Disordered Eating Behaviors Related to Food Addiction/Eating Addiction in Inpatients with Obesity and the General Population: The Italian Version of the Addiction-like Eating Behaviors Scale (AEBS-IT)

Purpose. The purpose of this research is to test the psychometric properties and factorial structure of the Addiction-like Eating Behaviors Scale (AEBS) in an Italian sample of adults with severe obesity seeking treatment for weight reduction and the general population, and to examine the measurement invariance of the tool by comparing a clinical and a nonclinical sample. Methods. A confirmatory factor analysis (CFA) was initially conducted to test the factorial structure of the Italian version of the AEBS (AEBS-IT) on a total of 953 participants. Following this, the measurement invariance and psychometric properties of the tool AEBS-IT were assessed on both inpatients with severe obesity (n = 502) and individuals from the general population (n = 451). Reliability and convergent validity analysis were also run. Results. CFA revealed a bi-factor structure for the AEBS-IT, which also showed good reliability and positive correlations with food addiction (through the mYFAS2.0 symptom count), binge-eating symptoms, compulsive eating behavior, and dysfunctional eating patterns and the individuals’ body mass index (BMI). Moreover, the tool was invariant across populations. Conclusion. This study provided evidence that the AEBS-IT is a valid and reliable measure of FA in both clinical and nonclinical samples.


Introduction
Obesity is on the rise, and it is predicted to rise to nearly half the global population by 2030. Recent statistics suggested that 50% of men and 55% of women, globally, are overweight or obese [1]. This is problematic given the adverse health and social factors associated with obesity, such as decreased quality of life and increased morbidity and mortality [2,3].
One of the central causes of obesity is an energy imbalance of calories consumed and expended [1]. However, the onset and development of obesity are complex and include several factors such as psychological, sociological, biological, evolutionary, economic, and institutional factors [4,5]. Obesity is a preventable and treatable condition, with strategies frequently focusing on reducing sedentary lifestyles and improving dietary intake, however, long-term success is limited, and relapse is frequent [6]. Therefore-to inform clinical interventions and contribute to health prevention strategies-it is important to concentrate on other explanations that may contribute to the behavioral phenomenon of food overconsumption. The last two decades have been characterized by increased interest in the construct of Food Addiction (FA), a substance-use disorder (SUD) characterized by The BES [41,42] is a questionnaire measuring binge-eating severity in both community [10] and clinical samples [43]. It comprises 16 items: eight items that describe behavioral manifestations of BED (i.e., eating fast or consuming large amounts of food) and eight items focused on associated feelings and cognitions (i.e., fear of not stopping eating).
Each question has 3-4 separate responses increasing in severity grouped into two subscales (FC-Feelings/Cognitions; and B-Behaviors) and a total score [43]. Assigning to each statement a numerical value ranging from 0 to 3 points (0 = no severity in BED symptoms, 3 = serious problems in terms of BED symptoms), the BES total score ranges from 0 to 46: a score of less than 17 points indicates minimal BE problems; a score between 18 and 26 points indicates moderate BED; and a score of more than 27 points indicates severe BED [44].
Studies carried out in the past decade, mainly with individuals with obesity, have shown that the BES is very sensitive and specific in distinguishing between compulsive and normal eaters [45,46], and a large number of investigations support its reliability and validity as a measure of eating-related pathology [41,42,47,48].
In this study, the BES showed adequate internal consistency. Indeed, Cronbach's alphas were 0.901, 0.816, and 0.835 for the BES Total scale, the FC scale, and the B scale, respectively.

The Measure of Eating Compulsivity-Italian Version (MEC10-IT)
The MEC10-IT [49,50] is a measure of compulsive eating within the FA framework. It comprises 10 items scored on a 5-point Likert scale with partial semantic autonomy (from 0 = "Very Untrue" to 4 = "Very True") and higher scores corresponding to a higher eating compulsivity. In its original validation study [49], the MEC10 displayed adequate reliability (α = 0.946). Similarly, in the present study, the internal consistency of the tool was adequate, with Cronbach's alpha equal to 0.941.

The Dutch Eating Behavior Questionnaire (DEBQ)
The DEBQ [51,52] is a questionnaire measuring behaviors and attitudes related to ED in both general [10,53] and clinical populations [54]. It comprises 33 items scored on a 5-point Likert scale (ranging from 1 = "never" to 5 = "very often") and grouped into three subscales: Restrained Eating (RE), Emotional Eating (EE), and External Eating (ExE), plus a total score. The DEBQ has been shown to be an adequately reliable and valid measure of eating-related pathology with a strong three-factor structure, high internal consistency, and high test-retest reliability after 4 weeks [51][52][53][54]. In this study, Cronbach's alphas were 0.934, 0.929, 0.969, and 0.835 for the total scale, the RE scale, the EE scale, and the ExE scale, respectively.
According to its original validation study [21], a first-order model comprising twofactor was specified: 9 items loaded onto the 'appetite drive' latent factor (from item#1 to item#9) and 6 items filled into the 'low dietary control' latent factor (from item#10 to item#15). Moreover, considering that the original validation study assumed a 'general factor' (i.e., total score), two alternative models were additionally tested. First, a single factor model was verified: all the 15 items of the AEBS-IT loaded onto the latent factor 'Addiction-like eating behavior' (general factor, only). Second, a bi-factor structure was set: 9 items loaded onto the 'appetite drive' latent factor (from item#1 to item#9), 6 items filled in the 'low dietary control' latent factor (from item#10 to item#15), and all the 15 items loaded onto an overall factor named 'Addiction-like eating behavior' (total score).
To select the best factorial structure (namely, the best model), a model comparison analysis was performed. The following set of criteria and cutoffs for model equality were employed: DIFFTEST (equal to ∆χ 2 ; p-value > 0.050) and ∆CFI (<0.010). Considering the great sensibility to the sample size of the χ 2 , the overpass of the ∆CFI cut-off criteria was considered to be evidence of model inadequateness-combined with worse fit indices [67,[72][73][74][75].
These nested models were sequentially compared. Model calculations were done by means of the aforementioned test differences: DIFFTEST and ∆CFI. Additionally, the same rule-of-thumb for model inadequateness was applied.
Moreover, the adjusted item-total correlation was calculated [81][82][83]. In addition, item discriminant power (IDP) was performed to assess the capacity of the items to differentiate between individuals with a low or high level of the construct that has been measured [84,85]. For each subject, the total score and its quartile rank were first computed. Subsequently, independent sample t-tests-and related effect sizes (Cohen's d) [86]-were determined to evaluate IDP by means of the scale-total score as the dependent variable and its lowest and highest quartile as the grouping variable [84,85].
Moreover, a Receiver Operating Characteristics (ROC) curve was run to evaluate the accuracy of the general dimension of the AEBS-IT ('addiction-like eating behavioral' scale) to differentiate between (A) individuals without FA and individuals with FA as well as (B) individuals without BED and individuals with BED [87,88]. The global accuracy-validity of the AEBS-IT was estimated with the area under the ROC curve (AUC; 5000 stratified bootstrap resamples). The AUC was interpreted with Swets' benchmarks: AUC = 0.50, null; AUC from 0.51 to 0.70, small; AUC from 0.71 to 0.90, moderate; AUC from 0.91 to 0.99, large; and AUC = 1.00, perfect accuracy [89,90]. Moreover, sensitivity (SE), specificity (SP), and accuracy (ACC) were computed for each AEBS-IT cut-off point [87,88].

Structural Validity and Model Comparisons
The first model of the AEBS-IT (two first-order latent factors) showed a non-adequate fit to the data for the two samples combined. The Chi-square statistic was statistically significant: S-Bχ 2 (89) = 2284.543; p < 0.001. The RMSEA was equal to 0.161; 90%CI: 0.155-0.167; p(RMSEA < 0.05) < 0.001. The CFI was equal to 0.945. The SRMR was equal to 0.109.
The second model of the AEBS-IT (one first-order latent factor) showed a non-adequate fit to the data for the two samples combined. The Chi-square statistic was statistically significant: S-Bχ 2 (90) = 4505.043; p < 0.001. The RMSEA was equal to 0.227; 90%CI: 0.221-0.233; p(RMSEA < 0.05) < 0.001. The CFI was equal to 0.891. The SRMR was equal to 0.152.
Model comparisons analysis showed the superiority of the bi-factor model ( Table 2). Thus, the third model was considered the best factorial structure of the AEBS-IT. Consequently, the bi-factor model was used for subsequent statistical analyses.   Notes: * = item reverse (reversed); SD = standard deviation; SK = skewness; K = kurtosis; ( . . . )sf = referred to the specific factor (item#1 to item#9: appetite drive; item#10 to item#15: low dietary control); ( . . . )gf = referred to the general factor; IDP = item discriminant power; t = t-test; d = Cohen's d; r (IT-TOT) = item-total correlation (adjusted); |λsf | = absolute value of the factor loading on the specific factor (item#1 to item#9: appetite drive; item#10 to item#15: low dietary control); |λgf | = absolute value of the factor loading on the general factor; R 2 = explained variance. Note. χ 2 = Satorra-Bentler scaled Chi-square test; df = degree of freedoms; ∆ = differences between indices; RMSEA = root mean square error of approximation; CFI = comparative fit index.

Measurement Invariance across Samples
Inpatients with severe obesity. The Chi-square statistic was statistically significant: Latent Means Invariance. The latent mean invariance model fitted the data well: χ 2 (225) = 1820.760, p < 0.001; the RMSEA = 0.122; the CFI = 0.962; and the SRMR = 0.073. The Chi-square showed a statistically significant decrease: DIFTEST (3) = 537.94; p < 0.001. Also, a statistically significant decreases in CFI (|∆CFI| = 0.013) was detected-signifying a difference in the expected latent mean of the traits of the two samples.
In addition, the adjusted item-total correlation revealed statistically significant associations between each item and their specific factor as well as the general dimension (Table 1).
Reliability analysis showed substantial outcomes: for the 'appetite drive' scale, Cronbach's alpha and McDonald's ω were equal to 0.880 and to 0.922, respectively; for the 'low dietary control' scale, Cronbach's alpha resulted equal to 0.787, and McDonald's ω resulted equal to 0.906; for the 'addiction-like eating behavior' scale (total score), Cronbach's alpha corresponded to 0.883, and McDonald's corresponded to 0.919.
Reliability analysis showed substantial outcomes: for the 'appetite drive' scale, Cronbach's alpha and McDonald's ω were equal to 0.880 and to 0.922, respectively; for the 'low dietary control' scale, Cronbach's alpha resulted equal to 0.787, and McDonald's ω resulted equal to 0.906; for the 'addiction-like eating behavior' scale (total score), Cronbach's alpha corresponded to 0.883, and McDonald's corresponded to 0.919.

Discussion
The use of the AEBS-IT is important in both research and clinical fields as it is the only tool available enabling the evaluation of behavioral addiction to eating. This study investigates, for the first time, the psychometric proprieties of this tool among the Italian population and tests its factorial structure, also in the comparison between a clinical sample of adults with obesity seeking treatment for weight reduction and the general population.
In terms of the two samples combined, results confirmed that the bi-factor model of the AEBS-IT has an excellent fit to the data-meaning the nature of the construct of behavioral addiction to eating is adequately measured with the items comprising the tool.
MI analysis also revealed this to be true in both clinical and community samples, separately. The AEBS-IT's items were equivalently related to the latent factor in each sample, and the two samples had the same expected item response at the same absolute level of the trait. These results suggest that inpatients with obesity and the general population interpreted the items in the same way (the factorial structure was equal across samples), with the same strength (items were related to the latent construct equally between the two samples), and with the same starting point (item thresholds were equal between the two samples). However, the latent trait was not equally distributed (latent means were different between the two samples). Thus, inpatients with severe obesity and the general population were perfectly comparable (equal items threshold), but with caution (different latent means) [91][92][93][94][95]. Therefore, the AEBS-IT can be employed in clinical and research practice to confront outcomes resulting from these two populations.
Moreover, IDP analysis exhibited that the 15 items of the AEBS-IT discriminated well between respondents with low and high addiction-like eating behaviors, and demonstrated the capacity of each item to signify its latent construct.
Further, reliability analyses were run, showing good results for both subscales and the AEBS-IT total score. Statistically significant positive correlations were also found between all the dimensions of the AEBS-IT, the mYFAS2.0 symptom count, the BES subscales, the MEC10-IT total score, the DEBQ factors, and the individuals' BMI-demonstrating the tool's good convergent validity.
These findings corroborate the link between the construct of addiction-like eating behaviors, FA, BED symptoms, compulsive eating, dysfunctional eating patterns, and the individuals' BMI.
Not surprisingly, the highest correlations were observed between the AEBS-IT scores, and both the mYFAS2.0 symptom count and BES scores-thus confirming the finding of the validation study carried out by Legendre et al. (2020) in both clinical and community samples among Canadian adults [26].
Further, in a recent examination of the YFAS in a clinical sample of patients with obesity with BED, a diagnosis of FA was met by 57% of patients, and a higher number of FA symptoms was related to more frequent binge-eating episodes [96,97].
Additionally, the compulsive overeating characteristic of BED has been shown to statistically predict FA diagnosis based on YFAS criteria in individuals with obesity in several investigations [98,99].
These results corroborate the partial overlap between the constructs of compulsive eating measured using the MEC10-IT, with a diagnosis of FA based on addiction criteria and BED already documented in the previous studies [26,49,100,101], and also highlight the important relationships between addiction-like eating behaviors with BED and compulsive eating patterns. Indeed, the AEBS-IT total score and its subscales showed higher correlations with all the BES dimensions and the construct of eating compulsivity measured using the MEC10-IT than those displayed by the mYFAS2.0 symptom count, while FA measured using the mYFAS2.0 and FA based on the behavioral models show a moderate correlation. The lowest correlations were, instead, observed between the AEBS-IT subscales and total score and the DEBQ restrained eating dimension. Specifically, a negative correlation was detected between the AEBS-IT low dietary control factor and the above DEBQ dimension. Accordingly, previous investigations showed small, significant negative correlations or non-significant correlation coefficients of FA based on the substance abuse model (measured with the YFAS) with dietary restraint (measured with the Three-Factor Eating Questionnaire-TFEQ or the Eating Disorder Examination-EDE) [12,96,102].
Still, for the first time, the construct of EA is measured here through the AEBS-IT, while the presence of eating disorder psychopathology was assessed using the DEBQ.
These findings further corroborate the association with mYFAS2.0 and AEBS-IT conceptualization of EA. Still, whether these conditions represent forms of compulsive eating supported by different mechanisms or represent two different facets of a unique underlying phenomenon needs to be further assessed in future studies. In fact, the heterogeneity characterizing the sample (i.e., diagnosis and demographics/clinical parameters) and the cross-sectional design of this study make it difficult to derive any conclusion or causal relation between variables.
Lastly, the ROC analyses revealed that the AEBS-IT represents a valid screening/ diagnostic tool for the detection of addiction-like eating behaviors in people with severe obesity. Indeed, it presented high accuracy (AUC = 0.819), sensitivity (0.807), and specificity (0.701) in discriminating between individuals with FA and those without FA. Similarly, the measure demonstrated to be able to successfully detect BED symptoms among adults with severe obesity seeking treatment for weight reductions-as it showed high accuracy (AUC = 0.895), sensitivity (0.885), and high specificity (0.738) in discriminating among individuals with and without BED.
In this regard, the magnitude of the effect size (AUC) suggests a substantial (but not total) overlap between food addiction, binge-eating behaviors, and eating addiction, suggesting that these three psychological constructs may be somewhat intersecting and not necessarily mutually exclusive. Indeed, food addiction-conceptualized as SUD-does not necessarily exclude overeating and binge-eating behaviors and-at the same time-does not necessarily exclude a drive toward hunger (appetitive drive) that can lead to low diet control through behavioral dependence on the act of eating.
Some limitations of this study should be highlighted. Despite the presence of a large number of subjects from the clinical population (inpatients with severe obesity), a convenience snowball sampling enrollment procedure was used for the individuals from the general population. However, MI analysis suggests that the factorial structure of AEBS is invariant across the two samples, at the level of the thresholds. Moreover, the cross-sectional research design and the use of self-report questionnaires did not allow for testing of the possible changes of the AEBS over time nor its temporal stability (e.g., longitudinal MI and test-retest reliability) and its predictive validity. Future studies may identify the recurring patterns of EA by creating latent psychological profiles. Another potential limitation is related to the variability in the age of both samples-which could lead to changes in the subjects' metabolism and, thus, their tendency to engage in EArelated behaviors. Future studies, including longitudinal ones, could control for this variable and use it as a predictor/outcome to create possible explanatory models of EA behaviors. In addition, future studies could investigate the interaction of confounding factors (e.g., age, BMI, gender, etc.) on the latent dimensions of AEBS. Lastly, future studies might also consider examining possible cross-cultural similarities and differences in the conceptualization of EA through AEBS (i.e., cross-cultural MI).
Despite these limitations, this study still has several strengths, both methodological and clinical. It is noteworthy that this is the first study validating the AEBS-IT, allowing for the assessment of EA to be conducted with accuracy and parsimony. Additionally, the relatively small number of items allows the AEBS-IT to be more easily included in longer assessment batteries, both in clinical and research practice. About the methodological strengths, the CFA revealed that the best factorial of the AEBS-IT is the one comprising two specific factors ('appetite drive' and 'low dietary control') and an overarching latent dimension ('eating addiction'). Regarding clinical strengths, the AEBS-IT has significant clinical impact and implications, as it represents a useful assessment tool for clinicians because EA seems to be a transdiagnostic construct shared by various EDs and psychological difficulties related to eating and feeding attitudes. Thus, assessing EA is important due to its multiple roles: EA may be the cause, the result, and a maintenance factor of psychological suffering and dysfunctional behaviors in EDs, and this provides useful information for both the conceptualization and treatment of clinical conditions. Moreover, the AEBS-IT provides useful information to allow for better understandings of psychological difficulties and the tailoring of specific psychological interventions.

Conclusions
Conceptualizing FA as a behavioral addiction-namely, 'eating addiction' (EA)-the AEBS-IT represents a psychometrically sound instrument able to measure the presence of addictive-like eating behavioral patterns in both clinical and nonclinical samples. Indeed, this tool demonstrated good validity and reliability in both patients with severe obesity and the community sample, and might be used by researchers and clinicians to assess FA [103]. Considering that the AEBS-IT-unlike the YFAS/YFAS2.0/mYFAS2.0-is not meant to be a diagnostic tool, its good sensitivity to clinical populations (i.e., a good capacity to detect people with FA) further supports its utilization. Additionally-besides being moderately correlated with the FA-substance-based model-it shows high associations with compulsive eating, BED, and dysfunctional eating patterns that are above those displayed by the mYFAS 2.0. This suggests that, despite the YFAS representing the most widely used measure of FA, the AEBS-IT nonetheless properly reflects the behavioral correlates of the EA phenomenon (i.e., compulsive overeating), its characteristics, and related psychiatric comorbidities. Moreover, the term "Eating Addiction" would be more appropriate for describing the behavioral phenomenon of continuous overeating of a variety of foods and avoiding the conflicting assumptions that certain food can lead to the development of a SUD.
Still, the scientific debate about "eating addiction" is in its infancy, and further studies should try to replicate these results by also employing cross-cultural designs and investigating the AEBS discrimination capability with a wider range of populations, including those with BED and other eating disorders. Institutional Review Board Statement: The study was conducted in accordance with the Declaration of Helsinki, and approved by the Ethics Committee of the IRCCS Istituto Auxologico Italiano (protocol no. 2020_02_18_04).
Informed Consent Statement: Informed consent will be obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy restrictions.

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