Bardos, Achilles N.
There is growing concern among both researchers and practitioners that mainstream psychopathological classification schemes inadequately meet the needs of individuals seeking treatment for mental illness. Recent efforts in childhood psychopathology have sought to derive new diagnostic subgroups that are less susceptible to the heterogeneity and comorbidity issues that have been reported with categories specified in more recent versions of the diagnostic and statistical manual of mental illness. These transdiagnostic subgroups are generated by using unsupervised machine-learning algorithms that separate cases into homogenous groups with information from multiple continuous indicators. This dissertation seeks to expand on this work by exploring natural groups of children and adolescents that emerge from the normative samples of a rating scale designed to measure the following domains: conduct, negative affect, cognitive/attention, social, and academic functioning. This project uniquely contributes to the previous literature by comparing solutions across teacher, parent, and child raters, by including two adaptive domains (social and academic functioning) and by exploring differences in item-level versus subscale-level methods for profile estimation. Profile membership was also contrasted with categorical diagnostic categories. Results indicated solutions that ranged from four to six profiles across samples. Out of 10 separate analyses across rater groups, developmental levels, and indicator types, 10 qualitatively different latent profiles were identified, six of which fell across a broad spectrum of severity ranging from an extreme level of psychopathology to above average functioning. Two profiles characterized by social dysfunction were identified across all rater groups but not all developmental levels. The remaining two profiles were defined by attention/executive dysfunction but were only present in teacher-rater samples. Thus, latent profiles were replicated across samples to a varying degree, indicating a low level of interrater stability overall. Findings further indicated that, while the item-level approach was helpful in identifying key profile symptoms and yielded broadly similar results, the subscale-approach tended to have higher stability during the enumeration phase and was a more reliable indicator of overall psychopathology. Low associations between profiles and specific categorical diagnoses were found. However, subscale profiles characterized by more extreme maladaptive estimated means were associated with a higher degree of overall psychopathology. Profile demographics are discussed, along with implications for latent profile analysis, the classification of mental health in children and adolescents, treatment and intervention, and psychological assessment. Overall, the results of this study support a broad and transdiagnostic view of child and adolescent behavioral and emotional functioning that continues to show a promising ability to address problems endemic to our current nosological paradigm, improve problematic practices in effectiveness research, and increase the precision and size of treatment effects.
Copyright is held by the author.
Rempe, Gary, "Transdiagnostic Classification of Behavior in Childhood: Profile Analysis and Interrater Stability" (2021). Dissertations. 795.