Cramer et al.'s account of comorbidity is a thought-provoking contribution to the methodological foundations of psychopathology research. While we agree in spirit with the authors' motivations (e.g., addressing problems with latent variable approaches to comorbidity), we will point out several prima facie challenges for the network approach to comorbidity.
Consider the following philosophical views concerning psychological disorders. Realism, the view suggested by latent variable models, represents psychological disorders as the unobservable causes of symptoms. On this view, as Cramer et al. note, disorders exist independently of our measurement and diagnostic practices. They are detected through measurement. Conventionalist approaches, on the other hand, represent psychological disorders as artifacts of our measurement practices. For example, operationism collapses the distinction between disorders and their symptoms: the symptoms are constitutive of the disorders. Operationists define a disorder such as generalized anxiety disorder (GAD) as nothing more than the satisfying of a certain set of diagnostic criteria. As Cramer et al. note, the differences between these positions can have practical consequences such as different prevalence rates for a disorder. Cramer et al. have sought to carve out a third position with respect to comorbidity and psychological disorders. On their view, disorders are “clusters” of symptoms that systematically covary. Hence, it seems that on the network view, disorders do not exist independently of their symptoms. Disorders are aggregations of symptoms; this is very close to operationism. One logical consequence of this account is that psychological disorders cannot meaningfully be appealed to in causal explanations of pathological behavior. If GAD is just a specific cluster of symptoms, explaining why a patient manifests those symptoms by citing GAD would be circular.
Cramer et al. note that the network view agrees with many clinicians' conceptualizations of psychological disorders; however, there seem to be noteworthy ways in which the network view conflicts with clinical practice. Specifically, this view seems not to leave room for special cases, where clinicians would say that the psychiatric disorder is “latent” or “silent” (Lovett & Hood, in press). For instance, say that someone with schizophrenia experiences a lengthy period (several months) without symptoms. Does this person not have schizophrenia during that time? It is consistent with the latent variable approach to say that, for a time, the schizophrenia was not producing symptoms, but the network model does not seem to allow a representation of such a statement. To take a similar case, say that a child diagnosed with attention deficit hyperactivity disorder (ADHD) is placed in an environment where the behavioral symptoms are not permitted to be expressed; on the network view, has the ADHD been cured?
There are additional concerns about the practice of diagnosis itself under the network approach. Individuals are diagnosed with psychological disorders; however, symptom clusters in Cramer et al.'s models are determined by population-level data. It is well established that structures implied by covariation across individuals may not correctly represent cognitive structures in individuals (Borsboom Reference Borsboom2005; Borsboom et al. Reference Borsboom, Kievit, Cervone, Hood, Valsiner, Molenaar, Lyra and Chaudary2009b; Hamaker et al. Reference Hamaker, Nesselroade and Molenaar2007; Molenaar Reference Molenaar2004; Molenaar et al. Reference Molenaar, Huizenga, Nesselroade, Staudinger and Lindenberger2003). This is the problem of local homogeneity, and it is a serious conceptual obstacle for realism with respect to latent variable models generally and psychometric models of intelligence, personality, and psychopathology in particular. As Cramer et al. note, this is a challenge for the network approach as well. They propose addressing it through the analysis of intra-individual time-series data (target article, sect. 8, para. 5), but there are conceptual and practical limitations to this approach. First, if intra-individual structures of covariation in symptoms vary across people, then each patient may have his or her own kinds of disorders, even if multiple patients share the same symptoms. In other words, in such a case, it would not be clear what the basis would be for saying two individuals have the same disorder. Second, suppose that this problem can be overcome and we have some criteria for issuing a diagnosis of, for example, delusional disorder. Cramer et al. would suggest that we diagnose a person as delusional on the basis of intra-individual time-series data. But such a diagnosis is then made relative to the patient's prior level of symptoms, and a patient who has long been consistently delusional would seem to be asymptomatic. Indeed, if a person with psychiatric symptoms seeks mental health services, and the mental health professional wishes to make a diagnosis, how long must the person's case be followed to note covariation in his or her different symptoms over time before a person-specific network structure can be developed and a disorder (defined as a cluster of symptoms) can be diagnosed?
These problems should be judged relative to the benefits of adopting a network approach. Certainly, the network approach can account for the relationships between symptoms that are not a product of common causes (i.e., violations of local independence), and it leads to an intuitive conception of disorders as having fuzzy boundaries. These virtues of the network approach are important conceptual benefits. However, the practical benefits claimed by Cramer et al. are less certain. For instance, Cramer et al. claim that symptom-symptom causal relationships are often the focus of clinical intervention. However, their examples on this point are not actually symptom-symptom relationships (unless “seeing a feared object is itself” a psychiatric symptom, as is “believing that one has not finished a list of tasks”); and in clinical practice, the symptoms themselves (and the consequent impairment in everyday life) are the focus of intervention rather than symptom-symptom relationships.
The challenges we raise here notwithstanding, Cramer et al.'s contribution is ambitious, and their proposal certainly warrants further consideration. We especially like how the authors endeavor to engage a wide audience of methodologists, clinical psychologists, and philosophers of science. In this they show the interdisciplinary relevance of comorbidity and the need to enlist the efforts of diverse specialists in addressing the nature of psychopathology and comorbidity.
You have
Access
Cramer et al.'s account of comorbidity is a thought-provoking contribution to the methodological foundations of psychopathology research. While we agree in spirit with the authors' motivations (e.g., addressing problems with latent variable approaches to comorbidity), we will point out several prima facie challenges for the network approach to comorbidity.
Consider the following philosophical views concerning psychological disorders. Realism, the view suggested by latent variable models, represents psychological disorders as the unobservable causes of symptoms. On this view, as Cramer et al. note, disorders exist independently of our measurement and diagnostic practices. They are detected through measurement. Conventionalist approaches, on the other hand, represent psychological disorders as artifacts of our measurement practices. For example, operationism collapses the distinction between disorders and their symptoms: the symptoms are constitutive of the disorders. Operationists define a disorder such as generalized anxiety disorder (GAD) as nothing more than the satisfying of a certain set of diagnostic criteria. As Cramer et al. note, the differences between these positions can have practical consequences such as different prevalence rates for a disorder. Cramer et al. have sought to carve out a third position with respect to comorbidity and psychological disorders. On their view, disorders are “clusters” of symptoms that systematically covary. Hence, it seems that on the network view, disorders do not exist independently of their symptoms. Disorders are aggregations of symptoms; this is very close to operationism. One logical consequence of this account is that psychological disorders cannot meaningfully be appealed to in causal explanations of pathological behavior. If GAD is just a specific cluster of symptoms, explaining why a patient manifests those symptoms by citing GAD would be circular.
Cramer et al. note that the network view agrees with many clinicians' conceptualizations of psychological disorders; however, there seem to be noteworthy ways in which the network view conflicts with clinical practice. Specifically, this view seems not to leave room for special cases, where clinicians would say that the psychiatric disorder is “latent” or “silent” (Lovett & Hood, in press). For instance, say that someone with schizophrenia experiences a lengthy period (several months) without symptoms. Does this person not have schizophrenia during that time? It is consistent with the latent variable approach to say that, for a time, the schizophrenia was not producing symptoms, but the network model does not seem to allow a representation of such a statement. To take a similar case, say that a child diagnosed with attention deficit hyperactivity disorder (ADHD) is placed in an environment where the behavioral symptoms are not permitted to be expressed; on the network view, has the ADHD been cured?
There are additional concerns about the practice of diagnosis itself under the network approach. Individuals are diagnosed with psychological disorders; however, symptom clusters in Cramer et al.'s models are determined by population-level data. It is well established that structures implied by covariation across individuals may not correctly represent cognitive structures in individuals (Borsboom Reference Borsboom2005; Borsboom et al. Reference Borsboom, Kievit, Cervone, Hood, Valsiner, Molenaar, Lyra and Chaudary2009b; Hamaker et al. Reference Hamaker, Nesselroade and Molenaar2007; Molenaar Reference Molenaar2004; Molenaar et al. Reference Molenaar, Huizenga, Nesselroade, Staudinger and Lindenberger2003). This is the problem of local homogeneity, and it is a serious conceptual obstacle for realism with respect to latent variable models generally and psychometric models of intelligence, personality, and psychopathology in particular. As Cramer et al. note, this is a challenge for the network approach as well. They propose addressing it through the analysis of intra-individual time-series data (target article, sect. 8, para. 5), but there are conceptual and practical limitations to this approach. First, if intra-individual structures of covariation in symptoms vary across people, then each patient may have his or her own kinds of disorders, even if multiple patients share the same symptoms. In other words, in such a case, it would not be clear what the basis would be for saying two individuals have the same disorder. Second, suppose that this problem can be overcome and we have some criteria for issuing a diagnosis of, for example, delusional disorder. Cramer et al. would suggest that we diagnose a person as delusional on the basis of intra-individual time-series data. But such a diagnosis is then made relative to the patient's prior level of symptoms, and a patient who has long been consistently delusional would seem to be asymptomatic. Indeed, if a person with psychiatric symptoms seeks mental health services, and the mental health professional wishes to make a diagnosis, how long must the person's case be followed to note covariation in his or her different symptoms over time before a person-specific network structure can be developed and a disorder (defined as a cluster of symptoms) can be diagnosed?
These problems should be judged relative to the benefits of adopting a network approach. Certainly, the network approach can account for the relationships between symptoms that are not a product of common causes (i.e., violations of local independence), and it leads to an intuitive conception of disorders as having fuzzy boundaries. These virtues of the network approach are important conceptual benefits. However, the practical benefits claimed by Cramer et al. are less certain. For instance, Cramer et al. claim that symptom-symptom causal relationships are often the focus of clinical intervention. However, their examples on this point are not actually symptom-symptom relationships (unless “seeing a feared object is itself” a psychiatric symptom, as is “believing that one has not finished a list of tasks”); and in clinical practice, the symptoms themselves (and the consequent impairment in everyday life) are the focus of intervention rather than symptom-symptom relationships.
The challenges we raise here notwithstanding, Cramer et al.'s contribution is ambitious, and their proposal certainly warrants further consideration. We especially like how the authors endeavor to engage a wide audience of methodologists, clinical psychologists, and philosophers of science. In this they show the interdisciplinary relevance of comorbidity and the need to enlist the efforts of diverse specialists in addressing the nature of psychopathology and comorbidity.