Cramer et al.'s network model represents a promising new approach for conceptualizing and quantifying comorbidity in psychiatric diagnosis, helping avoid the thorny challenge of operationalizing latent constructs, and shifting the focus of comorbidity research from syndrome to symptom. Scrutiny of Cramer et al.'s analysis reveals that the theoretical underpinnings and empirical evidence bearing on this model apply most directly to Axis I symptom disorders (e.g., major depression, generalized anxiety). Because Axis II personality disorders differ in myriad ways from Axis I symptom disorders, the degree to which the network comorbidity model generalizes to Axis II disorders remains open to question. This commentary addresses that issue, discussing issues that arise in applying the network model to DSM-diagnosed personality pathology (i.e., the personality disorder [PD] diagnoses offered in the Diagnostic and Statistical Manual of Mental Disorders, 4th edition or DSM-IV; American Psychiatric Association 1994).
As Cramer et al. have noted, diagnostic comorbidity evidence involving DSM-IV Axis I disorders can yield ambiguous, confusing patterns. Diagnostic comorbidity evidence bearing on DSM-IV Axis II is far worse. Consider: The number of differential diagnoses per DSM-IV PD ranges from 3 (dependent, obsessive-compulsive) to 7 (paranoid), with the mean number of differential diagnoses per PD being 4.5. Thus, on average each DSM-IV PD shows substantial overlap with 50% of the remaining PDs. When Ekselius et al. (Reference Ekselius, Lindstrom, Knorring, Bodlund and Kullgren1994) calculated correlations among interview-derived scores for PDs in a heterogeneous sample of psychiatric patients and nonclinical participants, they obtained a mean interscale correlation (r) of .41, and statistically significant interscale correlations in 41 of 45 comparisons (91%). Subsequent comorbidity studies have confirmed these results (Bornstein Reference Bornstein1998; Reference Bornstein2005).
Given these patterns, extending the network comorbidity model to Axis II presents some unique challenges, but it also involves some unique opportunities to gain new perspective on the model, its advantages, and its limitations. Two issues are germane in this context.
1. PD symptoms differ fundamentally from Axis I symptoms
Axis I disorders are sometimes called “symptom disorders” because they are characterized by prominent, psychologically painful symptoms (e.g., depressed mood, difficulty sleeping, binge eating). Whether or not patients choose to acknowledge them when asked, they are typically aware of experiencing these symptoms (even if not fully aware of the symptoms' negative impact). The situation is very different for Axis II PDs, which have traditionally been conceptualized as being “ego syntonic” (i.e., consistent with the patient's experience of self). As a result, personality-disordered patients typically have far less insight into their symptoms than do patients with Axis I disorders, which complicates diagnosis, decreases motivation for treatment, and reduces therapeutic efficacy (Peters Reference Peters1990; Shedler & Westen Reference Shedler and Westen1999).
In the context of the network model, these Axis I–Axis II differences have two noteworthy implications. First, although self-report assessment tools (e.g., questionnaires, diagnostic interviews) are the measures of choice for rendering Axis I diagnoses, such measures are of limited value in rendering Axis II diagnoses (see, e.g., Widiger & Samuel Reference Widiger and Samuel2005). Indirect measures (e.g., free-response tests) and reports from knowledgeable informants must be used in conjunction with self-report instruments to assess PDs reliably (Bornstein Reference Bornstein2007).
Second, these Axis I–Axis II differences in insight and self-awareness suggest that the definition of latent variable as conceptualized in the network model must be expanded when applied to Axis II. Here it is not only necessary to distinguish observable symptoms from unobservable latent constructs, as the network model suggests, but also to distinguish symptoms that are phenomenologically latent (i.e., ego syntonic) from those that are experienced as problematic by the patient (i.e., ego dystonic).
2. Unlike Axis I criteria, Axis II criteria are revised to minimize comorbidity
In many clinical settings the most common Axis II diagnosis is “mixed PD,” and epidemiological data indicate that a sizeable proportion of PD-diagnosed patients – more than 50% in some samples – receive two or more PD diagnoses (Bornstein Reference Bornstein2003; Widiger & Clark Reference Widiger and Clark2000). As a result, symptom revision across successive editions of the DSM entails somewhat different goals on Axis I and Axis II. On Axis I symptoms are revised to increase diagnostic accuracy, but on Axis II symptoms are revised to maximize accuracy while simultaneously reducing comorbidity.
As I have noted elsewhere (Bornstein Reference Bornstein2003), when Axis II symptoms are reworded or removed merely to limit escalating comorbidity rates, clinicians are choosing to alter reality (i.e., a high level of PD comorbidity) to fit some idealized conceptualization of PDs as distinct and separate syndromes. For example, “frantic efforts to avoid real or imagined abandonment” (see Diagnostic and Statistical Manual of Mental Disorders, 3rd edition, revised [DSM-III-R]; American Psychiatric Association 1987, p. 347) was removed from the dependent PD criteria in DSM-IV because patients with borderline PD also show this symptom, but every extant model of dependent PD would argue for inclusion of this symptom (Bornstein Reference Bornstein2005). Removing certain PD symptoms merely to minimize PD overlap is akin to arguing that labored breathing should no longer be considered a symptom of pneumonia because patients with emphysema also show this symptom. Clearly, the contrasting strategies used to revise symptoms on Axis I and Axis II present a challenge when the network comorbidity model is extended from symptom disorders to personality pathology.
Without question, Cramer et al.'s network model represents a promising new approach to conceptualizing and quantifying comorbidity in psychiatric diagnosis. This perspective not only captures dynamic features of psychopathology that traditional latent variable models cannot capture, but has the additional advantages of shifting the focus from surface behavior to underlying process, and the level of analysis from syndrome to symptom. Extending the network model to Axis II will be challenging, but likely to benefit the model over the long term by compelling researchers to confront conceptual and empirical challenges that do not arise when the model is applied to Axis I. Questions regarding the generalizability of symptom clusters, nodes, and bridge symptoms across culture, age, and gender are almost certain to emerge, and as research on the network model advances, it will not only be useful to extend this model from Axis I to Axis II, but to begin to address aspects of cross-axis comorbidity as well.
You have
Access
Cramer et al.'s network model represents a promising new approach for conceptualizing and quantifying comorbidity in psychiatric diagnosis, helping avoid the thorny challenge of operationalizing latent constructs, and shifting the focus of comorbidity research from syndrome to symptom. Scrutiny of Cramer et al.'s analysis reveals that the theoretical underpinnings and empirical evidence bearing on this model apply most directly to Axis I symptom disorders (e.g., major depression, generalized anxiety). Because Axis II personality disorders differ in myriad ways from Axis I symptom disorders, the degree to which the network comorbidity model generalizes to Axis II disorders remains open to question. This commentary addresses that issue, discussing issues that arise in applying the network model to DSM-diagnosed personality pathology (i.e., the personality disorder [PD] diagnoses offered in the Diagnostic and Statistical Manual of Mental Disorders, 4th edition or DSM-IV; American Psychiatric Association 1994).
As Cramer et al. have noted, diagnostic comorbidity evidence involving DSM-IV Axis I disorders can yield ambiguous, confusing patterns. Diagnostic comorbidity evidence bearing on DSM-IV Axis II is far worse. Consider: The number of differential diagnoses per DSM-IV PD ranges from 3 (dependent, obsessive-compulsive) to 7 (paranoid), with the mean number of differential diagnoses per PD being 4.5. Thus, on average each DSM-IV PD shows substantial overlap with 50% of the remaining PDs. When Ekselius et al. (Reference Ekselius, Lindstrom, Knorring, Bodlund and Kullgren1994) calculated correlations among interview-derived scores for PDs in a heterogeneous sample of psychiatric patients and nonclinical participants, they obtained a mean interscale correlation (r) of .41, and statistically significant interscale correlations in 41 of 45 comparisons (91%). Subsequent comorbidity studies have confirmed these results (Bornstein Reference Bornstein1998; Reference Bornstein2005).
Given these patterns, extending the network comorbidity model to Axis II presents some unique challenges, but it also involves some unique opportunities to gain new perspective on the model, its advantages, and its limitations. Two issues are germane in this context.
1. PD symptoms differ fundamentally from Axis I symptoms
Axis I disorders are sometimes called “symptom disorders” because they are characterized by prominent, psychologically painful symptoms (e.g., depressed mood, difficulty sleeping, binge eating). Whether or not patients choose to acknowledge them when asked, they are typically aware of experiencing these symptoms (even if not fully aware of the symptoms' negative impact). The situation is very different for Axis II PDs, which have traditionally been conceptualized as being “ego syntonic” (i.e., consistent with the patient's experience of self). As a result, personality-disordered patients typically have far less insight into their symptoms than do patients with Axis I disorders, which complicates diagnosis, decreases motivation for treatment, and reduces therapeutic efficacy (Peters Reference Peters1990; Shedler & Westen Reference Shedler and Westen1999).
In the context of the network model, these Axis I–Axis II differences have two noteworthy implications. First, although self-report assessment tools (e.g., questionnaires, diagnostic interviews) are the measures of choice for rendering Axis I diagnoses, such measures are of limited value in rendering Axis II diagnoses (see, e.g., Widiger & Samuel Reference Widiger and Samuel2005). Indirect measures (e.g., free-response tests) and reports from knowledgeable informants must be used in conjunction with self-report instruments to assess PDs reliably (Bornstein Reference Bornstein2007).
Second, these Axis I–Axis II differences in insight and self-awareness suggest that the definition of latent variable as conceptualized in the network model must be expanded when applied to Axis II. Here it is not only necessary to distinguish observable symptoms from unobservable latent constructs, as the network model suggests, but also to distinguish symptoms that are phenomenologically latent (i.e., ego syntonic) from those that are experienced as problematic by the patient (i.e., ego dystonic).
2. Unlike Axis I criteria, Axis II criteria are revised to minimize comorbidity
In many clinical settings the most common Axis II diagnosis is “mixed PD,” and epidemiological data indicate that a sizeable proportion of PD-diagnosed patients – more than 50% in some samples – receive two or more PD diagnoses (Bornstein Reference Bornstein2003; Widiger & Clark Reference Widiger and Clark2000). As a result, symptom revision across successive editions of the DSM entails somewhat different goals on Axis I and Axis II. On Axis I symptoms are revised to increase diagnostic accuracy, but on Axis II symptoms are revised to maximize accuracy while simultaneously reducing comorbidity.
As I have noted elsewhere (Bornstein Reference Bornstein2003), when Axis II symptoms are reworded or removed merely to limit escalating comorbidity rates, clinicians are choosing to alter reality (i.e., a high level of PD comorbidity) to fit some idealized conceptualization of PDs as distinct and separate syndromes. For example, “frantic efforts to avoid real or imagined abandonment” (see Diagnostic and Statistical Manual of Mental Disorders, 3rd edition, revised [DSM-III-R]; American Psychiatric Association 1987, p. 347) was removed from the dependent PD criteria in DSM-IV because patients with borderline PD also show this symptom, but every extant model of dependent PD would argue for inclusion of this symptom (Bornstein Reference Bornstein2005). Removing certain PD symptoms merely to minimize PD overlap is akin to arguing that labored breathing should no longer be considered a symptom of pneumonia because patients with emphysema also show this symptom. Clearly, the contrasting strategies used to revise symptoms on Axis I and Axis II present a challenge when the network comorbidity model is extended from symptom disorders to personality pathology.
Without question, Cramer et al.'s network model represents a promising new approach to conceptualizing and quantifying comorbidity in psychiatric diagnosis. This perspective not only captures dynamic features of psychopathology that traditional latent variable models cannot capture, but has the additional advantages of shifting the focus from surface behavior to underlying process, and the level of analysis from syndrome to symptom. Extending the network model to Axis II will be challenging, but likely to benefit the model over the long term by compelling researchers to confront conceptual and empirical challenges that do not arise when the model is applied to Axis I. Questions regarding the generalizability of symptom clusters, nodes, and bridge symptoms across culture, age, and gender are almost certain to emerge, and as research on the network model advances, it will not only be useful to extend this model from Axis I to Axis II, but to begin to address aspects of cross-axis comorbidity as well.