General Medicine

Machine Learning Predicts Psychosis in High-Risk Patients

Heather R. Johnson
|
shadow like figures
shadow like figures
The study authors evaluated whether psychosis transition can be predicted in patients with clinical high-risk or recent-onset depression using multimodal machine learning that integrates clinical and neurocognitive data, structural magnetic resonance imaging, and polygenic risk scores for schizophrenia.

Multimodal machine learning can predict psychosis in a broader range patients who meet clinical high-risk (CHR) criteria. Currently only 22% of CHR patients transition to psychosis during a 3-year period, indicating the CHR criteria and method may be limited. Machine learning models may improve accuracy.

Analysis included 334 patients in CHR states or who had recent-onset depression. The researchers used patients with an 18-month follow-up for model training and leave-one-site-out cross-validation. The researchers used 334 healthy volunteers to provide a normative sample. The researchers used NeuroMiner machine learning software and constructed various risk calculators.

The tool predicted psychosis transition with a balanced accuracy of 75.7% (sensitivity, 84.6%; specificity, 66.8%; P < .001). Notable predictors included attenuated positive symptom, motor disturbance, a nonsupportive family environment during childhood, and reduced facial emotion recognition. A second model achieved accuracy of 66.1% (sensitivity, 76.0%; specificity, 56.2%; P < .001). A model based on structural magnetic resonance imaging achieved accuracy of 70.7% (sensitivity, 88.0%; specificity, 53.5%; P < .001).

A limitation of this study was that psychosis transitions were limited to 26 individuals. This sample size increased the risk of producing overly optimistic prediction results owing to an accidental collection of well-classifiable cases.

“Our study showed for the first time, to our knowledge, that the augmentation of human prognostic abilities with algorithmic pattern recognition improves prognostic accuracy to margins that likely justify the clinical implementation of cybernetic decision-support tools,” the researchers concluded.

Related Content

Disclosure: Several study authors declared affiliations with industry. Please see the original reference for a full list of authors’ disclosures.

Reference

PRONIA Consortium. Multimodal machine learning workflows for prediction of psychosis in patients with clinical high-risk syndromes and recent-onset depression. JAMA Psychiatry. Published online December 2, 2020. doi:10.1001/jamapsychiatry.2020.3604