One well-established model of co-occurring mental disorders that may be relevant to the understanding of the structure of posttraumatic psychopathology proposes that patterns of behavioral disturbance and psychiatric symptoms cohere along latent dimensions termed externalizing (EXT) and internalizing (INT). This model, rooted in a tradition of research on childhood behavior disorders (Achenbach & Edelbrock, 1978, 1984), has recently come to the fore in the adult psychopathology literature as the result of a series of influential factor-analytic studies on the latent structure of adult mental illness (Kendler, Prescott, Myers, & Neale, 2003; Kessler, Chiu, Demler, Merikangas, & Walters, 2005; Krueger, 1999; Krueger, Caspi, Moffitt & Silva, 1998; Krueger, Chentsova-Dutton, Markon, Goldberg, & Ormel, 2003; Krueger, McGue, & Iacono, 2001; Vollebergh et al., 2001). Findings of these studies suggest that the alcohol- and drug-related disorders and antisocial personality disorder load on an EXT dimension characterized primarily by problems in the domain of impulsivity, whereas the unipolar mood and anxiety disorders fall on an INT dimension defined largely by heightened negative emotionality. In several studies, the latter has been subdivided into correlated factors termed “anxious-misery” (defined by major depression, dysthymia, generalized anxiety disorder) and “fear” (comprised of panic and phobic disorders; Cox, Clara, & Enns, 2002; Krueger, 1999; Vollebergh et al., 2001; Slade & Watson, 2006). This basic structure has demonstrated invariance across sex, time, and multiple samples drawn at random from a larger one (Kendler et al., 2003; Krueger, 1999; Vollebergh et al., 2001) and may offer a useful framework for incorporating a dimensional component into future editions of the Diagnostic and Statistical Manual (DSM; APA).

Emerging evidence suggests that the INT/EXT model may also be relevant to our understanding of the heterogeneity of posttraumatic psychopathology (Miller, Greif, & Smith, 2003; Miller, Kaloupek, Dillon, & Keane, 2004; Miller & Resick, 2007). Through a series of cluster analytic studies of personality inventories completed by individuals with PTSD and other disorders, Miller and colleagues (2003, 2004, 2007) have found and replicated evidence of INT and EXT subtypes in both male veteran and female civilian samples. Summarizing across these studies, one subtype, termed “externalizing”, was characterized by elevated levels of anger and aggression, substance-related disorders, antisocial and borderline personality disorder features, and personality inventory profiles defined by high disconstraint coupled with high negative emotionality. In contrast, the “internalizing” subtype was characterized by high rates of comorbid major depression and panic disorder, schizoid and avoidant personality disorder features, and personality profiles defined by high negative emotionality combined with low positive emotionality. Similar patterns that might now be interepreted as reflecting individual differences in internalizing and externalizing processes were described in earlier cluster analytic studies of U.S. and Australian veterans with PTSD (Hyer, Davis, Albrecht, Boudewyns, Woods, 1994; Forbes et al., 2003; Piekowski, Sherwood, & Funari, 1993). Together, these findings suggest that the INT/EXT model be may be a useful heurisitic for studying the structure of posttraumatic psychopathology. However, while cluster analysis is a reasonable starting point for classifying qualitatively different subgoups of individuals, it is not well suited for examining facets of psychopathology that are conceptualized as fundamentally dimensional in nature. Thus, a primary objective of this study was to test competing models for the latent structure of posttraumatic psychopathology using confirmatory factor analysis.

Three prior published studies have employed factor analysis to examine the relationship of PTSD to hypothesized INT and EXT dimensions of psychopathology. In the first, Cox et al. (2002) performed a principal components analysis on lifetime prevalence data from the National Comorbidity Survey (Kessler et al., 1994) and found that PTSD loaded rather weakly on Anxious-Misery (r = .39) but not on Fear or EXT in a three-factor (Fear, Anxious-Misery, and EXT) solution. This pattern was replicated in 12-month prevalence data and indicated that PTSD may share more common variance with disorders defined by anhedonic mood and anxious rumination than with those characterized primarily by pathological fear (e.g., panic disorder and phobias) or externalizing. In the second study, Kessler et al. (2005) performed an exploratory factor analysis on 12-month prevalence data from the National Comorbidity Survey Replication and found evidence for a two-factor (INT/EXT) solution with PTSD loading again on INT but not EXT. In the third one, Slade & Watson (2006) performed confirmatory factor analyses on 12-month prevalence data from the Australian National Survey of Mental Health and Wellbeing (N = 10,641) and found that PTSD loaded strongly on Anxious-misery (r = .83) but not on Fear or EXT in a three-factor solution.

Conclusions from these studies were limited by two concerns. First, although representative of the general population, the base rates of PTSD were rather low in these studies (7.8% for lifetime in the first, 3.7% for 12-month in the second, 1.3% for 12-month in the third). In the present study, we wondered whether the INT/EXT factor structure would replicate in a sample with higher rates of diagnosed psychopathology and how the pattern of factor loadings for PTSD might change. Only one prior study has examined this question in a treatment-seeking sample. In that one, which did not include PTSD in the analysis, Krueger (1999) examined data from a subsample (n = 251) of the National Comorbidity Survey but reported equivocal findings regarding the relative fit of 2, 3, or 4 factor solutions. A second limitation of prior studies that have examined the relationship of PTSD to INT/EXT was that two out of three of them employed analyses that do not permit a priori specification of the patterns of relationships between observed measures and latent variables so they were unable to test competing models/hypotheses for the associations between disorders. Furthermore, no prior study has focused on testing alternative representations of the relationship of PTSD to these dimensions.

This study was designed to address these limitations in two ways. First, we examined data drawn from a large sample of Vietnam veterans receiving services in the U.S. Department of Veterans Affairs Healthcare System (VA) who had a high rate of combat-related PTSD and other disorders. Second, we performed a sequence of hypothesis-driven model testing using confirmatory factor analysis that permitted comparison of alternative models for the relationship of PTSD to latent dimensions of psychopathology. We began by examining a one-factor general disturbance model in which all disorders were specified as indicators of a single latent psychopathology dimension. Next, we evaluated variations of the two-factor (INT/EXT) and three-factor (FEAR, anxious-misery [ANX-MIS], and EXT) models suggested by prior research. In each of these models, major depressive disorder, panic disorder/agoraphobia, and obsessive-compulsive disorder were conceptualized as indicators of INT, whereas alcohol abuse/dependence, substance abuse/dependence, and antisocial personality disorder were defined as indicators of EXT. In one variation, PTSD loaded exclusively on INT (two-factor model; cf. Kessler et al., 2005) or ANX-MIS (three-factor model; cf. Cox et al., 2002). In a second variation, suggested by prior evidence for high rates of comorbidity between PTSD and disorders of both the INT and EXT domain, PTSD cross-loaded on ANX-MIS and EXT (Breslau et al., 1991, 2000; Kessler et al., 1995; Kulka et al., 1990; Miller et al., 2003, 2004, 2007; Xian et al., 2000).

We also tested alternative representations of the relationship of alcohol and drug abuse/dependence to the hypothesized INT and EXT latent factors. Recent research linking substance-related disorders to an EXT factor defined also by adult and childhood antisocial behavior and the personality dimension disconstraint (e.g., Krueger et al., 2001) contrasts with another body of research demonstrating strong associations between these disorders and psychopathology in the domain of INT. For example, research has shown high rates of alcohol and drug dependence in individuals with major depression (e.g., Hasin, Goodwin, Stinson, & Grant, 2005; Kendler, Heath, Neale, Kessler, & Eaves, 1993) and twin and family studies suggest that the association between these disorders may be due to a common genetic factor (Kendler et al., 1993; Lin et al., 1996; Merikangas, Risch, & Weissman, 1994). In individuals with PTSD, substance-related disorders are among the most common comorbidities (Breslau et al., 1991; Kessler et al., 1995; Kulka et al., 1990) and evidence suggests that when these disorders co-occur, PTSD tends to predate the substance abuse (Kessler, 2000; Kessler et al., 1995). The dominant explanation for this association, known as the self-medication hypothesis, posits that individuals abuse substances in an effort to reduce or control distress-related symptoms and this behavior is maintained by negative reinforcement resulting from symptom relief (Baker, Piper, McCarthy, Majeskie, & Fiore, 2004; Stewart, 1996). This conceptually distinct model provided the rationale for testing alternative representations of the associations between alcohol abuse/dependence, drug abuse/dependence, and the latent factors INT and EXT. To do so, our model testing sequence evaluated the relative fit of models in which alcohol and drug abuse/dependence loaded exclusively on EXT, and two alternatives suggested by the self-medication hypothesis: In the first, alcohol abuse/dependence cross-loaded on INT (2 factor solution) or ANX-MIS (3 factor solution), in the second, alcohol abuse/dependence and drug abuse/dependence both cross-loaded on these factors.

The best fitting model, shown in Figure 1, was the three factor variation in which PTSD loaded on ANX-MIS and alcohol and drug abuse/dependence both cross-loaded on ANX-MIS and EXT. The diagram illustrates several important findings. First, examination of the associations among the latent factors showed a high correlation (.67) between the anxious-misery and fear factors. Second, there was a significant but substantially smaller correlation between the Fear and EXT factors (.20), but no such association between ANX-MIS and EXT. Third, factor loadings were generally large, with two exceptions: the loadings of alcohol and drug abuse/dependence on ANX-MIS (.37 and .26, respectively), while statistically significant, were considerably smaller than their loadings on EXT (.67 and .54, respectively).

These findings suggest that PTSD may share more in common with the melancholy, apathy, and rumination of major depression than with the disabling anxiety and avoidance that is characteristic of panic disorder/agoraphobia and obsessive-compulsive disorder—a finding that is inconsistent with the current placement of PTSD in the anxiety disorders category in DSM-IV. This finding is in line with evidence from a previous study of twin brothers of Vietnam veterans which showed that (a) combat veterans with PTSD had higher rates of major depression and dysthymia than combat controls and (b) the unexposed twins of veterans with combat-related PTSD had higher rates of both disorders than the unexposed twins of combat veterans without PTSD (Koenen et al., 2003). Taken collectively, these findings suggest that the affinity between major depression and PTSD observed in this and prior structural analyses (Cox et al., 2002; Kessler et al., 2005; Slade & Watson, 2006) may reflect an endophenotypic trait or common genetic factor that confers vulnerability for the development of posttraumatic psychopathology, i.e., negative emotionality (for a review, see Miller, 2003).

The second internalizing factor, termed Fear, was defined by OCD and panic disorder. OCD has been excluded from most previous studies of this type because of its low prevalence in the general population so knowledge about its structural relationships to other disorders is limited. Three prior studies that have addressed the issue produced inconsistent results. Brown et al. (1998) reported that OCD was more highly correlated with major depression and generalized anxiety disorder than with panic disorder/agoraphobia or social phobia. In contrast, Watson (2005) and Slade and Watson (2006) found that OCD loaded primarily on a Fear factor defined also by panic disorder, agoraphobia, social phobia. The relationship between panic disorder and PTSD has been the source of similar disagreement. Some investigations have suggested that a close affinity exists between these two disorders, possibly reflecting the involvement of pathological autonomic arousal in both (Brown, Campbell, Lehman, Grisham, & Mancill, 2001), while others suggest a dissociation between them. For example, Cox et al. (2002) and Slade and Watson (2006) both reported that panic disorder loaded primarily on a Fear factor, whereas PTSD loaded on an Anxious-Misery factor as it did in the present study.

Another noteworthy finding from this study was that the best fitting model included cross-loadings of alcohol and drug abuse/dependence diagnoses on both ANX-MIS and EXT. The relative strengths of these loadings indicated that these disorders cohered primarily with the EXT factor defined also by antisocial personality disorder (as they have in prior studies using epidemiologic samples), but they also showed a significant affinity for the ANX-MIS factor defined by major depression and PTSD. In addition, alcohol abuse/dependence loaded more highly on ANX-MIS than drug abuse/dependence did. These novel findings align empirically with prior evidence of high rates of comorbid alcohol and drug abuse in individuals with major depression and PTSD and they are conceptually consistent with the aforementioned self-medication hypothesis which posits a reciprocal relationship between negative emotional distress and addictive behaviors (Baker et al., 2004; Stewart, 1996). Theses associations are also consistent with prior evidence for a stronger relationship between alcohol-related problems and symptoms in the domain of negative emotionality than between illicit drug-related problems and this dimension (Miller, Vogt, Mozley, Kaloupek, & Keane, 2006). Most importantly, these findings appear to reconcile contrasting models of comorbid substance-related disorders and seem to capture the complex relationships between these disorders and broad but distinct (INT and EXT) classes of psychopathology. The fact that similar cross-loadings have not been reported in prior analyses of data from community studies implies that there may be differences in the structure of comorbidity between epidemiologic and clinical samples (Keane & Wolfe, 1990). Stronger links between substance-related disorders and psychopathology of the INT spectrum may be evident in pathological groups in which these disorders develop secondary to a primary INT syndrome. Unfortunately, the cross-sectional data from this study precluded a direct test of this temporal hypothesis.