Comorbidity between major depressive disorder (MDD), anxiety disorders, attention-deficit hyperactivity disorder (ADHD), and autism spectrum disorder (ASD) is high (Antshel, Zhang-James, Wagner, Ledesma, & Faraone, Reference Antshel, Zhang-James, Wagner, Ledesma and Faraone2016; Ghaziuddin, Ghaziuddin, & Greden, Reference Ghaziuddin, Ghaziuddin and Greden2002; Grevet et al., Reference Grevet, Bau, Salgado, Fischer, Kalil, Victor and Belmonte-de-Abreu2006; Hirschfeld, Reference Hirschfeld2001; Joshi et al., Reference Joshi, Wozniak, Petty, Martelon, Fried, Bolfek and Biederman2013; Kaufman & Charney, Reference Kaufman and Charney2000; Kessler et al., Reference Kessler, Adler, Barkley, Biederman, Conners, Demler and Zaslavsky2006a; Kessler, Merikangas, & Wang, Reference Kessler, Merikangas and Wang2006b; Schatz & Rostain, Reference Schatz and Rostain2006; van Loo, Romeijn, de Jong, & Schroevers, Reference van Loo, Romeijn, de Jong and Schroevers2013). To illustrate, 67–73% of the patients diagnosed with an anxiety disorder, are also diagnosed with MDD (Kaufman & Charney, Reference Kaufman and Charney2000; Lamers et al., Reference Lamers, van Oppen, Comijs, Smit, Spinhoven, van Balkom and Penninx2011). Comorbidity rates between ADHD and ASD are estimated between 14% and 78% (Gargaro, Rinehart, Bradshaw, Tonge, & Sheppard, Reference Gargaro, Rinehart, Bradshaw, Tonge and Sheppard2011) and comorbidity rates between these two neurodevelopmental disorders, MDD, and anxiety disorders are estimated between 25% and 75% (Grevet et al., Reference Grevet, Bau, Salgado, Fischer, Kalil, Victor and Belmonte-de-Abreu2006; Joshi et al., Reference Joshi, Wozniak, Petty, Martelon, Fried, Bolfek and Biederman2013; Kessler et al., Reference Kessler, Adler, Barkley, Biederman, Conners, Demler and Zaslavsky2006a; Schatz & Rostain, Reference Schatz and Rostain2006). This pattern of co-occurrence questions the validity of distinct diagnostic categories (Forbes, Tacket, Markon, & Krueger, Reference Forbes, Tacket, Markon and Krueger2016). Instead, comorbid disorders may represent manifestations of common underlying transdiagnostic psychopathology features that cut across diagnostic boundaries (Cuthbert, Reference Cuthbert2014; Forbes et al., Reference Forbes, Tacket, Markon and Krueger2016; Kessler et al., Reference Kessler, Ormel, Petukhova, McLaughlin, Green, Russo and Üstün2011).
The frequent comorbidity among supposedly distinct disorders motivated the National Institute of Mental Health (NIMH) to develop the Research Domain Criteria (RDoC) framework. The goal of RDoC is to explore transdiagnostic, underlying dimensions of psychopathology across a broad range of disorders (Cuthbert, Reference Cuthbert2014; Insel et al., Reference Insel, Cuthbert, Garvey, Heinssen, Pine, Quinn and Wang2010). The Negative Valence System is one such dimension and includes the subconstruct ‘loss’, which describes a state of deprivation, which may include behavior, status, loved ones or relationships (Cuthbert, Reference Cuthbert2014). This construct can be assessed at different levels (e.g. molecules, behavior, physiology). One element on the behavioral level is anhedonia, which can be understood as a loss of pleasure in formerly enjoyable activities and includes a loss of effort/motivation, desire, anticipation, and consummatory pleasure (Rizvi, Pizzagalli, Sproule, & Kennedy, Reference Rizvi, Pizzagalli, Sproule and Kennedy2016). Though traditionally linked to MDD, anhedonia is a common symptom of mental disorders as it is seen in e.g. ADHD, ASD, schizophrenia, substance use disorders, and anxiety disorders [American Psychiatric Association (APA), 2013; Bringmann, Lemmens, Huibers, Borsboom, & Tuerlinckx, Reference Bringmann, Lemmens, Huibers, Borsboom and Tuerlinckx2015; Chevallier, Grèzes, Molesworth, Berthoz, & Happé, Reference Chevallier, Grèzes, Molesworth, Berthoz and Happé2011; Gard, Kring, Gard, Horan, & Green, Reference Gard, Kring, Gard, Horan and Green2007; Hatzigiakoumis, Martinotti, Giannantonio, & Janiri, Reference Hatzigiakoumis, Martinotti, Giannantonio and Janiri2011; Kashdan, Zvolensky, & McLeish, Reference Kashdan, Zvolensky and McLeish2008; Meinzer, Pettit, Leventhal, & Hill, Reference Meinzer, Pettit, Leventhal and Hill2014; Sarramon, Verdoux, Schmitt, & Bourgeois, Reference Sarramon, Verdoux, Schmitt and Bourgeois1999]. Anhedonia is also a central part of the diagnosis of MDD, ASD, and anxiety disorders (APA, 2013).
Anhedonia is suggested to be related to dysfunctions in the reward system and/or affect regulation (Keedwell, Andrew, Williams, Brammer, & Philips, Reference Keedwell, Andrew, Williams, Brammer and Philips2005; Rizvi et al., Reference Rizvi, Pizzagalli, Sproule and Kennedy2016). To illustrate, MDD patients display deficits in reward responsiveness, reward anticipation and increased suppression of positive and negative affect (Beblo et al., Reference Beblo, Fernando, Klocke, Griepenstroh, Aschenbrenner and Driessen2012; Pizzagalli, Losifescu, Hallet, Ratner, & Fava, Reference Pizzagalli, Losifescu, Hallet, Ratner and Fava2008; Vrieze et al., Reference Vrieze, Pizzagalli, Demyttenaere, Hompes, Sienaert, de Boer and Claes2014b). Relatedly, decreased processing of rewarding cues in ADHD patients, could give rise to anhedonia (Meinzer et al., Reference Meinzer, Pettit, Leventhal and Hill2014). For anxiety patients, reduced acceptance of emotional distress may contribute to anhedonia (Kashdan et al., Reference Kashdan, Zvolensky and McLeish2008) and theory-of-mind deficits in ASD make social interactions less pleasurable and rewarding (Krach, Paulus, Bodden, & Kircher, Reference Krach, Paulus, Bodden and Kircher2010). Anhedonia is seen across these often-comorbid mental disorders and may present a transdiagnostic underlying feature for psychopathology (Keedwell et al., Reference Keedwell, Andrew, Williams, Brammer and Philips2005; Rizvi et al., Reference Rizvi, Pizzagalli, Sproule and Kennedy2016).
Anhedonia might also be linked to symptom severity of mental disorders. Anhedonia correlated positively with ASD symptom severity (Chevallier et al., Reference Chevallier, Grèzes, Molesworth, Berthoz and Happé2011; Novacek, Gooding, & Pflum, Reference Novacek, Gooding and Pflum2016) and interacted with ADHD severity and higher levels of depression severity (Babinski, Waschbusch, & Waxmonsky, Reference Babinski, Waschbusch and Waxmonsky2019). However, results regarding MDD and anxiety are mixed. Several studies demonstrate that anhedonia is predictive of MDD (Loas, Salinas, Guelfi, & Samuel-Lajeunesse, Reference Loas, Salinas, Guelfi and Samuel-Lajeunesse1992; Rey, Jouvent, & Dubal, Reference Rey, Jouvent and Dubal2009; Vrieze et al., Reference Vrieze, Demyttenaere, Bruffaerts, Hermans, Pizzagalli, Sienaert and Claes2014a) and anxiety severity (Keedwell et al., Reference Keedwell, Andrew, Williams, Brammer and Philips2005; Rey et al., Reference Rey, Jouvent and Dubal2009) whereas another study demonstrated that anhedonia did not correlate with MDD severity (e.g. Schrader, Reference Schrader1997) nor anxiety symptoms (Kashdan et al. Reference Kashdan, Zvolensky and McLeish2008). On the other hand, anxiety sensitivity (AS), a core feature of anxiety disorders (Mantar, Alkin, & Yemez, Reference Mantar, Alkin and Yemez2011; Taylor, Reference Taylor1995, Reference Taylor1996), was found to be associated with more severe anhedonia (Kashdan et al., Reference Kashdan, Zvolensky and McLeish2008). Nevertheless, Watson and Naragon-Gainey (Reference Watson and Naragon-Gainey2010) note that anhedonia has stronger associations with MDD than with anxiety disorders, but this might be because anhedonia is considered a core characteristic of MDD. Although anhedonia is implicated in a range of mental disorders and is related to poor treatment outcomes (Craske, Meuret, Ritz, Treanor, & Dour, Reference Craske, Meuret, Ritz, Treanor and Dour2016; McIntyre et al., Reference McIntyre, Woldeyohannes, Soczynska, Maruschak, Wium-Andersen, Vinberg and Kennedy2015), it's possible transdiagnostic role is underexplored.
A network approach is a promising way of investigating the relationships between anhedonia and validated symptom clusters characterizing depression, AS, ADHD, and ASD symptom severity, because it aims to identify plausible relations among symptom clusters that might be masked by traditional statistical approaches (Costantini et al., Reference Costantini, Epskamp, Borsboom, Perugini, Mottus, Waldorp and Cramer2015). More specifically, network analysis allows for the examination of how symptoms of one disorder (i.e. nodes) are associated with symptoms of another disorder (represented by edges; Borsboom & Cramer, Reference Borsboom and Cramer2013; Cramer, Waldorp, van der Maas, & Borsboom, Reference Cramer, Waldorp, van der Maas and Borsboom2010). In doing so, it is also possible to identify central features of the network that reflect transdiagnostic features that cut across diagnostic boundaries (Cramer et al., Reference Cramer, Waldorp, van der Maas and Borsboom2010).
The current study aims to explore the network structure and centrality of anhedonia severity and symptom severity clusters of depression, AS, ADHD, and ASD in a heterogeneous clinical sample with a high level of comorbidity (81.3%), close to clinical practice (Antshel et al., Reference Antshel, Zhang-James, Wagner, Ledesma and Faraone2016; McIntosh et al., Reference McIntosh, Kutcher, Binder, Levitt, Fallu and Rosenbluth2009). We computed a graphical LASSO network (Friedman, Hastie, & Tibshirani, Reference Friedman, Hastie and Tibshirani2008) to begin visualizing associations between symptom clusters, and a relative importance network (Grömping, Reference Grömping2006) to assess the directionality of associations. We first explored the strongest edges within the network and then focused on the nodes' centrality values.
MDD, anxiety disorders, ADHD and ASD are often comorbid, yet mechanisms that potentially underlie this comorbidity remain relatively underexplored. Research that did focus on the negative valence system, has often been limited to specific disorders. For example, anhedonia has thus far predominantly been investigated in the context of depression (e.g. Keedwell et al., Reference Keedwell, Andrew, Williams, Brammer and Philips2005; Loas et al., Reference Loas, Salinas, Guelfi and Samuel-Lajeunesse1992; McIntyre et al., Reference McIntyre, Woldeyohannes, Soczynska, Maruschak, Wium-Andersen, Vinberg and Kennedy2015). This study informs us about anhedonia as a transdiagnostic psychopathological feature in symptom clusters of depression, AS, ADHD and ASD, and which may partially explain their high comorbidity rates observed in clinical practice (Zbozinek et al., Reference Zbozinek, Rose, Wolitzky-Taylor, Sherbourne, Sullivan, Stein and Craske2012). Importantly, this study focused on anhedonia in the context of symptom severity clusters instead of diagnostic categories. As such, our study is in line with the goals of the RDoC framework to investigate the mechanisms pertaining to mental health dysfunctions from a dimensional point of view (Cuthbert, Reference Cuthbert2014; Insel et al., Reference Insel, Cuthbert, Garvey, Heinssen, Pine, Quinn and Wang2010).
Method
Participants
This study used data of the ongoing MIND-Set study (Measuring Integrated Novel Dimensions in neurodevelopmental and stress-related mental disorders), conducted at the Department of Psychiatry of the Radboud university medical center (Radboudumc) and the Donders Institute, Nijmegen, The Netherlands. The MIND-Set study (N = 559 at the time of writing the manuscript) is an observational, cross-sectional study of Dutch adult (age >17 years) patients who were diagnosed with one or more of the following disorders: mood disorders, anxiety disorders, ADHD, or ASD, and possibly a comorbid substance use disorder. The study included mostly outpatients (N = 555, 99.6%), but eligible inpatients were also allowed to participate.
The current study excluded two patients as they did not meet the inclusion criteria for the current study (i.e. no diagnosis of mood disorders, anxiety disorders, ADHD or ASD). This resulted in a final analytical sample of 557 patients. This sample included slightly more males (n = 302; 54.2%) than females. The mean age was approximately 39 years (s.d. = 14; range 17 to 79). Educational level was reasonably distributed being low for 81 participants (14.5%), medium for 209 participants (37.5%), high for 206 participants (36.9%), unknown for 36 participants (6.5%), and 19 participants (3.4%) did not fulfill any graduation. Within this sample, 232 participants (41.6%) were diagnosed with MDD, 176 participants (31.5%) were diagnosed with an anxiety disorder, 215 participants (38.5%) were diagnosed with ADHD, and 133 participants (23.8%) were diagnosed with ASD. Of the 176 participants with anxiety disorders, 38 participants (21.5%) were diagnosed with panic disorder, 6 participants (3.4%) were diagnosed with agoraphobia, 16 participants (9.0%) were diagnosed with obsessive compulsive disorder (OCD), 48 participants (27.2%) were diagnosed with PTSD, 36 participants (20.4%) were diagnosed with GAD, 50 participants (28.4%) were diagnosed with a social phobia, 15 participants (8.5%) were diagnosed with a specific phobia, and 19 participants (10.7%) were diagnosed with an anxiety disorder not otherwise specified (NOS).Footnote †Footnote 1 Note that comorbidity in the sample was common (81.3%) and that the sum score of the differential diagnoses exceeds the number of patients in total that were diagnosed with an anxiety disorder. See Table 1 for an overview of sample characteristics.
Note. The structured interviews did not provide information about which diagnosis was primary or secondary. Patients with substance use disorders were only included in the study if they were also diagnosed with comorbid MDD, anxiety, ADHD or ASD. This resulted in the exclusion of two patients with a substance use disorder only, leaving a total sample of 557 patients. Educational level was divided into four levels: (almost) no education (elementary education or education not finished), low (lower vocational and general secondary education), middle (intermediate vocational and higher secondary education) and high (higher vocational education and university). Categories of substance use disorders included: nicotine (n = 99), alcohol use disorders (n = 40), cannabis (n = 33), sedatives (n = 12), opiates (n = 5), stimulants (n = 5), cocaine (n = 4), gambling (n = 2), GHB (n = 2) and ecstasy (n = 1).
Procedure
Patients referred to the Department of Psychiatry of the Radboudumc received an invitation letter for their first appointment and a MIND-Set information letter at home. Before the first appointment, all patients filled out standard clinical questionnaires at home via an online secure questionnaire program. These included the symptom severity scales used in the current study. If preferred, a paper copy of the questionnaires was sent to the patients’ home address.
During the clinical diagnostic phase at the psychiatric department, patients were diagnosed in accordance with Dutch guidelines. The Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-IV-RV, referred to as SCID-I; First, Spitzer, Gibbon,& Williams, Reference First, Spitzer, Gibbon and Williams1996) was used to diagnose mood and anxiety disorders. The Structured Interview Measurements in the Addictions for Triage and Evaluation (MATE; Schippers & Broekman, Reference Schippers and Broekman2010) was used to assess substance use disorders. The Diagnostic Interview for ADHD in adults, second edition (DIVA 2.0; Kooij & Francken, Reference Kooij and Francken2010) was used to diagnose ADHD, and the Dutch Interview for Autism Spectrum Disorders in Adults (NIDA; Vuijk, Reference Vuijk2014) was used to diagnose ASD. Before the diagnostic interview of the DIVA and the NIDA were administered, a clinician explained the patient that it is necessary to administer this interview in the presence of a significant other. The clinician helped the patient in deciding who could join the interview and indicated the preference for a parent/brother/sister joining the interview (i.e. someone who knows the patient since they were a child). Thus, to ensure that clinicians could assess retrospective and hetero-anamnestic information, both the DIVA and the NIDA were always administered in the presence of a significant other. Their response was used in the diagnostic process. All clinicians involved in the clinical diagnostic procedure were experienced psychiatrists and psychologists who received extensive training in clinical interviewing.
Patients who were willing to participate in the MIND-Set study signed an informed consent form at the end of their diagnostic procedure. When patients consented to participation, the structurally collected questionnaire data could be used for research purposes. Also, biological measures (e.g. blood samples, hair samples), neuropsychological tests, and neuro-imaging measures followed. These measures are not part of the current study. The MIND-Set study has been approved by the local medical ethics committee (Commissie Mensgebonden Onderzoek Arnhem – Nijmegen). 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.
Measures
Severity of anhedonia
Seven items which represent an anhedonia factor in the Outcome Questionnaire (OQ-45-2;Footnote 2 Lambert et al., Reference Lambert, Smart, Campbell, Hawkins, Harmon and Slade2006), are used to assess the severity of anhedonia (Minami et al., Reference Minami, Davies, Tierney, Bettmann, McAward, Averill and Wampold2009). These seven anhedonia items are combined into the anhedonia factor, ranging from 0–27 (higher scores indicate more severe anhedonia). The items (e.g. ‘I am not interested in anything’ or the reversed item ‘I enjoy my leisure time’) were answered on a 5-point Likert scale, ranging from 0 (never) to 4 (almost always). The anhedonia factor correlated significantly with the SCID item representing anhedonia according to the DSM-IV, r s(441) = 0.536, p < 0.001. In the current sample, the internal consistency of the factor was good (α = 0.87).
Depressive symptoms
The Inventory of Depressive Symptomatology Self-Report (IDS-SR; Rush, Gullion, Basco, Jarrett, & Trivedi, Reference Rush, Gullion, Basco, Jarrett and Trivedi1996) assesses the severity of depressive symptoms in the past seven days. The 30 items (e.g. ‘Feeling sad’) are answered on a four-point scale (0–3) that describes different responses (e.g. with 0 indicating ‘I do not feel sad’ and 3 indicating ‘I feel sad nearly all the time’). In the current sample, the internal consistency of the IDS-SR was good (α = 0.86).
Anxiety sensitivity
The Anxiety Sensitivity Index (ASI; Reiss, Peterson, Gursky, McNally, Reference Reiss, Peterson, Gursky and McNally1986; Rodriguez, Bruce, Pagano, Spencer, & Keller, Reference Rodriguez, Bruce, Pagano, Spencer and Keller2004) assesses AS. As an individual trait concept, AS determines a person's proneness to become frightened by anxiety-related sensations (Reiss, Reference Reiss1997). We selected the ASI as it allows for a transdiagnostic and broad concept of the anxiety domain for both subclinical patients as well as for patients with comorbid anxiety disorders, in a sensitive way. Moreover, previous studies (e.g. Grös, Simms, Antony, & McCabe, Reference Grös, Simms, Antony and McCabe2007; Julian, Reference Julian2011) have shown that other measures, such as the STAI, have problems in differentiating anxiety from depression symptoms. By using the ASI, we aimed to use a ‘pure’ anxiety measure, without overlapping depressive symptoms. The 16 items (e.g. ‘It scares me when my heart beats rapidly’) are answered on a 5-point Likert scale, ranging from 0 (barely) to 4 (extremely much). In the current sample, the internal consistency of the ASI was good (α = 0.86).
ADHD symptoms
The Conners' Adult ADHD Rating Scale, Short Version (CAARS-SS; Connors, Erhardt,& Sparrow, Reference Connors, Erhardt and Sparrow1999) assesses the presence and severity of ADHD symptoms. The 26 items (e.g. ‘I am looking for fast exciting activities’) are answered on a 4-point Likert scale, ranging from 0 (not at all/never) to 3 (very much/very frequently). In the current sample, the internal consistency of the CAARS-SS was good (α = 0.88).
Autistic traits
The 50-item Autism Spectrum Quotient (AQ-50; Baron-Cohen, Wheelwright, Skinner, Martin, & Clubley, Reference Baron-Cohen, Wheelwright, Skinner, Martin and Clubley2001) assesses autistic traits. Items (e.g. ‘I notice sounds which other people don't seem to notice’) are answered on a 4-point Likert scale, ranging from 1 (totally agree) to 4 (totally disagree). In the current sample, the internal consistency of the AQ-50 was excellent (α = 0.90). The mean values and standard deviations for all measures are described in Table 2.
IDS-SR, Inventory of Depressive Symptomatology Self-Report; ASI, Anxiety Sensitivity Index; CAARS-SS, Conners' Adult ADHD Rating Scale, Short Version; AQ-50, 50-item Autism Spectrum Quotient. Note that scores on the ASI were positively skewed, this may be due to the high number of males in the sample who tend to underreport their fear and anxiety symptoms (Egloff & Schmukle, Reference Egloff and Schmukle2004; Pierce & Kirkpatrick, Reference Pierce and Kirkpatrick1992).
Data analyses
All analyses were conducted in the R programming environment (Version 3.6.3; R Core Team, 2020), the R code is available on the Open Science Framework (https://osf.io/6mjdf/). As data collection is still ongoing, the de-identified dataset is available upon request from the first author. We computed two types of networks: a graphical LASSO network to visualize the associations between the symptom clusters and a relative importance network to assess directionality.
Graphical LASSO network
The regularized partial correlation network was computed using a graphical Least Absolute Shrinkage and Selection Operator (LASSO), by using the R packages qgraph (Epskamp, Cramer, Waldorp, Schmittmann, & Borsboom, Reference Epskamp, Cramer, Waldorp, Schmittmann and Borsboom2012) and glasso (Friedman et al., Reference Friedman, Hastie and Tibshirani2008). Graphical LASSO first computes regularized partial correlations between all nodes, thereby controlling for the influence of all other nodes in the network. Next, it shrinks small associations to zero. This procedure removes potentially false-positive associations and returns a sparse network compromising only the edges that are likely genuine (Epskamp, Reference Epskamp2016; Epskamp & Fried, Reference Epskamp and Fried2018).
The qgraph package implements the graphical LASSO regularization in combination with extended Bayesian information criterion (EBIC) model selection. Graphical LASSO uses a tuning parameter (γ) to control the degree to which regularization is applied. This tuning parameter controls the trade-off between including false-positive edges and removing true edges. The tuning parameter is advised to be set between zero and 0.5 (Foygel & Drton, Reference Foygel and Drton2010). A tuning parameter close to 0.5 can be conservative and not reflect the true model, whereas a tuning parameter close to zero can be useful in exploratory and hypothesis-generating research (Hevey, Reference Hevey2018). We chose a moderately conservative value of γ to 0.25, to increase specificity and interpretability (Epskamp & Fried, Reference Epskamp and Fried2018).
As ASI scores were positively skewed, we computed an additional graphical LASSO network after performing a non-linear transformation of ASI scores, to test whether skewness was a potential confound.
Node centrality for the graphical LASSO network
The most central symptom clusters in the network were identified using betweenness, closeness, and strength indices (Costantini et al., Reference Costantini, Epskamp, Borsboom, Perugini, Mottus, Waldorp and Cramer2015). Betweenness is calculated by summing the number of times that a given node lies on the shortest path between two other nodes. Closeness is calculated as the inverse of all shortest path lengths between the given node and all other nodes in the network. Node strength calculates the direct connection of a given node to all other nodes in the network, by summing all edge weights a node is directly connected to. Each centrality index was computed using the R package qgraph (Epskamp, Borsboom, & Fried, Reference Epskamp, Borsboom and Fried2018). For all centrality measures, higher values reflect stronger associations in the network.
Stability analyses
To test whether the results allow for reliable interpretations, we examined the stability of the network using the R package bootnet (Epskamp et al., Reference Epskamp, Borsboom and Fried2018). Edge weight stability was tested using a non-parametric 1000-sample bootstrap technique to calculate 95% confidence intervals (CIs) for the edges. An additional 1000-sample dropping bootstrap was conducted to estimate the stability of the centrality indices. This procedure yields a correlation stability (CS) coefficient. The CS-coefficient reflects the maximum proportion of participants that could be dropped from the analyses, while still retaining a correlation of at least 0.7 with the centrality values estimated in the original network within a 95% CI. The recommended CS-coefficient should be at least 0.25, but preferably above 0.5. Finally, bootstrapped significance tests were used to test for significant differences between edge-weights or node centralities. A bootstrapped CI was constructed around the difference between bootstrap values of edge-weights or node centralities. Edge weights or node centralities differed from one-another when zero was not in the bootstrapped CI (Epskamp et al., Reference Epskamp, Borsboom and Fried2018).
Relative importance network
We then computed a relative importance network by using the R package relaimpo (Grömping, Reference Grömping2006). A relative importance network considers the amount of explained variance that each node makes to R 2 after controlling for multicollinearity (Johnson & LeBreton, Reference Johnson and LeBreton2004). These direct effects are adjusted for all other nodes in the network. Each edge is assigned a relative importance metric (lmg) between 0 and 1 (Grömping, Reference Grömping2006; Johnson & LeBreton, Reference Johnson and LeBreton2004). Edges in a relative importance network are weighted and directed. This procedure has been shown to be more stable and generalizable compared to other network procedures to assess directionality [i.e. a Bayesian directed acyclic graph (DAG); Forbes, Wright, Markon, & Krueger, Reference Forbes, Wright, Markon and Krueger2017).
Node centrality for the relative importance network
Similar to the Graphical LASSO network, we also computed centrality indices (Costantini et al., Reference Costantini, Epskamp, Borsboom, Perugini, Mottus, Waldorp and Cramer2015). Within a relative importance network, the strength parameter can be broken up into in-strength and out-strength. In-strength is calculated by summing the directed edge weights originating from other nodes and ending at a given node. It thus quantifies the extent to which a node is influenced by other nodes in the network. In contrast, out-strength is calculated by summing the directed edge weights originating from a given node and ending at all other nodes. It thus quantifies the extent to which a given node exerts predictive influence on other nodes in the network. Again, higher values reflect greater centrality (Epskamp et al., Reference Epskamp, Borsboom and Fried2018). We also assessed the stability of the relative importance network, similar to the graphical LASSO network.
Missing data
Overall, 0.017% of participants had missing data (Anhedonia factor: 2 patients, IDS-SR: 4 patients, AQ-50: 4 patients). There was no missing data on ASI and CAARS-SS items. The network was estimated using complete pairwise observations (i.e. using all available data).
Results
Graphical LASSO network
Figure 1 depicts the graphical LASSO network, which includes regularized partial correlations and limits spurious associations. Online Supplementary Fig. S1 depicts the network with the transformed measure of AS, this transformation did not change the network structure. Edge weights reflect the strength of associations and edge color the direction (green edges reflect positive associations). For all networks, node placement was determined by Fruchterman and Reingold's (Reference Fruchterman and Reingold1991) algorithm, which results in a visually appealing graph where nodes generally do not overlap and edges have approximately the same length (Jones, Mair, & McNally, Reference Jones, Mair and McNally2018).
Several features are immediately apparent. First, depression symptom severity emerged as a central node, connected to anhedonia, AS, and ADHD symptom severity. Anhedonia severity also emerged as a central node, connected to depression, ADHD, and ASD symptom severity. ADHD symptom severity and AS symptom severity were also connected, as well as AS and ASD symptom severity. The correlation matrix with all the regularized partial correlation coefficients is presented in online Supplementary Table S2. Bootstrapped CI's for the edges indicate that most edges were fairly stable, as demonstrated by relatively narrow CI's (online Supplementary Fig. S2). The bootstrapped significance test (online Supplementary Fig. S3) revealed that the edges between the symptom severity clusters of anhedonia and depression, depression and AS, depression and ADHD, and between ADHD and ASD were significantly stronger than at least half of the other edges in the network.
Node centrality
These dynamics are further reflected in the centrality indices (online Supplementary Fig. S4). Depression severity showed the highest level of betweenness, closeness and strength. ADHD symptom severity showed high levels of betweenness and closeness. Anhedonia severity showed a high strength value. The CS-coefficients were 0.75 for strength, 0.59 for closeness, and 0.21 for betweenness. The values of strength and closeness are above the recommended threshold between 0.25 and 0.5, suggesting that these are the most stable centrality indices. The CS-coefficient for betweenness is below the recommended threshold and should thus be interpreted more cautiously (Epskamp et al., Reference Epskamp, Borsboom and Fried2018). Results of the associated sample-dropping bootstrap are presented in online Supplementary Fig. S5. We tested whether nodes significantly differed in node strength (online Supplementary Fig. S6) and closeness (online Supplementary Fig. S7). As strength and closeness were the most stable centrality indices, we focused solely on them (Armour, Fried, Deserno, Tsai, & Pietrzak, Reference Armour, Fried, Deserno, Tsai and Pietrzak2017; Heeren, Bernstein, & McNally, Reference Heeren, Bernstein and McNally2018). Accordingly, depression severity and anhedonia severity were significantly more central than the other nodes.
Relative importance network
Figure 2 depicts the relative importance network. The relative importance values (i.e. lmg) are presented in online Supplementary Table S3. We found that depression severity was strongly predictive of anhedonia (lmg = 0.75), AS (lmg = 0.53) and ADHD (lmg = .36) symptom severity. Anhedonia was predictive of depression (lmg = 0.62), ASD (lmg = 0.25), ADHD (lmg = 0.20), and AS (lmg = .15) symptom severity. Although anhedonia strongly predicted AS (lmg = .15), AS predicted ADHD severity as well (lmg = 0.21), which in turn predicted anhedonia severity (lmg = 0.11). Moreover, ADHD symptom severity was strongly predictive of ASD symptom severity (lmg = 0.54). The bootstrapped CIs for the edges indicated again that the edges were fairly stable (online Supplementary Fig. S8).
Node centrality
These relationships are also reflected in the centrality indices (Fig. 3). Depression and ADHD symptom severity showed the highest levels of betweenness centrality. Depression and anhedonia severity showed the greatest closeness centrality. Depression and anhedonia severity yielded the highest out-strength. Depression severity, ADHD symptom severity and anhedonia severity demonstrated the highest in-strength values. The results of the associated case-dropping bootstrap suggest that all centrality indices were stable, except for in-strength. The CS-coefficients were 0.75 for out-strength, 0 for in-strength, 0.75 for closeness, and 0.59 for betweenness. As the CS-coefficient for in-strength is below the recommended cutoff between 0.25 and 0.5 (Epskamp et al., Reference Epskamp, Borsboom and Fried2018), in-strength cannot be interpreted reliably.
Additional analyses
The anhedonia and depression severity nodes may be measuring overlapping constructs (i.e. anhedonia is a core symptom of MDD and hence also assessed in the IDS-SR). This could have inflated edge weights and centrality, which makes it difficult to ensure that the strong connection between the nodes is not due to their conceptual overlap (Rodebaugh et al., Reference Rodebaugh, Tonge, Piccirillo, Fried, Horenstein, Morrison and Heimberg2018). To control for this possibility, we computed an additional network without the overlapping anhedonia items in the IDS-SR. This resulted in a similar network structure (online Supplementary Fig. S9). Additionally, we tested an exploratory network without the MDD node, to determine more unique relationships between anhedonia and the other symptoms clusters independent of MDD symptomatology. The resulting network demonstrates stronger relationships between anhedonia and symptom clusters of AS, ADHD and ASD while the overall pattern of results was highly similar to the original model. We present these additional exploratory results in the Supplemental materials.
Discussion
Mental disorders as defined by DSM classifications are highly comorbid (Antshel et al., Reference Antshel, Zhang-James, Wagner, Ledesma and Faraone2016; Ghaziuddin et al., Reference Ghaziuddin, Ghaziuddin and Greden2002; Grevet et al., Reference Grevet, Bau, Salgado, Fischer, Kalil, Victor and Belmonte-de-Abreu2006; Hirschfeld, Reference Hirschfeld2001; Joshi et al., Reference Joshi, Wozniak, Petty, Martelon, Fried, Bolfek and Biederman2013; Kaufman & Charney, Reference Kaufman and Charney2000; Kessler et al., Reference Kessler, Adler, Barkley, Biederman, Conners, Demler and Zaslavsky2006a, Reference Kessler, Merikangas and Wang2006b; Schatz & Rostain, Reference Schatz and Rostain2006; van Loo et al., Reference van Loo, Romeijn, de Jong and Schroevers2013). These high levels of comorbidity make it important to look for underlying constructs. In line with the RDoC framework, this study explored the role of anhedonia as an underlying feature associated with the severity of symptom clusters of depression, AS, ADHD, and ASD, by computing two types of networks. The most interesting results were the central role of depression severity and anhedonia severity within the network.
Anhedonia severity yielded a high strength centrality value in both networks. The relative importance network illustrated that anhedonia severity is predictive of the severity of symptom clusters of depression, AS, ASD, and ADHD, with the statistical prediction of depression severity being the strongest. Interestingly, the influence of anhedonia severity on these symptom clusters was stronger than the reverse influences on anhedonia severity. These results offer new insights into the ongoing discussion whether anhedonia is depression specific. In line with Watson and Naragon-Gainey (Reference Watson and Naragon-Gainey2010), we specifically found that anhedonia is more strongly predictive of depression than AS. Thus, anhedonia is not confined solely to depression, but also seems to characterize – to a lesser degree – AS and possibly anxiety disorders (Kashdan et al., Reference Kashdan, Zvolensky and McLeish2008; Keedwell et al., Reference Keedwell, Andrew, Williams, Brammer and Philips2005; Loas et al., Reference Loas, Salinas, Guelfi and Samuel-Lajeunesse1992; Rey et al., Reference Rey, Jouvent and Dubal2009; Schrader, Reference Schrader1997; Vrieze et al., Reference Vrieze, Demyttenaere, Bruffaerts, Hermans, Pizzagalli, Sienaert and Claes2014a, Reference Vrieze, Pizzagalli, Demyttenaere, Hompes, Sienaert, de Boer and Claes2014b). Moreover, the results suggest that anhedonia severity is predictive of and therefore potentially contributes to more severe ASD and ADHD symptom severity. This suggests that anhedonia may serve as a common underlying transdiagnostic psychopathology feature predictive of symptom severity of AS, ADHD, and ASD.
Although anhedonia severity was strongly predictive of more severe depression symptoms, depression symptom severity was slightly more predictive of anhedonia severity than vice versa. This suggests that more severe depression symptoms may foster anhedonia and that anhedonic individuals may be prone to experience more severe depression. This interplay between depression symptoms in general and anhedonia as a core depressive symptom might be a maintaining feature of depressive symptom severity. These relationships appear consistent with theory stating that anhedonia may be the most influential MDD symptom (Beard et al., Reference Beard, Millner, Forgeard, Fried, Hsu, Treadway and Björgvinsson2016; Fried, Epskamp, Nesse, Tuerlinckx, & Borsboom, Reference Fried, Epskamp, Nesse, Tuerlinckx and Borsboom2016; Rosenström et al., Reference Rosenström, Elovainio, Jokela, Pirkola, Koskinen, Lindfors and Keltikangas-Järvinen2015) and that more severe depression could also result in more severe anhedonia (Beard et al., Reference Beard, Millner, Forgeard, Fried, Hsu, Treadway and Björgvinsson2016). Interestingly, the central role of anhedonia corresponds to empirically based treatments for MDD. For example, behavioral activation, in which individuals are promoted to engage in positively reinforcing activities, may be particularly helpful for individuals who display anhedonia (Naragon-Gainey, Watson, & Markon, Reference Naragon-Gainey, Watson and Markon2009; Walsh et al., Reference Walsh, Eisenlohr-Moul, Minkel, Bizzell, Petty, Crowther and Dichter2019). Based on the central role of anhedonia in our analyses, behavioral activation may serve as an important initial treatment strategy for patients presenting with (comorbidity of) depression, anxiety disorders, ADHD, or ASD.
Depression severity was even more central in the network than anhedonia severity. Depression severity was predictive of more severe anhedonia, AS, ADHD, and ASD symptoms. Interestingly, these relationships were stronger than the reverse influences of the symptom clusters on depression severity. It is possible that more severe depression strengthens other present symptoms such as irritability, nervousness, feeling restless, and concentration problems (Beard et al., Reference Beard, Millner, Forgeard, Fried, Hsu, Treadway and Björgvinsson2016; Meinzer et al., Reference Meinzer, Pettit, Leventhal and Hill2014; Park & Kim, Reference Park and Kim2020), leading to more severe AS, ADHD or ASD symptoms (Breslau & Davis, Reference Breslau and Davis1985; Bron et al., Reference Bron, Bijlenga, Verduijn, Penninx, Beekman and Kooij2016; Loas et al., Reference Loas, Salinas, Guelfi and Samuel-Lajeunesse1992).
It is also possible that some of the relationships between depression severity and the symptom clusters may be explained by the high betweenness value of ADHD symptom severity (Figs. 3 and S4). This betweenness value could only be interpreted in the relative importance network, and therefore these results should be interpreted with caution. If the results will be replicated in future studies, this value could suggest that the importance of ADHD symptom severity is derived from its central position in the network, rather than the strength of relationships. Accordingly, ADHD symptom severity may perhaps serve as a symptom cluster that contains both symptoms related to emotional disorders and neurodevelopmental disorders (Antshel et al., Reference Antshel, Zhang-James, Wagner, Ledesma and Faraone2016). Previous studies already illustrated that MDD and ADHD, and ADHD and ASD are strongly associated, largely due to genetic factors (Ghirardi et al., Reference Ghirardi, Pettersson, Taylor, Freitag, Franke, Asherson and Kuja-Halkola2019; Lee et al., Reference Lee, Ripke, Neale, Faraone, Purcell, Perlis and Kähler2013), but there might be additional symptom dimensions underlying the comorbidity between them. For example, cognitive biases that are thought to underlie vulnerability for MDD have also been found in individuals with increased ADHD symptoms (Lemoult & Gotlib, Reference Lemoult and Gotlib2018; Vrijsen et al., Reference Vrijsen, Tendolkar, Onnink, Hoogman, Schene, Fernández and Franke2018), whereas individuals with ASD only are not found to display these cognitive biases (Bergman et al., Reference Bergman, Schene, Vissers, Vrijsen, Kan and van Oostrom2020). One possibility is that cognitive vulnerability factors (e.g. attentional difficulties, selective attention bias, or negative memory bias to all emotional cues; Antshel et al., Reference Antshel, Zhang-James, Wagner, Ledesma and Faraone2016; Lawson et al., Reference Lawson, Papadakis, Higginson, Barnett, Wills, Strang and Kenworthy2015) serve as a symptom cluster bridging symptoms related to MDD with those related to neurodevelopmental disorders, making ASD symptoms more pronounced. A relevant question for future studies is to gain more evidence for such an intermediate role of ADHD related symptom clusters and the potential clinical relevance for treating those cognitive vulnerability factors.
Interestingly, depression severity was more strongly predictive of AS and ADHD symptom severity than anhedonia severity. This suggests that another aspect of depression severity, other than the core symptom anhedonia, may be related to AS and ADHD symptom severity. Indeed, ADHD and anxiety symptoms are increased among individuals with more severe depression (Bron et al., Reference Bron, Bijlenga, Verduijn, Penninx, Beekman and Kooij2016; Rector, Szacun-Shimizu, & Leybman, Reference Rector, Szacun-Shimizu and Leybman2007). More studies that disentangle anhedonia and other depression symptoms are needed to verify whether other aspects of depression severity are related to AS and ADHD symptom severity. Thus, even though anhedonia seems to be a common underlying transdiagnostic psychopathology feature of depression, AS, ADHD, and ASD symptom severity, there are other emotional, (neuro-)cognitive, or biological factors at play driving high comorbidity between these symptom clusters and disorders.
Overall, our findings suggest that anhedonia may serve as a common underlying transdiagnostic psychopathology feature, associated with symptom severity of depression, AS, ADHD, and ASD (Chevallier et al., Reference Chevallier, Grèzes, Molesworth, Berthoz and Happé2011; Loas et al., Reference Loas, Salinas, Guelfi and Samuel-Lajeunesse1992; Meinzer et al., Reference Meinzer, Pettit, Leventhal and Hill2014; Novacek et al., Reference Novacek, Gooding and Pflum2016). On a biological level, this relationship may be explained by a dysfunction in the reward system and/or affect regulation (Chevallier et al., Reference Chevallier, Grèzes, Molesworth, Berthoz and Happé2011; Keedwell et al., Reference Keedwell, Andrew, Williams, Brammer and Philips2005; Krach et al., Reference Krach, Paulus, Bodden and Kircher2010; Meinzer et al., Reference Meinzer, Pettit, Leventhal and Hill2014; Pizzagalli et al., Reference Pizzagalli, Losifescu, Hallet, Ratner and Fava2008; Vrieze et al., Reference Vrieze, Demyttenaere, Bruffaerts, Hermans, Pizzagalli, Sienaert and Claes2014a, Reference Vrieze, Pizzagalli, Demyttenaere, Hompes, Sienaert, de Boer and Claes2014b). Future studies should therefore clarify the relation between dysfunctions in the reward system and/or affect regulation and symptom severity of depression, AS, ADHD, and ASD by using a different level of analysis as documented in the RDoC matrix (e.g. brain circuits or different research paradigms).
Strengths of this study include the use of a large clinical sample and the use of the network approach to investigate the transdiagnostic role of anhedonia related to the severity of symptom clusters, instead of a focus on separate diagnostic categories (Cuthbert, Reference Cuthbert2014; Insel et al., Reference Insel, Cuthbert, Garvey, Heinssen, Pine, Quinn and Wang2010). Moreover, we did not exclude participants with comorbid disorders. This approach is representative of clinical practice, where patients are often diagnosed with comorbid disorders that cannot be easily disentangled (Zbozinek et al., Reference Zbozinek, Rose, Wolitzky-Taylor, Sherbourne, Sullivan, Stein and Craske2012). Finally, the use of a heterogeneous clinical sample provided variability among all symptoms, including symptoms outside of patients' diagnosed disorder(s).
Our study has several limitations as well. First, anhedonia was measured using the anhedonia factor structure of the OQ-45 (Minami et al., Reference Minami, Davies, Tierney, Bettmann, McAward, Averill and Wampold2009), instead of a validated measure. Although the internal consistency of this factor was good (α = 0.87), its use limits the generalizability and perhaps the reliability and validity of our results because this is not validated. Hence, our results should be interpreted with caution and replication of these findings with validated anhedonia measures, such as the SHAPS (Snaith et al., Reference Snaith, Hamilton, Morley, Humayan, Hargreaves and Trigwell1995) or DARS (Rizvi et al., Reference Rizvi, Quilty, Sproule, Cyriac, Bagby and Kennedy2015) is warranted. However, at the time of the start of MIND-Set, no Dutch validated versions of these self-report measures were available. Second, our results cannot be generalized to non-help seeking patients and/or community-based samples, given that we used a clinical sample. In addition, it is possible that individuals with more severe symptoms are less likely to enter psychological treatment or that individuals with a milder form of psychopathology are too engaged in different activities to seek help. We unfortunately cannot evaluate whether this influenced the inclusion of patients in our study. However, the OQ-45 total score (M = 76.9, s.d. = 24.5) indicates that participants psychological distress is of clinical significance, based on the recommended Dutch cut-off (i.e. cut-off score = 55; De Jong, Nugter, Lambert, & Burlingame, Reference De Jong, Nugter, Lambert and Burlingame2009) and the American cut-off (i.e. cut-of score = 63; Lambert, Gregersen, & Burlingame, Reference Lambert, Gregersen, Burlingame and Maruish2004). Although a selection bias might be less potent in the study compared to studies where non-clinical care data is used, a selection bias might have a strong influence on our results which we cannot fully address or rule out. Third, clinicians assessed all diagnoses using validated structured interviews (Luderer et al., Reference Luderer, Sick, Kaplan-Wickel, Reinhard, Richter, Kiefer and Weber2020). Despite these high-standard efforts, we cannot completely rule out that ADHD is underdiagnosed, also given that ADHD is often underrecognized in patients with substance use disorders (Huntley et al., Reference Huntley, Maltezos, Williams, Morinan, Hammon, Ball and Asherson2012). Moreover, the structured interviews did not provide information about which diagnosis was primary or secondary and information from the clinician about the presumed primary or secondary diagnosis was not included in our prospective study design and we cannot access such information retrospectively. However, the aim of our study was to explicitly look across diagnostic boundaries to explore transdiagnostic symptom profiles, instead of diagnoses. Thus, we present the diagnoses of participants to give an overview of sample characteristics but did not use this information in our statistical analyses. Instead, we focused on symptom clusters based on questionnaire scores. Therefore, a possible underdiagnosis of ADHD as well as a lack of information about primary/secondary diagnoses, would not have affected our results. Fourth, a small number of patients in each diagnostic group rendered subgroup analyses of ‘pure’ diagnostic categories without comorbidity (e.g. those with solely MDD), impossible. Future studies interested in disorder-specific contributions of anhedonia could investigate such networks on the symptom level within one diagnostic category. Finally, the network approach as used in the current study is cross-sectional and therefore, our data suggests, but cannot confirm causal connections among nodes. In response to this limitation, we employed a relative importance network. The directionality does allow us to suggest potential hypotheses for future research into causal relations. The findings that anhedonia is related to symptom severity of depression, AS, ADHD, and ASD are consistent with theory (Chevallier et al., Reference Chevallier, Grèzes, Molesworth, Berthoz and Happé2011; Keedwell et al., Reference Keedwell, Andrew, Williams, Brammer and Philips2005; Novacek et al., Reference Novacek, Gooding and Pflum2016; Rey et al., Reference Rey, Jouvent and Dubal2009; Schrader, Reference Schrader1997). Longitudinal data and/or experimental data are needed to clarify the temporal and causal structure of the relations found in the present study.
These limitations notwithstanding, this study offers unique insights on plausible relationships between anhedonia severity and severity of symptom clusters that merit further study. While our results are mostly hypothesis-generating, they suggest anhedonia as a common underlying transdiagnostic psychopathology feature, predictive of the severity of symptom clusters of depression, AS, ADHD, and ASD. Thus, anhedonia may be associated with the high comorbidity between these symptom clusters and disorders. Moreover, our results suggest that depression severity is predictive of the severity of symptom clusters of AS and ADHD. If our results will be replicated in future studies, it is recommended for clinicians to be more vigilant about screening for anhedonia and/or depression severity in individuals diagnosed with an anxiety disorder, ADHD and/or ASD. Future studies should investigate whether a focus on anhedonia, as a potential underlying transdiagnostic psychopathology feature associated with symptom severity of depression, AS, ADHD and ASD, is therapeutically efficacious.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291722000575
Financial support
This work is based on the MIND-Set database, first established in Nijmegen, The Netherlands, in 2015. The database was initiated by the Department of Psychiatry, Radboud University Medical Center and the Donders Centre for Cognitive Neuroimaging. This work was supported by the Radboud University Medical Centre and the Psychiatry Foundation as well as from a European Union's Horizon 2020 research and innovation program grant (grant number 728018). These funding sources had no role in the study design, collection, analysis or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication. M. G. Guineau was supported by a BSI travel grant from the Behavioural Science Institute, Radboud University, Nijmegen.
Conflict of interest
There are no conflicts of interest.