Introduction

In the last decade, the construct of food addiction (FA) has gained increasing attention from researchers and clinicians as a dysfunctional eating pattern frequently associated with obesity and eating disorders (EDs) [1, 2]. Indeed, although FA is frequently observed also in normal-weight individuals (about 11%) [1], it is more commonly diagnosed in patients with obesity (around 25%) [1] and with EDs, especially in individuals with binge eating disorder (rates range from 5.9 to 87.2%) and bulimia nervosa (rates range from 5.3 to 89.1%) [3].

Although many controversies have emerged regarding both the definition of FA and its nosological status [3, 4], the most common approach to conceptualize FA derives from the overlap with the latest editions of the Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria [5, 6] for addictive disorders (e.g., craving, tolerance, withdrawal symptoms, etc.). The Yale Food Addiction Scale [YFAS; 7] has become the most widely used psychometric tool to assess FA in both clinical and non-clinical samples [1, 3]. The YFAS was originally developed in accordance with the DSM-IV-TR diagnostic criteria [5] for substance addiction. It was initially validated in U.S. undergraduate students, showing a single-factor structure and satisfactory psychometric properties [7]. Specifically, the internal reliability for the single factor was adequate (i.e., Kuder–Richardson α = 0.75) and the convergent validity with eating pathology (i.e., emotional eating and eating troubles) was satisfactory for both the symptom count and diagnostic version of the YFAS (i.e., Pearson correlation coefficient ranged from 0.46 to 0.61) [7]. Similarly, discriminant validity with other addictive related behaviors (i.e., alcohol-related problems) and incremental validity in predicting binge-eating behavior (i.e., b = 0.48, p < 0.001) was also adequate [7]. Good psychometric properties such as the one-factor structure, adequate internal consistency and convergent validity with eating pathology were also reported in clinical samples, such as patients with obesity and EDs [8,9,10,11], and in international studies [9, 11,12,13,14,15,16,17,18] (for a detailed overview about the psychometric properties of the YFAS and its adaptations, see [19]).

Following the changes in the section of the DMS-5 [6] related to substance-related and addictive disorders (e.g., the inclusion of craving and the use of a diagnostic continuum of severity) a revised version of the YFAS, the YFAS 2.0, was recently proposed [20]. The YFAS 2.0 is composed of 35 items rated on an eight-point Likert scale (from 0 = never to 7 = every day), and it includes two scoring options [20]: (i) a symptom-count version, assessing the number of diagnostic criteria met (from 0 to 11) and (ii) a categorical diagnostic version (FA is diagnosed when at least two symptoms and clinical impairment/distress from eating are present). Furthermore, the new scale provides a diagnostic continuum of severity according to three cutoffs [20]: mild (two to three symptoms), moderate (four to five symptoms), and severe (six or more symptoms).

The YFAS 2.0 was initially validated in the U.S. population, with a single-factor structure revealed through a confirmatory factor analysis (CFA). Although the root mean square error of approximation (RMSEA) suggested a less than optimal fit (i.e., 0.108), the confirmatory fit index (CFI) and the Tucker Lewis Index (TLI), respectively 0.96 and 0.97, suggested a good fit for the one-factor model. Furthermore, a satisfactory internal consistency (i.e., Kuder–Richardson α = 0.90) and good convergent validity with problematic eating behaviors (e.g., frequency of binge eating episodes) were reported (i.e., Pearson correlation coefficient ranged from 0.24 to 0.63) [20]. The one-factor structure and the good psychometric properties were recently replicated in clinical samples [21, 22] and in several European countries, including Spain [21], Italy [23], France [24], and Germany [22], although some cross-cultural differences have been observed in the prevalence of a diagnosis of FA. For example, the prevalence rate observed in US adults was 13.1% [25], similar to those reported in several European countries (i.e., France and Germany) [22, 24]. However, Italy, Spain (i.e., 3.3%), and some South American countries, such as Brazil (i.e., 4.3%), had lower prevalence rates [21, 26].

Based on the original version [27], a brief version of the YFAS 2.0, the so-called modified Yale Food Addiction Scale 2.0 (mYFAS 2.0) has been recently developed [25]. The mYFAS 2.0 consists of 11 of the diagnostic items (those with the highest factor loadings) from the original YFAS 2.0, and two impairment/distress questions. The scale is characterized by the same main features of the original YFAS 2.0 (i.e., two scoring options and a diagnostic continuum of severity), and all items are rated on an eight-point Likert scale (from 0 = never to 7 = every day). In the original validation study on 213 U.S. adults, the mYFAS 2.0 showed a single-factor structure (CFI = 0.96, TLI = 0.95, RMSEA = 0.08), revealed through an exploratory factor analysis (EFA). The mYFAS 2.0 also exhibited good internal consistency (i.e., Kuder–Richardson α = 0.86), as well as good convergent validity (i.e., a strong association with the frequency of binge-eating episodes). The one-factor structure (e.g., CFI = 0.93, RMSEA = 0.07) and the good psychometric properties (e.g., Cronbach’s α = 0.89) of the mYFAS 2.0 have been recently replicated in a large sample of Brazilian adults [26].

International research on the psychometric properties of the mYFAS 2.0 is still scarce, and, to the best of our knowledge, no previous studies have been conducted on the Italian population. Evaluating the psychometric properties of the mYFAS is considered an import topic in international FA research [25]. Indeed, the use of a brief measure for screening FA may be useful when time constraints impede the use of the original 35-item version, such as in large epidemiological cohort studies [25, 27]. Furthermore, the mYFAS could be also useful when investigating similarities and differences in eating behavior across large samples from different countries [28].

Therefore, the aims of the present study were to adapt an Italian version of the mYFAS 2.0 and to investigate its dimensionality and psychometric properties in a non-clinical sample. Furthermore, convergent and discriminant validities were assessed by administering measures assessing eating and emotional psychopathology (e.g., binge eating severity and anxiety and depressive symptoms), and distinct eating-related problems (i.e., dietary restraint), respectively. Finally, the incremental validity was investigated by the ability of the mYFAS 2.0 to predict binge eating severity over emotion dysregulation and psychopathology (i.e., depressive and anxiety symptoms).

Materials and methods

Participants

Two hundred and sixty-six participants of both sexes were recruited through advertisements (i.e., flyers, newspaper and online ads) posted for established community groups, such as universities, hospitals, shopping centers, churches, and gyms in Central Italy.

Inclusion criteria were age of 18 and higher (in line with previous studies on the mYFAS 2.0 [25, 26]) and the ability to understand written Italian. All subjects voluntarily participated in the study (i.e., they did not receive payment or other compensation). Participants were assessed between October 2017 and July 2018. The questionnaires were individually delivered to each participant and completed in the presence of a researcher who helped them fill out the questionnaires if questions arose. All participants also received information about the general purpose of the research and provided written consent. The study was approved by the European University’s ethics review board and was performed according to the Helsinki declaration standards.

Measures

All participants were administered the mYFAS 2.0 [25], the Binge Eating Scale (BES) [29], the Hospital Anxiety and Depression Scale (HADS) [30], and the Emotion Dysregulation Scale-short version (EDS-short) [31]. Furthermore, participants were also asked to complete a checklist assessing socio-demographic variables (e.g., sex, age, marital status, job, educational attainment) and clinical variables (e.g., tobacco and alcohol use in the last 6 months, use of recreational drugs in the last 6 months, dietary restraint in the last month) using dichotomous questions (Yes/No). Height and weight were also self-reported to calculate the body mass index (BMI).

Food addiction

The mYFAS 2.0 [25] is the short version of the YFAS 2.0 [20]. It is composed of 13 items, rated on an eight-point Likert scale (from 0 = never to 7 = every day) assessing addictive eating behaviors (e.g., “I had such strong urges to eat certain foods that I could not think of anything else”; “I ate to the point where I felt physically ill”). The mYFAS 2.0. provides two scoring options: a symptom count version (scores ranging from 0 to 11) and a diagnostic version based on the last edition of the DSM criteria [6] for substance-related and addictive disorder section (i.e., a categorical cut off diagnostic is met when at least two symptoms and clinically significant impairment/distress from eating are present).

In the 35-item version, each symptom is scored based on two or more theoretically related items [20]. In the short version, each symptom is scored based on one item with the exception of the “impairment and distress” criterion, which is assessed by two items. For the “diagnosis” scoring option, a participant can meet the criteria for mild FA (two to three symptoms), moderate FA (four to five symptoms) or severe FA (six or more symptoms). The mYFAS 2.0 items were translated as part of a research project whose objective was to create an Italian validated version of the YFAS 2.0 long-form. During the early phases of this project, two bilingual researchers, under the supervision of two coauthors (GMM and GC) of the present paper, adapted the Italian version of the questionnaire from the original English version using a back-translation procedure.

Binge eating severity

The BES is a 16-item self-report scale assessing behavioral (e.g., “I don’t have any difficulty eating slowly in the proper manner”) and cognitive/emotional (e.g., “Because I feel so helpless about controlling my eating, I have become very desperate about trying to get in control”) manifestations of binge eating [29]. The scores range from 0 to 46, with higher scores indicating greater binge eating severity. Marcus et al. [32] identified three different levels of severity: (1) non-clinical level of binge eating (i.e., total scores ranging from 0 to 17); (2) moderate level of binge eating (i.e., total scores ranging from 18 to 26); (3) and severe binge eating (i.e., total scores ranging from 27 to 46). Following Ricca et al., a cut-off value of 17 was used in the present study to discriminate between subjects with and without a clinical level of binge eating [33]. The psychometric properties of the BES have been investigated with satisfactory results in several countries [33,34,35,36,37]. In the present study, the Italian version of this scale was used [37] and Cronbach’s α was 0.88.

Anxiety and depressive symptoms

The HADS [30] is a 14-item self-report questionnaire assessing both anxiety (e.g., “Worrying thoughts go through my mind”) and depressive (e.g., “I feel as if I am slowed down”) symptoms in clinical and non-clinical samples [38]. The anxiety (HADS-A) and depression (HADS-D) subscales are composed of seven items each, which are rated on a four-point scale (0–3). Total scores range from 0 to 21 for both subscales, with higher scores indicating more severe depressive and anxiety symptoms. The psychometric properties of the HADS have been investigated with satisfactory results [39]. In the present study, we used the Italian version [40] of this scale, and Cronbach’s α in our sample was 0.82 and 0.70 for the anxiety and depression subscales, respectively.

Emotion dysregulation

The EDS-short [31] is an unidimensional self-administered questionnaire assessing behavioral and cognitive/emotional manifestations of emotion regulation (e.g., “Emotions overwhelm me”; “I have trouble soothing myself when I am upset”). It is composed of 12 items rated on a 7-point Likert scale (from 1 = not true to 7 = very true) [31]. Total scores range from 12 to 84, with higher scores indicating greater problems with emotion regulation. Good psychometric properties were reported in the original validation study [31]. Cronbach’s α for the present sample was 0.91.

Statistical analyses

As recommended [41], missing values were replaced with the individual’s mean for the relevant total scale/subscale for protocols with two missing items or fewer (seven participants). Four protocols with three or more missing items were excluded from the analyses.

All the analyses were performed with the Statistical Package for the Social Sciences (SPSS) 19.0 for Windows, and Mplus 7.0 (Muthén and Muthén, 1998–2010). The Bayesian confirmatory factor analysis (BCFA) approach was used to investigate the uni-dimensionality of the mYFAS. We applied a BCFA using a Markov chain Monte Carlo (MCMC) algorithm to investigate whether the one-factor model fits the structure of the mYFAS 2.0 with all the eleven symptomatic dimensions included in the original version loading significantly on the latent factor. In this analysis, we used informative priors for items loadings based on the results from the original validation study [25]. A sensitivity analysis of the prior distribution was assessed comparing different levels of certainty about the values of the factor loadings. We let the priors variance vary between 0.05 (less certainty) and 0.001 (more certainty), and also re-ran the model with no informative priors (all priors set to 0.00 ± 5.00). The model fit was evaluated using the Bayesian Posterior Predictive Checking (PPC) and the Posterior Predictive P value (PPP) [42]. BCFA simulates replicated data under the model of interest, and PPC compares the proportion of iterations for which the replicated χ2 exceeds the observed χ2. The fit of the model was based on PPC confidence interval crossing the zero and PPP > 0.05. The deviance information criterion (DIC) was not used to compare models because it was not available in MPLUS when using categorical variables, and the sensitivity analysis used only inspection of PPC confidence intervals and PPP. For each variable, we reported factor loadings, thresholds and R2, and their 95% Bayesian Credibility Intervals (95% BCI). The BCI can be interpreted as the probability that the population parameter is between the upper and lower bounds [43]. The Bayesian approach is preferred over the classical approach because it can incorporate previous knowledge into the analyses, such as the informative priors which contain numerical information that influences the final parameter estimate [43]. Furthermore, BCFA results could be more reliable with small samples, especially when it is possible to incorporate very informative priors in the models, as was possible in our study [43].

As a measure of reliability, we reported ordinal Cronbach α [44]. Convergent validity with the BES, BMI, emotion regulation and both anxiety and depressive symptoms was evaluated by calculating Pearson’s r correlation coefficients. As in previous studies [20, 22, 25], discriminant validity was assessed using the Gamma coefficient correlations between mYFAS 2.0 total score and dietary restraint, defined as “the intention to restrict food for weight loss purposes” [20]. Finally, because of the strong association between FA and binge eating behaviors [3, 45], and as in the original validation studies of the YFAS [7], YFAS 2.0 [20] and mYFAS 2.0 [25], the incremental validity of the mYFAS 2.0 in predicting BES total score over emotion dysregulation and psychopathology (i.e., depressive and anxiety symptoms) was evaluated by means of a hierarchical linear regression analysis.

Results

The final sample consisted of 262 Italian individuals (184 women and 78 men; mean age = 29.43, SD = 13.55 years; range 18–88). Participants had an average self-reported BMI of 22.45 kg/m2 (SD = 3.12: range 16.30–33.20). According to the standard BMI cut-off [46], there were 18 (6.9%) underweight participants (i.e., BMI < 18.5 kg/m2), 193 (73.7%) normal weight participants (i.e., BMI between 18.50 and 24.99 kg/m2), 45 (17.2%) overweight participants (i.e., BMI between 25 and 29.9 kg/m2), and 6 (2.3%) obese participants (i.e., BMI ≥ 30 kg/m2) participants. There were 15 participants (5.7%) who met the criteria for a diagnosis of FA according to the mYFAS 2.0, 26 (9.9%) who met the criteria for clinical-level binge eating (BES > 17), and 11 (4.2%) who met both criteria. Of those who met FA criteria, 73.3% (N = 11) also satisfied criteria for clinical-level binge eating, whereas, of patients with clinical-level binge eating, 42.3% (N = 12) also satisfied the criteria for FA. The clinical and socio-demographic characteristics of the sample are reported in Table 1.

Table 1 Descriptive statistics for the sample (N = 262)
Full size table

Dimensionality of the mYFAS 2.0

Table 2 reports PPC and PPP of the unidimensional model with different certainty about priors. A nonsignificant PPP indicated the fit of the model, while the sensitivity analysis suggested a prior data conflict, with more stringent prior variance being associated with wider Bayesian credibility intervals. The model with no informative priors had more stringent credibility intervals than other models. When comparing factor loadings for this model with the model with very informative priors (priors variance = 0.001), the factor loading estimates were lower for Time (0.345 vs. 0.596) and Situations (0.539 vs. 0.693) and higher for all the other variables. Despite the fact that the estimate of the factor loading was significant (posterior p = 0.039) (Table 3), credibility intervals for the variable Time crossed zero (95% BCI = − 0.034/0.664), indicating a high variability of the estimate with potential low correlation with the latent trait.

Table 2 Bayesian Posterior Predictive Checking (PPC) and Posterior Predictive P value (PPP) for the competing models
Full size table
Table 3 Standardized factor loadings, thresholds and R2, and their 95% Bayesian credibility intervals (95% BCI)
Full size table

Psychometric properties of the mYFAS

Ordinal α for the mYFAS 2.0 was satisfactory (α = 0.91), with corrected item-total correlations ranging between 0.48 for Time and 0.92 for Consequences. Convergent validity was also satisfactory. The mYFAS 2.0 total score (i.e., symptom count) was strongly and positively related to the BES total score (r = 0.67, p < 0.001). The mYFAS 2.0 total score was also positively associated with the EDS-short score (r = 0.35, p < 0.001), and with both anxiety (r = 0.31, p < 0.001) and depressive (r = 0.35, p < 0.001) symptoms. A small but significant positive correlation between mYFAS 2.0 total score and BMI (r = 0.16, p = 0.01) was also identified. Discriminant validity with dietary restraint was also good (Gamma = 0.11; p = 0.52).

In a hierarchical linear regression model, the mYFAS 2.0 total score (b = 0.52; t = 11.11; p < 0.001), EDS-short total score (b = 0.21; t = 3.82; p < 0.001) and HADS-D total score (b = 0.19; t = 3.48; p < 0.001) were independently associated with the BES total score. In the first block, anxiety symptoms, depressive symptoms and emotion dysregulation explained 32.0% of the variability of the BES total score (F = 39.98, p < 0.001). In the second block, when the mYFAS 2.0 was added, the model explained 52% of the variability of the data (F = 75.64, p < 0.001; F Change = 123.39, p < 0.001).

Discussion

The aim of the present study was to assess the dimensionality and psychometric properties of the mYFAS 2.0 in an Italian non-clinical sample. Consistent with previous studies [25, 26] our data support a single-factor solution, although credibility intervals for the variable “Time” suggested a high variability of the estimate with a potential low correlation with the latent trait.

Nevertheless, our results showed that the psychometric properties of the Italian mYFAS 2.0 was satisfactory. In particular, consistent with previous reports investigating the psychometric properties of both the original [20,21,22,23, 25] and the short version of the YFAS 2.0 [25, 26], good internal consistency reliability was reported. Specifically, as reported elsewhere [25], our results suggest that the mYFAS 2.0 and full YFAS 2.0 performed similarly with respect to reliability and validity. For example, compared to the Italian version of the YFAS 2.0 [23], the reduced version showed a comparable reliability (α = 0.87 and α = 0.91, respectively).

Convergent validity with problematic eating behaviors (i.e., binge eating severity) and difficulties in emotion regulation was also good [20,21,22,23, 25, 26]. Furthermore, it is interesting to note that a small but significant association was observed between the mYFAS 2.0 score and self-reported BMI, which may be consistent with the hypothesis of a non-linear relationship between FA and BMI [47]. Another possibility is that the low number of obese participants in the present sample, and consequently a small BMI range, attenuates the correlation with the mYFAS 2.0 total score. Nevertheless, the strong association between the mYFAS 2.0 and binge eating severity, as well as the small but significant association with BMI, further confirm that the mYFAS 2.0 has similar convergent validity as the full YFAS 2.0. Indeed, Aloi and colleagues [23], reported that the Italian version of the YFAS 2.0 was strongly related to the BES total score (r = 0.69) and significantly associated with self-reported BMI (r = 0.17), although the effect size was small.

Finally, both discriminant validity with dietary restraint and incremental validity in predicting the BES total score over emotion dysregulation and psychopathology (i.e., depressive and anxiety symptoms) were also confirmed [20, 22, 25]. In line with previous findings [10, 45, 48,49,50], our results (i.e., the high prevalence of clinical binge eating in individuals with FA as well as the strong correlation between the mYFAS 2.0 and binge eating severity) seem to suggest that both the behavioral and emotional/cognitive features of uncontrolled eating could be considered key psychopathological elements of FA [45].

In the present sample, according to the mYFAS 2.0 diagnostic score, 5.7% of the study participants received a FA diagnosis, which is similar (i.e., 3.4%) to that reported using the YFAS 2.0 with Italian undergraduate students [23], but it is lower when compared to the prevalence rates found in other countries [20, 22, 24, 25]. For example, the prevalence rate observed in US adults was 13.1% [25], and in French and German samples 8.2% [24] and 9.7% [22], respectively. Instead, similar to the Italian figure, the prevalence rates were lower in non-clinical samples from other Latin countries, such as Brazil (i.e., 4.3%) and Spain (i.e., 3.3%) [21, 26], suggesting, as reported [51, 52] for other addictive eating behaviors (i.e., food craving), possible cross-cultural differences [2]. Although this interpretation remains speculative, it might be useful as a future research topic.

The study’s contributions should be considered in light of some limitations. First, we did not investigate the psychometric properties of the mYFAS 2.0 in patients with EDs and/or with obesity, which make our interpretations specific to non-clinical individuals. Furthermore, in our sample there was a high proportion of female participants, young adults and normal weight subjects, limiting the generalizability of these findings to other populations (e.g., older adults and obese individuals). For example, the low number of male participants did not allow us to test the structural invariance of the mYFAS 2.0 by sex. This could be relevant because previous research has found sex differences in addictive eating behaviors, such as food craving [53]. Therefore, further studies should be conducted to investigate the dimensionality of the mYFAS 2.0 across different groups, especially in males and females, as well as in obese and normal weight individuals. Third, we did not assess the stability of mYFAS 2.0 over time and its predictive validity for future disordered eating behaviors. Therefore, further longitudinal research using a test–retest procedure in large clinical samples should be conducted. Finally, our data are all self-reported. This could be particularly problematic in relation to the BMI. Indeed, although the self-reported measures of height and weight are widely used in studies investigating FA in non-clinical samples [23, 25, 45, 54,55,56,57], their validity is still debated [58,59,60,61,62].

Despite these limitations, to the best of our knowledge, this is one of the few international studies ever conducted, and the first study conducted in Italy, which investigated the dimensionality and psychometric properties of the mYFAS 2.0. Taken together, our results confirm that the Italian mYFAS 2.0 had satisfactory psychometric properties, performing similarly on measures of reliability and validity as the full YFAS 2.0. Therefore, our data suggest that the mYFAS 2.0 may be a useful brief instrument for the assessment of addictive eating behaviors when time constraints impede the use of the original YFAS 2.0 (e.g., large epidemiological cohorts studies). Furthermore, our study has also some clinical implications. FA is commonly diagnosed in both patients with obesity [1] and with EDs [3], and it is known that the co-occurrence of FA and other dysfunctional eating patterns is associated with worse clinical conditions (e.g., BMI) and psychopathological symptoms [55, 63]. Therefore, it is important for clinicians to screen obese and EDs patients for FA using a well-validated instrument. Identifying patients with addictive eating symptoms could help improve effectiveness of psychological treatments [64].