Introduction

Eating disorders (EDs) are serious and difficult-to-treat mental illnesses, often showing ego-syntonic features and resistance to treatments. Epidemiological studies usually underestimate the occurrence of EDs in the general population, since individuals are rarely aware of their illness and only occasionally refer to mental health care [1]. Many factors conspire to impede the treatment of EDs, including entrenched thinking, ambivalence about change, avoidant and perfectionistic personality traits, and comorbidity of trauma symptoms [2, 3].

Cognitive behavioural therapy (CBT) is widely recognized as the treatment of choice for adults with EDs [4]. Despite the widespread support for its efficacy [5, 6], therapy is often hampered by the well-known phenomenon of dropout [5].

Schema therapy (ST) is an integrative and multi-modal approach developed to address deeper levels of cognition and entrenched behaviours that do not respond to first-line treatments [7].

The goal of the ST treatment for EDs is to enable core psychological (and physiological) needs to be met [8], and to bring about change in eating habits by breaking enduring and self-defeating patterns of thinking, feeling, and behaving that typically begin early in life as a result of the interaction between temperament and unmet core emotional needs—referred to as early maladaptive schemas (EMS)—whilst developing healthy coping mechanisms [9, 10]. Indeed, research, suggests that those who suffer from EDs experience significantly higher levels of maladaptive modes than community samples [11, 12]. The ST treatment for EDs includes recognizing and challenging Internalized Critic Modes, re-parenting to heal the vulnerable child mode, and bypassing the resulting coping modes that are linked to the over-evaluation of shape, weight, and self-starvation. Limits are also set on Angry and Impulsive Child Modes that drive a self-destructive “acting out” of needs (i.e., bingeing). Cognitive and behavioural techniques are considered core aspects of ST, but the model gives equal weight to emotion-focused work and experiential techniques, in addition to the basic healing components of the therapeutic relationship. As with CBT, ST is structured, systematic and specific, following a sequence of assessment and treatment procedures. However, the pace and emphasis on aspects of treatment may vary depending on the individual needs.

To facilitate more precise measurement of mode states within the ED population, the schema mode inventory for eating disorders (SMI-ED) was recently developed, showing adequate validity and reliability [13]. Given the large number of items in the SMI-ED (n = 190)—which make it cumbersome for everyday clinical practice—the purpose of the present study was to develop a shortened Italian version of the SMI-ED, to assess its psychometric proprieties, and to determine the internal reliability of its subscales. The relationship between ED symptoms (restraint, binge eating and purging) and schema modes was also explored.

Materials and methods

Participants

The sample comprised 649 participants [181 males (27.9%) and 468 females (72.1%)] aged from 18 to 91 years (mean = 40.66, SD = 18.27). The study was open to individuals (1) aged over 18 years old, (2) who were Italian-speaking and that (3) signed digital informed consent to participate in the study. Exclusion criteria included the inability to complete the questionnaire due to visual or cognitive impairments. Participation was voluntary, and respondents did not receive remuneration.

Sample size calculation

Sample size calculation was based on two recommendations: first, that 500 or more observations can be considered “very good” for conducting a confirmatory factor analyses [14]; second, using the rule of ten subjects per item [15].

Measures

Demographics Information including age, gender, education, relationships, and employment status were collected.

Biomedical data Data on height and weight were registered and BMI was calculated as weight (kg) divided by height squared (m2). Participants were also asked to report on the presence of existing diagnosis of eating disorders through a multiple choice question (“Have you ever been diagnosed with one of the following eating disorder?”) [16].

The Italian version of the Eating Disorder Examination Questionnaire (EDE-Q) [17] The EDE-Q 6.0 is a 28-item self-report measure of ED attitudes psychopathology and behaviours in both community and clinical populations. The questions concern the frequency of key behavioural features of EDs in which the person engages over the preceding 28 days. The questionnaire is scored on a 7-point Likert scale (0–6), rated using four subscales (restraint—R; eating concern—EC; shape concern—SC; and weight concern—WC) and a global score.

The EDE-Q has generally received support as an adequately reliable and valid measure of eating-related pathology [13]. Similarly, in the present sample, the dimensions of the EDE-Q have demonstrated acceptable internal consistency (R-α = 0.804; EC-α = 0.822; SC-α = 0.900; WC-α = 0.800; General/Total-α = 0.944).

The Italian version of schema mode inventory for eating disorders—short form (SMI-ED-SF) The item-pool (n = 64) for the new SMI-ED-SF was first created independently by two clinicians/researchers specialized both in ST and in the treatment of ED (authors GP and SS), who listed the items under each of the 16 modes in order of relevance in observance of the ST conceptualization for EDs.

Simultaneously, and blinded from the other authors, a third researcher (not specialized in ST; author AR) identified those items showing higher factor loading for each dimension of the original SMI-ED [13]. Conclusions from the authors were matched and discussed until agreement on the final set of items for the SMI-ED-SF was reached. Four items (three general, and one EDs-specific statement—where applicable) per mode were retained—thus to overcome the limitation of the previous version of the tool—where the number of items was highly heterogeneous between modes.

The SMI-ED is a 190-item self-report questionnaire with sixteen different modes clustered thematically: (A) five innate child modes (1. vulnerable child—VC, 2. angry child—AC, 3. enraged child—EC, 4. impulsive child—IC and 5. undisciplined child—UC); (B) two maladaptive (internalized/introject) modes (6. punitive mode—PM and 7. demanding mode—DM); (C) seven maladaptive coping modes (8. compliant surrenderer—CS, 9. helpless surrenderer—DS, 10. detached protector—Det.P, 11. detached self-soother—Det.SS, 12. self-aggrandizer—SA, 13. bully and attack—BA 14. eating disorder overcontroller—EDO); and (D) two healthy factors (15. happy child—HC and 16. healthy adult—HA). Notably, two modes (IC and EC) only included items retrieved from the original version of the SMI [18], while the HS and the EDO modes exclusively consisted of new ED-specific statements.

The SMI-ED revealed acceptable internal consistency, with Cronbach’s alpha coefficients ranging from 0.807 (Det.SS) to 0.976 (PM) across subscales (meanα-factors = 0.914; SDα-factors = 0.048).

Contrary to its full-length version—in which the number of items between scales varies from 5 (DS) to 20 (VC)—a fixed list of four statements was ensured for each of the SMI-ED-SF subscales (n = 16). Specifically, except for those modes only including either items retrieved from the original SMI or consisting of EDs-specific statements, the remaining subscales comprised three general statements and one item representative of the ED population.

Consistent with the previous versions of the tool [13, 18], items were scored on a six-point Likert scale ranging from 0 (“never or hardly ever”) to 5 (“all of the time”) and the score for each mode was computed dividing the sum scores by the number of items in each subscale. The higher the score, the more frequent were the manifestations of the modes.

Translation and cross-cultural adaptation

The SMI-ED-SF was independently translated from the original English version into Italian by two bilingual experts in the field, with one of them also having good knowledge of the measure. Any inconsistencies were revised and adjusted by a third investigator independent from the study using culturally and clinically fitting expressions. Also, to ensure conceptual equivalence between translations, a blind back translation of the Italian version of the SMI-ED-SF into English was conducted by an independent bilingual translator. Prior to the main study, the approved Italian version of the questionnaire was trialed with a random sample of 15 patients with EDs and 23 non-clinical participants, to assess item comprehensibility for the target population. No further adjustment was required.

Procedure

This study was completed entirely online, hosted by the questionnaire tool Qualtrics. Recruitment advertisements included a link placed on the main social networks (i.e., Facebook, Twitter) and websites of various local clinical centers specialized in the treatment and rehabilitation of EDs in Italy. In addition, flyers were placed around University campuses and in clinical waiting rooms of local ED services. The initial page contained a detailed description of the study, inclusion, and exclusion criteria along with any potential risks that may occur as a result of participation. Subjects were then asked to acknowledge they had read the terms and conditions and were aware of any potential risks by signing an informed consent form. Following informed consent, participants were asked to report demographic information and to answer the study questionnaires. After completing the survey, they were given access to a debriefing page of the study aims, and methodology, and received contact details for support services.

Statistical analyses

To test the factorial structural model of the SMI-ED-SF a Confirmatory Factor Analysis (CFA) was performed using ‘lavaan’ package [19, 20] for R software (R-core project [21, 22]). All the other statistical analysis were carried out with SPSS software (version 20.0, SPSS Inc., Bologna, Italy) [23].

As reported in Table 2, items’ descriptive statistics showed a non-normal distribution of some indicators. Therefore, in line with the previous study [13], the robust maximum likelihood method (MLM) [24,25,26,27] was chosen as estimator for the CFA. The MLM is a robust variant of the Maximum likelihood [27] that provides robust standard errors and is also referred to as the Satorra–Bentler Chi square (S–Bχ2) [19, 28, 29] to assess the model fit. Other fit indexes used to assess the model fit [30] were: the root mean square error of approximation (RMSEA) [31, 32], the comparative fit index (CFI) [33], and the standard root mean square residual (SRMR) [27], and the ratio of S–Bχ2 to the degrees of freedom (df) [34]. A S–Bχ2 test non-significant is desirable [35]. The RMSEA expresses fit per degrees of freedom of the model, with values lower than 0.08 suggesting an acceptable model fit [36] and values below 0.05 indicating a good fit [37]. The CFI designates the amount of variance and covariance accounted by the model compared with a baseline model, with values between 0.90 and 0.95 considered an acceptable fit [38, 39], and values > 0.95 indicating a good fit [36].

However, Kenny and McCoach mathematically demonstrate that a higher number of indicators analyzed negatively affects this fit index [40,41,42]. The SRMR derives from the residual correlation matrix and represents the average discrepancy between the correlations observed in the input matrix and those predicted by the model [27, 38]. A cutoff value higher than 0.08 is considered good [26, 36]. Also, the χ2/df ratio is considered as an easily computable measure of fit [26, 43], and a χ2/df ratio value of 3 or less indicates good fit [44,45,46,47].

The Cronbach’s alpha coefficient was used as measure of internal consistency for each SMI-ED-SF subscale—and values higher than 0.7 are deemed acceptable [48]. However, considering the differences in the magnitude of SMI-ED-SF’s factor loadings, Cronbach’s alpha was supported by Raykov’s maximal reliability (MR) [49] and the Bentler’s “Model-Based Internal Consistency Coefficient” (MBICC) [50]. These two indices were, respectively, chosen as measures of internal consistency of each single factor and multidimensional (overall) reliability: values higher than 0.6 suggest good reliability [51].

In addition, a MANCOVA was conducted to assess for possible statistical differences between the disordered eating subgroups simultaneously, on the SMI-ED-SF subscales, while adjusting for differences in age and gender.

Results

Sample characteristics

Participants’ self-reported BMI ranged from 13.71 to 65.31 (mean = 28.26; SD = 10.54), with 15.7% of the sample having a BMI below 18.5 and 38.4% of the respondents having a BMI above 30.1.

Of 649 participants, 46 self-reported a diagnosis of anorexia nervosa (AN), 31 were diagnosed with bulimia nervosa (BN), 64 suffered from binge eating disorder (BED), and 58 declared eating disorders not otherwise specified (EDNOS)—while the remaining 450 participants did not self-report a diagnosis of EDs. Descriptive statistics are presented in Table 1.

Table 1 Descriptive statistics for the EDE-Q 6.0 subscales
Full size table

Structural validity

Item analysis revealed a non-perfect normal distribution, with Kolmogorov–Smirnov and Shapiro–Wilk tests being significant (p < .001). Skewness ranged between − 1.18 and 2.76 (meansk = 0.79, SDsk = 0.81), and kurtosis ranged between − 1.03 and 8.09 (meank = 0.64, SDk = 2.01) (Table 2).

Table 2 Factor loading of the SMI-ED-SF items
Full size table

In line with the SMI-ED validation study [13], results from the CFA suggested an acceptable 16-correlated-factors solution for the SMI-ED-SF, despite not all the model’s fit indexes reaching the desired value [36]. Indeed, the Satorra–Bentler Chi square model for fit was statistically significant [S–Bχ2 (1832) = 3324.799; p < .001] and the CFI value did not achieve the threshold (CFI > 0.90 [38, 39]: CFI = 0.880). However, the RMSEA showed a good approximation fit of the model to the data [RMSEA = 0.045 (90% CI from 0.043 to 0.048), p(RMSEA < 0.05) = 1], and the SRMR also accounted for the goodness of the model (SRMR = 0.066 [36]). By dividing the χ2 for the degrees of freedom (df) of the model [34, 36], the model further resulted acceptable (χ2/df = 1.81; < 3) [26].

As reported in Table 2, each item loaded significantly on its associated factor (p < .001), meanloadings = 0.698; SDloadings = 0.122; ranging from 0.339 (item#22) to 0.901 (item#11). Correlations between the 16 factors ranged from |0.065| to |0.654|; meanr-factors = 0.238; SDr-factors = 0.297 (Table 3).

Table 3 Mean values, standard deviations, correlations between subscales of the SMI-ED-SF and reliability of each subscale (Cronbach’s Alpha and MR)
Full size table

Concurrent validity: correlation between SMI-ED-SF factors and eating disorder variables

Most SMI-ED-SF factors were significantly associated (ranging from |0.088| to |0.855|) with the EDE-Q subscales and ED symptoms (Table 4). In line with the original SMI-ED the adaptive modes (happy child and healthy adult) were negatively correlated with all the ED variables.

Table 4 Correlations between SMI-ED-SF subscales
Full size table

Correlation between SMI-ED-SF factors, gender, age, and BMI

Most of the SMI-ED-SF factors were not significantly associated with gender, age and BMI (Table 5). Regarding gender, significant associations ranged from |0.084| (angry child) to |0.235| (vulnerable child). Considering age, statistically significant correlations ranged from |0.079| (happy child) to |0.197| (helpless surrenderer). Also, significant correlations between the SMI-ED-SF factors and BMI ranged from |0.099| (self-aggrandizer) and |0.168| (eating disorder overcontroller).

Table 5 Correlations between SMI-ED-SF subscales, gender, age, and BMI across EDs
Full size table

Mode scores across disordered eating subscales

While controlling for age and gender as possible confounding variables, the MANCOVA revealed a significant difference between the presence of a self-reported diagnosis of ED and most of the SMI-ED-SF subscales: Wilks’s Λ = 0.638, F = 4.587, p < .001, partial η2 = 106. No differences emerged between ED diagnoses and the enraged child mode measured by the SMI-ED-SF. Also, to test differences between groups within the SMI-ED-SF subscales, ANCOVAs with focused contrasts were conducted for each dependent variable (Table 6).

Table 6 Mean (SD) for the ED diagnosis resulting from the MANCOVA
Full size table

Participants with no self-reported diagnosis of EDs showed lower means for each maladaptive mode as well as higher means for the adaptive modes, thus suggesting the goodness of the SMI-ED-SF in discriminating between the clinical and the general population.

Discussion

This study tested the psychometric properties of the shorter version of the Schema Mode Inventory for disordered eating both for the general population and a clinical sample, in Italy.

Findings confirmed an adequate fit for the 16-factor model, with moderate intercorrelations between subscales. However, the Satorra-Bentler Chi square was statistically significant and the CFI values did not achieve the desired cutoff score (CFI > 0.90 [38, 39]: CFI = 0.880). They may have been affected by the sample size (i.e., Chi square [34, 35, 52,53,54]) and the number of considered indicators, (i.e., CFI [36, 40,41,42, 46, 54,55,56]) respectively, but, since both the SRMR and RMSEA accounted for the goodness of the model, this is not reason for concern [40]. Also, internal consistency within subscales was high, and the scale showed good overall reliability.

As expected, disordered eating behaviours were positively correlated with most of the negative coping modes, and negatively related to the healthy modes (healthy adult and happy child). Specifically, the overcontroller mode and the helpless surrenderer dimensions (explicitly designating the presence of disordered eating patterns) showed moderate-to-high correlations with the eating/weight/shape concerns subscales of the EDE-Q, as well as with the EDE-Q global score. Consistently, higher mean scores for the Healthy Modes were noticed in respondents with no self-reported diagnosis of EDs.

Findings from this study reflect those observed by testing the psychometric properties of the Schema Mode Inventory for eating Disorders (SMI-ED) [13]—the adapted version of the Schema Mode Inventory (SMI) for the measurement of mode states within a population with self-reported disordered eating behaviours [18]—but overcome some of its methodological and practical limitations. In fact, unlike for the SMI-ED validation study, participants were recruited from both clinical and non-clinical populations, thus supporting the discriminatory power of the tool and its ability to identify individuals at risk/with disordered eating behaviours. By assessing the psychometric proprieties of the questionnaire in Italian—and demonstrating their goodness of fit—further evidence was also reached for both its construct and external validity. Moreover, a meaningful item reduction resulting in the development of a new shorter instrument in Italian increases the scale usability for both clinical and research purposes.

Nonetheless, these results should be considered a first step in the validation process of the SMI-ED-SF, and as a promising starting point for future research on the topic. In fact, as the sample was purely recruited via online survey, it has its limitations. First, it was not possible to ensure gender homogeneity among respondents—although a smaller proportion of males is representative of the gender ratio usually found in clinical settings [57]. Also, a relatively low proportion of participants revealed binge eating behaviours compared with other dysfunctional eating patterns, and the percentage of respondents who had never been diagnosed with an ED doubled its counterpart. In addition, asking people to self-report an existing diagnosis of EDs may have led to under-represent both those with reduced capacity to acknowledge their ED patterns, and individuals with severe EDs but avoidant of support services.

Future studies should ideally include a larger percentage of males in the sample, and all ED subgroups should be adequately represented within the sample to more precisely determine whether specific profiles of schema modes exist within a given diagnostic group, and the degree to which this is statistically feasible. The measurement invariance between clinical and non-clinical populations should also be tested to ascertain whether the questionnaire is valid to measure schema modes in each group separately.

Conclusion

This scale is of significant value for clinicians and researchers in identifying and exploring mechanisms through which schema modes are expressed within the ED population—both quantitatively and qualitatively. In fact,—as the SMI-ED—the SMI-ED-SF not only provides information regarding modes that would not be otherwise accessible in the original SMI [18], but—because of its reduced number of items—it facilitates the capacity to make important links between ED symptoms and schema modes, and in developing individually tailored case conceptualizations and treatments.

In fact, although CBT is widely recognized as the gold standard intervention for adults with EDs, it is still restricted to the ineffective coping mechanisms maintaining the problem [58], without adequately addressing early life experiences often at the root of the painful or unhelpful ways of thinking, feeling and behaving typical of clients with EDs. Evidence supports the effectiveness of ST in facilitating behavioural change both through diminishing the emotional intensity of memories linked to EMS [and associated ED symptoms], alongside direct behavioural pattern-breaking. The development of a measure specifically aimed at facilitating a more precise measurement of mode states within the ED population will enable clinicians to provide more sophisticated conceptualizations and therapeutic opportunities for those with EDs, and to enhance long-term maintenance of the achieved results [10].