Introduction

Children with conduct problems are one of the most frequently referred groups to mental health clinics and a tremendous financial burden to the social system [1, 2]. This is consistent with the negative short- and long-term impacts of conduct problems in childhood [3–5]. Children with conduct problems are prone to co-occurring mental health concerns, including anxiety, mood difficulties, and adjustment difficulties [6, 7]. Moreover, social impairments including aggression and peer problems are more prevalent in children with conduct problems compared to typically developing peers [8, 9]. Approaches to best characterize the specific clinical and therapeutic needs of these children are necessary to inform implementation of best practice treatments.

Children referred to mental health clinics because of conduct problems demonstrate numerous and pervasive social and emotional problems [10, 11]. As a result, characterization of these children’s deficits, in order to select a course of treatment to meet their therapeutic needs, is a major clinical concern. Of recent interest is the identification of specific childhood characteristics that influence the severity of behavior and conduct problems [7, 12, 13]. Frick and colleagues described two childhood pathways for the emergence of conduct problems [4, 14]. The first pathway of “callous-disruptive” children includes those with low levels of anxiety, apparent guilt, remorse, and empathy. The second pathway of “emotionally-disruptive” children show elevated emotional difficulties, behavioral reactivity and temper outbursts. These two pathways have important implications for the social and behavioral outcomes of these children [15, 16] and have potential utility for characterization of the dimensions of psychopathology to target in treatment.

Children who show “callous-disruptive” behaviors demonstrate a constellation of characteristics classified as callous and unemotional [17, 18]. Callous–unemotional (CU) traits are common in clinic-referred children with conduct problems, with prevalence estimates of 30–50 % [19]. Research over the past two decades has documented negative outcomes associated with callous-disruptive children [14, 20–22]. Children with high levels of CU traits appear to lack empathy or remorse in social contexts and appear to react in a seemingly less caring manner to peers’ distress compared to children without CU traits [14, 23, 24]. Importantly, these traits are associated with elevated severities of conduct problems and aggression [14, 25, 26]. Children with CU traits show higher rates of proactive (i.e., planned) and reactive (i.e., retaliatory) aggression [21, 27, 28], which are associated with behavioral impairment and social problems [29, 30]. Moreover, CU traits emerge early in childhood and remain relatively stable with development [4, 18]. As such, CU traits have a pervasive negative impact on children’s social and emotional development [7, 25, 31].

The last two decades have seen tremendous growth in research on CU traits. However, recent comprehensive reviews attest to the need to investigate the behavioral and emotional processes associated with CU traits that influence the development of conduct problems in children [17, 32]. Doing so may provide avenues for novel assessment and treatment for children with these difficulties. Although much research has used variable-centered methodology to compare the functional and behavioral impairments among groups of children and adolescents with and without conduct problems and CU traits, few studies have used analytic approaches that account for the co-occurrence of CU traits with other dimensions of psychopathology in youth with conduct problems [28, 32]. These person-centered approaches are extremely important because unlike variable-centered approaches they allow categorization based on the constellation of risk factors at the level of the person. The person-centered approach is especially relevant for application with children with disruptive behaviors who have varying levels of conduct problems, CU traits, and emotional and behavioral difficulties. The approach is “bottom-up” and, as opposed to variable-centered approaches, does not assume that the same processes apply to all individuals. As such, person-centered methodology is relevant to applied clinical research.

Recent research using person-centered methodology with a community sample of adolescents showed that the severities of CU traits and conduct problems interact to produce five risk profiles [28]. These include low risk (48.7 %), average risk (33.8 %), high conduct problems and CU traits (5.4 %), high conduct problems and low CU traits (5.2 %), and low conduct problems and high CU traits (6.9 %). Findings showed that adolescents clustered into profiles with high CU traits showed more persistent, severe, and aggressive patterns of antisocial behavior [28]. Furthermore, Kahn and colleagues used person-centered methodology to identify profiles based on CU traits, anxiety, and a history of trauma in a sample of clinic-referred adolescents. This analysis yielded three profiles; a high CU traits and low anxiety/trauma group, a high CU with high anxiety/trauma group, and a low CU traits with high anxiety/trauma group. These groups differed in their levels of impulsivity, aggression, and externalizing behavior, with the high CU traits and high anxiety group showing the most severe problems [33]. These studies highlight the utility of the person-centered analytic approach; however, many important questions remain unanswered.

First, it is not clear whether the association among CU traits and social and behavioral functioning varies with age. The influence of CU traits on the severity of social and behavioral problems may be different in children than in adolescents. The present study used a childhood sample in order to uncover potential differences.

Second, determining the interaction among CU traits and emotional and behavioral difficulties in children with disruptive behaviors using person-centered methodology may help to clarify a greater spectrum of psychopathology associated with children’s presenting problems. This would build upon previous findings such as those of Kahn and colleagues (2013). From an intervention perspective this work is extremely important given that children with conduct problems with non-normative CU traits show diminished response to typical treatments compared to children with conduct problems without CU traits [19, 22, 34]. However, studies suggest that children with CU traits may benefit most from specialized combinations of intensive behavioral and pharmacological treatments or treatments that are targeted to their emotional and behavioral profiles [19, 34–36]. Evidently children’s levels of CU traits and emotional difficulties may alter their profile of psychosocial needs that contribute to their conduct problems, emphasizing the importance of considering CU traits along with emotional and behavioral indicators when assessing children’s clinical and therapeutic needs.

Finally, in addition to CU traits, “emotionally disruptive” children with conduct problems show elevated emotional dysregulation and a different pattern of social and behavioral risk compared to those children with conduct problems and CU traits [17, 37]. As such, a number of community-based studies have identified emotional processes associated with elevated levels of conduct problems [4, 20, 38]. Emotional dysregulation is related to disruptive behaviors, aggression and clinical diagnoses of oppositional defiant (OD) disorder and conduct disorder [20, 38, 39]. Children who have difficulty managing their emotions demonstrate behavioral problems in situations that are uncomfortable, unpredictable or lack adequate structure [15, 40]. Although behaviors are manifested as problems with conduct, these may partly result from dysregulated emotional processes [41–43]. As such, knowledge of emotional-functioning in children with disruptive behavior (with or without CU traits) points to possible therapeutic components to include in clinical assessment batteries and treatment programs.

Given the aforementioned evidence that highlights the importance of CU traits and emotional functioning for the emergence of conduct problems, it follows that identifying profiles that include CU traits and emotional functioning indicators may capture children’s specific clinical needs. This approach is consistent with recent expert calls by the National Institutes of Mental Health (NIMH) that emphasize characterizing the underlying processes that contribute to mental disorders using a dimensional system [44]. Knowledge of clinical profiles that range in severity of CU traits, emotional difficulties, and conduct problems may inform tailored approaches to assessment and treatment. Additionally, identification of these profiles may inform clinical-trials research and development of treatments to best match the needs of these exceptionally challenged children. To date, no study has determined whether profiles emerge based on CU traits, emotional difficulties and conduct problems in children referred to a mental health clinic because of disruptive behavior. Furthermore, no study has tested associations between identified profiles and children’s behavioral and social impairment.

This study used a person-centered analytic approach to identify clusters (i.e., profiles) of children referred to a mental health clinic because of disruptive behaviors based on the severity of CU traits, emotional difficulties (i.e., internalizing difficulties such as worrying, low mood, etc.), and conduct problems. Guided by Kahn et al. 2013 and Fanti et al. 2013 in which three and five profiles were identified with adolescents with disruptive behaviors, it was hypothesized that up to five clusters would be identified in the present study with a sample of children [6, 28, 45, 46]. Moreover, consistent with Kahn et al. [33] it was expected that children whose profiles included the most elevated CU traits would show the most severe problems, whereas children whose profiles included elevated emotional difficulties without elevated CU traits would show less severe problems.

Method

Participants and Procedure

Participants were 166 children and their parents referred because of disruptive behavior to a children’s mental health program in an urban mental health hospital in Canada. The specialized service for children with behavioral and related difficulties receives referrals from parents, schools, and physicians from a large metropolitan area. The program offers comprehensive multi-disciplinary assessment and treatment services. Parents of children aged 6–12 years were invited to participate in the study at assessment. Twenty-nine parents did not consent to the study and were provided clinical services as usual (assent rate of 85 %). If consent was received, parents were invited to complete study related measures. All procedures were approved by the institutional research ethics board.

Children ranged in age from 6.01 to 12.8 years (M = 8.64, SD = 1.72), with 132 males and 34 females (see Table 1). As is shown in Table 1, and typical of a clinical sample, children showed high levels of Oppositional/Defiant (O/D) and Inattentive/Overactive behavior. Parents identified primarily as Caucasian origins and reported level of education as follows: “did not complete high school” (.9 %), “completed high school” (10.3 %), “some college/university” (16.8 %), and “completed university” (72.0 %).

Table 1 Participant characteristics
Full size table

Measures

Strengths and Difficulties Questionnaire (SDQ)

Conduct problems and emotional difficulties were assessed using independent scales on the Strengths and Difficulties Questionnaire (SDQ; [47, 48]. The SDQ is a brief parent or teacher completed screener which enquires about 25 attributes that are evenly divided among five behavioral dimensions; prosocial behaviors, emotional difficulties, conduct problems, hyperactivity-inattention, and peer problems. Subscales do not overlap, and each produces a total score. The conduct problems scale includes questions such as “often lies or cheats” and “steals from home, school or elsewhere.” The emotional difficulties scale includes items such as “many worries, often seems worried” and “often unhappy, downhearted or tearful.” Each item is rated on a 3-point Likert scale ranging from 0 (not true), 1 (somewhat true) to 2 (certainly true). The SDQ shows strong psychometric properties [49]. Internal consistency of the conduct problems and emotional difficulties subscales in the present study were .63 and .75, respectively.

Callous–Unemotional Traits: Brief Measure

The CU scale used is a three item parent-report measure derived from previous research [50, 51]. Items include (1) appears to lack remorse, (2) seems to enjoy being mean, and (3) is cold or uncaring. These items were embedded within a broader measure to assess behavioral dysfunction. Items are rated on a 4-point Likert scale: 0 (not at all), 1 (just a little), 2 (pretty much), 3 (very much). Items show strong internal consistency in this sample (α = .80).

Aggression Rating Scale (ARS)

A six-item aggression rating scale was used to measure reactive and proactive aggression (PA) [52]. The scale was included within a broader measure to assess behavioral dysfunction. Items were rated on a 4-point Likert scale: 0 (not at all), 1(just a little), 2 (pretty much), 3 (very much). Reactive aggression (RA) items include (1) when teased, strikes back, (2) blames others in fights, (3) overreacts angrily to accidents. Items that measure PA include (1) uses physical force to dominate, (2) gets others to gang up on peers, (3) threatens and bullies others. Items show high internal consistency in this study (overall α = .81, reactive α = .77, proactive α = .67) and past research with these items show similar values (proactive α = .87, reactive α = .86) [52, 53].

Impairment Rating Scale (IRS)

The IRS measures the child’s current functioning and need for treatment in several developmentally important areas [54]. Parent’s respond to visual-analogue scales that are scored using a 0 (no problems/no need for treatment) to 6 (severe problems/definitely needs treatment) metric. Alphas are not reported for the IRS because each item is scored separately, but the reliability and validity of the IRS have been supported in several samples. For example, Fabiano et al. (2006) reported criterion validity correlations ranging from .44 to .80, 1-year test–retest reliability correlations from .40 to .67 and inter-rater (parent and teacher) reliability correlations ranging from .47 to .64. The overall impairment rating was used in the present study.

IOWA Conners Rating Scale (IOWA)

The IOWA is a brief measure of child behavior. The scale examines inattentive-impulsive-overactive (IO) and OD domains. Independent and non-overlapping scales derived from responses include the five-item IOWA Inattention/Overactivity (I/O) subscale and five-item O/D subscale. Items that make up the I/O scale include “Fidgeting”, “Hums and makes other odd noises”, “Excited, impulsive”, “Inattentive, easily distracted”, and “Fails to finish things he or she starts (short attention span)”. The O/D scale includes, “Quarrelsome”, “Acts ‘smart’”, “Temper outburst, behavior explosive and unpredictable”, “Defiant”, and “Uncooperative”. Items are scored on a four-point Likert scale: 0 (Not at All), 1 (Just a Little), 2 (Pretty Much), 3 (Very Much). The psychometric properties of the IOWA have been demonstrated [55] and in this study showed high internal consistency (IO α = .78 and OD α = .84). The IO and OD scales were used in this study to provide additional description of the level of severity of behavioral difficulties of this clinically-referred sample of children.

Analysis Plan

Hierarchical cluster analysis was performed using SPSS v18. This form of cluster analysis involves successive steps to identify clusters and was chosen based on previous studies with similar designs and suitability for small to moderate sample sizes [56]. Variables entered were mean CU traits score, mean Emotional Difficulties subscale score, and mean Conduct Problems subscale score. Correlations were computed among all cluster variables. The assumptions of hierarchical cluster analysis were investigated (e.g., normality of variables). Due to differing numbers of items per variable, mean scores were used for all continuous variables.

Following recommended guidelines for hierarchical cluster analysis, the Squared Euclidian Distance was applied as the distance metric [57], while Ward’s Method was the algorithm used to combine cases. This approach has been shown to be robust [58]. The cluster analytic approach comprised two phases. In the first phase a hierarchical cluster analysis was performed and the statistical output visually appraised (i.e., dendogram, scree plot) to determine the number of clusters to be retained. Second, Hoeve and colleague’s methodology (2008) was followed for cluster validation. A k-means cluster analysis was computed to derive cluster solutions. The number of clusters specified for the k-means cluster analyses was based on the initial hierarchical cluster analysis. Kappa values were calculated to assess cluster membership agreement among the k-means and hierarchical cluster solutions. The final cluster solution was selected based on kappa values and theoretical interpretation.

The stability of the selected cluster solution was examined in the second phase in three ways. First, the sample was randomly split into half and hierarchical analyses run on each sample. The resulting cluster means from each half were compared using ANOVA in order to investigate the consistency of the cluster solutions across halves [59]. The process of splitting the sample in half was repeated five times. Second, the stability of clusters was further established by inputting centroids from the hierarchical solution into a k-means cluster procedure [60]. The extent to which the final cluster solution from this k-means cluster analyses is consistent with that generated in phase one is evidence of the robustness of the retained cluster solution [60]. Last, the external validity of clusters was examined. ANOVAs were computed to compare clusters on outcome variables (RA, PA and behavioral impairment).

Results

Descriptive Statistics

Correlations

Mean CU traits, Emotional Difficulties and Conduct Problems were correlated (see Table 2). As is shown, low to moderate correlations were found among cluster variables. Furthermore, with the exception of PA and emotional difficulties, all cluster variables were significantly correlated with outcome variables [proactive and RA and overall behavioral impairment (OBI)].

Table 2 Correlations between cluster variables
Full size table

Cluster Analysis

Normality of variables was established by examining skewness and kurtosis. As expected given the relatively low frequency of severe CU traits, mean CU traits was positively skewed; skewness = 1.17(.19). As such, a square root transformation was applied to reduce skewness. The square root transformed mean CU traits score was used in further analyses.

Appraising the hierarchical cluster analysis output visually (i.e., dendogram, scree plot) revealed three-, four-, and five-cluster solutions that fit the data. Computation of a k-means cluster analysis showed that the three-, four-, and five-cluster solutions each obtained substantial agreement with the originally derived hierarchical solution (κ = .793, κ = .648, and κ = .715, respectively). Upon further inspection, the five-cluster solution was imbalanced with respect to number of participants in each group (ranging from 8 to 55). Imbalanced groups is not consistent with the use of Ward’s Method for combining cases [60] or ANOVAs for comparing groups on outcome variables (see below). As such, this solution was determined unsatisfactory. Both the three- and four-cluster solutions were equally balanced; however, the four-cluster solution was selected because it was most theoretically and clinically meaningful, as well as consistent with foundational research in the area [4].

Two methods were applied to establish the stability of the four-cluster solution. First, the sample was randomly split into halves and a four-cluster hierarchical analysis run on each sample. The majority of differences among cluster means across halves were not significant. A few differences emerged during these split-halve analyses of the four-cluster solution that were attributed to uneven sample sizes among comparison groups (i.e., group with eight participants). Regardless, consistency among cluster means in the two halves demonstrates stability of the cluster solution [60]. Second, a k-means cluster procedure was computed using centroids calculated from the hierarchical solution [59]. Applying this procedure revealed minimal changes among the initial and final cluster centers, providing further evidence of cluster stability [59]. The cluster mean of each variable for the final solution is reported in Table 3.

Table 3 Means and standard deviations for each cluster group
Full size table

Defining the Clusters

Inspection of cluster centers revealed distinct groups. First, as can be seen in Table 3, cluster one was characterized by CU traits, Emotional Difficulties, and Conduct Problems, each below their respective means. This group of children was classified as the Low cluster (n = 36). Clusters two and three were characterized by symptom severity above the mean in two domains and below the mean in the third. Thus, these were classified as High Emotional/Conduct cluster (i.e., High Emotional Difficulties and High Conduct Problems; n = 34) and High CU/Conduct cluster (i.e., High CU traits and High Conduct Problems; n = 57), respectively. The fourth cluster was defined by symptom severity above the mean in all domains and was classified as the High cluster (n = 39). Post-hoc ANOVA’s were computed to further describe how CU traits, Emotional Difficulties and Conduct Problem severity differed between clusters (see Table 3).

Differences Between Clusters on Outcome Variables

First, clusters were compared on demographic variables—age, gender, ethnicity, medication status, and socioeconomic status (as indexed by parental education level), yielding no significant differences on these variables between clusters. To establish external validity, clusters were compared on RA, PA and OBI (see Table 4). All comparisons reached significance, suggesting important differences between clusters. Post-hoc analyses revealed that the High CU/Conduct and High clusters showed the most severe levels of PA (see Table 4). The High CU/Conduct cluster showed significantly more PA compared to the High Emotional/Conduct and Low clusters. Additionally, the High cluster showed more severe RA compared to the High Emotional/Conduct and Low clusters. The High and High CU/Conduct clusters did not statistically differ on RA severity. The Low cluster showed the least severe reactive and PA.

Table 4 Difference between cluster groups on outcome variables and standardized effect sizes
Full size table

Post-hoc comparisons of OBI ratings showed that the Low cluster had significantly less impairment compared to other clusters. Impairment ratings for the High CU/Conduct, High Emotional/Conduct, and High clusters did not differ significantly.

Discussion

This study tested whether clinic-referred children with disruptive behavior cluster based on severity of CU traits, emotional difficulties, and conduct problems. Results show four reliable and clinically useful profiles that include (1) children below the mean in severity on all domains; Low cluster, (2) children below the mean in severity on CU traits but above on emotional difficulties and conduct problems; High Emotional/Conduct cluster, (3) children below the mean in severity on emotional difficulties but above on CU traits and conduct problems; High CU/Conduct cluster, and (4) children above the mean in severity on all domains; High cluster. Importantly, clusters showed differences on reactive and PA and OBI.

A first novel finding of the present study is the application of person-centered methodology to demonstrate that children referred to a mental health clinic because of disruptive behaviors can be reliably grouped into clinically-relevant profiles. Two profiles of children with conduct problems were identified that comprised children with either high levels of emotional difficulties or CU traits. This finding in a clinical sample of children adds to previous research that describes developmental pathways of conduct problems distinguished by emotionally reactive or CU processes [20]. Further, in the present study the CU/Conduct profile showed higher levels of PA and behavioral difficulties compared to the Emotional/Conduct profile. Consistent with past research, this finding highlights the importance of assessing emotional processes and CU traits to best determine the dimensions of psychopathology that are associated with children’s conduct problems.

Overall, identified clusters differentiated children based on key aspects of social behavior (i.e., reactive and proactive aggression) and functional difficulties (i.e., OBI). The finding that children in the High and High CU/Conduct clusters showed the most elevated levels of aggression adds to a growing body of knowledge [26, 27, 61–63]. Furthermore, consistent with past research, the High CU/Conduct cluster showed greater PA compared to the High Emotional/Conduct and Low clusters [64]. This finding is conceptually important given that PA and CU traits share similarities in that actions perceived by others to be “cold or uncaring” (i.e., CU) may also display power or achieve social goals through aggression (i.e., PA). This finding has important implications for the social and behavioral development of this cluster of children and for interventions to reduce children’s conduct problems and aggressive behaviors.

Also of clinical relevance, clusters differed in slightly different ways on severity of RA. Although the High cluster showed the most severe RA, this was not significantly different than the High CU/Conduct cluster. This similarity may be attributed to high levels of CU traits demonstrated by children in both clusters. Of note, the High cluster showed significantly more severe levels of RA compared to clusters of children with elevated emotional difficulties and conduct problems without elevated CU traits (i.e., High Emotional/Conduct cluster) and children below the mean in all domains (i.e., Low cluster).

The four profiles identified in this study highlight important differences in clinical needs of children referred because of disruptive behavior. Findings indicate that overt disruptive behavior and conduct problems may be best understood when considered along with associated clinical characteristics such as CU traits and emotional difficulties; each of which may qualitatively change the child’s profile of concerns. For example, children with conduct problems and emotional difficulties (i.e., High Emotional/Conduct) showed less severe aggression compared to those with high CU traits. As such, consideration of conduct problems in the context of CU traits and emotional difficulties further describes potential clinical needs and areas for assessment and treatment.

Importantly, this study used a person-centered approach for analyses, which grouped children into clusters according to research-based clinical indicators. When clinical indicators are considered as a continuum, this approach adds to determination of children’s specific clinical needs. Using a person-centered dimensional approach to identifying factors that are most closely associated with disruptive behavior is consistent with NIMH guidelines and much emerging research to clarify the negative outcomes associated with mental and behavioral health disorders in childhood [6, 65, 66]. Further uses of person-centered analytic approaches to study processes that contribute to disruptive behavior in childhood are needed.

Limitations and Considerations

This study included children referred to a specialized mental health clinic because of challenging behaviors. As such, findings from this study are most generalizable to clinical populations of children. Further, this study used a previously developed and internally consistent measure of CU traits. It was necessary to include a brief measure to maximize clinical feasibility (i.e., the measure was included among other measures to assess children’s functioning). Further person-centered research with children that includes more comprehensive measurement of psychopathology that are also consistent with DSM-5 criteria, may be beneficial to highlight additional complexities within clinical profiles [23, 67]. Lastly, this study used one informant per child. Future studies that incorporate multiple informants may be useful to gather additional perspectives on child characteristics and behavior.

Clinical Implications

Imperative within clinical settings is the efficient determination of therapeutic needs to specify suitable treatment. This study provides initial support for screening CU traits and emotional difficulties to guide treatment selection of children with conduct problems. For example, children with high levels of emotional difficulties and conduct problems without pronounced CU traits may be best served by interventions to manage underlying emotional dysregulation. However, children with elevated CU traits and conduct problems may require treatments that target cognitive, behavioral, and emotional processes related to CU traits.

Evidence-based treatments that address children’s cognitive, emotional, and behavioral skills may benefit from modules that specifically target underlying emotional difficulties and cognitions and behaviors associated with CU traits. Although current programs target development of emotional regulation, problem-solving and behavioral skills, few explicitly target CU traits [17, 19, 34, 68, 69].

Additionally, clinical trials research to investigate the impacts of best-practice interventions may be strengthened by comparing treatments based on profiles that include children classified based on their severity of CU traits and emotional difficulties. Some work to test the moderating influence of these domains of psychopathology has begun; however, further research is needed [19, 35, 36]. Identification of profiles of children with conduct problems based on severity of CU traits and emotional difficulties is another step towards considering unique aspects of children’s psychological make-up that influence behavior and development. This multi-component perspective may be essential to tailor intervention approaches to best match the specificity of children’s clinical concerns.

Summary

This study identified profiles of referred children based on the severity of CU traits, emotional difficulties, and conduct problems that may have application for assessment and treatment selection procedures. Person-centered analysis was used to identify four distinct profiles (1) Children low in severity on the three domains, (2) Children high in severity on the three domains, (3) Children high in severity in conduct problems and CU traits with minimal emotional difficulties, and (4) Children high in severity in conduct problems and emotional difficulties with minimal CU traits. Profiles differed in degree of reactive and PA and behavioral impairment. Despite having similar levels of conduct problems, profiles that included children with higher levels of CU traits showed the most PA and behavioral impairment. Findings show that clinic-referred children with disruptive behaviors can be grouped based on these important indicators into profiles that have important implications for assessment and treatment selection.