Multimodal Machine Learning Workflows for Prediction of Psychosis in Patients With Clinical High-Risk Syndromes and Recent-Onset Depression
Nikolaos Koutsouleris,Dominic B. Dwyer,Franziska Degenhardt,Carlo Maj,Maria Fernanda Urquijo-Castro,Rachele Sanfelici,David Popovic,Oemer Oeztuerk,Shalaila S. Haas,Johanna Weiske,Anne Ruef,Lana Kambeitz-Ilankovic,Linda A. Antonucci,Susanne Neufang,Christian Schmidt-Kraepelin,Stephan Ruhrmann,Nora Penzel,Joseph Kambeitz,Theresa K. Haidl,Marlene Rosen,Katharine Chisholm,Anita Riecher-Rössler,Laura Egloff,André Schmidt,Christina Andreou,Jarmo Hietala,Timo Schirmer,Georg Romer,Petra Walger,Maurizia Franscini,Nina Traber-Walker,Benno G. Schimmelmann,Rahel Flückiger,Chantal Michel,Wulf Rössler,Oleg Borisov,Peter M. Krawitz,Karsten Heekeren,Roman Buechler,Christos Pantelis,Peter Falkai,Raimo K. R. Salokangas,Rebekka Lencer,Alessandro Bertolino,Stefan Borgwardt,Markus Noethen,Paolo Brambilla,Stephen J. Wood,Rachel Upthegrove,Frauke Schultze-Lutter,Anastasia Theodoridou,Eva Meisenzahl,,
DOI: https://doi.org/10.1001/jamapsychiatry.2020.3604
2021-02-01
JAMA Psychiatry
Abstract:ImportanceDiverse models have been developed to predict psychosis in patients with clinical high-risk (CHR) states. Whether prediction can be improved by efficiently combining clinical and biological models and by broadening the risk spectrum to young patients with depressive syndromes remains unclear.ObjectivesTo evaluate whether psychosis transition can be predicted in patients with CHR or recent-onset depression (ROD) using multimodal machine learning that optimally integrates clinical and neurocognitive data, structural magnetic resonance imaging (sMRI), and polygenic risk scores (PRS) for schizophrenia; to assess models' geographic generalizability; to test and integrate clinicians' predictions; and to maximize clinical utility by building a sequential prognostic system.Design, Setting, and ParticipantsThis multisite, longitudinal prognostic study performed in 7 academic early recognition services in 5 European countries followed up patients with CHR syndromes or ROD and healthy volunteers. The referred sample of 167 patients with CHR syndromes and 167 with ROD was recruited from February 1, 2014, to May 31, 2017, of whom 26 (23 with CHR syndromes and 3 with ROD) developed psychosis. Patients with 18-month follow-up (n = 246) were used for model training and leave-one-site-out cross-validation. The remaining 88 patients with nontransition served as the validation of model specificity. Three hundred thirty-four healthy volunteers provided a normative sample for prognostic signature evaluation. Three independent Swiss projects contributed a further 45 cases with psychosis transition and 600 with nontransition for the external validation of clinical-neurocognitive, sMRI-based, and combined models. Data were analyzed from January 1, 2019, to March 31, 2020.Main Outcomes and MeasuresAccuracy and generalizability of prognostic systems.ResultsA total of 668 individuals (334 patients and 334 controls) were included in the analysis (mean [SD] age, 25.1 [5.8] years; 354 [53.0%] female and 314 [47.0%] male). Clinicians attained a balanced accuracy of 73.2% by effectively ruling out (specificity, 84.9%) but ineffectively ruling in (sensitivity, 61.5%) psychosis transition. In contrast, algorithms showed high sensitivity (76.0%-88.0%) but low specificity (53.5%-66.8%). A cybernetic risk calculator combining all algorithmic and human components predicted psychosis with a balanced accuracy of 85.5% (sensitivity, 84.6%; specificity, 86.4%). In comparison, an optimal prognostic workflow produced a balanced accuracy of 85.9% (sensitivity, 84.6%; specificity, 87.3%) at a much lower diagnostic burden by sequentially integrating clinical-neurocognitive, expert-based, PRS-based, and sMRI-based risk estimates as needed for the given patient. Findings were supported by good external validation results.Conclusions and RelevanceThese findings suggest that psychosis transition can be predicted in a broader risk spectrum by sequentially integrating algorithms' and clinicians' risk estimates. For clinical translation, the proposed workflow should undergo large-scale international validation.
psychiatry