April 1, 2013

Evidence for a Multi-Dimensional Latent Structural Model of Externalizing Disorders

Evidence for a Multi-Dimensional Latent Structural Model of Externalizing Disorders

Katie Witkiewitz & Kevin King & Robert J. McMahon & Johnny Wu & Jeremy Luk & Karen L. Bierman & John D. Coie & Kenneth A. Dodge & Mark T. Greenberg & John E. Lochman & Ellen E. Pinderhughes & the Conduct Problems Prevention Research Group


·         Like most forms of psychopathology, externalizing disorders appear to be dimensional in nature.
·         Latent variable models can be useful in understanding and modeling dimensionality, including hybrid models that mix categorical and dimensional latent variables.
·         Krueger and colleagues have proposed a single, genetically mediated vulnerability (a latent factor) that contributes to the development of conduct disorder (CD), antisocial personality disorder (ASPD), and substance use disorders (SUD).
·         However, Krueger et al. have not included other childhood externalizing disorders (oppositional defiant disorder [ODD] and attention deficit/hyperactivity disorder [ADHD]) in this single-factor model, and when other studies have included ODD and ADHD in models of externalizing behavior, a one-factor model doesn’t fit as well as two- or three-factor models (e.g., models that distinguish “oppositional behaviors” and “social norm violation behaviors”).
·         The goal of the current study was to examine the factor structure of all of these externalizing disorders using models that fit categorical (latent class), dimensional (latent factor), and hybrid (factor mixture) structures.


·         Externalizing psychopathology assessed prospectively through the Fast Track project in children who were selected to be at high risk for conduct problems. Parent and child self-report at grades 3, 6, 9, and 12, and two years post-high school were used to determine lifetime diagnoses of ODD, CD, ADHD, adult antisocial behavior (AAB), and three categories of SUD (alcohol, marijuana, or other drug abuse/dependence).
·         Externalizing psychopathology was assessed retrospectively in a nationally representative sample through the National Comorbidity Survey-Replication (NCS-R). Self-report was used to determine lifetime diagnoses of ODD, CD, ADHD, AAB, and two categories of SUD (alcohol or other drug abuse/dependence).
·         The Fast Track data were used to fit the models, and the NCS-R data were used to test the generalizability of the models via confirmatory factor analysis.
·         The authors fit the following latent factor models:
o   A one-factor model in line with Krueger et al.’s work, proposing one latent factor (“externalizing”) that contributes to all of the externalizing disorders.
o   Four two-factor models: a model dividing SUDs from all of the other disorders (“disruptive behaviors” vs. “substance use disorders”); a model dividing ADHD and ODD (“oppositional disorders”) from all others (“social norm violation disorders”); a model dividing “child-onset externalizing” (ADHD, ODD, CD) from “antisocial/substance use” (AAB and SUDs); and a model which proposes that AAB loads on a factor along with ADHD, ODD, and CD (“disruptive/antisocial behaviors”) as well as a factor with the SUDs (“antisocial/substance use”)
o   A three-factor model, proposing separate factors for “oppositional behavior” (with ADHD and ODD loading), “social norm violation” (CD, AAB), and “substance use disorders.”
·         Several other models were fit as well (see paper for details).


·         The two-factor model in which AAB loaded on a factor with other disruptive behaviors (ADHD, ODD, and CD) as well as on a separate factor with SUDs fit the Fast Track data best out of all of the models. In the later confirmatory factory analysis, this model also provided an excellent fit to the NCS-R data.
·         Notably, AAB was found to have the lowest factor loadings on both factors that were fit. The authors suggested that, while AAB is associated with the other externalizing disorders, it is not necessary to define the structure externalizing behavior.
·         The authors concluded that latent structure of externalizing disorders is likely to involve at least two factors, rather than a single factor. Moreover, a dimensional, rather than categorical, model of externalizing disorders was supported.
·         The authors also noted that the best-fitting model is consistent with previous work that groups ADHD, ODD, and CD together, as well as research that different developmental pathways (e.g., through associations with internalizing disorders) appear to be functioning in the etiology of SUDs in comparison to the other externalizing disorders.
·         Limitations of the study include the fact that different measures and reporters were used over time in the Fast Track study; some groups of participants were more likely to have missing data than others in the Fast Track study (African Americans, those with ADHD or ODD diagnosis); recall in the NCS-R study may have been inaccurate; the AAB variable was not based on full DSM symptoms of ASPD; measurements more sensitive to dimensionality in symptoms might have been more appropriate; and many other well-fitting models could have been fit to the data.

Implications/Conclusions/Future Directions

·         The authors state that a proposal had been made to include CD, SUD, and possibly borderline personality disorder and ADHD in an “externalizing disorders” section of DSM-5. They argue that not considering the commonalities of CD, ODD, and ADHD, while grouping CD with the distinct category of SUDs, may make research into the developmental pathways of these disorders difficult.
·         The authors encourage further exploration of the factors involved in the development of externalizing psychopathology in order to gain advances in etiology, prevention, and treatment research.


·         The model that was supported (which separated disruptive behaviors from substance use disorders) works intuitively for me, since SUDs seem so different clinically from the other externalizing disorders (and of course there’s the commonly voiced call for non-violent drug offenders to be released from prison or have reduced sentences, which strikes me as different from the attitude usually taken toward offenders). Does anyone have any arguments against this separation?
·         In the model that was supported, adult antisocial behavior loaded on both factors, but it didn’t fit particularly well with either. Is it possible that the low loading of ASPD with disruptive behavior disorders is due to how these various disorders are measured? That is, the disruptive behaviors are measured using items that are more suitable for kids – the symptoms of ADHD and ODD are clearly more descriptive of child behavior, and while many of the symptoms of CD can also apply to adults, the “status violation” symptoms (e.g., running away from home over night, being truant from school, etc.) cannot, by definition, apply to adults. If they only included symptoms that apply equally to adults and children, do you think the factor loadings for adult antisocial behavior would increase?
·         Are there any disorders/symptoms/behaviors that you think should be included in future investigations of the latent structure of externalizing disorders? For example, the authors noted that borderline personality disorder has been considered for inclusion in a section of the DSM on externalizing disorders.

1 comment:

  1. Thanks for the interesting article and great review. In terms of your second question, it is an interesting thought and I agree that if the
    authors only included symptoms that apply equally to adults and children, the factor loadings for adult antisocial behavior would
    potentially increase, but that would omit the important developmental aspect of these disorders. Another way to think of the model is as a developmental segregation of pathways to externalizing disorders where the disruptive/antisocial folks start out with ADHD/ODD/CD/AAB early on, the antisocial/substance use folks develop their externalizing disorders later in life, and the correlation between these two factors represents the disruptive kids who go on to develop substance use and antisocial behavior later in life. They may have talked about this in the article (I just skimmed it), but just an additional thought that came to mind.