Browsing by Subject "external validation"

Aim: Psychosis risk has been researched a lot during last couple of decades and different tools for predicting psychosis have been formed and taken in the clinical practice. About one third of those in clinical high risk for psychosis develop psychosis, but only in help-seeking population. There is ongoing research aiming to improve on the existing psychosis prediction tools and to develop new ones with better prognostic accuracy and greater generalizability to the more varied populations. It has been also argued that psychosis transition is not the most useful outcome to focus on and that functioning is more clinically relevant. However, lack of replication or external validation studies remains problem. The current study aimed to test the performance of the published psychosis prediction models in general psychiatric sample of adolescents. In addition to this, same prediction models were also used to predict functioning using psychiatric hospitalization as an proxy. Methods: The current study utilized Helsinki Prodromal Study cohort, an ongoing prospective psychosis risk study. The sample (N=146) consisted of adolescents aged 15-18 at baseline who had been starting as new patients in public psychiatric care. The psychosis prediction models tested in the current study were selected from recent exhaustive review. Final amount of prediction models was 16 from 13 published articles. Predictors included parts of Structured Interview for the Prodromal Syndromes (SIPS), cognitive functions, substance use, and functioning ability and outcome variables were psychosis diagnose and psychiatric hospitalization. Analyses methods were logistic regression and area under the receiver operator characteristics curve. The participants were followed 7 years or until they developed psychosis. Findings and conclusions: Performance of the models was not as good as in original studies, which was to be expected since the sample was more varied than in original studies. However, almost all of the models predicted both psychosis and hospitalization better than high psychosis risk assessed by the SIPS did. For psychosis predictors that would perform well regardless of the model couldn’t be found, but models including disorganized communication performed worst. For hospitalization, models including positive symptoms, functioning, and duration of (untreated) symptoms performed best. Accuracy of the models was not clearly connected to the models ability to explain variance. More research aiming to validate and replicate suggested psychosis prediction models is needed.