The network perspective on major depressive disorder (MDD) and generalized anxiety disorder (GAD) articulated by Cramer et al. in the target article captures a good deal of the scientific data on, and clinical experience with, those most ubiquitous and protean of Axis I disorders (cf. Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) for definitions of Axis I; American Psychiatric Association 1994). MDD and GAD are not only strongly comorbid with one another, but with a wide range of psychiatric afflictions. It is highly plausible that they designate the most common clusters in the network of co-occurring symptoms of serious mental distress generally. Equally worthy of further study and development is the authors' suggestion that the causal relationships among these symptoms are typically directly mediated, in part, by external factors (perhaps including the communication of psychiatric diagnoses to patients, a potential “etiological node” [sect. 4, para. 5] that Cramer et al. don't mention). It is this aspect of the network perspective that renders it a radical and welcome challenge to standard latent variable models. The authors are also importantly right in identifying the implicit commitments of such models with the psychometric methods used in their application, rather than with associated philosophical rhetoric.
Cramer et al. are not explicit about the way in which they expect their radical picture to generalize across the suite of mental and personality disorders. We may distinguish two possible interpretations that are compatible with their remarks. On the one hand, it might be that the network model applies to most disorders directly, just as it does to MDD and GAD. On this interpretation, most disorders are clusters of nodes in a single super-network of psychiatric conditions. Alternatively, the network structure of MDD and GAD, along with the ubiquity of their constitutive symptoms, might explain comorbidities involving other Axis I and Axis II disorders which are themselves based on latent variables. Evidently, the second “hybrid” picture would be more complicated and inelegant. However, I believe that the limited current evidence runs more strongly in its favor.
A major class of Axis I disorders that seems most likely to demand this hybrid conception is that of the addictions. As is distinctly not true of MDD or GAD, recent progress in understanding addiction has consisted mainly in progressive isolation and refinement of a neural pathways model of its etiology and maintenance (Everitt & Robbins Reference Everitt and Robbins2005; Everitt et al. Reference Everitt, Dickinson and Robbins2001; Goldstein & Volkow Reference Goldstein and Volkow2002; Koob Reference Koob2006; Koob & LeMoal Reference Koob and Le Moal2000). Summarizing very broadly, it seems to have been discovered that addiction arises through the dopamine reward circuit's learning a rich but entrenched set of cues that activate representation of highly valued and strongly salient addictive targets and that prepare motor response to consume them, which, when frustrated, are experienced as cravings. A crucial step in the etiology of addiction appears to be adaptation of frontal and prefrontal serotonin and gamma-aminobutyric acid (GABA) circuits that weakens the strength of opponent processes to impulsive consumption.
If learned reward system hysteresis and neuroadaptation that weakens cortical control are taken to be jointly necessary and sufficient conditions for addiction, then many people who consume addictive substances, or gamble, to problematic levels are probably not addicted. “Problem drinking” and “problem gambling” are very plausibly behavioral syndromes with network structures. However, what would precisely distinguish true addicts, according to the perspective being suggested, are specific ranges of values for neural processing variables in the anterior cingulate, ventral tegmental area (VTA), the ventral striatum including nucleus accumbens, and the orbitofrontal cortex (OFC) and dorsolateral prefrontal cortex (PFC). Preliminary evidence for this view comes from the first taxometric analyses of mixed populations of substance-abusing and substance-dependent people. Goedeker and Tiffany (Reference Goedeker and Tiffany2008) find strong convergence among three taxometric methods in identifying a taxon of heavy daily cigarette smokers, who also meet traditional clinical criteria for dependence, that excludes the large group of less regular nicotine users. That is, nicotine dependence does not appear to have a dimensional structure, in strong contrast to the findings of the same taxometric approaches as applied to analogue depression (Ruscio & Ruscio Reference Ruscio and Ruscio2002) and attention-deficit hyperactivity disorder (ADHD) (Haslam et al. Reference Haslam, Williams, Prior, Haslam, Graetz and Sawyer2006).
A disorder's being characterized by a network model does not entail that, under the guise of a latent variable perspective, its structure should emerge as dimensional. Nor does the inverse dependence hold: from the fact that a disorder's latent structure suggests a taxon one cannot infer rejection of a network model. However, Cramer et al. identify the view opposed to the network perspective as “essentialism” (sect. 6, para. 7), which in the case of mental disorders is the idea that people can be separated into disorder and no disorder classes on the basis of presence or absence of specific “defining features” (sect. 6, para. 1). In the case of addiction, such defining features which are not symptoms have been proposed on the basis of neuroscientific evidence; in this context, confirmation of taxonic structure buttresses the hypothesis that the features in question are indeed defining – that is, that addiction resembles Down's syndrome, Cramer et al.'s example of a medical condition that is not based on a network. It might thus be suggested that Cramer et al. overstate what the evidence will bear when they say baldly that “this line of reasoning [applied to Down's syndrome] is unlikely to hold for mental disorders” (sect. 6, para. 2). On the other hand, the issue might be thought to turn partly on semantics: someone might want to maintain that if addiction is a distinctive, identifiable pathology of the VTA-to-PFC/OFC circuit, then it is not really a mental disorder after all.
Addictions are highly comorbid with other Axis I and Axis II disorders, including MDD, ADHD, bipolar spectrum disorder, and antisocial personality disorder. A possible way of representing this is to conceive of addictions, themselves modeled using latent variable structures, as nodes in the overarching network of mental disorders. On this representation, however, comorbidities among addictions would not be best modeled by network analysis. Comorbidity among addictions might be grounded etiologically in a genetic predisposition, and manifestly in the common attenuation of cortical control circuits in addicts.
Cramer et al. have supplied a promising and liberating way of thinking about psychiatric comorbidity. They likely would not want to be taken as suggesting that it is the whole story.
You have
Access
The network perspective on major depressive disorder (MDD) and generalized anxiety disorder (GAD) articulated by Cramer et al. in the target article captures a good deal of the scientific data on, and clinical experience with, those most ubiquitous and protean of Axis I disorders (cf. Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) for definitions of Axis I; American Psychiatric Association 1994). MDD and GAD are not only strongly comorbid with one another, but with a wide range of psychiatric afflictions. It is highly plausible that they designate the most common clusters in the network of co-occurring symptoms of serious mental distress generally. Equally worthy of further study and development is the authors' suggestion that the causal relationships among these symptoms are typically directly mediated, in part, by external factors (perhaps including the communication of psychiatric diagnoses to patients, a potential “etiological node” [sect. 4, para. 5] that Cramer et al. don't mention). It is this aspect of the network perspective that renders it a radical and welcome challenge to standard latent variable models. The authors are also importantly right in identifying the implicit commitments of such models with the psychometric methods used in their application, rather than with associated philosophical rhetoric.
Cramer et al. are not explicit about the way in which they expect their radical picture to generalize across the suite of mental and personality disorders. We may distinguish two possible interpretations that are compatible with their remarks. On the one hand, it might be that the network model applies to most disorders directly, just as it does to MDD and GAD. On this interpretation, most disorders are clusters of nodes in a single super-network of psychiatric conditions. Alternatively, the network structure of MDD and GAD, along with the ubiquity of their constitutive symptoms, might explain comorbidities involving other Axis I and Axis II disorders which are themselves based on latent variables. Evidently, the second “hybrid” picture would be more complicated and inelegant. However, I believe that the limited current evidence runs more strongly in its favor.
A major class of Axis I disorders that seems most likely to demand this hybrid conception is that of the addictions. As is distinctly not true of MDD or GAD, recent progress in understanding addiction has consisted mainly in progressive isolation and refinement of a neural pathways model of its etiology and maintenance (Everitt & Robbins Reference Everitt and Robbins2005; Everitt et al. Reference Everitt, Dickinson and Robbins2001; Goldstein & Volkow Reference Goldstein and Volkow2002; Koob Reference Koob2006; Koob & LeMoal Reference Koob and Le Moal2000). Summarizing very broadly, it seems to have been discovered that addiction arises through the dopamine reward circuit's learning a rich but entrenched set of cues that activate representation of highly valued and strongly salient addictive targets and that prepare motor response to consume them, which, when frustrated, are experienced as cravings. A crucial step in the etiology of addiction appears to be adaptation of frontal and prefrontal serotonin and gamma-aminobutyric acid (GABA) circuits that weakens the strength of opponent processes to impulsive consumption.
If learned reward system hysteresis and neuroadaptation that weakens cortical control are taken to be jointly necessary and sufficient conditions for addiction, then many people who consume addictive substances, or gamble, to problematic levels are probably not addicted. “Problem drinking” and “problem gambling” are very plausibly behavioral syndromes with network structures. However, what would precisely distinguish true addicts, according to the perspective being suggested, are specific ranges of values for neural processing variables in the anterior cingulate, ventral tegmental area (VTA), the ventral striatum including nucleus accumbens, and the orbitofrontal cortex (OFC) and dorsolateral prefrontal cortex (PFC). Preliminary evidence for this view comes from the first taxometric analyses of mixed populations of substance-abusing and substance-dependent people. Goedeker and Tiffany (Reference Goedeker and Tiffany2008) find strong convergence among three taxometric methods in identifying a taxon of heavy daily cigarette smokers, who also meet traditional clinical criteria for dependence, that excludes the large group of less regular nicotine users. That is, nicotine dependence does not appear to have a dimensional structure, in strong contrast to the findings of the same taxometric approaches as applied to analogue depression (Ruscio & Ruscio Reference Ruscio and Ruscio2002) and attention-deficit hyperactivity disorder (ADHD) (Haslam et al. Reference Haslam, Williams, Prior, Haslam, Graetz and Sawyer2006).
A disorder's being characterized by a network model does not entail that, under the guise of a latent variable perspective, its structure should emerge as dimensional. Nor does the inverse dependence hold: from the fact that a disorder's latent structure suggests a taxon one cannot infer rejection of a network model. However, Cramer et al. identify the view opposed to the network perspective as “essentialism” (sect. 6, para. 7), which in the case of mental disorders is the idea that people can be separated into disorder and no disorder classes on the basis of presence or absence of specific “defining features” (sect. 6, para. 1). In the case of addiction, such defining features which are not symptoms have been proposed on the basis of neuroscientific evidence; in this context, confirmation of taxonic structure buttresses the hypothesis that the features in question are indeed defining – that is, that addiction resembles Down's syndrome, Cramer et al.'s example of a medical condition that is not based on a network. It might thus be suggested that Cramer et al. overstate what the evidence will bear when they say baldly that “this line of reasoning [applied to Down's syndrome] is unlikely to hold for mental disorders” (sect. 6, para. 2). On the other hand, the issue might be thought to turn partly on semantics: someone might want to maintain that if addiction is a distinctive, identifiable pathology of the VTA-to-PFC/OFC circuit, then it is not really a mental disorder after all.
Addictions are highly comorbid with other Axis I and Axis II disorders, including MDD, ADHD, bipolar spectrum disorder, and antisocial personality disorder. A possible way of representing this is to conceive of addictions, themselves modeled using latent variable structures, as nodes in the overarching network of mental disorders. On this representation, however, comorbidities among addictions would not be best modeled by network analysis. Comorbidity among addictions might be grounded etiologically in a genetic predisposition, and manifestly in the common attenuation of cortical control circuits in addicts.
Cramer et al. have supplied a promising and liberating way of thinking about psychiatric comorbidity. They likely would not want to be taken as suggesting that it is the whole story.