Despite five decades of quantitative research into early phases of psychoses, early recognition still depends on a human expert. Unfortunately, the availability of clinical services that provide this expert knowledge is not the only limitation. Early recognition is also a unique clinical challenge - accurately estimating the risk profile of an individual with diverse, frequently nonspecific psychopathology, leading to biased reasoning given different clinical traditions and gut feeling. Recently, opportunities to address the clincial challenge have emerged with machine learning, multi-site prospective study designs and international collaborations merging into a powerful methodology for precision psychiatry. Previous studies have provided preliminary evidence regarding the feasibility of stratifying at-risk and first-episode patients according to the odds of adverse outcomes by extracting candidate predictive models from diverse data. Candidate models are under further validation and analysis based on multi-site datasets collected within the NAPLS, PRONIA, PsySCAN and PNC projects. Should candidate markers generalize well, the outcomes would translate into significant increases in predictive and prognostic certainty. Such progress would allow for individualized risk-based stratification of patients and clinical trials, novel targets for drug development and tools for individualized neuromonitoring of preventive treatments. The symposium will present new data from the PRONIA project which has collected clinical, neurocognitive, blood-based and MRI data from 1600 persons in at-risk and early stages of psychoses and mood disorders, and healthy controls. PRONIA is currently generating machine-learning markers intended to predict clinically relevant outcomes, identify vulnerable subgroups and combine data for diagnosis and prediction across heterogeneous domains.
