We examined whether a two-factor model positing internalizing and externalizing factors represented the structure and co-occurrence of six psychiatric disorders (MDD, SAD, GAD, ODD, ADHD, and CD) in adolescents. Additionally, we examined the etiology (i.e., the magnitude of genetic and environmental influences) of the latent internalizing and externalizing factors and the covariation between the internalizing and externalizing factors. Phenotypic correlations suggested significant covariation among all six psychiatric disorders. Correlations among internalizing disorders and among externalizing disorders were higher on average than correlations between the two sets of disorders, and the results of factor analyses confirmed that there are two latent factors (internalizing and externalizing) underlying these common psychiatric disorders. In general, within-trait/cross-sibling correlations and cross-trait/cross-sibling correlations were higher in the MZ twin pairs than in DZ twin pairs or sibling pairs, suggesting genetic influences on the variations in individual disorders as well as on covariation between disorders.
The latent internalizing and externalizing factors explained 30% to 41% of the variance of the individual internalizing disorders and 26% to 52% of the variance of the individual externalizing disorders, respectively. These results are consistent with prior reports that found evidence for a two-factor model positing internalizing and externalizing factors (Achenbach, 1991
; Krueger, Caspi, Moffitt, & Silva, 1998
; Krueger, McGue, & Iacono, 2001
The etiology of the common internalizing and externalizing factors was best explained by additive genetics and non-shared environment (h2
= .60 and e2
= .40 for the internalizing factor and h2
= .65, e2
= .35 for the externalizing factor). The common internalizing and externalizing factors have higher heritabilities than each of the six individual disorders and are not influenced by shared environmental influences. In contrast, Ehringer et al.’s (2006)
univariate study of the same sample reported heritabilities of .36, .31, .56, and .30 for SAD, ADHD, CD, and ODD, respectively and was unable to distinguish between genetic and shared environmental influences on GAD and MDD because of inadequate power.
Most of the familial influences on the six psychiatric disorders examined were those on the latent internalizing and externalizing factors (see Figure 3). There was little evidence of disorder-specific familial influences, with the exception of shared environmental influences on MDD (disorder-specific c2 = .11) and additive genetic influences on CD (disorder-specific h2 = .38). Disorder-specific influences were limited to nonshared environmental influences (i.e., environmental influences leading to differences in sibling pairs or measurement error) for SAD, GAD, ADHD, and ODD.
The correlation between latent internalizing and externalizing factors (r = .72) was due to both common genetic (62%) and nonshared environmental (38%) influences. Although our results provide strong support for two separate internalizing and externalizing factors, common genetic and environmental factors confer risk across internalizing and externalizing factors. Since this result suggests the presence of a broad child psychopathology factor, we tested a broad childhood psychopathology model with a single latent psychopathology factor. This model did not represent psychopathology in our sample as well as the model including two correlated but separate internalizing and externalizing factors.
Of studies examining comorbidity between internalizing and externalizing disorders, our results most resemble those of O’Connor and colleagues (1998)
who found strong evidence for a common genetic liability underlying comorbidity between depression and antisocial behavior in adolescents. However, our results are inconsistent with those of Gjone and Stevenson (1997)
, who concluded that covariation between internalizing and externalizing behavior in childhood mainly results from common shared environmental influences.
As noted above, there are several methodological differences (e.g., the construct examined, the assessment method, and the age of the participants; see Introduction) that may have led to discrepancies in the results of these studies. First, it is possible that the use of parent report in Gjone and Stevenson (1997)
may have increased the estimate of common shared environmental influences. Parental ratings may be subject to rater biases (i.e., overestimating or underestimating scores consistently), and if such biases inflate the correlations for both MZ and DZ twin pairs, the evidence of shared environmental influences will be inflated (Hewitt et al., 1992
). Second, the statistically significant common nonshared environmental influences found in the present study may be inflated given the use of self-report interviews in the present study. Rater biases or measurement error specific to individuals may have increased the correlation between nonshared environmental influences on internalizing and externalizing disorders. Third, some studies have noted problems with the reliability and validity of the DISC (McMahon & Frick, 2005
; Silverman & Ollendick, 2005
), which was used in the present study. Specifically, the test-retest reliability kappa coefficient for MDD (.92) is higher than that for SAD (.46) and GAD (.58), suggesting that the DISC may be a better assessment of depressive than anxiety-related symptomatology (Shaffer et al., 2000
). If diagnoses made with the DISC in the present study are unreliable, the estimates of the comorbidity between internalizing and externalizing also may be unreliable. Furthermore, symptoms for diagnoses assessed earlier in a DISC interview may be more valid than those assessed later because endorsement of symptoms tends to diminish over the course of an interview (Jensen et al., 1999
). The degree to which methodological differences explain the discrepancies in the results is unclear, as systematic studies examining whether these methodological differences lead to differing conclusions regarding the etiology of the covariance between internalizing and externalizing disorders have not been conducted.
Our findings also differ from those of Kendler et al. (2003)
, and there are notable methodological differences between our study and Kendler et al.’s. Their internalizing factor included fear disorders and their externalizing factor included substance use disorders, whereas in the present study, the internalizing factor did not include fear disorders and the externalizing factor included ADHD and ODD. Kendler et al. examined an adult sample, whereas the present study examined an adolescent sample. Third, their multivariate model was quite different, with orthogonal internalizing and externalizing genetic factors.
Genetic contributions to developmental psychopathology are well recognized in contemporary research. Our results suggest that internalizing and externalizing disorders may in part share genetic influences. In other words, adolescents genetically predisposed to internalizing disorders may also be predisposed to externalizing psychopathology. Based on similar assumptions, recent reports have examined multiple genes as candidates for both internalizing and externalizing disorders (e.g., serotonin transporter gene – Caspi et al., 2003
; Sakai et al., 2006
, serotonin 2A receptor gene – Eley et al., 2004
; Mik et al., 2007
; dopamine D4 receptor gene – Lopez Leon et al., 2005
; Rowe et al., 2001
). While definitive roles for specific genes in the development of internalizing, externalizing, or co-occurring disorders have not yet been established, these candidate genes may confer risk for both classes of psychopathology.
Nonshared environment influenced latent internalizing and externalizing psychopathology factors as well as their covariation significantly. Nonshared environment includes measurement error and all influences that result in differences between individuals growing up in the same family. Possible sources of common non-shared environmental influences affecting development include family composition, sibling interactions, and peer influences (e.g., Plomin et al., 2001
Environmental conditions that are not shared between siblings appear to contribute more to the development of psychopathology than shared environment. Our findings of minimal shared environmental influences on either latent internalizing or externalizing factors are consistent with many behavioral genetics findings of psychopathology (Turkheimer et al., 2000
) and are not surprising. Shared environment seems to be particularly influential in high-risk environments such as disadvantaged neighborhoods or unstable family life (Burt, 2009
). However, ours is a community-based sample where prevalence rates for internalizing and externalizing psychopathology are relatively low and a majority of children were likely being raised in safe, stable environments.
On average, we observed higher correlations between MDD and externalizing disorders than between MDD and the other internalizing disorders (although MDD was also significantly correlated with the other internalizing disorders). In contrast, GAD and SAD were more highly correlated with the other internalizing disorders than with the externalizing disorders. This result is somewhat consistent with results from Hewitt et al. (1997)
, who reported higher factor correlations between MDD and ADHD, CD, and ODD (.34, .25, and .37, respectively) than between SAD or overanxious disorder and externalizing disorders (.07 to .22) in a sample of 8- to 16-year-olds. Similarly, Angold et al.’s (1999)
meta-analysis concluded that the odds ratio for the comorbidity between depression and CD/ODD (OR = 6.6) and between depression and ADHD (OR = 5.5) was higher than that between anxiety and externalizing disorders (OR = 3.0 to 3.1). It is possible that the higher than expected correlation between MDD and externalizing disorders is spurious or due to measurement error. Another hypothesis for the higher comorbidity between MDD and externalizing disorders is that in children and adolescents, irritability is a symptom common to both major depression as well as externalizing psychopathologies. Ideally, we would have tested a model examining whether MDD is still highly correlated with externalizing psychopathology after symptoms of “irritability” are removed. However, we could not test this hypothesis, as we did not have information regarding all symptoms for all CADD participants since the DISC-IV was stopped after diagnosis was established (i.e., enough symptoms were present).
The results of the present study should be interpreted while considering the following limitations. The present study examined only three internalizing disorders, MDD, GAD, and SAD, and did not examine any of the fear and panic disorders. Therefore, we were not able to test the validity of models making finer distinctions among the internalizing disorders, such as the distinction between the Anxious-Misery and Fear subfactors (Kendler et al., 2003
; Krueger, 1999
). Additional behavior genetic studies examining a larger number of internalizing disorders in children and adolescents are needed.
The data analyzed were collected from the general population, where the prevalence of psychiatric disorders is low (especially for internalizing disorders in males; see Ehringer et al, 2006
). Caution should be used when interpreting prevalence rates (see ) for CD and ADHD diagnoses in our sample since they do not reflect DSM-IV criteria with respect to age of onset of symptoms or symptom clustering. The data were based on self-report interviews, which may have led to underestimates of heritability (Eaves et al., 1997
). For example, the heritability of .31 for ADHD in the present sample is lower than those reported by researchers examining parent reports (although there is evidence that rating bias may be present in parent reports of ADHD symptoms [Simonoff et al., 1998
As noted above, the DISC may have imperfect reliability and validity (McMahan & Frick, 2005; Silverman & Ollendick, 2005
), leading to incorrect estimates of comorbidity between internalizing and externalizing disorders. However, it is important to note that significant covariances were observed among internalizing and externalizing disorders, and that these covariances were stronger within than between internalizing and externalizing disorders. Furthermore, there were significant genetic influences on the latent internalizing and externalizing factors (which would not be possible with an unreliable, invalid measure). These findings suggest that the DISC was reliable and valid in the assessment of these disorders.
Conclusion and Future Directions
This is the first study to examine alternative behavioral genetic models of co-occurrence among DSM-IV psychiatric disorders in a genetically informative sample of adolescents. We found support for a model positing two latent internalizing and externalizing factors explaining interrelationships among six psychiatric disorders in adolescents. Latent internalizing and externalizing disorder factors were moderately heritable and were influenced by common genetic and nonshared environmental factors.
suggested that comorbidity in psychiatric disorders results from common, underlying core psychopathological processes and argued for “focusing research on these core processes themselves, rather than on their varied manifestations as separate disorders” (p. 921). The present study’s results (i.e., the higher heritabilities for the latent internalizing and externalizing factors than for individual disorders and the relative lack of familial disorder-specific influences) support Krueger’s statements. Personality and temperament may underlie some of these core processes. Varying levels of constructs such as negative emotionality, neuroticism, effortful control, and low fear response (e.g., Krueger et al., 2001
; Lahey & Waldman, 2003
; Lilienfeld, 2003
; Nigg, 2006
; Muris & Ollendick, 2005
) may confer particular risk for development of psychopathology in adolescents. For example, Lahey and Waldman suggest that negative emotionality is a risk factor common to internalizing and externalizing disorders, whereas daring is a risk factor specific to externalizing disorders. Similarly, Wolff and Ollendick’s (2006)
developmental model of comorbid conduct problems and depression suggest a set of common risk factors such as negative emotionality and unique risk factors such as undercontrol of emotions for conduct problems and negative self-concept for depression.
These findings have significant implications for ongoing research examining developmental risk factors contributing to adolescent liability for internalizing, externalizing, and co-occurring psychopathology. First, research efforts must identify specific genes or combinations of genes that place adolescents at risk for the development of one or more psychiatric disorders. Simultaneous research efforts must continue to identify temperament and personality dimensions that predispose particular individuals to psychopathology. Finally, environmental risk factors common and specific to internalizing and externalizing disorders must be identified. Greater understanding of genetic and environmental influences that increase the risk of onset of psychopathology in adolescents may inform the development of effective intervention and prevention efforts (Jaffee & Price, 2007
; Knapp & Mastergeorge, 2009