Transdiagnostic investigation of the latent structure of psychosis in a nation-wide clinical cohort

Psychotic episodes are marked by substantial misperceptions in reality, bizarre or catatonic behaviors, and disorganized thinking and speech. Psychotic symptoms are most closely associated with schizophrenia, but psychosis can present in other situations, including as part of a mood disorder, as a sequelae of substance use, or in conjunction with other medical conditions like dementia. The purpose of this study was to explore the nosological dimensions of psychosis using complementary person-centered and item-centered analytic approaches. All phenotypic data came from Genomic Psychiatry Cohort participants diagnosed with schizophrenia [SCZ], schizoaffective disorder [SAD], and bipolar disorder I [BD1]. Trained clinicians collected detailed measures of symptoms, clinical characteristics, premorbid functioning using the Diagnostic Interview for Psychosis and Affective Disorders [DIPAD]. The poLCA and lavaan R packages were used to apply latent class analysis [LCA] and factor analysis [FA] to 28 symptom-level items. Items for the LCA were recoded to match the OPCRIT items used in the Roscommon Family Study. For the FA, exploratory and confirmatory FA was conducted on separate training and testing subsets. Latent classes and factors were validated using measures of clinical characteristics and premorbid functioning. Data was available for N = 17,715 individuals diagnosed with SCZ (n = 10,425), BD1 (n = 4,520), or SAD (n = 2,770). Models with two through six latent classes were tested. Model fit improved as class number increased with the 6-class solution having the best overall model fit indices (e.g., AIC). Two of the classes in the 6-class solution resembled established diagnostic categories for BD1 (15.9%) and SCZ (22.2%). No single term concisely describes the remaining classes, but they reflected 1) a subgroup with high levels of psychotic symptoms and poor outcomes (23.1%), 2) a subgroup composed primarily of individuals with SCZ and SAD with high levels of hallucinations, depressive symptoms, and poor outcomes (16.5%), 3) a subgroup with the overall highest symptom load and earliest average age of onset (14.0%), and 4) a subgroup with the lowest levels of psychotic symptoms and overall best outcomes (8.3%). Significant inter-class differences were observed in symptom prevalence, age of onset, current and premorbid employment status, proportion of men and women, and family history of a psychiatric disorder. The confirmatory FA identified five latent factors corresponding to depressive symptoms, delusions, disorganized behaviors and speech, manic symptoms, and hallucinations. Factor scores related to delusions and hallucinations most strongly predicted poor clinical outcomes (r = .38 and .47, respectively), and factor scores for the depressive and manic symptoms predicted family history of a psychiatric disorder other than SCZ (r = .19 and .20, respectively). These results broadly replicate the complex transdiagnostic dimensions reported in the Roscommon Family Study and suggest that these empirically derived phenotypes may identify cross-disorder subgroups that differ in clinical and behavioral correlates and patient family history psychiatric disorders. Future analyses will use genotype measures to examine the relationship between factor scores and genetic variants and aggregate metrics of genome-wide risk (i.e., polygenic risk scores) for psychiatric disorders and cognitive traits.