Latent classes of emotional and behavioural problems in epidemiological and referred samples and their relations to DSM-IV diagnoses

The Child Behaviour CheckList/6–18 (CBCL/6–18) [1] is a parent report form to screen for emotional, behavioural, and social problems in children and adolescents aged 6–18. It is an empirically based questionnaire, and its reliability and validity have been widely demonstrated [1]. The checklist returns a profile of scores on eight syndrome scales and six DSM-oriented scales. Several studies have examined the associations between CBCL/6–18 scales and diagnoses based on DSM criteria [2, 3]. In recent years, however, researchers’ interests have moved toward the identification of recurrent profiles characterized by concurrent elevations on more scales, and among these profiles one that has received particular attention is the Child Behaviour Checklist-Dysregulation Profile (CBCL-DP) [4]. This profile is characterized by co-occurring elevations in attention problems, aggressive, and anxious/depressed scales, and it reflects the condition of emotional and behavioural dysregulation described for the first time by Biederman et al. [5]. The CBCL-DP has demonstrated its utility in terms of diagnosis and outcome prediction [4, 6‐10]. The majority of these studies used an a priori identified profile that sums the T-score of the three scales, or they alternatively used person-centred statistical approaches, such as Latent class analysis (LCA), applying the method exclusively to the items of these identified three scales. Only a small number of studies used a bottom-up approach, which explores the different profiles that emerge within a population while considering all the syndromic scales of the CBCL. This approach can be found in the work of Connell et al. [11] and of Basten et al. [12]. These studies used LCA to identify profiles of psychopathological traits from CBCL/1.5–5 scales in a sample of economically disadvantaged families and in a population-based sample. Despite coming from different starting points, both authors identified a 4-class solution with an externalizing class, an internalizing class, a normative group, and a highly problematic and dysregulated class. Only one study [13] applied this methodology on a sample of older children and adolescents. These authors performed an LCA on CBCL/6-18 syndrome scales of a combined sample of general-population subjects and referred children and they identified six classes of subjects, including a dysregulated class. The main limit of this interesting work, however, is the small sample size. Interestingly, Bonadio et al. [14] in a similar way used Latent Profile Analysis (i.e. LCA with continuous indicators) to identify psychopathological profiles from the Ohio Scales-Problem Severity Scale (OS PS) filled out by youth and parent dyads of a community mental health sample. In this case, using a different instrument, authors identified five classes of subjects, including a high risk class with elevated scores both on internalizing and externalizing symptoms, which very closely resemble the CBCL-DP.

The present study had two major aims. The first aim was to replicate and extend the preliminary results presented by De Caluwé et al. [13] by examining whether it was possible to identify different psychopathological profiles in a very large mixed sample (including both general and referred children and adolescents). To achieve this goal, we performed an LCA, including the full spectrum of the CBCL syndromic scales. The second aim was to evaluate the relationship between identified classes and DSM-IV diagnoses, also taking the presence/absence of comorbidity into consideration.

In the present study, we examined whether different profiles of psychopathology could be identified using a person-centred statistical approach in a large and inclusive sample of Italian children and adolescents combining general-population and referred subjects. Moreover, we investigated the relationship between these profiles and DSM-IV diagnoses as well as the comorbidity of diagnoses.

Latent class analysis performed on CBCL/6–18 syndrome scales for our samples identified 4 classes. One class grouped subjects without a significant risk of elevation on any scales. A second class was characterized by children and adolescents with a high risk of being in the clinical range for Attention problems. The third class grouped subjects with a high risk of elevation on the Anxious/depressed scale. Finally, the last and more compromised class included subjects with an elevated probability of being in the clinical range for all CBCL scales, with the exception of Somatic complaints and Rule-breaking behaviour. These results are very similar to those presented by Connell et al. [11] and by Basten et al. [12]. Considering the differences between the version of the checklist used (1.5–5 vs. 6–18), the overlap between the macro-structure of our classes and the one identified by these authors appears clearly. Furthermore, observing the similarities between the profiles identified in early childhood and those identified in children and adolescents, we might guess that this way of grouping on the basis of psychopathological traits is so strong that it is independent of the developmental stage. It would be of interest to use the same approach on the Adult Behaviour Checklist 18–59 (ABCL/18–59) [32] to check whether these profiles emerge again. Finally, considering that sample features may influence the structure and the prevalence of identified classes [33], our results seem to be reliable as they were obtained from a large and heterogeneous sample, including both referred subjects and general-population subjects. Furthermore, these results are also supported by the observation that the same structure of classes has been obtained by performing separate LCAs on the two sub-samples, which yielded a substantial classification agreement.

The second significant result was the identification of a profile characterized by elevated levels of dysregulation, which was similar to previously published works. De Caluwé et al. [13] identified a higher number of classes (six), but it is worth noting that their ‘No symptoms’ class and ‘CBCL-DP’ class appear very similar to the classes identified in this work. Looking at the profile of dysregulation, we can note that in both cases it was primarily characterized by a very elevated probability of being in the clinical range for the Anxious/depressed, Attention problems, and Aggressive behaviour scales, and secondly in the other scales with the exception of Somatic complaints and, only in our sample, of Rule-breaking behaviour scale. We believe that our study confirms the existence of a dysregulated profile, with limited differences linked to the substantial differences in sample size.

The term Dysregulation Profile (DP) usually refers to the specific profile identified by Althoff [6, 34], which is characterized by the elevation of the three scales, Anxious/depressed, Attention problems, and Aggressive behaviour. In our study, we identified a larger profile of dysregulation that included the three scales used by Althoff as well as the Withdrawn/depressed, Social problems, and Thought problems scales. A possible explanation for this difference is that Althoff et al. identified the DP by performing an LCA specifically on the items of only these three scales, while our analysis was conducted on a broader spectrum of emotional and behavioural problems, including all scales of the CBCL/6–18. Our results were probably more realistic, as they were not based on an a priori hypothesis, but rather on empirical findings. Moreover, these results are not in contrast with those of Althoff. On the contrary, they confirm and extend Althoff’s results, particularly if we consider that our profile had a frequency of 7.82% in our sample. This was consistent with other studies that used LCAs and identified a prevalence of DP ranging from 4 to 8% [4, 6, 34]. Finally, there is no evidence that children with DP have elevations only in these three scales. On the contrary, several studies have claimed that these children show elevations in other scales [7, 35‐37]. In conclusion, we can say that our study provides further evidence of the replicability of DP in different countries, samples, and methodologies [38].

In the second step of our study, we highlighted the diagnostic sensibility of each class. For all diagnostic categories, there was a very low percentage of subjects in the LOW class (between 4 and 20%, approximately). In the INT class, the higher percentages of diagnosis were for Anxiety Disorders (42.8%) and Mood disorders (26.4%). In the ADHD class, about half of the subjects (49.0%) had a diagnosis of Attention-Deficit/Hyperactivity Disorder, but we also found that 23.8% of subjects had an Anxiety Disorder. This result is not surprising considering that Anxiety Disorders in both children and adults are among the disorders that most commonly co-occur with Attention-Deficit/Hyperactivity Disorder [39]. Finally, in the DYS class the highest percentages we found were for Anxiety Disorders (54.3%), Mood Disorders (44.5%), and Attention-Deficit/Hyperactivity Disorder (30.7%), underlining and confirming previous results suggesting a high heterogeneity of diagnoses assigned to individuals with emotional and behavioural dysregulation [8, 34, 40].

Finally, when we analysed how the four classes were able to predict three different diagnostic profiles (absence of diagnosis vs. one diagnosis vs. two or more diagnosis), we found that the LOW class better identified an absence of diagnosis, while all three other classes were able to predict the presence of comorbidity, with the DYS class being the significantly stronger predictor as reported by other studies [41, 42]. The same results were confirmed when analyses were repeated with a more conservative approach (i.e. by incorporating outcomes into the latent class model) to correctly account for uncertainty in class membership.

There are several limitations in this study. First, it is based only on parent reports (CBCL/6–18). An interesting future direction would be to consider the CBCL not only being completed by the parents but also the Youth Self Report (YSR/11–18) [1] completed by the child/adolescent or the Teacher’s Report Form (TRF/6–18) [1] filled out by teachers. Second, the incidence of certain diagnostic outcomes was low, necessitating the collapse of several diagnoses into combined categories. Third, although the results are comparable with those of other studies, the participants were recruited from an Italian sample, which potentially limits the replicability of the results in other countries. Fourth, since we used a cross-sectional design, our findings provide only a static view of the classes. Thus, they do not provide information on the onset, progress, or changes in the class types and their syndromes over time.