Replicating latent structural models of psychopathology: Reply to Kotov, Ruggero, Krueger, Watson, and Zimmerman (2018)

Abstract

In this issue, we reported that an established dimensional model of emotional and substance use disorders fit poorly to diagnostic correlations in a large sample of outpatients diagnosed with anxiety and depressive disorders (Conway & Brown, 2018). Kotov, Ruggero, Krueger, Watson, and Zimmerman's (2018) commentary on our article raised the possibility that hierarchical exclusion rules embedded in the diagnostic and statistical manual of mental disorders (DSM) were to blame for this nonreplication. They argued that these rules-many of them rooted in outmoded etiological theory or clinical heuristics-are misaligned with the natural structure of mental illness. For example, although depressive and (hypo)manic signs and symptoms frequently co-occur, DSM prohibits comorbidity among major depressive and bipolar disorder diagnoses. In our study, diagnostic assessments followed these DSM conventions, whereas other studies disregarded them (e.g., Kotov et al., 2011), possibly leading to cross-study variation in the joint distribution of diagnoses.Kotov, Ruggero, Krueger, Watson, and Zimmerman (2018) also underscored the differences across studies in how assessors parse the clinical picture at the symptom level. In studies of clinical samples, highly trained interviewers following DSM diagnostic guidelines often assign symptoms to one disorder only, instead of "double counting" the symptoms to more than one condition (i.e., the DSM requirement that a diagnosis should not be assigned if its features are better accounted for another disorder). Some research projects, in contrast, permit symptoms to count toward multiple diagnoses. This may particularly be the case for community-based studies that rely on lay interviewers and highly structured (fully standardized) interviews that leave little room for clinical judgment. It is important to note, however, that in clinical samples, such as ours, symptoms can be and are often double counted as indicated by careful functional analysis during the interview process. For instance, situational fear and avoidance of air travel will count toward separate diagnoses of panic disorder/agoraphobia and specific phobia if there is fear of both panic attacks and crashing.We think these are sensible and important points. They underline how important it is for investigators to publish details of their assess-ment routines, including the logic behind ultimate diagnostic decisions. The research community should be able to evaluate-through quantitative or qualitative review-whether these study attributes affect the structural integrity of internalizing, externalizing, and other latent dimensions of psychopathology.Nevertheless, there was other evidence in our diagnostic correlation matrix that differential diagnosis issues do not completely explain our results (see Conway & Brown, 2018, table 1). For instance, the correlation between major depression and generalized anxiety disorder was the largest in our sample, despite being subject to a DSM exclusion rule that was most frequently applied in our ...