Numerous community studies [reviewed in (1)] using various versions of the ICD or DSM diagnostic systems and structured psychiatric interviews have documented the ubiquity of comorbidity among child and adolescent disorders [e.g., (2–10)]. These studies show that conduct disorder (CD), oppositional disorder (ODD), attention-deficit/hyperactivity disorder (ADHD), depressive disorders, and anxiety disorders co-occur more often than would be expected from their prevalences in the general population. The vast majority of studies to date have concentrated on pairwise, or occasionally three-way, diagnostic associations. Little is known, however, about which of these links persist after accounting for other comorbidities.

In a simplified scenario with just three disorders – anxiety, conduct, and depressive – all of which are significantly associated, it is possible that one of the three associations between disorders is merely a product of the other two: CD may be correlated with anxiety disorders only because both CD and anxiety disorders are correlated with depressive disorders. Thus, in this scenario, the apparent association (or comorbidity) is “indirect” and secondary to the direct associations with depressive disorders (1). This is an example of confounding (11) or of one disorder being a proxy for the other (12). The term “epiphenomenal” comorbidity (1) has been previously suggested, but some have been concerned that it could imply that such comorbidity was not “real.” We use that term “indirect” because it carries no such connotation.

The possibility of some comorbidities being indirect is important for a number of reasons. First, the apparent comorbidity of every disorder with all other disorders undermines the validity of the boundaries between diagnostic entities. Differential patterns of comorbidity would serve as evidence of divergent validity for the diagnostic system. Second, studying the “causes” of comorbidity between two disorders may be futile if, in fact, the observed link is indirect. If some combinations were found to be indirect, then the problem of identifying the causes of comorbidity would be simplified, since the number of combinations of interest would be smaller than the number of possible combinations. Third, identifying indirect comorbidity may help to resolve the question of why comorbidity is found between conditions with very disparate behavioral and emotional profiles. For instance, anxious and conduct disordered children lie at opposite ends of the reactivity spectrum (13–15), but the two disorders tend to co-occur.

Findings from the British Mental Health Survey of over 10,000 children ages 5–15 suggested that associations between ADHD and both anxiety and depression, CD and anxiety, and ODD and depression may be examples of indirect comorbidity (10). The authors concluded that “comorbidity is selective, being particularly evident between anxiety and depression, between ADHD and behavioral disorders, and between depression and at least some behavioral disorders (p. 1209).” An earlier analysis of sex-specific comorbidity patterns in the longitudinal Great Smoky Mountains Study (GSMS) study supported indirect associations of ADHD with depression and CD with anxiety, but found direct associations of ADHD and anxiety and ODD with depression (9).

Our aim here is to clarify which disorder associations are statistically independent and which associations attenuate after accounting for other disorders by combining data from three community samples with distinct racial and urban/rural configurations. Furthermore, this study will extend previous work by (1) treating common childhood anxiety disorders separately, rather than in aggregate (16, 17) and (2) grouping depressive disorders and generalized anxiety (18–20). Recent studies suggest ODD and irritability may play a key role in the associations between putative behavioral and emotional disorders (17, 21, 22). As such, we hypothesize that apparent associations between other putative behavioral disorders (CD and ADHD) and anxiety and depression may be indirect associations accounted for by ODD. As such, this study does not apply the hierarchical exclusion rules for diagnosing ODD only in the absence of CD.

A number of prior studies have look at the related issue of how common childhood disorders cluster, typically by using factor analysis (23–25). While these studies tell us which disorders are most strongly associated with one another, they do not clarify which disorders may account for apparent or indirect associations between pairs of disorder. These studies, however, converge with other evidence to support a strong association between depression and generalized anxiety with Lahey et al. (24) finding that “the latent factors of GAD and MDD were correlated near unity (p. 201).” Therefore, our analysis will test comorbidity patterns with the current DSM disorder structure and with also with depressive and generalized anxiety grouped into a single “distress disorders” category.

Prevalence rates of disorder groupings in each sample are provided in prior publications (9, 26, 27). Table 2 displays the unadjusted and adjusted associations between seven common childhood disorders with depressive disorders and generalized anxiety disorder treated separately. Un-shaded cells show associations among putative behavioral disorders (ADHD, CD, and oppositional defiant disorder). Cells shaded in dark gray show associations amongst putative emotional disorders (depressive disorders, generalized anxiety, social phobia, and separation anxiety). Cells shaded in light gray show associations between emotional and behavioral disorders.

All 21 unadjusted pairwise associations were positive and statistically significant and 19 were large effects (OR ≥ 4.0). The adjusted model provides pairwise associations after accounting for comorbidities. Of the significant pairwise associations, nine were no longer statistically associated in the adjusted model and others were attenuated to small/moderate effects. Many of these involved an association between an emotional disorder and a behavioral disorder (e.g., ADHD and depressive disorder or CD with social phobia), but there was also evidence of associations between behavioral or emotional disorders that were attenuated (e.g., GAD and social phobia, ADHD and CD). A number of these attenuated associations involved either depressive disorders or generalized anxiety. The adjusted association between depression and generalized anxiety, however, was very high with an OR of 37.9. This high level of overlap may adversely affect the accuracy of individual associations within a multivariate model with associations with one disorder very high and with the overlapping disorder very low (11).

The higher overlap between depression and generalized anxiety is consistent with prior studies suggesting that the association between generalized anxiety and depression is sufficiently strong to recommend grouping these categories. Therefore, we reran the adjusted model with generalized anxiety and depressive disorders grouped in one category as distress disorders (Table 3). All unadjusted pairwise associations continued to be significant and all but 1 were strong effects (exception was CD and separation anxiety).

Almost all adjusted associations between emotional disorders and behavioral disorders were either non-significant or attenuated to small/moderate effects, thus meeting our criterion for indirect comorbidity. The lone exception was the association of the distress disorders category with ODD which was still strong suggesting a direct association. The association of CD with social phobia was the only instance in which an adjusted association had an OR of less than 1.0. The only instance of indirect comorbidity between like disorders involved associations that were attenuated but still statistically significant (CD and ADHD).

In simple bivariate analyses, comorbidity appeared to be the rule: Each childhood disorder was associated with increased risk for every other disorder, albeit to varying degrees. This pattern of universal comorbidity is precisely the pattern expected based upon a meta-analysis of comorbidity patterns in childhood and adolescence (1). At the same time, it is not at all clear why disorders as different as CD and anxiety disorders should co-occur.

The question for the current study was whether there were certain pairwise associations where the apparent association was the result of other disorders with which both disorders co-occurred. Generalized anxiety and depressive disorders overlapped so much that it made sense to group these disorders as has been suggested by others (18). With this grouping, direct comorbidity was uncommon. All associations association of a behavioral disorder with an emotional disorder met criteria for indirect comorbidity. The lone exception to this pattern was the direct comorbidity of ODD with distress disorders.

One of the longstanding contradictions of comorbidity research has been the persistent evidence that disorders that lie at opposite ends of the reactivity spectrum (e.g., CD and anxiety disorders) commonly co-occur (1). This study suggests that such associations are accounted for by the proxy risk factors of oppositional defiant disorder or the distress disorder category. Factor analytic studies have helped to clarify how disorders tend to cluster (24, 25), and such studies have tended to support DSM-based distinctions while recommending a category that combines major depressive disorder and generalized anxiety. This analysis, however, clarifies the apparent overlap between putative externalizing and internalizing disorders by identifying the role of proxy risk factors (12). Previous studies have either suggested that the association between CD and anxiety is indirect (10) or observed a lack of association (35, 36). These studies and others looking at comorbidities have typically aggregated multiple anxiety disorders and combined CD and ODD into a single group (1). Findings from our study do not support this approach, and patterns of comorbidity differed in part for CD and oppositional defiant disorder. For example, CD was not associated with social phobia in the adjusted models but oppositional defiant disorder continued to have a modest positive association. It may be unnecessary, therefore, to pursue any explanation for the co-occurrence of either CD with select individual anxiety disorders beyond the joint associations of CD and anxiety disorders with oppositional defiant disorder or distress disorder. Others have also found similar evidence of indirect comorbidity between ADHD and depression (10) or simply noted relative weak pairwise associations (4, 7). Our analysis supports this conclusion.

Our results showed difference between ODD and CD in their associations with other disorders (e.g., ADHD, distress disorders). This serves to inform meta-analytic findings suggesting strong associations of CD/ODD with ADHD and depression (1). The meta-analysis was limited by the convention in many studies of combining CD with ODD. It is easy to presume that the putatively “more severe” CD is driving the associations, but this study suggests that this not the case and that ODD may account for much of the associations between ADHD and depression with CD. This contradicts findings in clinical samples that children with ADHD and CD represent a distinct subtype [e.g., (37)] or that ODD is a subsyndromal form of CD (38). This assumption is similarly misguided in looking at prediction from childhood to young adulthood where ODD, not CD, predicts later anxiety and depressive disorders (17). Together, this evidence suggests that there are problems with both the ICD-10 practice of placing CD and ODD together in a single diagnosis of CD, with “ODD” as a subtype (39) and the DSM-IV exclusionary criterion for ODD which only allows the diagnosis if the criteria for CD are not met. Perhaps more importantly, our study emphasizes the relationship between ODD and depression/generalized anxiety as substantially responsible for driving the association between distinct emotional and behavioral disorders.

This, in turn, leads to the question of why the relationship between ODD and depression/generalized anxiety (or distress disorders) should be so strong. A prime suspect must surely be the role of irritability. Factor analyses of the symptoms of ODD reliably yield an irritability factor, and irritable mood is common in depression (and included in the diagnostic criteria for children and adolescents) (21). Perhaps it is irritability that provides the key link between the emotional and disruptive disorders. ODD and depression (or distress disorders) sit in the middle of the web of associations between emotional and behavioral disorders. This observation informed the inclusion of Disruptive Mood Dysregulation Disorder in DSM 5.0 to capture children with persistent negative mood coupled with frequent behavioral outbursts (40). This disorder seems to do a good job of accounting for severe, non-episodic irritability (41), but still leaves questions about the more general of irritability in common childhood psychiatric disorders.

Despite findings of indirect comorbidity, it is important to be clear that the unadjusted associations indicate that these disorders do indeed often co-occur. That apparent co-occurrence, however, is often explained by comorbidity of both disorders with an intermediary disorder(s). This observation affects a significant literature that has attempted to understand within-disorder heterogeneity by parsing different comorbid subgroups [e.g., (42, 43)]. This study suggests that the number of such comorbid subgroups that merit study is more limited than that of all possible combinations. This is not a trivial concern as comorbidity research for the past 20 years has consistently reported positive, significant associations between CD and anxiety (1).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.