Disruptive Behavior Disorders: The Challenge of Delineating Mechanisms in the Face of Heterogeneity

Disruptive behavior disorders (DBDs) in childhood are classified at the level of behavior. Although current diagnostic categories identify clinically disturbed functioning, they arguably do not identify etiologically delineated groups with distinct risk factor profiles (1, 2)—something that would come as no surprise to mental health practitioners and teachers who work with children and young people with DBDs. Children can be diagnosed with the same disorder with limited symptom overlap (3). Even those with relatively comparable behavioral presentations can differ in the underlying mechanisms and developmental trajectories associated with their DBD (1, 4). Furthermore, not all children who are diagnosed with DBD in childhood will develop into adults with antisocial behavior. Some remain on a stable DBD trajectory, but others remit in their disruptive behaviors (5, 6), and many migrate to different diagnostic categories over development (7, 8).

In light of the heterogeneity in presentation of DBDs, their underlying mechanisms, and their outcomes, how can we best advance our understanding? DBDs have spawned research into genetic, environmental, and neurocognitive risk factors, but we are far from truly integrating our understanding across these different levels of analysis. Here we outline several key interrelated areas that need to be considered if we are to advance a genuinely holistic explanatory model that has the potential to account for multiple developmental trajectories (and indeed variation in treatment approaches and outcomes) in children with DBDs.

Equifinality and Multifinality

The term “equifinality” refers to the notion that a particular outcome or end state can be reached by many potential means (9). In other words, two individuals may display comparable DBD symptoms, but may do so for different underlying reasons. If these individuals are treated with a single intervention, possibly targeting the underlying causal factors important for person A, it is by no means inevitable that person B will be helped.

We can use conduct disorder as an example of the challenge of advancing our understanding of DBDs in the face of equifinal outcomes. Typically, when we try to study mechanisms that are related to conduct disorder, we select individuals on the basis of their behavioral symptoms and try to find an underlying cause or causes for their presentation. There is a clear circularity of argument here. We already know that children and young people who qualify for a conduct disorder diagnosis are not all alike. In fact, they may have very different symptom profiles or they may display the same symptoms for different reasons. Despite this heterogeneity, we use behaviorally defined diagnostic criteria to select individuals for our studies, and then look for causes and mechanisms. If we accept that in all likelihood there are conduct disorders, rather than a conduct disorder, then such an approach is inherently limited.

One way researchers, ourselves included, have attempted to get around this problem has been to use differences in behavior/trait indicators to classify individuals with conduct disorder into subgroups and then investigate these subgroups at different levels of analyses. Although this approach explicitly recognizes heterogeneity within the disorder, it does not resolve the circularity problem. This is because the heterogeneity is still defined at the behavioral level, and we cannot assume that the behavioral indicators we have chosen, or our ability to observe them, are sufficiently accurate or discriminating with respect to underlying causes or mechanisms. We should therefore view work in this tradition as one approach for triangulating the “heterogeneity problem” and reducing the problem space—while continuing to recognize its inherent limitations.

Research into children with conduct disorder with high versus low levels of callous-unemotional traits can be used to more fully illustrate this approach and its limitations (1, 10, 11). DSM-5 (12) included measurement of callous-unemotional traits (termed “limited prosocial emotions”) as a specifier for children with conduct disorder, based on findings from different methodological traditions indicating divergence between children with conduct disorder and high callous-unemotional traits and their peers with conduct disorder and low callous-unemotional traits (1, 10, 11). For example, twin data from our group have indicated that conduct problems may be more heritable in the presence of high callous-unemotional traits, while reflecting predominantly environmental risk in the presence of low callous-unemotional traits (13, 14). Neuroimaging and experimental data across different research groups have suggested that children with conduct disorder and high callous-unemotional traits show diminished neural, psychophysiological, and behavioral responses to other people’s pain, fear, and laughter (1, 15, 16), while those with conduct disorder and low callous-unemotional traits may have exaggerated neural and behavioral responses to threat (1, 17, 18). These findings are consistent with the view that there are distinct subgroups of children presenting with conduct disorder that can be differentiated at multiple levels of analysis and may have different underlying vulnerabilities for their disruptive behavior. This gives some cause for optimism for the use of subgrouping as one approach for triangulating the heterogeneity problem.

However, as noted above, even with this approach it is not possible to avoid the circularity that stems from basing grouping on behavioral criteria or traits that are ultimately inferred from behavior. In recent years, data have accumulated indicating that not all children presenting with conduct disorder and high callous-unemotional traits are the same. For some children with conduct disorder, high callous-unemotional traits are accompanied by high levels of anxiety, a history of trauma, and a different cognitive/affective presentation compared with other children, in whom high callous-unemotional traits are associated with an absence of anxiety symptoms and prior trauma histories (19–21). In other words, children may present with conduct disorder and high callous-unemotional traits as a result of different etiological pathways.

The recognition of the problems of relying on behavioral diagnoses has contributed to a range of initiatives, including the National Institute of Mental Health’s Research Domain Criteria (RDoC) (22), which encourages researchers to focus on basic dimensions of functioning (e.g., threat reactivity, variously specified) rather than diagnostic criteria. The aim of such initiatives is to elucidate how individual differences in a particular domain or construct, such as negative valence representation probed by paradigms targeting the functioning of threat circuitry, may increase the risk (for example) of developing the aggressive symptoms that we know characterize some children with DBDs.

The research focus advocated by the RDoC initiative is important, but has its own set of challenges. Most practitioners recognize that particular shared risk indicators can characterize two individuals who in fact qualify for different diagnoses or may even be present in some individuals who appear largely free from mental health problems. Genetically informative studies indicate that a substantial amount of genetic risk is common across different DBDs, as well as DBDs and psychopathological outcomes more broadly (23, 24). This is in line with the notion of there being general vulnerability to psychopathology that is likely to explain why most risk indicators for DBDs are transdiagnostic (25). How particular DBDs that share risk indicators with other disorders canalize over development, or why some individuals with initial DBD symptoms remit over time, is likely to depend on the degree of general psychopathology risk, as well as unique genetic and environmental risk and protective factors, over and above those related to general psychopathology. Cicchetti and Rogosch (9) described the phenomenon of common risk indicators being associated with different outcomes as “multifinality.” They proposed the following: “The principle of multifinality (Wilden, 1980) suggests that any one component may function differently depending on the organization of the system in which it operates” (9, pp. 597–598). Individuals who share particular characteristics—for example, higher reactivity of the amygdala to threat—often differ in their genetic and environmental endowments in multiple ways. This accounts for individual differences in the ability to regulate the amygdala response and the range of likely choices and behaviors available to a particular person. In other words, developmental outcomes for people with similar amygdala responsivity in childhood may vary considerably, with only a subset of those with exaggerated amygdala response to threat developing a persistent DBD.

Both equifinality and multifinality need to be considered as we seek to understand development of DBDs. Cicchetti and Rogosch (9, pp. 599) wrote the following:

In the nearly 25 years since Cicchetti and Rogosch’s article was published, we have not made as much progress in entertaining more complex and varied approaches to conceptualizing and investigating DBDs as we might have hoped for. Below we outline a set of issues that need meaningful attention if we are to realize this ambition. Given the constraints of space, we have chosen to focus on those issues that appear to us most salient at the current time. Of course there are other factors that are important and require further consideration as the field progresses.

Challenges Around Finding Risk Genes

Despite robust and well-replicated findings from twin and adoption studies indicating substantial heritability of DBDs (23, 26, 27), success in molecular genetic research has been modest (28–30). In other words, we have not yet found a substantive proportion of genes that increase the risk of developing DBDs or identified how these might vary for different patterns of DBD development. This limits both our ability to understand how genetic risk for DBDs operates and constrains our capacity for effective multilevel study of DBDs. There is no doubt that much larger samples will be needed to find common genetic variants with small effect size that probabilistically increase risk of DBDs (30). Studies of other phenotypes indicate that the ability to find common variants rises exponentially as sample sizes increase from tens of thousands to hundreds of thousands or millions of participants (31); to date, the largest studies of DBDs have involved samples in the thousands (e.g., 28). Studies should also be conducted to search for rare variants associated with DBD risk. However, the success of molecular genetic research will not just depend on large samples or the latest analytic technologies, although both are important.

If molecular genetic studies focus on generic measures of disruptive behavior and do not pay attention to the particular presentation or age of the participants, we are biasing our studies for finding those variants that are responsible for what is common across subgroups and ages. One might argue that if we want to find those genes, we should focus on screening for variants that increase risk for general psychopathology, as twin data indicate that these will also increase risk of DBDs (24). We should additionally consider that partly divergent risk genetic factors may be important when we seek to understand different DBD developmental profiles (high callous-unemotional traits, low callous-unemotional traits, persistent, remitting, increasing, heterotypically continuous, comorbid with anxiety, etc.) (5, 13, 27).

Recent findings in relation to “genetic innovation” serve as a sobering illustration of this issue. Genetic innovation refers to novel heritable effects that become apparent over the course of development with genes that were previously inactive coming “online.” It has been shown that those genetic factors that increase early risk of developing conduct problems are largely independent of those genetic factors that explain subsequent change in these behaviors (32). A comparable pattern is seen in relation to callous-unemotional traits (33). We have speculated that genetic risk factors influencing the baseline levels of conduct problems and callous-unemotional traits may be related to the temperamental makeup of the child, including those genetic variants that influence emotional reactivity or drive social affiliation and resonating with other people (15, 34). A second set of genetic factors influencing the developmental course of DBDs may relate more specifically to traits and capacities that mature in childhood and adolescence and are likely to have an impact on expression of behavior and trait profile over time. As an example, the capacity to engage in complex, goal‐oriented thinking substantially increases across childhood and adolescence (35), as does sensitivity to what other people think (36). Both are thought to be linked to changes in adolescent brain structure and function (35, 36). These processes may be important for assessing the best strategies for executing one’s own goals, which may result in less or more adaptive ways of interacting with others. Developmental changes such as these could lead to genuine changes in the appreciation and understanding of others’ emotions, for example, or might reflect masking or unmasking of baseline dispositional traits as superior or inferior (compared with the age group) planning and regulatory capacities emerge.

In conclusion, advances in our understanding of DBDs will be impeded if the granularity of molecular genetic studies does not mirror the granularity of how we operationalize etiology, presentation, and trajectories. This will in turn constrain our capacity for effective multilevel study of different DBD trajectories where molecular genetic risk is successfully integrated into any analytic approach.

Challenges Around Harmonizing and Improving Neuroimaging and Experimental Studies

At present, there is a burgeoning body of functional neuroimaging and experimental research aimed at elucidating the information-processing patterns associated with DBDs (1, 37). The extant studies have documented, for example, atypical affect perception, empathy, affect regulation, and decision making in DBD populations and suggest that the specific patterns may differ by subgroup (1). While some of these findings have been replicated, in many instances it is difficult to interpret the not infrequent inconsistencies reported in the literature. There is no doubt that some of these inconsistencies are driven by variation in whom these studies sample, and the field needs more consistency in participant selection practices. We also want to highlight three further challenges if we want to build a stronger evidence base of neurocognitive risk for DBDs.

First, task parameters and task demands often vary considerably between studies claiming to assess the same cognitive/affective constructs. Just because two investigators both state that they assess, for example, emotion regulation, this does not mean that they are actually quantifying the same information-processing parameters. We need more precision in definitions of constructs and an agreed-upon set of measures for quantifying those constructs. A related issue concerns the degree of inference afforded by the choice of paradigm. For example, if a task conflates a number of cognitive processes without parsing them, it is not possible to use it as unequivocal evidence of atypical processing in a single domain. It would substantially advance the field to agree on a core set of paradigms that more precisely and reliably measure a set of clearly defined candidate cognitive/affective functions.

Second, we need considerably more work on psychometric validation of functional neuroimaging and experimental measures if we want to advance the longitudinal study of DBD development. Unfortunately, the functional neuroimaging and experimental paradigms that are currently available have largely been adopted from cognitive neuroscience and experimental psychology studies that were originally developed to study “species universals.” That is, these paradigms are designed to minimize between-individual variation and to reliably capture effects across all humans or within a specific group. In other words, they are optimized to capture group effects. These paradigms have not, as a rule, been psychometrically validated to sensitively and reliably capture individual differences (38). This currently limits their utility for inclusion in large-scale longitudinal studies of individual differences in developmental trajectories—particularly our ability to relate functional neuroimaging and experimental data to behavioral (including clinical) outcomes.

A third challenge relates to the dearth of work validating paradigms that could be used to assess the same neurocognitive domains across the lifespan. For example, how we might experimentally index a process such as emotion regulation is unlikely to be the same for preschool children as it would be for adolescents. There is an intrinsic challenge here in how we can have confidence that our various measures employed at different ages are indexing the same underlying cognitive processes. Furthermore, as developmental researchers, we must grapple with the reality that such processes themselves will, in almost all instances, evolve and change over time.

Until these challenges are addressed, the most reliable indices of individual differences in neurocognitive development are likely to be indices of brain structure. These do sensitively chart individual differences, as evidenced by the utility of brain structural measures in charting heritable individual differences in brain development (39, 40). Recent studies have shown that longitudinal structural brain phenotypes can be reliably associated with development of behaviors related to DBDs, such as impulsivity (e.g., 41). However, measures of brain structure are naturally limited in the insight they can offer regarding information-processing patterns that may underlie development of particular DBD profiles (or remission thereof).

Challenges Around Embedding the Study of the Brain into the Social Context

One critical shortcoming in our understanding of DBDs has been the failure to systematically consider the complex and reciprocal relationships between the brain/cognition and the social world. Research has either focused on genetic or neurocognitive vulnerability, often conceptualized as “located within the child,” or in social/environmental risk factors, often conceptualized as “external to the child.” However, the complex psychological and behavioral features that characterize DBDs are clearly emergent phenomena that are the product of dynamic interplay between these domains. Brain changes do not mean that vulnerability is simply located in the child. Rather, vulnerability unfolds in a relational context—that is, through the interaction of a child’s social behavior and capacities, and the responses of peers, adults, and systems around them. We need to advance study of the “embedded brain,” which neither denies biology nor adopts a biologically reductionist approach in the study of DBDs. We need to better understand how neurocognitive endowments or adaptations affect specific aspects of social functioning in order to inform approaches to prevention and intervention.

DBDs should not be seen as just an outcome, but rather should be viewed through a developmental lens, whereby atypical behaviors (by children themselves and those around them) shape and maintain social interactions. Five decades ago, Patterson (42) introduced the notion of “coercive cycles” as a model of how disruptive behaviors escalate and are reinforced within the family ecology. His theory described a process of mutual reinforcement, where caregiver behaviors reinforce the child’s disruptive behaviors, which in turn evoke anger and hostility in the caregiver, which then escalate the child’s behaviors (43). Genetically sensitive studies have since demonstrated that many social risk factors associated with DBDs include genetic confounding—providing some insight into sources of individual differences in family social interactions (44). For example, harsh and inconsistent discipline is associated with higher levels of DBDs, but this in part reflects shared genetic vulnerabilities between parents and children (passive gene-environment correlation) and reactions that a child with difficult DBD evokes in parents (evocative gene-environment correlation) (44). Although gene-environment correlation in the context of DBDs has not been studied using neuroimaging or experimental probes, it is not unreasonable to propose that genetic endowments calibrate children’s (and caregivers’) cognitive and affective functioning in ways that constrain subsequent building and maintaining of social relationships (15, 34). For example, difficulty in empathizing with other people’s distress (1) or sharing in their joy (45), as seen in children with conduct disorder and high callous-unemotional traits, may evoke fear, discomfort, and even hostility in child-caregiver interactions. The genetic endowments of the parent may also constrain their ability to respond to a challenging child.

Environmental risk—for example, extreme childhood adversity—can also lead to neurodevelopmental adaptations that may confer latent vulnerability to subsequent development of DBDs (46). In brief, the brain may adapt to an adverse environment; however, these calibrations may mean that a child may be less well equipped to function in more normative environments. For example, children who have experienced maltreatment in the past show heightened reactivity to threat (47, 48). Being alert to potential threat will clearly serve a purpose in an environment that is not safe. However, it may result in threat-reactive aggression in response to perceived threat, even if threat was not intended, causing problems in school, for example. We have argued that this “mismatch”—where brain systems calibrated for an adverse environment function less well in a more normative environment—may lead to mental health vulnerability that is socially mediated—that is, neurocognitive adaptations associated with early adverse environments may have an impact on how a child shapes and experiences the social world in ways that become problematic. For example, altered brain functioning may contribute to the generation of new stressful events (“stress generation”) that in turn contribute to an increased risk of future internalizing symptoms and, plausibly, externalizing symptoms (49). For example, if conflict escalates for a child at school, and this ultimately leads to exclusion, this would likely result in significant stress for the child and their family in addition to that which the child may have experienced earlier in life. Equally, brain adaptations associated with exposure to childhood adversity may lead to what has been termed “social thinning” (50). Here, the range and quality of a child’s social interactions are reduced over time (50, 51), attenuating the protective effects associated with social support.

It is regrettable that there is such a dearth of research into how different cognitive/affective biases may feed into generating and maintaining atypical social interactions, and how the social interactions in turn calibrate future brain/cognitive development in children with DBDs. In order to achieve longitudinal, multilevel study of the “embedded brain,” we will need advances in our understanding of the etiology of different DBD trajectories and refinement of neuroimaging and experimental study protocols, as outlined in the previous sections. We will also need improvements in sensitive measurement of social functioning over development (via, for example, observational, experience sampling, and social network measures) so that we can examine how particular cognitive and affective biases shape social experiences at different developmental stages, and how those social experiences in turn shape brain development in ways that either increase risk of, or protect against, developing a DBD.

Improving our ability to study the “embedded brain” is a challenging but important task. Social learning principles used in many therapeutic programs of DBDs emphasize the ways in which adult behavior can have an impact on the child outcome. However, children also play a key role in shaping the responses of adults around them, often evoking particularly negative or conflictual reactions. Furthermore, caregivers may share some of the vulnerabilities of their child, augmenting the challenge of delivering a systemic intervention. Helping caregivers and teachers understand the child’s cognitive and affective biases may help them reframe the child’s behavior and change how they consequently experience and respond to that behavior. Therapeutic programs also target child cognitions, aiming to shift the way in which a child with DBD processes information (e.g., emotion regulation or empathy training). A more precise understanding of the neurocognitive processes that contribute to a particular child’s disruptive behavior could help clinical formulation. If the child displays aggressive responses in the face of perceived threat, then shifting how affective cues are perceived (52) or emotion regulation training (53) may be appropriate and feed into reduction of aggression and subsequent improvement in social functioning. If the child displays diminished responses to other people’s positive affect (45), it may be possible to pair positive affect stimuli with something that the child finds rewarding. This could, over time, make the child more receptive to adult positive affect and feedback and improve the quality of social interactions. In both cases, the breakdown in social relationships or the outcome after intervention may be very similar, but the reasons for the breakdown or who benefits from which approach is not.

Conclusions

There are structural barriers inherent in the scientific world, where we all specialize in particular methods, schools of thought, or disorders. Increasingly, data are accumulating that challenge our traditional notions of diagnoses and shine a light onto the complexities of developmental risk and resilience. Causal pathways to DBDs, even within the same diagnostic category, are varied. If we are to advance the study of DBDs, we do not just need bigger samples or larger quantities of data. We need a more systematic approach to uncover underlying mechanisms, and we need to creatively address the current reliance on behavior as the primary organizing framework. The field needs to work together to generate an integrated conceptual framework that articulates the relationships between levels of explanation, but also across development and across domains of functioning. Genetic advances in our understanding of DBDs will be impeded if they do not occur alongside improved granularity in how we operationalize presentation and developmental trajectories. Improvement in measurement is essential, particularly in relation to neurocognitive and social risk factor indices. Unless we achieve this goal, we will not be able to place what we are learning about brain function in a systemic, multilevel context, where brain changes are understood not simply as outcomes of genetic and environmental influences, but also as factors that reciprocally influence future social environments in ways that are important in understanding risk and resilience. The ultimate goal of advancing the field in this way is to improve our ability to develop effective preventive approaches and more targeted—and therefore more effective—approaches to intervention.