Introduction
Anorexia nervosa (AN) is a severe psychiatric illness characterized by an intense fear of gaining weight and a distortion of body image, leading to malnutrition that is often denied. Weight controlling strategies such as food restriction, purging (vomiting and laxative misuse), and inappropriate physical exercise are its core behavioral symptoms [1–Reference Treasure, Duarte and Schmidt3]. The paradoxical persistence of these behaviors along with ongoing malnutrition, and in spite of self-damage [Reference Walsh4], mirrors a tendency to prefer the immediate effect of these maladaptive behaviors over their long-term negative consequences [Reference Brogan, Hevey and Pignatti5, Reference Tenconi, Degortes, Clementi, Collantoni, Pinato and Forzan6].
Tasks exploring decision-making ability have shown impaired performances in patients with AN [Reference Guillaume, Gorwood, Jollant, Van den Eynde, Courtet and Richard-Devantoy7]. This inability to make the most advantageous decisions could be underpinned by different factors, such as a dysregulation of reward processing [Reference Guillaume, Gorwood, Jollant, Van den Eynde, Courtet and Richard-Devantoy7–Reference O’Hara, Campbell and Schmidt9], anxiety, poor enteroception, intolerance to uncertainty [Reference Bechara and Damasio10–Reference van Elburg, Danner, Sternheim, Lammers and Elzakkers12], and cognitive rigidity [Reference Tchanturia, Anderluh, Morris, Rabe-Hesketh, Collier and Sanchez13]. If decision-making impairment is widely recognized as a cognitive feature of AN, it is still debated whether it represents an endophenotype (i.e., associated with the illness, heritable, trait-associated, and present in unaffected family members of patients at higher rates than in the general population) [Reference Gottesman and Gould14], and whether it is a state-associated alteration (i.e., being explained by acute symptoms and malnutrition) or trait-associated feature (associated to illness and present at any stage of the disease, independently from acute symptoms). Past explorations of decision-making ability in AN provide contrasting findings and make difficult to close the debate. Two studies on remitted patients found normal decision-making in patients after weight restoration [Reference Steward, Mestre-Bach, Agüera, Granero, Martín-Romera and Sánchez15, Reference King, Bernardoni, Geisler, Ritschel, Doose and Pauligk16], suggesting this alteration as state-associated, while others support a trait-associated nature for decision-making alteration, as independent from nutritional status [Reference Guillaume, Gorwood, Jollant, Van den Eynde, Courtet and Richard-Devantoy7, Reference Bodell, Keel, Brumm, Akubiro, Caballero and Tranel17, Reference Giannunzio, Degortes, Tenconi, Collantoni, Solmi and Santonastaso18]. The only study including healthy relatives proposed impaired decision-making as an endophenotype of AN [Reference Galimberti, Fadda, Cavallini, Martoni, Erzegovesi and Bellodi19].
Most of the previous studies limited the exploration to the raw scores of the neuropsychological test used, that is, in most cases, the Iowa gambling task (IGT) [Reference Bechara, Damasio, Damasio and Anderson20]. Raw IGT scores measure risk-related decision-making under conditions of uncertainty [Reference Kim, Kang and Lim21], but do not identify the psychological processes underneath [Reference Steingroever, Wetzels and Wagenmakers22]. Cognitive process modeling, informed by cognitive neuroscience and Bayesian theories [Reference Steingroever, Pachur, Šmíra and Lee23], allows a more precise analysis of the multiple processes involved in a cognitive task by deconstructing task performance into theorized underlying psychological processes [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24].
The prospect valence learning (PVL) model is a performant and widely used model to explain the mechanisms underlying decision-making [Reference Ahn, Busemeyer, Wagenmakers and Stout25], already employed in studies on patients with AN [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24, Reference Na, Kang and Kim26].
Previous studies employing the PVL model found discrepant results [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24, Reference Na, Kang and Kim26, Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27]. Small samples of patients with AN [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24, Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27] or at risk of AN [Reference Na, Kang and Kim26] were compared with patients with bulimia nervosa [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24], and healthy controls (HC) [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24, Reference Na, Kang and Kim26, Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27]. Patients with AN reported an alteration of the decision-making parameters identified by the PVL, such as lower learning scores [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24, Reference Na, Kang and Kim26], response consistency [Reference Na, Kang and Kim26], and loss aversion [Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27]. Neither patients remitted from AN, nor unaffected relatives (UR) of patients have ever been tested with this model.
Several questions on decision-making in AN are thus still open. It is still to be determined whether altered decision-making is an endophenotype of AN, and whether it is associated with the illness as a trait or a state. Moreover, nothing is known about the endophenotypic, trait, or state nature of the cognitive dimensions underlying decision-making in AN.
In this context, studies addressing the nature of decision-making alteration in AN are still necessary, and should overcome the limitations of small sample sizes, and analyses limited to raw IGT scores. Another problem is represented by the heterogeneity of the cognitive models employed. Comparability with previous studies will be ensured by choosing one of the already-used cognitive models in previous research.
The aim of the current study was to assess whether impaired decision-making ability is an endophenotype of AN, and whether it is a state- or trait-associated feature of the illness. We also assumed different decision-making profiles in patients with acute AN (A-AN), distinguishing them from patients with remitted AN, UR, and HC.
Methods
Participants
The present study utilizes data collected as part of a monocentric, cross-sectional open trial. Four hundred and eleven participants were offered the possibility to participate in this protocol. Three were excluded, two because of previous knowledge of the test and one because of registration failure of results. In total, 408 participants were included in this study, of whom 91 were patients with A-AN, 90 were UR of patients with AN (UR) unrelated to the patients included (79 mothers, 11 sisters), 204 were HC subjects and 23 were patients remitted from AN (R-AN).
Patients with A-AN and R-AN were recruited from the inpatient and outpatient unit of the Clinique des Maladies Mentales et de l’Encéphale of Sainte Anne hospital in Paris. Inclusion criteria for patients with A-AN were a diagnosis of A-AN according to the DSM-5 [1] and the Diagnostic interview for genetic studies (DIGS) [Reference Preisig, Fenton, Matthey, Berney and Ferrero28, Reference Nurnberger29], and body mass index (BMI) < 18.5 kg/m2. Patients with R-AN were included if they had a diagnosis of past AN at the DIGS and of remitted AN at the DSM-5, with a BMI > 18.5 kg/m2. All patients were assessed only once.
UR were recruited from family groups and family interviews provided in the same hospital. HC were recruited at universities, at research laboratories and by ear-to-mouth and should report no family history of eating disorders. Exclusion criteria for both groups were BMI < 18.5 kg/m2 and a diagnosis of past or present personal history of eating disorders according to the DIGS.
All participants were Caucasian females. Previous knowledge of the IGT, impaired performance at the National Adult Reading Test (NART), poor understanding of French language, and a diagnosis of psychosis and/or bipolar disorder represented exclusion criteria for all groups.
All study procedures were approved by the Ile-de-France III ethics committee (#A01636-49) and the Commission Nationale de l’Informatique et des Libertés. In accordance with the Helsinki declaration, written informed consent was obtained from each participant before inclusion.
Variables
Age, BMI, and NART score [Reference Bovet30] were collected for each participant. All participants were measured in height and weight after night starvation, and BMI was calculated in kg/m2.
The DIGS is a semistructured interview developed to collect comprehensive databases of psychiatric symptoms, signs, current, and lifetime psychiatric history. This instrument showed good validity, interrater, and test–retest reliability in all translated versions. Moreover, it proved capable features of collecting both cross-sectional and longitudinal data with clear diagnostic criteria, and extensive information on the course and chronology of comorbid conditions, thereby reducing their contamination on the psychiatric phenotypes of interest [Reference Preisig, Fenton, Matthey, Berney and Ferrero28, Reference Nurnberger29]. Illness duration, lifetime BMI, and duration of remission from AN were collected by the DIGS, given their potential impact on cognitive performance [Reference Miranda-Olivos, Testa, Lucas, Sánchez, Sánchez-González and Granero31].
The NART consists of a list of irregularly spelled words, graded in difficulty [Reference Mackinnon, Ritchie and Mulligan32]. This test is adopted as a screening measure for verbal intelligence, as the number of words pronounced correctly shows a high correlation with the Wechsler Adult Intelligence Scale [Reference Wechsler33] verbal and total IQ scores (r = 0.80 and r = 0.77, respectively).
Eating disorder psychopathology was assessed by the Eating disorder inventory-2 (EDI-2), using a validated French version [Reference Criquillon-Doublet, Divac, Gaillac, Dardennes and Guelfi34]. This widely used instrument explores attitudinal and behavioral dimensions relevant to eating disorders. Internal consistency is >0.60 in healthy subjects and >0.80 in patients with AN for the eight EDI original dimensions [Reference Garner, Olmstead and Polivy35], and 0.65 to 0.75 for the three additional dimensions [Reference Eberenz and Gleaves36].
Since previous studies of decision-making on eating-disordered samples found a correlation of poor decision-making with anxiety [Reference Bishop and Anxiety37] and impulsivity [Reference Garrido and Subirà38], we employed two scales to measure these dimensions.
Anxiety was measured by the French version of the State–Trait Anxiety Inventory-Y (STAI-Y) [Reference Gauthier and Bouchard39]. The STAI-Y is a self-reported inventory composed of 40 questions based on a 4-point Likert scale. The STAI-Y measures two types of anxiety—state anxiety, or situational anxiety, and trait anxiety, as a personal characteristic. Higher scores are positively correlated with higher levels of anxiety [Reference Spielberger40]. This scale shows good internal consistency both for the state (Cronbach’s alpha 0.90) and the trait (0.91) aspects [Reference Gauthier and Bouchard39].
Impulsivity was assessed by the French version of the Barratt Impulsivity Scale (BIS-10) [Reference Baylé, Bourdel, Caci, Gorwood, Chignon and Adés41]. The BIS-10 is a gold standard measure of impulsiveness by the analysis of the subtracts of cognitive impulsiveness, motor impulsiveness, and nonplanning impulsiveness [Reference Barratt, Spence and Itard42]. Internal consistency for the total score of the French-validated version is 0.82 [Reference Baylé, Bourdel, Caci, Gorwood, Chignon and Adés41].
Iowa gambling task
The most widely used instrument to assess decision-making is the IGT. This test profiles the capacity to learn from past experience, the tendency for risk-taking, and impulsive behavior, characterized by the preference of immediate gains despite negative consequences. The IGT assesses decision-making alterations [Reference Bechara, Damasio, Damasio and Anderson20], simulating real-life decision-making by factoring reward and punishment [Reference Linnet43]. Players have to choose between cards providing immediate rewards, with the risk of important losses, and cards associated with moderate gains, but also moderate losses.
A computerized version of the IGT [Reference Bechara44] was used to evaluate decision-making. Participants have to choose from 4 decks of 25 cards (100 cards in total) presented on a computer monitor, with the aim of achieving monetary gain. While decks A and B give large immediate gains but also high losses, (high-risk), decks C and D give smaller rewards but also lower losses (low-risk). No information is provided on the content of each deck.
The 100 experimental trials can be divided into blocks of 20 trials each, where the block net score is calculated by subtracting the number of trials choosing advantageous decks, minus the number of trials choosing disadvantageous decks [(C + D) − (A + B)]. Performance is also assessed by the net score calculated on the 100 sorts (net IGT score).
Cognitive modeling analysis by PVL model
This model was developed from the expectancy valence model (EVL), that assumes “attention to winning,” “recency,” and “consistency” as the three determinants of IGT performance [Reference Busemeyer and Stout45]. The PVL has the advantage to account for the influence of the frequency of gains and losses on the formation of expectancy, and to be more performant in short-term prediction and more fit for complex choice behaviors [Reference Kim, Kang and Lim21, Reference Ahn, Busemeyer, Wagenmakers and Stout25]. In this model, the motivational process is analyzed by two proxies: “aversion to losses” and “sensitivity to feedback” [Reference Steingroever, Wetzels and Wagenmakers46]. Deck selection in each trial of the IGT is based on the expectancy of valence, based on learning on the experience of the previous gains and losses [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24]. The expectancy of valence [u(t)] is formed by the ensemble of the magnitude of gains/losses, aversion to loss, and sensitivity to feedback as described by the equation, where x(t) is the net gain on the tth trial.
$$ u\left(\mathrm{t}\right)=\left\{\begin{array}{c}\mathrm{x}\;{\left(\mathrm{t}\right)}^{\alpha}\\ {}\hbox{-} \unicode{x03BB} {\left|\mathrm{x}\;\left(\mathrm{t}\right)\right|}^{\unicode{x03B1}}\end{array}\right.\hskip0.24em {\displaystyle \begin{array}{c}\hskip2em if\;\mathrm{x}\left(\mathrm{t}\right)\ge 0\\ {}\hskip2em if\;\mathrm{x}\left(\mathrm{t}\right)<0\end{array}} $$
x(t) α determines the feedback sensitivity (α) (ranging from 0 to 1): the nonlinear relationship between the magnitude of the valence expected and that of the actual gain or loss. The closer to 1, the more proportional the expectance to the actual gain/loss. Loss aversion (λ) (0 to 5) measures the tendency for losses to influence the expectancy valence as opposed to gains. Values higher than 1 indicate greater sensitivity to losses than to gains. In addition to the α and λ parameters present in the equation, learning (A) (0 to 1) indicates at what degree the earlier experience influences the current expectancy valence. Response consistency (c) (0 to 5) represents the influence of expectancy on deck choice as opposite to random deck selection.
The four parameters of the PVL model were estimated by using the computational model developed and described by Anh et al. [Reference Ahn, Haines and Zhang47], who provided an R package hBayesDM (hierarchical Bayesian modeling of Decision-Making tasks) [Reference Ahn, Krawitz, Kim, Busmeyer and Brown48, 49]. This package offers a hierarchical Bayesian modeling which implements a Markov Chain Monte Carlo (MCMC) algorithm called Hamiltonian Monte Carlo (HMC). A more comprehensive description of HMC can be found at (https://mc-stan.org/documentation/) and [Reference Krusckhe50]. Using the HMC sampling scheme, 1000 samples were drawn after a 800-sample burn-in with four chains.
Statistical analysis
Normality of distribution was assessed by the inspection of skewedness and kurtosis. Values of skewedness between −2 and + 2, and values of kurtosis between −7 and + 7, were considered acceptable to prove normal univariate distribution [Reference George and Mallery51, Reference Hair52].
A-AN, R-AN, UR, and HC groups were compared on sociodemographic and clinical variables by analyses of variance (ANOVAs) with Bonferroni corrections for multiple testing.
The following decision-making indicators were compared between the four groups: IGT scores, PVL-A (learning), PVL-alpha (feedback sensitivity), PVL-C (consistency), and PVL-lambda (loss aversion), and included in the analysis of covariance (ANCOVA) as dependent variables. Age, intellectual level (NART score), impulsivity (BIS-10 score), and state anxiety (STAI-A score) were included as covariates in the model [Reference van Elburg, Danner, Sternheim, Lammers and Elzakkers12, Reference Giannunzio, Degortes, Tenconi, Collantoni, Solmi and Santonastaso18, Reference Bishop and Anxiety37, Reference Garrido and Subirà38, Reference Davis, Fox, Patte, Curtis, Strimas and Reid53].
All tests were submitted to post hoc Bonferroni correction for multiple testing. A variable distinguishing patients and UR from HC would be retained as an endophenotype. One distinguishing A-AN and R-AN from UR and HC would be considered as trait-associated.
To rule out the potential role of depression [Reference Abbate-Daga, Buzzichelli, Marzola, Aloi, Amianto and Fassino54], all analyses were repeated after excluding all participants with a diagnosis of depression at the DIGS. The effect of psychotropic drugs on cognitive performance was assessed by comparing IGT performance of patients with versus without psychotropic medications.
All statistical analyses were performed by SPSS (IBM Corp. Released 1989, 2021. IBM SPSS Statistics for Macintosh, Version 28.0.0.0 (190) Armonk, NY: IBM Corp).
Results
Descriptive results
All data were normally distributed. Table 1 presents sample characteristics, IGT performance, and PVL model parameters for each group.
Clinical, demographic, psychometric characteristics, and IGT parameters of patients with acute anorexia nervosa (A-AN), healthy controls (HC) unaffected relatives of patients with AN (UR) and patients remitted from AN (R-AN).
Note: The variables marked with an asterisk were calculated by ANCOVAs with age, NART, BIS-10, and STAI-A scores as covariates.
Abbreviations: BIS, Barratt impulsivity scale; BMI, body mass index; EDI, eating disorder inventory; IGT, Iowa gambling test; NART, national adult reading test; PVL, prospect valence learning; STAI, state–trait anxiety inventory.
Average illness duration was of 87.81 ± 78.64 months for A-AN patients, and of 63.91 ± 48.36 months for the R-AN group, with remission duration of 77.21 ± 83.36 months. Fifty patients had daily benzodiazepine administration, 39 had antidepressants, and 17 had antipsychotics (Supplementary Table S1).
Age was significantly different across groups, and patients had the highest levels of anxiety (p < 0.01). No difference was found in impulsivity among groups.
IGT performance and PVL model parameters
A-AN patients had the lowest total net IGT scores (p < 0.01). Feedback sensitivity (PVL-alpha) was significantly lower in A-AN than in HC, whereas no difference emerged for the three other PVL parameters (p > 0.05). A-AN patients also showed higher aversion to losses (PVL-lambda) than R-AN (p < 0.01) (Figure 1).
Decision-making parameters according to the prospect valence learning model across four groups of patients with acute anorexia nervosa (A-AN), remitted anorexia nervosa (R-AN), unaffected relatives (UR), and healthy controls (HC). The four decision-making parameters are: learning (A), feedback sensitivity (alpha), consistency (C), and loss aversion (lambda). A-AN is characterized by low feedback sensitivity. R-AN are characterized by low feedback sensitivity and lower loss aversion than A-AN. Significant inter-group differences are shown by asterisks (p < 0.05).
UR had comparable performance with HC on all parameters (p > 0.05), except for feedback sensitivity that was higher than in the three other groups (p < 0.01).
Like A-AN, R-AN had low values of feedback sensitivity (p < 0.01), but their performance at the IGT was comparable to HC (p > 0.05).
Significance of results was preserved after excluding the 22 patients with depression for the five-block IGT score (F(3,376) = 3.06, p = 0.03), the IGT total net score (F(3,376) = 2.97, p = 0.03), and the PVL parameter analyses (F(3,376) = 28.00, p < 0.01 for PVL-alpha, and F(3,376) =3.37, p = 0.02 for the PVL-lambda).
Benzodiazepines, antidepressants, and antipsychotics did not affect IGT performance in patients with AN (p > 0.05) (Supplementary Table S1). Supplementary Figure S1 shows mean IGT block scores for each group.
Post hoc correlations
To explore whether any clinical indicator was related to IGT performance, we calculated post hoc correlations between total net IGT score and duration of illness (in A-AN and R-AN) and of remission (in R-AN), and minimum and current BMI (in the whole sample), controlling for age, NART, STAI-A, and BIS-10. We found no correlation between IGT performance and duration of illness (r = −0.08, n = 114, p = 0.42), of remission (r = 0.07, n = 23, p = 0.08), and minimum lifetime BMI (r = 0.07, n = 408, p = 0.79), and a significant correlation between total net IGT score and current BMI (r = 0.11, n = 408, p = 0.04).
Discussion
This study explored whether decision-making deficits could represent a third cognitive endophenotype of AN and/or a trait-associated feature, by a cognitive modeling approach. The present results suggest decision-making impairment as a state-associated feature of AN, partially underpinned by a trait-associated diminution of feedback sensitivity.
This study dovetails with previous research confirming impaired performance at the IGT in patients with A-AN [Reference Guillaume, Gorwood, Jollant, Van den Eynde, Courtet and Richard-Devantoy7]. Normal performance of UR suggests that this impairment is not an endophenotype, respecting no criteria of an endophenotype as listed by Gottesman and Gould [Reference Gottesman and Gould14]. Moreover, in line with previous studies including remitted patients [Reference Steward, Mestre-Bach, Agüera, Granero, Martín-Romera and Sánchez15, Reference King, Bernardoni, Geisler, Ritschel, Doose and Pauligk16], we found comparable IGT scores between R-AN and HC, suggesting decision-making impairment as a state-related feature of AN, unaffected in patients who previously suffered from AN and are at normal nutritional status.
The intriguing finding of low PVL-feedback sensitivity in remitted patients suggests this cognitive feature as trait-dependent in AN. Low sensitivity to feedback may underpin some alterations observed in patients with AN, such as the difficulty in discriminating between reward and punishment [Reference O’Hara, Campbell and Schmidt9], and reduced feedback learning [Reference Foerde and Steinglass55]. Altered responses to reward, and difficulties in discriminating positive from negative feedback, were already described in remitted patients [Reference Kaye, Fudge and Paulus56], even though monetary reward processing was investigated by means of other cognitive tasks, such as the delay discounting test. Studies of functional magnetic resonance supporting this hypothesis showed a lack of differential activation in reward-related neural circuits in response to monetary wins and losses [Reference Wagner, Aizenstein, Venkatraman, Fudge, May and Mazurkewicz57].
The coexistence of trait-associated cognitive dimensions underpinning decision-making, and of successful decision-making after remission from AN, may suggest the existence of compensatory mechanisms occurring at a normal nutritional status and/or relief from symptoms.
For instance, a functional magnetic resonance study on women recovered from AN found exaggerated activation in the caudate–dorsal striatum and in the “cognitive” cortical regions that project to this area, such as the dorsolateral prefrontal and the parietal cortex [Reference Wagner, Aizenstein, Venkatraman, Fudge, May and Mazurkewicz57]. One could assume that the hyperactivation of these brain regions could represent a compensatory mechanism allowing increased cognitive control on decision-making in remitted patients. Additionally, remitted patients showed enhanced cognitive control allowing for “strategic” (as opposed to hedonic) approaches to improve the ratio of wins to losses, despite altered reward system activation to monetary stimuli [Reference Kaye, Fudge and Paulus56, Reference Wierenga, Bischoff-Grethe, Melrose, Irvine, Torres and Bailer58]. Learning capacity may also play a role in the improvement of decision-making and response to therapy [Reference Lucas, Miranda-Olivos, Testa, Granero, Sánchez and Sánchez-González59]. Impaired IGT performance in A-AN could be interpreted as a consequence of weaker prefrontal efficiency secondary to malnutrition, and/or state-dependent neurobiological alterations such as reduced amygdala gray matter volume [Reference Friederich, Walther, Bendszus, Biller, Thomann and Zeigermann60] and altered fronto-amygdalar response [Reference Steward, Martínez-Zalacaín, Mestre-Bach, Sánchez, Riesco and Jiménez-Murcia61].
Post hoc analyses allowed an investigation of the clinical factors involved in decision-making performance, highlighting the role of poor nutritional status. The association between cognitive parameters of IGT performance and state markers of AN, already found in previous literature [Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27], suggests that the achievement of a normal nutritional state and long-term remission may reverse these deficits [Reference Steward, Mestre-Bach, Agüera, Granero, Martín-Romera and Sánchez15]. Further cognitive parameters that seem significantly different in acute versus remitted AN, such as high loss aversion, may play an additional role in decision-making alterations in A-AN by a differential evaluation of gains and losses.
This study presents discrepant results from those that previously applied the PVL model in AN [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24, Reference Na, Kang and Kim26]. For example, higher sensitivity to feedback was found in patients with BN but not with AN, who were characterized by reduced learning ability [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24]. Verharen et al. [Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27], had found low loss aversion in patients with AN. Some explanations could be the use of a different cognitive model, or a different approach to the task: as underlined by the authors, the values of the cognitive parameter of consistency rather suggested a more random choice of decks as the session progressed [Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27], reflecting fatigue or boredom.
This work proposes a neurocognitive interpretation of the persistence of symptoms of AN, as a result of the progressive deterioration of decision-making along with illness progression. Disadvantageous behaviors, such as pursuing weight-loss behaviors despite their long-term negative effects on health, could be seen as a result of this impairment of decision-making and reward evaluation [Reference Cavedini, Zorzi, Bassi, Gorini, Baraldi and Ubbiali62]. For instance, impaired feedback sensitivity could explain the pursuit of certain behavioral excesses, devoted to weight loss, and insensitive to some consequences that healthy subjects would consider punishing, such as hunger and fatigue [Reference Haynos, Lavender, Nelson, Crow and Peterson63].
This research presents a set of limitations. First, this protocol did not include any quantitative assessment of depression besides eliminating qualitative major depressive episodes, thanks to the DIGS, which can be considered as a major limitation given the central role of mood variability in AN psychopathology [Reference Monteleone and Cascino64]. The role of depression on IGT performance is still debated, but the most recent systematic review on the role of depression on cognitive performance in AN mainly showed no influence of depressive symptoms on decision-making [Reference Abbate-Daga, Buzzichelli, Marzola, Aloi, Amianto and Fassino54]. More recent studies replicate this finding [Reference Chan, Ahn, Bates, Busemeyer, Guillaume and Redgrave24, Reference Fornasari, Gregoraci, Isola, Laura Negri, Rambaldelli and Cremaschi65], and that differences in model fit between patients with AN and HC were not driven by baseline differences in depression [Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27]. We ruled out the influence of depression anyway, repeating all analyses after excluding patients with current depressive disorder according to the DIGS, and showing that results did not change.
A more consistent body of evidence supports a role for anxiety as a potential confounder of decision-making in eating disorders [Reference Fornasari, Gregoraci, Isola, Laura Negri, Rambaldelli and Cremaschi65], with contrasting findings [Reference Verharen, Danner, Schröder, Aarts, van Elburg and Adan27]. Unsurprisingly, A-AN and R-AN had the highest levels of anxiety. Even though separating anxiety from AN psychopathology remains delicate, given the high entanglement between the two syndromes [Reference Lavender, De Young, Wonderlich, Crosby, Engel and Mitchell66], we excluded the contaminating effect of state anxiety on decision-making [Reference Miu, Heilman and Houser67], by including this covariate in the ANCOVA.
Another limitation is represented by the disparity between the sample size of R-AN and that of the three other groups. This limitation may have affected significance of some results, but was sufficient to replicate results of previous literature, in particular, that of significant differences between acute and recovered patients. The relatively scarce number of R-AN patients is due to recruitment difficulties, but is in line from that of previous studies on the subject [Reference Steward, Mestre-Bach, Agüera, Granero, Martín-Romera and Sánchez15, Reference Foerde and Steinglass55]. Accordingly, the PVL model provided intriguing results that are complementary to, but do not explain, decision-making impairment in A-AN.
The cross-sectional design of the study limits the generalizability of results. It has been shown that A-AN is characterized by great interindividual variability in decision-making, with better performances predicting response to therapy [Reference Lucas, Miranda-Olivos, Testa, Granero, Sánchez and Sánchez-González59]. A prospective evaluation is indispensable to disentangle the relationship between remission achievement and decision-making performance, and to confirm the impact of renutrition on the latter.
Finally, it could be criticized that no distinction has been made between restricting-type and binge-purging AN. On the other hand, previous studies found comparable performance between the two groups [Reference Giannunzio, Degortes, Tenconi, Collantoni, Solmi and Santonastaso18, Reference Cavedini, Zorzi, Bassi, Gorini, Baraldi and Ubbiali62] even though this result is still debated [Reference Guillaume, Gorwood, Jollant, Van den Eynde, Courtet and Richard-Devantoy7]. Nonetheless, altered performance at the IGT has been demonstrated in both subtypes [Reference Guillaume, Gorwood, Jollant, Van den Eynde, Courtet and Richard-Devantoy7] and additional analyses distinguishing the two groups were out of the scope of this research.
Conclusion
Decision-making performance seems a state-associated cognitive feature of AN, prevalent at acute stages of the disease. Decreased sensitivity to feedback was found in both acute and remitted patients, suggesting trait-associated features. Decision-making process in AN may be the result of the interplay between trait-associated decisional features and state-associated compensatory mechanisms, allowing for unimpaired performance in some remitted patients.
Prospective studies are necessary to confirm that weight gain and reduction of symptoms facilitate a more appropriate decision-making process. These two dimensions could serve as important leverages for the improvement of cognitive functioning in patients with AN.
Supplementary Materials
To view supplementary material for this article, please visit http://doi.org/10.1192/j.eurpsy.2022.2327.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author, upon reasonable request.
Author Contributions
Conceptualization: L.D.L., P.G., N.R.; Data curation: L.D.L., A.V., M.L., N.R.; Formal analysis: L.D.L., M.L., J.M.; Funding acquisition: P.G.; Investigation: A.V., P.G.; Supervision: P.G., N.R.; Validation: P.G.; Writing—original draft: L.D.L., M.L., J.M.; Writing—review and editing: L.D.L., P.G.
Financial Support
This work was supported by Sainte Anne Hospital as a project financed by the Programme Hospitalier de Recherche Clinique (PHRC).
Conflicts of Interest
The authors declare none.
Ethical Standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Open access
Comments
No Comments have been published for this article.