Average accuracy hides where a brain-imaging psychosis risk model fails across clinical high-risk cohorts
Year: 2026
Type: preprint
Abstract: Brain-imaging models predicting psychiatric outcomes are rarely tested where they would be used. Clinical high risk for psychosis is a demanding test case because services differ in whom they recruit and how they record transition. We re-estimated a published structural magnetic resonance imaging (MRI) risk model and applied it unchanged to four independent cohorts (685 participants, 106 transitions). Discrimination was weak (area under the curve, 0.56) and near chance in the two cohorts not previously used to evaluate the model (0.53 and 0.56). Predicted risks did not transfer (calibration slope, 0.17). Exploratory performance landscapes, which map local discrimination across participant characteristics, showed that discrimination varied with negative-symptom severity, also in external cohorts, whereas apparent age- and brain-age-related variation depended on single cohorts; no global test survived multiple-testing correction. Reporting where a model works, not only its average accuracy, can reveal when one diagnostic label hides several prediction problems.
Source: medRxiv
Authors: Clara S. Vetter, Alessa Grund, Lisa Hahn, Sergio Mena, Charlotte Fraza, Julia Fietz, Nikos F. Diederichs, Dominic B. Dwyer, Paris Lalousis, Nora Penzel, Maria Fernanda Urquijo, Frauke Schultze-Lutter, Scott Woods, Tyrone D. Cannon, Shalaila S. Haas, Lana Kambeitz-Ilankovic, Joseph Kambeitz, Stephan Ruhrmann, Theresa Lichtenstein, Katharine Chisholm, Georg Romer, Linda A. Antonucci, Anita Riecher-Rössler, Christos Pantelis, Jarmo Hietala, Raimo K. R. Salokangas, Rebekka Lencer, Udo Dannlowski, Alexandra Korda, Alessandro Bertolino, Stefan Borgwardt, Paolo Brambilla, Stephen J. Wood, Eva Meisenzahl, Rachel Upthegrove, Peter Falkai, Karsten Heekeren, Wulf Rössler, Anastasia Theodoridou, Daniel Rueckert, Nikolaos Koutsouleris
Institutions: Helmholtz Munich, LMU Klinikum, Max Planck Institute of Psychiatry, Klinik und Poliklinik für Psychiatrie und Psychotherapie, Munich Center for Machine Learning, Ludwig-Maximilians-Universität München, King's College London, Deutsches Zentrum für Psychische Gesundheit, University of Birmingham, Harvard University, Mass General Brigham, University of Bern, Universitas Airlangga, Heinrich Heine University Düsseldorf, Yale University, Icahn School of Medicine at Mount Sinai, University of Cologne, University Hospital Cologne, Max Planck Institute for Metabolism Research, University of Sussex, University of Münster, University of Bari Aldo Moro, University of Basel, Universitäre Psychiatrische Kliniken Basel, The University of Melbourne, Florey Institute of Neuroscience and Mental Health, Monash University, University of Turku, Bielefeld University, University of Lübeck, University of Milan, Fondazione IRCCS Ca' Granda Ospedale Maggiore Policlinico, Swinburne University of Technology, Warneford Hospital, University of Oxford, University Hospital Zurich, Psychiatrische Universitätsklinik Zürich, St. Hedwig-Krankenhaus, Charité – Universitätsmedizin Berlin, TUM Klinikum, Imperial College London, Technical University of Munich
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: green
Awards: 3/9- Predictors and Mechanisms of Conversion to Psychosis, "1/9-Predictors and Mechanisms of Conversion to Psychosis, 8/9 Predictors and Mechanisms of Conversion to Psychosis, 9/9-Predictors and Mechanisms of Conversion to Psychosis, Personalised Prognostic Tools for Early Psychosis Management, "7/9" Predictors and Mechanisms of Conversion to Psychosis, 4/9-Predictors and Mechanisms of Conversion to Psychosis, 5/9 Predictors and Mechanisms of Conversion to Psychosis, 6/9-Predictors and Mechanisms of Conversion to Psychosis, 2/9 - Predictors and Mechanisms of Conversion to Psychosis
OpenAlex ID: W7222943143