Is set-shifting and central coherence in anorexia nervosa influenced by body mass index, anxiety or depression? A systematic review

All primary studies were peer-reviewed and written in English. There were no limitations in terms of time frame. Studies were included if they met the following inclusion criteria:

Literature search was performed in PubMed, PsychInfo and Embase databases. A search for [anorexia nervosa] AND [central coherence] resulted in 73 (PubMed), 54 (PsychInfo) and 73 (Embase) articles. A search for [anorexia nervosa] and [set shifting] resulted in 92 (Pubmed), 101 (PsychInfo) and 117 (Embase) articles. See Additional file 1 for search strategy in PubMed. In total, the search resulted in 510 articles. After the initial searches in the different databases, duplicates were removed. Then, all titles and abstracts of 189 articles were examined and evaluated for relevance, which resulted in removal of 76 articles. Then, full text for all potential relevant articles (113), were obtained and screened thoroughly for relevance. Seventy-two articles were then removed due to lack of statistical analyses between neuropsychological function and clinical measures, or they were not relevant for the current review. In all, a total of 41 articles were included in the present study. Figure 1 illustrates a PRISMA flow chart of the search strategy (see Fig. 1).

The following variables were extracted from the articles: Authors, publication year, sample (size and diagnosis), neuropsychological test, measures of anxiety, depression and eating disorder symptoms, cognitive domain (set-shifting or central coherence), medical treatment. Outcome measures included analyses examining the relationship between set-shifting and/or central coherence and depression/anxiety/BMI/weight for height/eating disorder symptoms.

Risk of bias of the included studies was assessed by using a critical appraisal checklist for case control studies and cross sectional studies from the Joanna Briggs Institute (https://​joannabriggs.​org/​critical-appraisal-tools), which purpose is to assess the methodological quality of a study and to determine the extent to which a study has addressed the possibility of bias in its design, conduct and analysis. See Additional files 2 and 3.

Due to the heterogeneity of the primary studies, especially considering the wide variety of neuropsychological tests and measures of psychopathology, and that different statistical methods have been used, a narrative synthesis approach was used for this current review.

Seven studies included an adolescent sample (age < 18 years) [18‐24], and four studies included both adults and adolescents [25‐28]. The remaining 26 studies included an adult sample (age ≥ 18 years).

A variety of tests were used to measure set-shifting: Haptic Illusion Task, Cognitive Shift Task, Trail Making Test, Brixton Spatial Anticipation Test, Picture Set Test, Verbal Fluency Task, CatBat Task, Intra/Extra Dimensional Set Shift Test, Wisconsin Card Sorting Test, Visual Set-Shifting Task, Berg’s Card Sorting Test, Parametric Go/NoGo Test, Category Learning Task, Task Switching Paradigm, Design Fluency Task, Cued Color-Shape Switching Task and Comprehensive Trail Making Test. The following tests were used to measure central coherence: Rey Complex Figure Test, Embedded Figures Test, Sentence Completion Task, Homograph Reading Task, Overlapping Figures Test, Fragmented Pictures Task, Block Design, Object Assembly and Group Embedded Figures Test.

Different measures were used to assess anxiety and depression. The following measures were used to measure anxiety: State-Trait Anxiety Inventory (n = 14), Hospital Anxiety and Depression Scale (n = 8), Beck’s Anxiety inventory (n = 1), Depression, Anxiety and Stress Scale (n = 2), Spence Children’s Anxiety Scale (n = 1) and Beck Youth Inventory (n = 1). The most common measure of depression was Beck’s Depression Inventory (n = 18), then HADS (n = 8), Depression, Anxiety and Stress Scale (n = 2), Hopkins Symptoms Checklist (n = 1), Quick Inventory of Depressive Symptomatology, Children’s Depression Inventory (n = 1) and German depression inventory for children and adolescents (n = 1).

To examine the associations between cognitive performance and clinical characteristics, different regression analyses accounting for covariates as well as correlation analyses, both Pearson and Spearman’s, were used.

The majority of the studies included in the review had a low risk of bias, i.e. the groups were comparable other than the presence of disease, and the groups were properly matched. Valid, standardized tests were used, and testing procedures were the same for all participants. Confounding variables were identified and controlled for, and appropriate statistical tests were used. However, in some studies, issues concerning representativeness of the patient group and lacking control of possible confounding factors, led to an increased risk of bias. In one study, there was a low participation rate, which could affect the representativeness of the AN group [53]. Another study included a specific sample of AN with strict inclusion criteria, suggesting that this group was not representative for the AN population [55]. One study had a very heterogeneous AN group in terms of differences in duration of illness, age of onset, treatment experience and different levels of illness severity [50]. Furthermore, some of the studies did not control adequately for confounding factors. Most commonly was not considering the effect of pharmacological treatment on test performance [18, 39, 54, 56, 57], and lack of controlling for intelligence quotient [35, 51]. Other potential confounding factors are comorbidity, which was not assessed and controlled for in several studies [44, 56, 57]. These factors contribute to an increased risk of bias in these studies.

The studies that included associations between BMI and neuropsychological function show variable results. The majority of studies (n = 28) found no significant associations between set-shifting or central coherence and BMI [20, 22‐24, 27‐32, 34‐36, 39‐45, 48, 52‐58]. Some studies (n = 7) found an association between BMI and neuropsychological function. One study found that BMI correlated with the number of errors in the Homograph Reading Task, a measure of central coherence [5]. A negative correlation was detected between BMI and the Fragmented Picture Task, but the correlation was not significant after correcting for multiple comparisons [26]. Another study found that significant low central coherence index became non-significant when adjusting for nadir BMI [50]. Herbrich et al. [21] found a negative correlation between BMI and Group Embedded Figures Test in the AN-binge/purge group. In addition, a negative correlation was found between BMI and Trail Making Test – 4 in the AN restrictive group. Studies including measures of set-shifting show that BMI was significantly associated with poor cognitive flexibility, were extreme weight conditions performed worse than healthy controls [46]. Øverås et al. demonstrated significant correlations between BMI and Wisconsin Card Sorting Test [47], and Roberts et al. [49] reported that the group with poor set-shifting abilities and global processing had a lower lifetime BMI than patients with only one neurocognitive inefficiency.

Most studies (n = 16) did not find an association between neuropsychological function and anxiety [5, 19‐22, 25, 26, 28, 30, 32, 34, 39, 42, 52, 53, 56, 59], while nine studies did find a significant association. One study showed that group differences in a Cognitive Shift Task disappeared when controlling for anxiety [59]. Another study found that anxiety scored correlated with the Haptic Illusion task, a measure of set-shifting [31]. Roberts et al. [35] found increased anxiety in the poor set-shifting group, however the effects were small. Another study showed that state anxiety had a moderate correlation with Trail Making Test and Overlapping Figures Test [36]. Øverås et al. and Perpiña et al. also found that set-shifting correlated with anxiety [47, 51]. A better copy accuracy on the Rey Complex Figure Task was related to higher levels of anxiety [38], while trait anxiety in the AN-BP group correlated with Object Assembly, a measure of central coherence. Brown and colleagues [18] suggest that higher levels of anxiety could affect cognitive function.

A total of 20 studies reported that depression was not associated with neuropsychological function [5, 19‐23, 25, 26, 28, 30, 33, 36, 38, 39, 42, 43, 46, 52, 53, 56, 59]. However, nine studies did find significant correlations between depression and measures of set-shifting and central coherence. These studies demonstrate correlations between depression and the Haptic Illusion Task [31], and set-shifting measured by Trail Making Test [34]. Another study found increased depression in the poor set-shifting group, however with small effects [35]. Medium correlations between set-shifting and depression was found in both patients with AN and patients with unipolar depression [40]. Abbate Daga et al. [45] found that AN individuals with poor set-shifting reported higher depression scores, measured by Beck’s Depression Inventory, than those patients intact set-shifting. Furthermore, studies by Øverås et al. [47] and Perpiña et al. [51] show significant correlations between depression and Wisconsin Card Sorting Test. Brown et al. [18] also suggest that higher levels of depression could affect cognitive function such as central coherence and set-shifting.

Some studies investigated the associations between eating disorder severity and neuropsychological functions. Individuals with poor set-shifting had a higher score on the Yale-Brown-Cornell Eating Disorder Scale rituals [35]. Those with most fragmented perseverative cognitive style and the least global flexible style had more chronic and more severe illness, in terms of higher Eating disorder examination questionnaire (EDE-Q) global score and lower BMI [26]. Significant correlations were found between recall accuracy on the Rey Complex Figure Test and total EDE-Q score and scores on the Clinical and Research Inventory for Eating Disorders. In addition, an increase in eating disorder symptoms was related to an improvement in set-shifting measured by the trail making test/TMT [44].

Conversely, other studies showed no correlations between set-shifting performance [32, 40] or central coherence measured by the Rey Complex Figure Task [42] on scores on the Eating Disorder Inventory. Another study found no evidence that symptoms of eating disorders, as measured by EDE-Q, was associated with set-shifting ability [34]. Tenconi et al. [36] divided individuals with AN into three groups, acute AN, weight-recovered but symptomatic AN and long-term fully recovered AN. There were no differences between the groups on measures of set-shifting and central coherence. Furthermore, Danner et al. [38] found no correlations between Iowa Gambling Task and clinical characteristics, while Shott et al. [28] found no significant correlations between shift-cost index on the Category Learning Task and duration of illness. In another study, severity, assessed by the Yale-Brown-Cornell scale scores, BMI, onset and duration of illness, did not correlate with any neurocognitive performance [41]. One study found no significant associations between the Eating Attitudes Test scores and the Iowa Gambling Task total net scores or the Wisconsin Card Sorting Test [51].

The majority of studies report that neither BMI, eating disorder severity, anxiety nor depression is associated with neuropsychological task performance. Deficits in set-shifting and central coherence could thus be related to characteristics of the disorder, and are not a direct consequence of underweight, ED severity, anxiety or depression. However, although most findings points to the same direction, findings are inconsistent. The studies are heterogeneous and methodological differences needs to be considered. Various tests have been used to measure set-shifting and central coherence, one might therefore question whether these results are comparable, as different cognitive tests have different levels of complexity and might target multiple cognitive capabilities. For example, the WCST is primarily a measure of cognitive flexibility, however other cognitive functions such as working memory, inhibition and sustained attention are also necessary to perform this task. The Embedded Figures test appears to the participants to be a measure of attention, but is rather a visuospatial perceptual task measuring central coherence ability. A standardized approach to neuropsychological testing of individuals with AN could contribute to increased consistency across studies, and a better understanding of the neuropathology. Another important issue is that there are numerous measures of psychopathology, and especially for anxiety and depression. Even if they are supposed to measure the same constructs, results from different instruments can be difficult to compare [60].

Furthermore, different statistical methods have been used to determine the relationship between neuropsychological function and clinical measures. Most studies have used correlation analyses (i.e. Pearson or Spearman’s), however some studies did multivariate regression analyses including one or more covariates. The different methods could lead to different conclusions, as a correlation analysis does not adjust for the effect of other covariates.

Neuropsychological performance could also depend on medical treatment. A meta-analysis suggest that deficits in executive functions are greater in patients with more severe current depression, and those taking psychotropic medications [61]. In this review, a few studies accounted for medical treatment and whether it could influence neuropsychological function. Most of these studies suggested that medication did not influence test performance [26, 28, 31, 34, 37], while two studies found differences between medicated and non-medicated patients [36, 42]. Some studies had long-term use of medication [23, 24, 58] and current treatment with psychotropic drugs [25] as exclusion criteria.

It is important to note that cognitive development most likely impact neuropsychological function. Most neuropsychological studies have an adult sample, and studies with an adolescent AN sample are inconsistent, showing both weak and normal set-shifting [27, 62, 63] normal central coherence [21] as well as lower scores on style and central coherence index measured by Rey Complex Figure Test [22]. One study directly compared set-shifting performance in adolescents versus adults with AN and controls [28]. The results showed that set-shifting is normal in adolescents with AN, but impaired in adults, suggesting that neuropsychological difficulties do not precede the disorder, however are rather accompanied by the eating disorder symptoms. Whether neuropsychological function is influenced by clinical measures in adolescents remains unclear. Some studies showed a negative correlation between BMI and both set-shifting and central coherence [21], and that both anxiety and depression could affect cognitive function [18]. Conversely, quite a few studies did not find any significant relationship between set-shifting or central coherence and BMI, anxiety and depression [19, 22‐24, 28].

The main purpose of cognitive neuroscience research in AN is to explain the link between symptoms/behaviour and underlying neural systems. Abnormalities in the fronto-striatal and limbic systems have been suggested to be involved in restrictive eating, which is one of the core symptoms of AN [64]. Central coherence and set-shifting are both executive functions, which rely on frontal lobe activity. More specifically, during a set-shifting task, patients with AN have shown activation in ventrolateral prefrontal cortex, dorsolateral prefrontal cortex, anterior cingulate, superior parietal cortex, cerebellum and precuneus. When performing a central coherence task, parietal, occipital, ventral temporal and premotor cortex is usually activated [65]. Maladaptive behaviours are associated with specific traits, which in turn are associated with neuropsychological processes that are related to specific neural circuits [66]. Clinically, impairments in set-shifting in AN could be linked to excessive control of eating and weight, but also cognitive and behavioural rigidity, which is a common characteristic of the disorder. Patients with AN are often “stuck” in maladaptive behaviours, and have difficulties to change these behavioural patterns. Cognitive and behavioural rigidity could be expressed in terms of being unable to shift perspective from detailed rules and rituals concerning food and eating [6]. Another typical characteristic of individuals with AN, is the increased attention to details, which could be linked to poor global processing. They often have detailed rituals concerning food, caloric intake, exercise and body checking. Taken together, inflexibility and an excessive control of details concerning food and weight, could contribute to the unusual ability to reduce weight and sustain an extreme low weight over a long period.

AN can be very difficult to treat, and there is no proven treatment that can reverse the core symptoms. A complete understanding of neuropsychological functioning in AN, as a link between neural mechanisms and AN symptoms, could contribute to the development of better treatment strategies. As suggested in this review, neuropsychological functions are not influenced by low weight or comorbid disorders, but might be a trait of the illness and should be targeted in treatment with that in mind.

As a result of the emerging neurobiological models of AN, it has been suggested that there is a need for a paradigm shift in the development of new treatment strategies, with more focus on brain-directed treatments [67]. Findings from neuroscientific research has opened up for new avenues for understanding the underlying mechanisms and neuropsychological functions for AN, and for developing new treatments. Temperament based treatment with support (TBT-S) [68] is based on models explaining the underlying neurobiological mechanisms that cause AN. The goal of this treatment is to teach individuals with AN to develop coping strategies for managing their temperament traits that are related to risk factors to develop an eating disorder. The treatment is delivered in a way that is compatible with their neuropsychological thinking style, i.e. detailed oriented, well-structured and predictable. The Maudsley Model of Anorexia Nervosa Treatment for Adults (MANTRA), aims to address the neuropsychological, emotional, relational and biological aspects of AN. It is based on research that has identified the key factors that tend to maintain eating difficulties [69]. The treatment helps patients to understand what causes AN, as well as to encourage patients to change their behaviour when they are ready for it. Cognitive remediation therapy (CRT) aims at improving cognitive flexibility and central coherence, as well as reducing perfectionism and improving the awareness of dysfunctional thinking styles. This treatment is focusing on the process rather than the content of thinking [70].

Neuropsychological research in AN is distinguished by large varieties of tasks within the same cognitive domain and a variety of measures of psychopathology, which impact and limit interpretation of findings and comparison of results across studies. In a systematic review, a quantitative synthesis is considered to be superior methodologically compared to a narrative synthesis of the results. However, due to the heterogeneity of the primary studies, especially considering the wide variety of neuropsychological tests and measures of psychopathology and that different statistical methods have been used, a narrative synthesis approach was used. Another limitation with the existing literature is that few studies include an adolescent sample, who have a short duration of illness, and addressing this group could contribute to an increased understanding of predisposing factors of AN. Additional issues still needs to be addressed, such as when do the neuropsychological impairments occur? Are neuropsychological impairments related to age or to the progression of the disorder? Do neuropsychological impairments contribute to maintain the disorder? Future studies should target these issues, which could contribute to an increased understanding of the neuropathology of AN.

Future studies would also benefit from controlling for the potential effect of medical treatment on neuropsychological performance, include associations between neuropsychological functioning and eating disorder severity, by including other relevant measures that might have clinical relevance in addition to BMI, e.g. EDE-Q, or other measures of eating disorder symptoms.