Anorexia nervosa (AN) is a serious mental disorder characterized by restricted ... One key area for future research highlighted by NICE (2017) is the impact of ...
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Central Symptoms Predict Post-Treatment Outcomes and Clinical Impairment in Anorexia Nervosa: A Network Analysis Introduction Anorexia nervosa (AN) is a serious mental disorder characterized by restricted food intake, an intense fear of weight gain, and high levels of discomfort with one's own body. Rates of psychological (e.g. depression, anxiety) and physical comorbidity are high and mortality rates from suicide or physical complications are the highest among psychiatric disorders (Zipfel et al., 2015; Treasure et al., 2015). The last few years have seen a dramatic acceleration in the generation of new treatment evidence in AN (Brockmeyer et al., 2017; National Institute for Health and Care Excellence, 2017). Despite this, in adults with AN, treatment outcomes with best available psychological therapies remain poor, with remission/recovery rates in recent trials ranging from 13% to 42.9% at 12 to 24 months post-randomization (Brockmeyer et al., 2017). One key area for future research highlighted by NICE (2017) is the impact of comorbidities on treatment outcomes in AN and other eating disorders, and what approaches are effective in managing these comorbidities. Comorbidity in AN Depression and anxiety are the most common psychiatric comorbidities in AN, affecting over 50% of patients (Spindler & Milos, 2007). In general, across different ED diagnoses, having higher levels of comorbidity is associated with higher overall ED psychopathology, longer duration of illness, and more severe ED symptoms (McDermott, Forbes, Harris, McCormack & Gibbon, 2006; Spindler & Milos, 2007). AN patients with comorbid depression report higher scores on Eating Disorder Inventory (EDI) items including body image dissatisfaction, desire for thinness, bulimia, perfectionism, and fear of maturity (Bizeul, Brun & Rigaud, 2003). Comorbid
--------------------------------------------------PREPRINT VERSION--------------------------------------------------depression and anxiety disorders in AN have also been associated with more hospitalizations and suicide attempts (Brand-Gothelf, Leor, Apter & Fennig, 2014). While short-term weight restoration in AN seems to go along with improvements in depression, some symptoms such as anhedonia (i.e. the inability to experience pleasure) persist and are thought to be a traitcharacteristic of AN (Boehm et al., 2018). Symptoms of depression and anxiety at the start of treatment have been found to predict treatment outcomes at end of treatment and follow-up (Wild et al., 2015; Fewell et al., 2017). In one study, a comorbid diagnosis of major depression at the start strongly predicted outcome 22 years later, suggesting that comorbidity with depression is particularly pernicious in AN (Franko et al., 2018) Although extant research elucidates the problematic nature of comorbidity within AN, the mechanisms of comorbidity are poorly understood. One possibility is that specific symptoms of AN, such as poor self-image or anhedonia, play a role in connecting AN to symptoms of depression, anxiety, and other psychological disorders. Identifying such symptoms could provide more effective targets for treatment of AN, especially in the context of reducing comorbidity and might thus improve longer term outcomes. Prognostic Variables in AN Prognostic variables are indicators collected prior to treatment which are predictive of treatment outcome. Prognostic variables are useful for predicting an individual's illness course and for determining the level of care needed. In some cases, prognostic variables can be used to determine which type of treatment is optimal (e.g. DeRubeis et al., 204; Lorenzo-Luaces, DeRubeis, van Straten, & Tiemens, 2017).
--------------------------------------------------PREPRINT VERSION--------------------------------------------------While comorbidity with anxiety and depression has been associated with a more severe course of illness in patients with AN, substantial evidence for other prognostic variables for treatment outcomes is lacking. Proposed prognostic indices include self-esteem, global assessment of functioning scores, presence of a comorbid mood or personality disorder, severity of ED symptoms, social problems, anemia, and body mass index (BMI) as predictors of treatment outcome, treatment utilization, and illness severity in patients with AN (Halmi et al., 2005; Keel et al., 2002; Löwe, Zipfel, Buchholz, Dupont, Reas, & Herzog, 2001). Although the number of studies examining outpatient treatments for AN has increased over the past thirty years, meta-analyses of older studies which examined outpatient treatments for AN have proven inconclusive (Watson & Bulik, 2013). Identification of prognostic variables could improve prediction of which methods of treatment will be most effective for individual patients with AN. The Network Approach to Psychopathology Network analysis offers an alternative perspective to conceptualize psychiatric disorders. The traditional medical model conceptualizes symptoms as consequences of an underlying disease entity. In contrast, the network approach posits that symptoms constitute self-sustaining feedback loops which, when activated, result in a disorder. Psychiatric disorders, therefore, emerge from the complex, causal interactions of symptoms or other related variables (Borsboom, 2017; Jones, Heeren, & McNally, 2018). In other words, symptoms are not caused by the disorder, but are instead constitutive of the disorder. Networks of psychiatric disorders can be modeled as dynamic systems of nodes and edges, where nodes represent symptoms and edges represent the associations between them (Borsboom and Cramer, 2013). Nodes with high centrality represent symptoms that are highly
--------------------------------------------------PREPRINT VERSION--------------------------------------------------connected to other symptoms. If associations can be interpreted as causal, nodes with high centrality, when activated, have a high probability of triggering other symptoms in the network. Treatment targeted towards the central symptoms of a network could hypothetically deactivate both those symptoms and the symptoms associated with them, resulting in recovery from the disorder (McNally, 2016). Bridge symptoms are symptoms which connect one psychiatric disorder to another and may be important in treatments which intend to reduce or prevent comorbid problems (Cramer, Waldorp, van der Maas, & Borsboom, 2010; Jones, Ma, & McNally, 2018). Central symptoms are not necessarily equivalent to “hallmark” symptoms of a given disorder. For example, Fried and colleagues (2016) found that the symptoms constituting the DSM-5 criteria for major depressive disorder (MDD; e.g. energy loss, interest loss, and appetite change) were not more central to the MDD network than non-DSM criteria (e.g. palpitations, tremors, and panic). The network approach has shown some evidence of potential prognostic utility in MDD (van Borkulo, Boschloo, Borsboom, Penninx, Waldorp, & Schoevers, 2015; Wichers, 2016), and bipolar disorder (Koenders, de Kleign, Giltay, Elzinga, & Spinhoven, 2016). The eating disorders clinical research community has recently seen a rise in research utilizing network analysis. The majority of studies have examined samples of individuals with transdiagnostic eating disorders (Dubois, Rodgers, Franko, Eddy, & Thomas 2017; Forbush, Siew, & Vitevitch, 2016; Goldschmidt et al., 2018; Olatunji, Levinson, & Calebs, 2018). Researchers have also examined disorder-specific network analyses. Dubois et al. (2017) estimated separate networks for individuals with AN-related disorders, bulimia nervosa (BN), and binge eating disorder (BED), although these disorder-specific analyses were conducted with small sample sizes given the high-dimensionality of network analyses (n = 73, 52, 31). Levinson
--------------------------------------------------PREPRINT VERSION--------------------------------------------------et al. (2017) examined symptom networks in individuals with BN, also examining BN comorbidity with depression and anxiety from a network perspective. Forrest, Jones, Ortiz, & Smith (in press) examined AN and BN separately with large sample sizes (n = 604, 477), also generating a combined ED network. Notably, network analyses have used different measurement techniques to generate networks, including the Eating Disorders Examination Interview (EDE; Fairburn, Cooper, & O’Connor, 2008), the EDE Questionnaire (EDE-Q), the Eating Disorder Inventory (EDI-2; Garner, 1991), and the Eating Pathology Symptom Inventory (ESPI; Forbush et al., 2013). Studies have focused on primarily adult samples, with the exception of Goldschmidt et al. (2018). Despite these differences in measurement and sample, some correspondence has arisen from the various studies, with several studies emphasizing the central role of overvaluation of weight and shape (Dubois et al., 2017; Forrest et al., in press; Levinson et al., 2017) or dissatisfaction with shape and weight (Goldschmidt et al., 2018) in eating disorders transdiagnostically. Studies have also indicated a central role of fear of weight gain (Forrest et al., in press; Goldschmidt et al., 2018; Levinson et al., 2017) and desire to lose weight (Forrest et al., in press; Levinson et al., 2017). Only one study to date (Forrest et al., 2018) has specifically examined a network of AN with sufficient sample size. Using data from the EDE-Q, they note that fear of weight gain, desiring weight loss, restraint, shape & weight preoccupation, and shape overvaluation were central in the AN network. The Current Study The purpose of the current study was to use network analysis in the context of a randomized controlled trial to (1) determine the central symptoms of AN (2) identify key bridge
--------------------------------------------------PREPRINT VERSION--------------------------------------------------symptoms between AN, depression, and anxiety and (3) ascertain which AN symptoms are associated with differential recovery. To accomplish these objectives, we drew data from the Maudsley Outpatient Study of Treatments for Anorexia Nervosa and Related Conditions (MOSAIC). The MOSAIC study compared patient outcomes from the Maudsley Model of Anorexia Nervosa Treatment for Adults (MANTRA), a novel, targeted psychological therapy for AN, to Specialist Supportive Clinician Management (SSCM), a comparison treatment designed specifically for clinical trials which emphasizes resumption of normal eating and weight gain (Schmidt et al., 2015). We selected this dataset for several reasons. First, the MOSAIC study is one of the largest RCTs of first line outpatient treatments for patients with AN worldwide. Second, treatment completion and acceptability rates were high, with 75% of MANTRA patients and 59% of SSCM patients completing treatment. Third, primary and secondary outcomes were measured at baseline, 6 month, 12 months, and two years post-randomization. Fourth, AN patients in both conditions showed significant improvement in both primary and secondary outcomes, though outcomes were not significantly different between the two treatments. Lastly, the longitudinal design of this study allows for more direct conclusions about temporality, which draw us closer to causal conclusions (Schmidt et al., 2015; Schmidt et al., 2016). Methods Procedure Data were drawn from the MOSAIC study, a multi-site RCT that compared outcomes of a targeted outpatient treatment (MANTRA) to a comparison treatment designed for clinical trials (SSCM). The study was approved by the National Research Ethics Service in London. Full
--------------------------------------------------PREPRINT VERSION--------------------------------------------------details of the study protocol have been published separately (Schmidt et al. 2013). We used outcomes at baseline, 6-months, 12-months, and 2-years in this study. Participants 142 patients were included in primary outcome analysis (139 female, 3 male). Inclusion criteria were being age 18 to 60 years, having a BMI of 18.5 kg/m2 or below, and a DSM-IV (American Psychiatric Association, 2000) diagnosis of Anorexia Nervosa or Eating Disorder Not Otherwise Specified (EDNOS; criteria following Thomas, Vartanian, & Brownell, 2009). These inclusion criteria ensured patients with wide variations in AN severity were represented in the sample. Measures All measurements were collected at baseline, 6-months, 12-months, and 24-months following enrollment into the study. ED Psychopathology. ED psychopathology was assessed using the Eating Disorders Examination (EDE) Interview (Fairburn, Cooper, & O’Connor, 2008). The EDE is a widely supported measure that generates a global score based on dietary restraint, eating concern, weight concern, and shape concern (Schmidt et al., 2013). Depression and Anxiety. Symptoms of depression and anxiety were assessed using the Depression, Anxiety and Stress Scale-21 (DASS-21; Lovibond & Lovibond, 1995), a 21-item self-report measure that assesses mood state over the past seven days using a 4-point Likert scale. Items include statements such as “I found it hard to wind down,” “I tend to over-react to situations,” and “I felt that life was meaningless.”
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Body Mass Index (BMI) & Recovery Status. BMI is a weight-to-height ratio commonly used as an indicator of healthy weight. BMI was combined with ED psychopathology to determine recovery status at follow-ups (see Schmidt et al,. 2013 for computational details). Clinical Impairment. Clinical impairment was assessed using the Clinical Impairment Assessment (CIA; Bohn & Fairburn, 2008a), a self-report measure of impairment resulting from the individual’s ED. This is a 16-item questionnaire that generates a single global score to signify level of impairment. The CIA has shown high internal consistency, test-retest reliability, and construct validity in transdiagnostic treatment trials (Bohn & Fairburn, 2008b). Analyses Network Generation & Stability. Network models were estimated using the qgraph package in R (Epskamp, Cramer, Waldorp, Schmittmann, & Borsboom, 2012). Networks were generated using the graphical LASSO method, which estimates regularized partial correlations between nodes. The graphical LASSO shrinks small partial correlations to 0 in order to reduce false positive errors. The degree of shrinkage can be altered by using a hyperparameter; in this case, we chose a low hyperparameter (gamma = 0.0) in order to maximize the stability of the network and balance sensitivity and specificity (Epskamp & Fried, 2018). The stability of each network we generated was assessed using a modified version of the bootnet package (Epskamp, Borsboom, & Fried, 2018) which allowed us to assess the stability of expected influence and bridge expected influence (code is available in supplemental materials). For each network, a correlation-stability (CS) coefficient is computed in order to assess stability. This coefficient indicates the maximum proportion of the original sample that can be dropped while confidently retaining centralities which correlate highly (r > 0.7) with the original sample.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------It is suggested that a CS coefficient of 0.25 or above indicates adequate stability, and a coefficient of 0.5 or above indicates good stability (Epskamp, Borsboom, & Fried, 2018). To measure the centrality of nodes in the network, we used expected influence (Robinaugh, Millner, & McNally, 2016). Expected influence is defined as the sum of all edges which extend from a given node, accounting for both positive and negative values. Expected influence shows superior performance to other types of centrality when networks include both positive and negative edges (Robinaugh, Millner, & McNally, 2016; McNally, 2016). Since many of our networks included both positive and negative edges, we opted to use expected influence for all of our analyses. Expected influence values were calculated with the networktools R package (Jones, 2018). Centrality estimations can occasionally be affected by problems with restricted ranges of variance. To test for this problem, we tested the correlation between expected influence values and node variances. Anorexia Nervosa Networks. Nodes in the AN network were drawn from the EDE and were chosen to be reflective of eating disorder symptomatology. Our item selection was guided in order to produce the maximal degree of comparability with past network studies of eating disorders using the EDE and the EDE-Q (e.g., Levinson et al., 2017). One item we chose (fasting) had to be eliminated due to a high proportion of missing values. We generated network models of eating disorder symptomatology for each time-point in the study. Anorexia Nervosa Comorbidity Networks. We also generated a network model of comorbid eating disorder symptoms with depression and anxiety at baseline using all items from the DASS in addition to the chosen EDE measures. To determine the community structure of the combined EDE/DASS network, we used a spinglass detection algorithm, setting the maximum number of detectable communities to four (i.e., AN, depression, anxiety, and stress; four spins).
--------------------------------------------------PREPRINT VERSION--------------------------------------------------The algorithm detected that items separated into three communities, aggregating stress items with anxiety items into a single community. Thus, in the comorbidity network, we utilize a threecommunity structure (AN, depression, and anxiety/stress) to calculate bridge centrality statistics. In the comorbidity networks, we aimed to identify bridge symptoms between ED and comorbid depression and anxiety symptoms. Bridge symptoms are those which play a primary role in connecting two or more psychiatric disorders (Cramer et al., 2010). Bridge symptoms can be assessed using bridge centrality statistics (Jones, Ma, & McNally, 2018; Heeren, Jones, & McNally, 2018); specifically, we used bridge expected influence. Bridge expected influence is defined as the sum of all edges that connect a given node to all nodes in a community outside its own. For example, the bridge expected influence of the node "avoideat" in the AN network with depression would be measured by taking the sum of all edges which connect "avoideat" to nodes in the depression community. Network Connectivity. Some networks are more densely connected, whereas other networks have fewer and weaker connections between nodes. There are various ways to assess network connectivity. One popular measure is the global strength (GS), which is the sum of the absolute value of all edge weights in the network. Because the global strength treats negative edges and positive edges similarly, it can be interpreted in the sense of the overall predictability of a given node, given information about other nodes. In contrast the global expected influence (GEI) measures the sum of all edge weights, while taking into account negative edges. The interpretation of global expected influence differs: it indicates that the nodes in the network tend to cohere – that is, if one node is active, it is likely that other nodes in the network will be active. It is also unclear whether graphical LASSO networks are appropriate for assessing network connectivity, as regularization directly impacts the number of present versus absent edges in a
--------------------------------------------------PREPRINT VERSION--------------------------------------------------network. To address the wide range of potentially valid connectivity measures and methods, we assessed connectivity using both GS and GEI across graphical LASSO networks, partial correlation networks, and simple correlation networks. Prognostic Value of Central Symptoms. We hypothesized that central symptoms might have greater prognostic value in predicting recovery relative to symptoms which were less central in the network. To test this hypothesis, we first computed correlations between the severity of each symptom at baseline with two outcome measures: treatment recovery status (Recovered/Partially Recovered/Non-recovered) and clinical impairment. These correlations are interpreted as the "prognostic value" of symptoms. Subsequently, we used linear regression to determine whether symptoms with higher expected influence also had greater prognostic value. Results Network Generation & Stability The AN networks had high stability at all time points (CS = 0.75, 0.71, 0.64, 0.64). Expected influence values were not significantly related to node variances at any time point (p = 0.96, 0.69, 0.21, 0.58). Stability of bridge centrality metrics was adequate in the baseline network (CS = 0.27) and ranged from inadequate to adequate in the follow-up networks (CS = 0.19, 0.23, 0.30). Accordingly, we present and interpret centrality values from the AN network at all time points, but only present and interpret bridge centrality values from the comorbidity network at baseline, where stability was adequate and sample size was at its highest. Anorexia Nervosa Networks Figure 1 depicts the GLASSO network (γ = 0.0) of the baseline AN network. Data from all participants (n = 142) were included at baseline.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Participants completed follow-up measures at 6-months, 12-months and 24-months posttreatment. Some participants dropped out over the course of treatment, resulting in smaller sample sizes over time (n = 142, 119, 113, 105 for each time point respectively). Figures depicting follow-up network structures can be found in the supplemental materials. Figure 2 depicts the expected influence of nodes at all time points. Expected influence values were consistent at all four time points, showing high correlations across all possible combinations of networks (r > 0.75). Feeling fat (feelfat), fear of weight gain (fearwtgain), discomfort seeing one's own body (bodydisc), dissatisfaction with weight (wtdis), and a strong desire to lose weight (losewt) had the highest expected influence (EI). These symptoms may be core to AN psychopathology. Overeating (overeat), eating in secret (secreteat), and excessive exercise (excesexer) had consistently low expected influence values.
Anorexia Nervosa Comorbidity Network
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Figure 4 depicts the comorbidity network between AN, depression, and anxiety & stress. Figure 5 depicts bridge centralities between the AN symptom network and depression or anxiety/stress. Several symptoms that bridge AN with depression or anxiety were identified. The bridge centrality of ED symptoms is defined here as the bridge expected influence (bridge EI) of eating disorder symptoms on both depression and anxiety. The bridge centrality of anxiety or depression symptoms is defined as their bridge EI on ED symptoms. Feelings of worthlessness had the highest bridge centrality connecting depression and eating disorders. Many of the central AN symptoms (e.g. fearwtgain and feelfat) were not highly associated with anxiety/stress or depression. Body discomfort, which was central in the EDE network, and having a negative reaction to wanting to weigh oneself weekly were associated with anxiety and depression. Additionally, several items that were less central in the EDE network were strongly correlated with depression and anxiety. Avoidance of eating (i.e., number of days where participants went at least 8 hours avoiding eating), and not wanting to eat in social situations had high bridge centrality between AN and depression, anxiety, and stress. Bridge centralities between each comorbid group separately (e.g., bridges between AN and depression only, bridges between AN and anxiety only) can be found in the supplemental materials.
--------------------------------------------------PREPRINT VERSION---------------------------------------------------
Network Connectivity We assessed the network connectivity at each time point using global strength (GS) and global expected influence (GEI) within graphical LASSO, partial correlation, and correlation networks. GS is a measure of the overall connectivity of the network defined as the sum of the absolute values of edges in the network. GEI is a measure of the positive connectivity or coherence of the network, and is defined as the sum of the weights of the positive edges minus the sum of the absolute weights of the negative edges. We normalized the values of connectivity within each condition. The results of this analysis appear in Figure 3. Network connectivity, regardless of the type of measure or network, appeared to increase over time. Prognostic Value of Central Symptoms We tested whether central symptoms would be prognostic indicators of treatment response. First, association of symptoms at baseline was compared with recovery status at 12 months. Severity of each symptom at baseline was predictive of failure to recover. More central
--------------------------------------------------PREPRINT VERSION--------------------------------------------------items, however, were significantly more inhibitory to recovery. Central items were also more strongly associated with clinical impairment at 12 months. These relationships are shown in Figure 6. To properly understand this figure, it is important to recognize that each point represents a symptom (i.e., node) in the network. A point that is high on the X-axis represents a symptom that was highly central. A point that is high on the Y-axis represents a symptom that was highly prognostic of recovery failure or clinical impairment. The relationship thus represents an association between a symptom's centrality and the prognostic value of that symptom. Association of central items with change in BMI was nonsignificant.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Table 1. Prognostic Value of Anorexia Symptoms Failed Recovery Symptom Status Clinical Impairment BMI (Reverse) avoideat 0.23 0.28 0.09 binge 0.13 0.04 -0.08 bodydisc 0.19 0.27 -0.09 dietrule 0.19 0.07 0.16 emptystom 0.34 0.40 0.00 excesexer 0.07 0.06 -0.01 expodisc 0.05 0.21 -0.13 fearwtgain 0.27 0.34 -0.16 feelfat 0.34 0.40 -0.07 flatstom 0.36 0.42 -0.03 foodavoid 0.36 0.37 0.03 guilt 0.28 0.39 -0.02 losewt 0.35 0.34 -0.02 losscontrol 0.11 0.27 -0.15 overeat 0.04 0.05 0.02 preoc 0.29 0.38 -0.05 restrict 0.29 0.32 0.00 secreteat 0.01 0.01 0.01 shapedis 0.20 0.29 -0.04 shapeimport 0.15 0.22 -0.01 socialeat 0.19 0.23 0.08 vomit 0.15 0.21 0.01 weigh 0.10 0.32 -0.16 wtdis 0.24 0.19 -0.01 wtimport 0.17 0.22 0.01 preoc_fd 0.24 0.33 -0.03 Note: Prognostic value was determined by computing Spearman correlations between symptom severity at baseline with post-treatment outcomes. Outcomes are coded such that higher values indicate poorer outcome.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Discussion Using network analysis, we identified core symptoms of AN psychopathology across the stages of treatment in a randomized clinical trial. We identified bridge symptoms between AN and anxiety/stress and depression, which may be important in the mechanisms of comorbidity. Finally, we demonstrated that central symptoms of AN at baseline are important prognostic indicators for recovery and clinical impairment at follow-up. Anorexia Nervosa Network Feeling fat, body discomfort, a strong desire to lose weight, discomfort seeing one's own body, and fear of weight gain were central to the AN psychopathology network. These symptoms had the highest expected influence in the network, which suggests that they had strong positive connections to other nodes in the network. These results are consistent with current diagnostic criteria for AN, which includes an intense fear of weight gain and strong feelings of discomfort with one’s body weight and shape (American Psychiatric Association, 2013). These results share some overlap with the AN network reported in Forrest et al. (in press), where fear of weight gain and desiring weight loss also appeared as central symptoms. Behavior aimed at preventing weight gain was less central. Targeting therapy towards the most central symptoms identified in the AN psychopathology network could potentially improve clinical outcomes. Anorexia Nervosa Comorbidity Network Feelings of worthlessness, avoidance of eating, not wanting to eat in social situations, body discomfort, and negative reaction to weighing oneself weekly were identified as central bridge symptoms between AN and anxiety/stress and depression symptoms. These results are particularly important given the high rates of comorbidity between AN and anxiety disorders (53%) and affective disorders (51%) (Spindler & Milos, 2007). Identification of critical
--------------------------------------------------PREPRINT VERSION--------------------------------------------------associations between AN and anxiety/depression could help develop a treatment protocol that can more effectively integrate existing treatments for AN and its comorbid disorders by targeting specific, clinically relevant symptom domains. Such targeted therapy has shown success in anxiety disorders, ADHD, conduct related disorders, and mood disorders (Weisz et al., 2012) as well as social anxiety disorder, generalized anxiety disorder, and social phobia (Chiu et al., 2013), and holds promise for use in treatment of other psychiatric disorders. The high centrality of feelings of worthlessness was particularly interesting. Worthlessness could provide a potential link between the negative self-referential cognitive distortions seen in AN and other symptoms of depression. Our network structure provided some support for this hypothesis: feelings of worthlessness were strongly connected to the depression symptoms of meaninglessness and anhedonia, and connected to several AN symptoms, including negative reaction to weighing oneself weekly. Of note, negative self-beliefs (e.g. I am a failure) or negative self-schemata (e.g. sense of defectiveness, incompetence, failure) centering on the theme of worthlessness are common in AN and are thought to contribute to illness maintenance (Oldershaw et al., 2015). Network Connectivity Network connectivity measures the degree of overall connection between symptoms. Unexpectedly, we found that network connectivity, regardless of the measurement strategy, increased over time. This was surprising, as network theory would predict that a highly coherent network (e.g., high values of global expected influence) would be predictive of greater psychopathology. One possibility is that network connectivity was impacted by attrition: a total of 37 patients dropped out of the study by the last time point, accounting for about 26% of the original sample. If drop-outs had fundamentally different symptom patterns, this may have
--------------------------------------------------PREPRINT VERSION--------------------------------------------------reduced network connectivity at the first time points and artificially increased connectivity once the participants dropped out. Another possibility is that the prediction that coherent networks will lead to greater psychopathology is true at the level of the individual, but does not apply to crosssectional networks estimated across a group. Future research is needed to clarify this finding. Prognostic Value of Central Symptoms It has long been hypothesized that identification of highly central nodes can help tailor treatments to target the “key symptoms” of a disorder, thereby improving its effectiveness. Therefore, perhaps the most important of our findings is that the more central items in the AN network were more associated with failure to recover. In addition, these highly central items were more strongly associated with clinical impairment at 12-month follow-up. Previous research comparing pharmacological and psychological outpatient treatments for adults with AN have been inconclusive (Wade et al., 2017), and predictors of differential treatment outcomes is lacking (Halmi et al., 2005). Identification of the core AN symptoms that predict differential treatment response may be helpful in early intervention efforts and further improving patient outcomes in clinical practice. It is important that these results demonstrate a connection between central symptoms and failure to recover at the group level. Further evidence is needed to determine whether personalized network profiles would correspond with prognostic indicators of recovery at the individual level (Fisher, Reeves, Lawyer, Medaglia, & Rubel, 2017). Limitations While data were drawn from a large, multi-site RCT, the sample size at baseline was still relatively small (n = 142). Stability analyses indicated high stability for centrality estimates at all time points, and adequate stability for bridge centrality estimates at baseline. Bridge centrality estimates at other timepoints were not presented due to stability concerns. Additionally, 139 of
--------------------------------------------------PREPRINT VERSION--------------------------------------------------142 participants at baseline were female, making generalizations to non-female patients difficult. Further research should incorporate higher numbers of males and participants of non-Caucasian ethnicities to determine differences in AN presentation across gender and culture. The individuals included in this study were also from a treatment-seeking population, which limits the generalizability of our results to non-treatment seeking or community samples of individuals with AN. Our inclusion criteria, however, ensured the representation of patients with variations in AN severity. While network theory is particularly useful in identifying the most central symptoms of psychological disorders, it has several limitations that future research should address. Such limitations include network generalizability and stability, proper node selection, and power analysis (Smith et al., 2018). In addition, future studies should consider integrating several different measures of each symptom to avoid bias from single measures (McNally, 2016). In sum, this study was one of the first to use network analysis to examine the core symptoms of AN and begin to elucidate the symptom-level mechanisms for comorbidity in AN. Our results also suggest predictive relationships between central AN symptoms and posttreatment outcomes and clinical impairment. Further research is necessary to validate the clinical utility of these findings in developing more tailored interventions that target the central symptoms of AN.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------References Agras, W. S., Brandt, H. A, Bulik, C. M., Dolan-Sewell, R., Fairburn, C. G., Halmi, K. A., … Wilfley, D. E. (2004). Report of the National Institutes of Health workshop on overcoming barriers to treatment research in anorexia nervosa. International Journal of Eating Disorders, 35(4), 509-21. American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders (4th ed., text rev.). Washington, DC: Author. American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.). Arlington, VA: American Psychiatric Publishing. Arcelus, J., Mitchell, A. J., Wales, J., & Nielsen, S. (2011). Mortality Rates in Patients With Anorexia Nervosa and Other Eating Disorders: A Meta-analysis of 36 Studies. Archives of General Psychiatry, 68(7), 724-31. Bizuel, C., Brun, J. M., & Rigaud, D. (2003). Depression influences the EDI scores in anorexia nervosa patients. European Psychiatry, 18(3), 119-23. Boehm I, Flohr L, Steding J, Holzapfel L, Seitz J, Roessner V, & Ehrlich, S. (2018) The Trajectory of Anhedonic and Depressive Symptoms in Anorexia Nervosa: A Longitudinal and Cross-Sectional Approach. European Eating Disorders Review, 26(1), 69-74. Bohn K, Fairburn CG: The Clinical Impairment Assessment questionnaire (CIA). In Cognitive Behavioral Therapy for Eating Disorders. Edited by Fairburn CG. New York: Guildford Press; 2008a. Bohn, K., Doll, H. A., Cooper, Z., O’Connor, M., Palmer, R. L., & Fairburn, C. G. (2008b). The measurement of impairment due to eating disorder psychopathology. Behaviour Research
--------------------------------------------------PREPRINT VERSION--------------------------------------------------and Therapy, 46(10), 1105-1110. Borsboom, D. & Cramer A. O. J. (2013). Network Analysis: An Integrative Approach to the Structure of Psychopathology. Annual Review of Clinical Psychology, 9, 91-121. Borsboom, D. (2017). A network theory of mental disorders. World Psychiatry, 16(1), 5-13. Brand-Gothelf, A., Leor, S., Apter, A., & Fennig, S. (2014). The Impact of Comorbid Depressive and Anxiety Disorders on Severity of Anorexia Nervosa in Adolescent Girls. The Journal of Nervous and Mental Disease, 202(10), 759-62. Burgess, P., & Shallice, T. (1997). The Hayling and Brixton Tests. Bury St. Edmunds, UK: Thames Valley Test Company. Chiu, A. W., Langer, D. A., McLeod, B. D., Har, K., Drahota, A., Galla, B. M., ... & Wood, J. J. (2013). Effectiveness of modular CBT for child anxiety in elementary schools. School Psychology Quarterly, 28, 141. Cramer, A. O., Waldorp, L. J., van der Maas, H. L., & Borsboom, D. (2010). Complex realities require complex theories: Refining and extending the network approach to mental disorders. Behavioral and Brain Sciences, 33, 178-193. DeRubeis, R. J., Cohen, Z. D., Forand, N. R., Fournier, J. C., Gelfand, L. A., & Lorenzo-Luaces, L. (2014). The Personalized Advantage Index: Translating Research on Prediction into Individualized Treatment Recommendations: A Demonstration. PLoS ONE, 9(1), e83875. DuBois, R. H., Rodgers, R. F., Franko, D. L., Eddy, K. T., & Thomas, J. J. (2017). A network analysis investigation of the cognitive-behavioral theory of eating disorders. Behaviour Research and Therapy, 97, 213-221. Epskamp, S., Borsboom, D., & Fried, E. I. (2018). Estimating psychological networks and their accuracy: A tutorial paper. Behavior Research Methods, 50, 195-212.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Epskamp, S., Cramer, A. O. J., Waldorp, L., Schmittmann, V. D., & Borsboom, D. (2012). qgraph: Network Visualizations of Relationships in Psychometric Data. Journal of Statistical Software, 48(4). Epskamp, S., & Fried, E. I. (2018). A tutorial on regularized partial correlation networks. Psychological Methods. Fairburn C, Cooper Z, O’Connor M: Eating disorder examination (edition 16.0D). In Cognitive Behavioral Therapy for Eating Disorders. Edited by Fairburn C. New York: Guilford Press; 2008. Fewell, L., Levinson, C., & Stark L. (2017). Depression, worry, and psychosocial functioning predict eating disorder treatment outcomes in a residential and partial hospitalization setting. Eating and Weight Disorders, 22(2), 291-301. Fisher, A.J., Reeves, J. W., Lawyer, G. L., Medaglia, J. D. & Rubel, J. A. (2017). Exploring the Idiographic Dynamics of Mood and Anxiety via Network Analysis. Journal of Abnormal Psychology. Fried, E. I., Epskamp, S., Nesse, R. M., Tuerlinckx, F., & Borsboom, D. (2016). What are ‘good’ depression symptoms? Comparing the centrality of DSM and non-DSM symptoms of depression in a network analysis. Journal of Affective Disorders, 189, 314-320. Foa, E. B., Huppert, J. D., Leiberg, S., Langner, R., Kichic, R., Hajack, G., & Salkovskis, P. M. (2002). The Obsessive-Compulsive Inventory: Development and validation of a short version. Psychological Assessment, 14, 485-496. Forbush, K., Wildes, J., Pollack, L., Dunbar, D., Luo, J., Patterson, K., . . . Reynolds, Cecil R. (2013). Development and validation of the Eating Pathology Symptoms Inventory (EPSI). Psychological Assessment, 25, 859-878.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Franko, D. L., Tabri, N., Keshaviah, A., Murray, H. B., Herzog, D. B., Thomas, J. J., … & Eddy, K. T. (2018). Predictors of long-term recovery in anorexia nervosa and bulimia nervosa: Data from a 22-year longitudinal study. Journal of Psychiatric Research, 96, 183-188. Garner, D. M. (1991). Eating disorder inventory-2 manual. Psychological Assessment Resources, Odessa, FL. Golan, O., Baron-Cohen, S., Hill, J. J., & Golan, Y. (2006). The “reading the mind in films” task: Complex emotion recognition in adults with and without autism spectrum conditions. Social Neuroscience, 1, 111–123. Goldschmidt, A. B., Crosby, R. D., Cao, L., Moessner, M., Forbush, K. T., Accurso, A. C., & Le Grange, D. (2018). Network Analysis of Pediatric Eating Disorder Symptoms in a Treatment-Seeking, Transdiagnostic Sample. Journal of Abnormal Psychology, 127(2), 251-264. Halmi, K. A., Agras, W. S., Crow, S., Mitchell, J., Wilson, G. T, Bryson, S. W., & Kraemer, H. D. (2005). Predictors of Treatment Acceptance and Completion in Anorexia Nervosa. Archives of General Psychiatry, 62(7), 776-81. Hay, P., Claudino, A., Touyz, S., & Elbaky, A. G. (2015). Individual psychological therapy in the outpatient treatment of adults with anorexia nervosa. Cochrane Database of Systematic Reviews, 7. Heeren, A., Jones, P. J., & McNally, R. J. (2018). Mapping network connectivity among symptoms of social anxiety and comorbid depression in people with social anxiety disorder. Journal of Affective Disorders, 228, 75-82. Jones, P. J. (2018). networktools: Tools for identifying important nodes in networks. R package version 1.1.1. https://CRAN.R-project.org/package=networktools
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Jones, P. J., Ma, R., & McNally, R. J. (2017). Bridge centrality: A network approach to understanding comorbidity. Retrieved from osf.io/c5dkj Jones, P. J., Heeren, A., & McNally, R. J. (2017). Commentary: A network theory of mental disorders. Frontiers in Psychology, 8, 1305. Keel, P. K., Dorer, D. J., Eddy, K. T., Delinsky, S. S., Franko, D. L., Blais, M. A., … & Herzog, D. B. (2002). Predictors of Treatment Utilization Among Women With Anorexia and Bulimia Nervosa. American Journal of Psychiatry, 159(1), 140-142. Koenders, M., de Kleijn, R., Giltay, E., Elzinga, B., & Spinhoven, P. (2016). A network approach to bipolar symptomatology in patients with different course types. PLoS ONE, 10(10). https://doi.org/10.1371/journal.pone.0141420 Levinson, C. A., Zerwas, S., Calebs, B., Forbush, K., Kordy, H., Watson, H., … & Bulik, C. M. (2017). The Core Symptoms of Bulimia Nervosa, Anxiety, and Depression: A Network Analysis. Journal of Abnormal Psychology, 126(3), 340-354. Lorenzo-Luaces, L., DeRubeis, R. J., van Straten, A., & Tiemens, B. (2017). A prognostic index (PI) as a moderator of outcomes in the treatment of depression: A proof of concept combining multiple variables to inform risk-stratified stepped care models. Journal of Affective Disorders, 213, 78-85. Löwe, B., Zipfel, S., Buchholz, C., Dupont, Y., Reas, D. L., & Herzog, W. (2001). Long-term outcome of anorexia nervosa in a prospective 21-year follow-up study. Psychological Medicine, 31(5), 881-90. Lovibond, S. H., & Lovibond, P. F. (1995) Manual for the Depression-Anxiety Stress Scales (2nd ed.) Sydney, Australia: Psychology Foundation. McDermott, B., Forbes, D., Harris, C., McCormack, J., & Gibbon, J. (2006). Non-eating
--------------------------------------------------PREPRINT VERSION--------------------------------------------------disorders psychopathology in children and adolescents with eating disorders: Implications for malnutrition and symptom severity. Journal of Psychosomatic Research, 60(3), 257-61. McNally, R. (2016). Can network analysis transform psychopathology? Behavior and Research Therapy, 86, 95-104. Oldershaw A, Lavender T, Sallis H, Stahl D, Schmidt U. (2015) Emotion generation and regulation in anorexia nervosa: a systematic review and meta-analysis of self-report data. Clinical Psychology Review, 39, 83-95. Olatunji, B. O., Levinson, C., & Calebs, B. (2018). A network analysis of eating disorder symptoms and characteristics in an inpatient sample. Psychiatry Research, 262, 270-281. Psychological Assessment Resources. (2003). Computerized Wisconsin Card Sort Task Version 4 (WCST). Odessa, FL: Author. Ramos-Grille I., Gomà-i-Freixanet M., Aragay N., Valero S., & Vallès V. (2017). Predicting treatment failure in pathological gambling: The role of personality traits. Addictive Behaviors, 43, 54-59. Rey, A. (1941). L’examen psychologique dans les cas d’encephalopathie traumatique [The psychological examination in cases of traumatic encephalopathy]. Archives de Psychologie, 28, 215–285. Robinaugh, D. J., Millner, A. J., & McNally, R. J. (2015). Identifying Highly Influential Nodes in the Complicated Grief Network. Journal of Abnormal Psychology, 344(6188), 1173– 1178. Schmidt, U., Renwick, B., Lose, A., Kenyon, M., DeJong, H., Broadbent, H., … & Landau, S.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------(2013). The MOSAIC study - comparison of the Maudsley Model of Treatment for Adults with Anorexia Nervosa (MANTRA) with Specialist Supportive Clinical Management (SSCM) in outpatients with anorexia nervosa or eating disorder not otherwise specified, anorexia nervosa type: study protocol for a randomized controlled trial. U.S. National Library of Medicine, 14, 160.
Schmidt, U., Magill, N., Renwick, B., Keyes, A., Knyon, M., Dejong, H., … & Landau, S.
(2015). The Maudsley Outpatient Study of Treatments for Anorexia Nervosa and Related Conditions (MOSAIC): Comparison of the Maudsley Model of Anorexia Nervosa Treatment for Adults (MANTRA) with specialist supportive clinical management (SSCM) in outpatients with broadly defined anorexia nervosa: A randomized controlled trial. Journal of Consulting and Clinical Psychology, 83(4), 796-807. Schmidt, U., & Treasure, J. (2006). Anorexia nervosa: Valued and visible. A cognitiveinterpersonal maintenance model and its implications for research and practice. British Journal of Clinical Psychology, 45, 343–366. Schmidt, U., Ryan, E. G., Bartholdy, S., Renwick, B., Keyes, A., O’Hara, C., … & Treasure, J. (2016). Two-year follow-up of the MOSAIC trial: A multicenter randomized controlled trial comparing two psychological treatments in adult outpatients with broadly defined anorexia nervosa. International Journal of Eating Disorders, 49(8), 793-800. Smink, F. R. E., Van Hoeken, D., & Hoek, H. W. (2013). Epidemiology, course, and outcome of eating disorders. Current Opinion in Psychiatry, 26, 543-48. Smith, K. E., Crosby, R. D., Wonderlich, S. A., Forbush, K. T., Mason, T. B., & Moessner, M.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------(2018). Network Analysis: An Innovative Framework for Understanding Eating Disroder Psychopathology. Spindler, A. & Milos, G. (2007). Links between eating disorder symptom severity and psychiatric comorbidity. Eating Behaviors, 8(3), 364-73. Steinhausen, H. C. (2009). Outcome of Eating Disorders. Child and Adolescent Psychiatric Clinics of North America, 18(1), 225-242. Thomas, J. J., Vartanian, L. R., & Brownell, K. D. (2009). The relationship between eating disorder not otherwise specified (EDNOS) and officially recognized eating disorders: Meta-analysis and implications for DSM. Psychological Bulletin, 135, 407-33. van Borkulo, C., Boschloo, L., Borsboom, D., Penninx, B. W., Waldorp, L. J., & Schoevers, R. A. (2015). Association of symptom network structure with the course of depression. JAMA Psychiatry, 72, 1219-1226. Wade, T. D., Treasure, J., Schmidt, U., Fairburn, C. G., Byrne, S., Zipfel, S., & Cipriani, A. (2017). Comparative efficacy of pharmacological and non-pharmacological interventions for the acute treatment of adult outpatients with anorexia nervosa: study protocol for the systematic review and network meta-analysis of individual data. Journal of Eating Disorders, 5(1). Watson, H. & Bulik, C. M. (2013). Update On the Treatment of Anorexia Nervosa: Review of Clinical Trials, Practice Guidelines and Emerging Interventions. Weisz, J. R., Chorpita, B. F., Palinkas, L. A., Schoenwald, S. K., Miranda, J., Bearman, S. K., ... & Gray, J. (2012). Testing standard and modular designs for psychotherapy treating depression, anxiety, and conduct problems in youth: A randomized effectiveness trial. Archives of General Psychiatry, 69, 274-282.
--------------------------------------------------PREPRINT VERSION--------------------------------------------------Wichers M., & Groot P. C. (2016). Critical Slowing Down as a Personalized Early Warning Signal for Depression. Psychotherapy and Psychosomatics, 85(2), 114-116. Wild B, Friederich HC, Zipfel S, Resmark G, Giel K, Teufel M, … & Herzog W. (2016). Predictors of outcomes in outpatients with anorexia nervosa - Results from the ANTOP study. Psychiatry Research, 244, 45-50. Zipfel, S., Löwe, B., Reas, D. L., Deter, H. C., & Herzog, W. (2000). Long-term prognosis in anorexia nervosa: lessons from a 21-year follow-up study. The Lancet Psychiatry, 355, 721-722.
Acknowledgement: The MOSAIC trial was supported by the National Institute of Health Research (NIHR) under its Programme Grants for Applied Research (RP‐PG‐0606‐1043) and Research for Patient Benefit (PB‐PG‐0613‐31050) initiatives. Ulrike Schmidt is supported by an NIHR Senior Investigator Award and receives salary support from the NIHR Mental Health Biomedical Research Centre at the South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed herein are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
Supplemental Figure S1. Network structures at all time points
Supplemental Figure S2. Bridge Centrality Between AN and Depression
Supplemental Figure S3. Bridge Centrality Between AN and Anxiety/Stress
Baseline, EI expectedInf wtimport wtdis weigh vomit socialeat shapeimport shapedis secreteat restrict preoc_fd preoc overeat losscontrol losewt guilt foodavoid flatstom feelfat fearwtgain expodisc excesexer emptystom dietrule bodydisc binge avoideat wtimport
wtdis
weigh
vomit
socialeat
shapeimport
shapedis
secreteat
restrict
preoc_fd
preoc
overeat
losscontrol
losewt
guilt
foodavoid
flatstom
feelfat
fearwtgain
expodisc
excesexer
emptystom
dietrule
bodydisc
binge
avoideat
6 Months, EI expectedInf wtimport wtdis weigh vomit socialeat shapeimport shapedis secreteat restrict preoc_fd preoc overeat losscontrol losewt guilt foodavoid flatstom feelfat fearwtgain expodisc excesexer emptystom dietrule bodydisc binge avoideat wtimport
wtdis
weigh
vomit
socialeat
shapeimport
shapedis
secreteat
restrict
preoc_fd
preoc
overeat
losscontrol
losewt
guilt
foodavoid
flatstom
feelfat
fearwtgain
expodisc
excesexer
emptystom
dietrule
bodydisc
binge
avoideat
12 Months, EI expectedInf wtimport wtdis weigh vomit socialeat shapeimport shapedis secreteat restrict preoc_fd preoc overeat losscontrol losewt guilt foodavoid flatstom feelfat fearwtgain expodisc excesexer emptystom dietrule bodydisc binge avoideat wtimport
wtdis
weigh
vomit
socialeat
shapeimport
shapedis
secreteat
restrict
preoc_fd
preoc
overeat
losscontrol
losewt
guilt
foodavoid
flatstom
feelfat
fearwtgain
expodisc
excesexer
emptystom
dietrule
bodydisc
binge
avoideat
24 Months, EI expectedInf wtimport wtdis weigh vomit socialeat shapeimport shapedis secreteat restrict preoc_fd preoc overeat losscontrol losewt guilt foodavoid flatstom feelfat fearwtgain expodisc excesexer emptystom dietrule bodydisc binge avoideat wtimport
wtdis
weigh
vomit
socialeat
shapeimport
shapedis
secreteat
restrict
preoc_fd
preoc
overeat
losscontrol
losewt
guilt
foodavoid
flatstom
feelfat
fearwtgain
expodisc
excesexer
emptystom
dietrule
bodydisc
binge
avoideat
Baseline, Bridge EI, AN & Depression/Anxiety/Stress bridge_expectedInf wtimport wtdis wpanic worthless winddown weigh vomit unenthused tremble touchy socialeat shapeimport shapedis secreteat scared restrict preoc_fd preoc panic overreact overeat nerveng meaningless losscontrol losewt lookforward initi heartrate guilt frustr foodavoid flatstom feelfat fearwtgain expodisc excesexer emptystom drymouth drelax dietrule breath bodydisc blue binge avoideat anhed agitate wtimport wtdis wpanic worthless winddown weigh vomit unenthused tremble touchy socialeat shapeimport shapedis secreteat scared restrict preoc_fd preoc panic overreact overeat nerveng meaningless losscontrol losewt lookforward initi heartrate guilt frustr foodavoid flatstom feelfat fearwtgain expodisc excesexer emptystom drymouth drelax dietrule breath bodydisc blue binge avoideat anhed agitate
Baseline, Bridge EI, AN & Depression bridge_expectedInf wtimport wtdis worthless weigh vomit unenthused socialeat shapeimport shapedis secreteat restrict preoc_fd preoc overeat meaningless losscontrol losewt lookforward initi guilt foodavoid flatstom feelfat fearwtgain expodisc excesexer emptystom dietrule bodydisc blue binge avoideat anhed wtimport
wtdis
worthless
weigh
vomit
unenthused
socialeat
shapeimport
shapedis
secreteat
restrict
preoc_fd
preoc
overeat
meaningless
losscontrol
losewt
lookforward
initi
guilt
foodavoid
flatstom
feelfat
fearwtgain
expodisc
excesexer
emptystom
dietrule
bodydisc
blue
binge
avoideat
anhed
Baseline, Bridge EI, AN & Anxiety/Stress bridge_expectedInf wtimport wtdis wpanic winddown weigh vomit tremble touchy socialeat shapeimport shapedis secreteat scared restrict preoc_fd preoc panic overreact overeat nerveng losscontrol losewt heartrate guilt frustr foodavoid flatstom feelfat fearwtgain expodisc excesexer emptystom drymouth drelax dietrule breath bodydisc binge avoideat agitate wtimport
wtdis
wpanic
winddown
weigh
vomit
tremble
touchy
socialeat
shapeimport
shapedis
secreteat
scared
restrict
preoc_fd
preoc
panic
overreact
overeat
nerveng
losscontrol
losewt
heartrate
guilt
frustr
foodavoid
flatstom
feelfat
fearwtgain
expodisc
excesexer
emptystom
drymouth
drelax
dietrule
breath
bodydisc
binge
avoideat
agitate
