A Thorough Investigation of the Bifactor Model of Psychopathology in a Representative Birth Cohort: Testing Internal and Predictive Validity to Inform Models of Comorbidity

This study used symptom dimensions reflecting DSM-V internalizing, externalizing, eating disorders, and substance use (SU) and related problems to thoroughly investigate the structure of psychopathology in mid-adolescence (15 and 17 years, N = 1,515, 52% female). Compared to other hierarchical configurations (unidimensional, correlated factors, or higher-order model), a bifactor model of psychopathology wherein all first-order symptom dimensions loaded onto a second-order general psychopathology factor (P factor) and one of three, second-order specific internalizing, externalizing, or SU factors, best captured the structure of the psychopathology in mid-adolescence. This bifactor model was then used to predict several distinct mental health disorders and alcohol use disorder (AUD) at 20 years, via a structural equation model (SEM). The P factor (bifactor model) was associated with all but one outcome (suicidal ideation without an attempt), at 20 years. Controlling for the P factor, there were no additional, positive, temporal cross-associations (i.e., between mental health (mid-adolescence) and AUD at 20 years, or between SU (mid-adolescence) and mental health problems at 20 years). These results are bolstered by findings from a well-fitting correlated factors model. Namely, when mid-adolescent psychopathology was modeled using an adjusted correlated factors model, associations with outcomes at 20 years were largely masked, with no significant partial, temporal cross-associations. Thus, collectively, findings indicate that comorbidity between SU and mental health in youth may be largely attributable to an underlying liability to experience both problems (i.e., P factor). Ultimately, results support targeting the common liability to psychopathology in the prevention of later mental health problems and AUD. General Scientific SummaryHierarchical models of mental health disorders have become increasingly present in research over the past decade; however, uncertainty still exists on the best way to model this hierarchy. The current study suggests that comorbidity between different mental health disorders (e.g., depression, conduct problems) and substance use (e.g., cannabis use) may be more useful in predicting future mental health disorders and alcohol use disorder (at age 20 years), than multiple distinct (but related) disorders.