M121. CLINICAL PREDICTION MODELS FOR TRANSITION TO PSYCHOSIS: AN EXTERNAL VALIDATION STUDY IN THE PRONIA SAMPLE. (18th May 2020)

• Record Type: Journal Article
• Authors: Rosen, Marlene; Betz, Linda; Bertolino, Alessandro; Borgwardt, Stefan; Brambilla, Paolo; Chisholm, Katharine; Kambeitz-Ilankovic, Lana; Haidl, Theresa; Lencer, Rebekka; Meisenzahl, Eva; Ruhrmann, Stephan; Salokangas, Raimo K R; Schultze-Lutter, Frauke; Upthegrove, Rachel; Wood, Stephen J; Koutsouleris, Nikolaos; Kambeitz, Joseph
• Is Part Of: Schizophrenia bulletin. Volume 46(2020)Supplement 1
• Page Start: S181
• Page End: S181
• Publication Date: 2020-05-18
• Subjects: Schizophrenia -- Periodicals; Schizophrenia -- Research -- Periodicals; 616.898005
• DOI: 10.1093/schbul/sbaa030.433
• Languages: English
• ISSNs: 0586-7614

Abstract

Abstract: Background: A multitude of clinical models to predict transition to psychosis in individuals at clinical high risk (CHR) have been proposed. However, only limited efforts have been made to systematically compare these models and to validate their performance in independent samples. Therefore, in this study we identified psychosis risk models based on information readily obtainable in general clinical settings, such as clinical and neuropsychological data, and compared their performance in the PRONIA study (Personalised Prognostic Tools for Early Psychosis Management, www.pronia.eu ) as an independent sample. Methods: Of the 278 CHR participants in the PRONIA sample, 150 had available data until month 18 and were included in the validation of eleven psychosis prediction models identified through systematic literature search. Discrimination performance was assessed with the area under the receiver operating characteristic curve (AUC), and compared to the performance of the prognosis of clinical raters. Psychosocial functioning was explored as an alternative outcome. Results: Discrimination performance varied considerably across models (AUC ranging from 0.42 to 0.79). High model performance was associated with the inclusion of neurocognitive variables as predictors. Low model performance was associated with predictors based on dichotomized variables. Clinical raters performed comparable to the best data-driven models (AUC = 0.75). Combining raters' prognosis and model-based predictions improved discrimination performance (AUC = 0.84), particularly for less experienced raters. One of the tested models predicted transition to psychosis and psychosocial outcomes comparably well. Discussion: The present external validation study highlights the benefit of enriching clinical information with neuropsychological data in predicting transition to psychosis satisfactorily and with good generalizability across samples. Integration of data-driven risk models and clinical expertise may improve clinical decision-making in CHR for psychosis, particularly for less experienced raters. This external validation study provides an important step toward early intervention and the personalized treatment of psychotic disorders.