants met the criteria for FA. In total, 20.9% of the sample had a depression score of 50 or higher suggesting the presence of depression. See Table 1 for mean scores on questionnaires.

### *2.2. Measures*

### 2.2.1. Food Addiction

FA was assessed by the Modified Yale Food Addiction Scale 2.0 (mYFAS2) which is designed to evaluate indications of "addiction" toward foods according to the DSM-5 criteria for substance use disorder but with fewer questions than the YFAS 2.0 [13]. The mYFAS2 has been validated against questionnaires that measure related constructs, and has a Kuder–Richardson's alpha of 0.86. In the present study, Kuder–Richardson's alpha was 0.77. The mYFAS2 has 13 items (11 for FA symptoms and 2 for distress) and is scored by counting the number of diagnostic criteria that are met. A person is considered to have FA when 3 or more of the criteria are met, and there is impairment or distress present. In analyses presented below, the mYFAS2 was entered as a continuous variable (number of symptoms).

### 2.2.2. Irrational Beliefs

Irrational beliefs were measured by the Shortened General Attitude and Belief Scale (SGABS) [40]. The SGABS is a 26-item scale that uses Likert-type ratings with responses ranging from 1 (strongly disagree) to 5 (strongly agree). The SGABS has one rational beliefs subscale and six irrationality subscales (need for achievement, need for approval, need for comfort, demand for fairness, self-downing, and other downing) that are summed to create a total irrationality score (higher scores indicate stronger irrational beliefs). Several instruments are available to measure irrational beliefs, some of which may be more sensitive to affect than to cognitions (for a review see [41]). The SGABS is based on the Ellis model of psychopathology [41] and its score has been shown to be less affected by mood than some previous measures [40]. The reliability for this sample was very good (Cronbach's α = 0.86).

### 2.2.3. Eating Styles

Eating styles were measured by the Dutch Eating Behavior Questionnaire (DEBQ) which contains three subscales: restrained eating (DEBQr), emotional eating (DEBQe), and external eating (DEBQx) [42]. All 33 questions are rated on a 5-point Likert-type scale with "never" and "very often" as the anchors. The restraint (cognitive restraint of eating) and external eating (eating in the presence of external cues) scales each have 10 items while the DEBQe contains 13 items. Score for each subscale is the mean

rating. In the present sample, the Cronbach's α for the DEBQe, DEBQr, and DEBQx were 0.94, 0.79, and 0.94 respectively.

### 2.2.4. Trait Anxiety

Trait anxiety was measured using the State-Trait Anxiety Inventory for Adults (STAI) [43]. The STAI differentiates between trait anxiety and state anxiety with 40 items (4-point Likert-type scale) regarding how participants generally feel (trait) and how they feel at this moment (state). Only the trait measure was used in the statistical analysis. In the present study, the Cronbach's α was 0.93.

### 2.2.5. Depression

Depression was assessed using the Self-report Depression Scale (SDS) [44]. The SDS score is the sum of responses to 20 questions to which the participant responds on a 4-point Likert-type scale. The total score can range from 20 to 80 with most depressed persons scoring between 50 and 69 [45]. The Cronbach's α for the SDS was 0.87 for this sample.

### *2.3. Procedure*

The hypotheses were pre-registered with the Open Science Framework after data collection had commenced but prior to examination of the data (doi: 10.17605/OSF.IO/QWSRD). The procedure was approved by the Wagner College Human Experimentation Review Board (code F18–10).

All participants completed questionnaires using an online platform (Qualtrics, Provo, UT, USA). Questionnaires were presented in randomized order after informed consent was obtained. Questions regarding age, height and weight, gender, and ethnicity were presented after questionnaires. All participants were debriefed as to the purpose of the study after the survey was completed. In the debriefing, none of the participants appeared to be aware of the study's purpose. Students participated in groups at scheduled times in a computer laboratory as one way to complete a research requirement for an introductory psychology course (*n* = 174). Other participants were recruited via the university daily email bulletin and via a link (which took them to the same website on the university server as that used by the students) posted on Facebook. These participants were entered into a lottery to win one of four \$25 gift cards if they wished.

### *2.4. Data Analysis*

To ensure quality of data, records were screened for inappropriate responses to open-ended questions, lack of response variation (e.g., giving the same answer to all questions), and unusually short survey completion times.

Statistical analysis was performed using IBM SPSS (version 24). The variables were first correlated to establish whether there was a relationship between irrational beliefs and FA and whether each was correlated to trait anxiety, depression, eating styles, and BMI. Then, mediation multiple regression analysis was conducted using the PROCESS plug-in for SPSS (release 2.16.3) [46] using 5000 bootstrap samples. Variables were mean-centered and heteroscedasticity-consistent standard errors were used. Residuals were checked for stochasticity prior to the analysis. The unstandardized beta coefficients (B), confidence intervals (CI) and adjusted R<sup>2</sup> are reported. Planned analysis included testing whether the relationship between irrational beliefs and FA was mediated by emotional eating, trait anxiety, and/or depression and whether these would be moderated by BMI. That is, the causal hypothesis was that irrational beliefs produce an elevation in FA by increasing trait anxiety, depression, and/or emotional eating. Furthermore, the mediation of the effect of irrational beliefs on emotional eating by trait anxiety and/or depression was also examined (the hypothesis that irrational beliefs cause elevated emotional eating by increasing trait anxiety and/or depression). The results of these analyses led to an exploratory analysis of the indirect pathway between irrational beliefs, trait anxiety, depression, emotional eating, and FA in that order (serial mediation). This final analysis tested the hypothesis that

irrational beliefs increase FA symptoms via a pathway from higher trait anxiety and/or depression to higher emotional eating.

### **3. Results**

### *3.1. Confirmatory Analyses*

### 3.1.1. Correlations

To examine the relationships among the questionnaire variables and test hypotheses H1 and H2, Pearson correlation coefficients were performed. The results indicated that irrational beliefs were positively correlated with all measures (most strongly with depression and trait anxiety; see Table 1). Correlation coefficients indicated that body mass index was significantly positively correlated only with dietary restraint and FA and not with irrational beliefs or other measures (see Table 2).

**Table 1.** Pearson correlation coefficients among and descriptive statistics for the questionnaire measures.


<sup>1</sup> *p* ≤ 0.01, <sup>2</sup> *p* ≤ 0.001, <sup>3</sup> *p* < 0.0001, <sup>4</sup> *p* < 0.00001. SGABS: Shortened General Attitude Belief Scale. DEBQe: Dutch Eating Behavior Questionnaire Emotional Eating. DEBQx: Dutch Eating Behavior Questionnaire External Eating. DEBQr: Dutch Eating Behavior Questionnaire Restrained Eating. SDS: Self-report Depression Scale. mYFAS2: Modified Food Addiction Scale 2.0. STAI: State-Trait Anxiety Inventory.

**Table 2.** Pearson correlation coefficients among the questionnaire measures and body mass index (BMI).


<sup>1</sup> *p* < 0.05, <sup>2</sup> *p* < 0.0001. SGABS: Shortened General Attitude Belief Scale. DEBQe: Dutch Eating Behavior Questionnaire Emotional Eating. DEBQx: Dutch Eating Behavior Questionnaire External Eating. DEBQr: Dutch Eating Behavior Questionnaire Restrained Eating. SDS: Self-report Depression Scale. mYFAS2: Modified Food Addiction Scale 2.0. STAI: State-Trait Anxiety Inventory.

### 3.1.2. Multiple Regression Mediation Analysis

Multiple Mediation of the Effect of Irrational Beliefs on FA

Multiple mediation analysis (PROCESS model 4) was used to determine whether the effect of irrational beliefs on FA was mediated by depression, trait anxiety, and/or emotional eating (H1a). BMI was included as a covariate because of its correlation with the criterion variable, FA. The results (see Figure 1) indicated that, although irrational beliefs significantly predicted elevated scores on depression, trait anxiety, and emotional eating measures, only emotional eating mediated the effect; there was a significant indirect effect of irrational beliefs on FA through emotional eating (*B* = 0.02; 95%CI: 0.010–0.034) but not though trait anxiety (*B* = 0.00; 95%CI: −0.021–0.021) or depression (*B* = 0.02; 95%CI: −0.001–0.035). In the mediation model, there was no significant direct effect of irrational beliefs on FA (*B* = 0.02, *t* = 1.17, *p* = 0.242; 95%CI: −0.010–0.040). The total effect was statistically significant (*B* = 0.05, *t* = 4.20, *p* < 0.001; 95%CI: 0.027–0.074).

**Figure 1.** The relationship between irrational beliefs and food addiction (FA) is mediated by emotional eating but not depression or trait anxiety. Solid arrows indicate statistically significant regression coefficients. Nonsignificant BMI effects on mediators are not included for simplicity.

Moderated Mediation by BMI of the Effect of Irrational Beliefs on FA

In order to examine whether BMI moderated the direct and indirect (e.g., through emotional eating) effects of irrational beliefs on FA (H3), BMI was added to the model presented above as a moderator (a moderated mediation model, PROCESS model 15). The results of this analysis again showed emotional eating to be the only significant mediator of the effect of irrational beliefs on FA and indicated that there were no significant interactions between BMI and other effects on FA (see Figure 2). There was no significant conditional direct effect of irrational beliefs on FA at mean-1SD, mean, and mean+1SD values of BMI. In addition, there was a statistically significant conditional indirect effect of irrational beliefs on FA at mean-1SD BMI (95%CI: 0.008–0.037), mean BMI (95%CI: 0.009–0.034) and mean+1SD BMI (95%CI: 0.007–0.038); each coefficient was virtually identical (*B* = 0.019). The index for moderated mediation was not statistically significant (95%CI: −0.0017–0.0017). Finally, there was no moderation of the nonsignificant mediated effects of depression and trait anxiety on FA.

**Figure 2.** The statistical model for the moderated mediation of the effect of irrational beliefs on FA with BMI as moderator. Solid arrows indicate statistically significant regression coefficients. Nonsignificant mediator variables and BMI effects on mediators are not shown for simplicity.

Multiple Mediation of the Effect of Irrational Beliefs on Emotional Eating

Mediation analysis (PROCESS model 4) was used to determine whether the effect of irrational beliefs on emotional eating was mediated by depression and/or trait anxiety (H2a). The results indicated that, although irrational beliefs significantly predicted elevated scores on depression and trait anxiety, only the latter mediated the effect of irrational beliefs on FA (see Table 3). There was a significant indirect effect of irrational beliefs on emotional eating through trait anxiety (*B* = 0.02; 95%CI: 0.006–0.027) but not through depression (*B* = 0.00; 95%CI: −0.007–0.007; see Figure 3). In the mediation model, there was no significant direct effect of irrational beliefs on emotional eating (*B* = 0.01, *t* = 0.83, *p* = 0.406; 95%CI: −0.007–0.016). The total effect was statistically significant (*B* = 0.02, *t* = 4.69, *p* < 0.001, 95%CI: 0.012–0.030).

**Table 3.** Multiple mediation model predicting DEBQe from SGABS. Total effect: R<sup>2</sup> = 0.181, F(3,235) = 15.98, *p* < 0.0001.


STAI: State-Trait Anxiety Inventory. SDS: Self-report Depression Scale. DEBQe: Dutch Eating Behavior Questionnaire Emotional Eating. SGABS: Shortened General Attitude Belief Scale.

**Figure 3.** The relationship between irrational beliefs and emotional eating is mediated by trait anxiety but not depression. Solid arrows indicate statistically significant regression coefficients.

### *3.2. Exploratory Analyses: Serial Mediation of the E*ff*ect of Irrational Beliefs on FA*

The results presented above indicate a strong relationship between irrational beliefs and depression and trait anxiety but neither predicted FA. Because irrational beliefs are associated with emotional eating via trait anxiety and anxiety and depressed mood have been associated with emotional eating, an additional serial mediation analysis was conducted. Based on the results of the planned correlational and mediation analyses presented above, the hypothesis that the indirect pathway between irrational beliefs and FA would include trait anxiety and depression was examined. Specifically, it was proposed that irrational beliefs would increase trait anxiety which, in turn, would increase depression. Depression and/or trait anxiety was expected to increase emotional eating which, in turn, would increase higher food addiction. BMI was included as a covariate in the model because of its correlation with FA. The results of this analysis (PROCESS model 6; see Table 4) largely supported this hypothesis with the exception that depression was not associated with elevated emotional eating. As depicted in Figure 4, there was a significant indirect effect of irrational beliefs on food addiction through trait anxiety and emotional eating (*B* = 0.02; 95%CI: 0.007–0.028). While trait anxiety was a significant predictor of higher depression score (*B* = 0.65, *t* = 15.56, *p* < 0.001; 95%CI: 0.568–0.732), the indirect path including depression was not statistically significant (*B* = 0.00; 95%CI: −0.006–0.007).


**Table 4.** Serial mediation analysis predicting mYFAS2 from SGABS. Total effect: R2 = 0.155, F(2,236) = 16.93, *p* < 0.0001.

STAI: State-Trait Anxiety Inventory. SGABS: Shortened General Attitude Belief Scale. SDS: Self-report Depression Scale. DEBQe: Dutch Eating Behavior Questionnaire Emotional Eating. mYFAS2: Modified Food Addiction Scale 2.0.

**Figure 4.** The relationship between irrational beliefs and FA is mediated by the serial pathway through trait anxiety and emotional eating. Solid arrows indicate statistically significant regression coefficients. Nonsignificant BMI effects on mediators are not included for simplicity.

### **4. Discussion**

The results show that, as hypothesized, FA and emotional eating were each positively associated with irrational beliefs. The results of this study are consistent with cognitive behavioral theory and confirm previous findings using different measures of both irrational beliefs and psychopathology that irrational beliefs are associated with elevated trait anxiety and depression. While irrational beliefs did predict higher trait anxiety, depression, and emotional eating, only emotional eating mediated the effect of irrational beliefs on FA. This mediated effect was the same across values of BMI; thus, contrary to prediction, we were unable to show that BMI moderated the mediation of the effect of irrational beliefs on FA by emotional eating. The results also confirmed that the effect of irrational beliefs on emotional eating was mediated by trait anxiety. These findings suggested examination of a serial mediation which found that the indirect effect of irrational beliefs on FA also included trait anxiety. That is, the only significant pathway indicated that irrational beliefs increased trait anxiety which, in turn, increased emotional eating, which finally led to higher number of FA symptoms. The findings that trait anxiety and not depression mediated the associations between irrational beliefs and FA and emotional eating are consistent with the suggestion that anxiety is more important than depression of symptoms of problem eating [see 29]. Finally, we have confirmed that irrational beliefs were positively associated with restrained eating (using a measure other than the RRS) and have found that irrational beliefs were positively correlated with external eating (eating in the presence of food) both of which have been implicated in problem eating and control of body weight. The role of irrational beliefs in restraint and external eating warrants additional exploration as these relationships may be mediated by anxiety or depression.

The research method employed does not allow for causal relationships to be determined with certainty. However, mediation analysis depends upon a theory of causality in order to determine the order of placement of variables in statistical models. In the present analysis, the assumptions were that irrational beliefs are the cause of the psychopathologies and coping behaviors measured because, in CBT, irrational beliefs are considered prime sources of psychopathology. In traditional psychotherapy, it was often believed that activating events (i.e., negative occurrences) result in emotional consequences (i.e., psychological distress). However, according to Ellis [47], this theory is inaccurate or at best incomplete. In the research that led to the development of Rational Emotive Behavior Therapy (REBT), Ellis found that the emotional consequences of an activating event were primarily dictated by the belief a person holds about the activating event. For example, if an employee is reprimanded by an employer for a minor mistake, she or he could think "I am incompetent and will surely be fired

from my job, and then no one will ever want to hire me!" The emotional consequence of such a belief would most likely be a significant level of anxiety. If that same person had the more rational thought "It is unfortunate that I made a mistake, but I am human so it will happen sometimes. It is highly unlikely that I will be fired if I make a minor mistake once in a while.", the emotional consequence would likely be one of mild concern or annoyance. Hence, one's emotional state is usually the result of how he or she interprets an event, rather than the event itself. Ellis [47] found that individuals who frequently interpret reality from a distorted, or irrational, perspective are likely to have anxiety or depressive disorders. He also found that when people are emotionally disturbed, they seek out ways to cope with the distress. Coping mechanisms can be adaptive (i.e., make appropriate changes, practice acceptance, exercise, etc.) or maladaptive (i.e., substance use, self-harm, uncontrolled eating). Indeed, some persons may eat to regulate mood and escape anxiety [48,49].

Irrational beliefs-based uncontrolled eating may not necessarily lead to weight gain; in the present study, irrational beliefs were not correlated with BMI. The research on the relationship between irrational beliefs and alcohol consumption indicates that irrational beliefs are associated with problems with alcohol use and not amount of alcohol used [22,23] or frequency of use or getting drunk [24]. Furthermore, perceived lack of control over alcohol use is correlated with irrational beliefs [24]. This is interesting in relation to FA as, unlike alcohol consumption, everyone needs to eat food but those with FA constitute a subset of people who have problem eating who often feel that they cannot control eating. Indeed, lack of control over eating is the most commonly reported FA symptom [50]. Irrational beliefs may lead to FA; FA is common in those with high BMI [7]. It is important to determine whether irrational beliefs are associated with energy consumption from food which may lead to higher BMI in some people. Recent findings suggest that psychological distress is associated with elevated BMI via higher FA and emotional eating [51]. Irrational beliefs may be a source of that psychological distress.

This study has several limitations. While mediation models often use causal wording (i.e., direct and indirect "effects"), the results are correlational and the direction of effect speculative. The sample is composed mostly of students who report having normal body weight. While students are of interest in the study of anxiety and other psychopathology due to high rates of anxiety, depression [52,53] and problem eating [54], their overrepresentation in the present study may limit the generalizability of the findings. Furthermore, while significant, the conditional effects are somewhat weak which may be due to the relatively low percentage of people with BMI greater than 25. In addition, while nearly a fifth of the sample meets the criterion for FA, the number of symptoms is rather low in the sample as a whole. Given the association of emotional eating and FA in persons with high BMI, the relationships reported here would be expected to be higher in those with high BMI.

### **5. Conclusions**

In conclusion, these results suggest for the first time that irrational beliefs may underlie problem eating such as emotional eating and FA via trait anxiety. Additional research may consider whether anxiety sensitivity, the fear that somatic arousal leads to catastrophic consequences, more strongly mediates the relationship between irrational beliefs and FA than trait anxiety. Anxiety sensitivity is distinct from trait anxiety and has been related to substance misuse and emotional eating, particularly in those with high BMI [55]. The results of the present study also suggest that irrational beliefs may be an appropriate target for clinicians when treating problem eating; CBT is an effective approach in treatment of addictive behaviors and problem eating [56].

**Author Contributions:** Conceptualization, L.J.N. and S.M.J.; Formal analysis, L.J.N.; Investigation, L.J.N.; Methodology, L.J.N. and S.M.J.; Project administration, L.J.N.; Visualization, L.J.N.; Writing—original draft, L.J.N.; Writing—review and editing, L.J.N. and S.M.J.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review*
