Participants
Consecutive patients attending the Eating Disorder Center of the Department of Psychiatry at the University of Campania “Luigi Vanvitelli” and of the Rambam Health care campus, Psychiatric division and Eating Disorder center, Haifa, Israel were recruited. Inclusion criteria for the study were: 1) current diagnosis of BN or BED, according to DSM-5 and confirmed by the Structured Clinical Interview for DSM-5 Disorders–Research Version (SCID-RV) [26]; 2) absence of current/lifetime comorbid diagnosis of schizophrenia, bipolar disorder or substance abuse disorder; 3) willingness to cooperate in the experimental procedures and to sign a written informed consent.
Diagnostic assessment was made by a senior (in Israel) or trained (in Italy) psychiatrist (A.M.M.), who made the diagnosis first through a face-to-face clinical interview and then employing the SCID-RV, to confirm the ED diagnosis and psychiatric comorbidity.
Procedure
Sociodemographic and clinical data were collected as part of the routine assessment of patients with EDs.
Participants in the study were asked to complete the following questionnaires before entering specific treatment programs: 1) the Eating Disorders Inventory-2 (EDI-2) [27]; 2) the Childhood Trauma Questionnaire (CTQ) [28].
The EDI-2 [27] evaluates ED symptoms and psychopathology. The questionnaire includes 11 subscales: ineffectiveness, social insecurity, drive to thinness, interoceptive awareness, maturity fear, body dissatisfaction, perfectionism, interpersonal distrust, impulsivity, bulimia and ascetism. Cronbach’s values ranged from 0.72 (maturity fear) to 0.92 (ineffectiveness).
The CTQ [28] investigates a self-report recall of childhood maltreatment experiences. It is a 28-item questionnaire which differentiates five types of CM: emotional neglect (EN) (cut-off score ≥15; Cronbach’s α = 0.89), emotional abuse (EA) (cut-off score ≥10; Cronbach’s α = 0.84), sexual abuse (SA) (cut-off score ≥ 8; Cronbach’s α = 0.86), physical neglect (PN) (cut-off score ≥ 8; Cronbach’s α = 0.72) and physical abuse (PA) (cut-off score ≥ 8; Cronbach’s α = 0.87). The cut-off scores have been defined to indicate the occurrence of each CM type [30]. To the purposes of this study, a dimensional approach was adopted by employing the sum-score of each sub-scale [29]. Indeed, the use of continuous measures to evaluate trauma is largely documented in literature studies [9] and also employed in network analysis studies [31].
The study was approved by the Institutional Board of the University of Campania L. Vanvitelli and by the Ministry of Health and the Rambam Hospital's Helsinki Ethics committee (40-09RAM).
Statistical Analysis
Differences between people with BN and those with BED were analyzed by an independent samples t-Test and the Chi-square test, where appropriate.
Network Analysis (NA) was performed through R, version 3.4.4, using qgraph package [32]. A network is composed of nodes, which represent each variable included in the network, and edges, which are the connections among them. We have included in the network the EDI-2 sub-scores and the CTQ items. The thickness of an edge graphically represents the magnitude of the association. We have estimated partial-correlation networks, where the association between two nodes is controlled for the influence of all other variables [32]. We have run two separate networks: one for the sample with BN and one for the BED group. In order to retain only meaningful associations, we applied a “least absolute shrinkage and selection operator” (LASSO) regularization [33], which shrinks small partial correlations and sets them to zero [34]. The Extended Bayesan Information Criterion (EBIC) [35], a parameter that sets the degree of regularization/penalty applied to sparse correlations, was set to 0.5 in these analyses. The accuracy of edge-weights was estimated by drawing bootstrapped confidence intervals calculated through “nonparametric” bootstrapping (nboots = 2500) [36]. The bootnet package [37] was used for this analysis.
Subsequently, we performed the shortest pathways analysis in each network. This kind of network detects the shortest path between two nodes, i.e., the quickest out of all the routes connecting these two nodes. The shortest path between two nodes represents the minimum number of steps needed to go from one node to the other [38], computed using Dijkstra’s algorithm [39]. We employed this method in order to identify the shortest paths between each CM node, as assessed by the CTQ, and the ED core symptoms (body dissatisfaction and bulimia), evaluated by means of the EDI-2. The undirected edges assessed in this network point to conditional dependence between two variables: the edge-weight parameters reflect the strength of the associations between variables, which in turn suggests potential causal relationships [31; 36].