Using AI to Detect Psychosis Relapse: Scoping Review.
JMIR Ment Health June 16, 2026 Lorenzo Ghelfi, Jack Healy, Francesco Piacenza et al.
A scoping review of 10 studies found that artificial intelligence methods, including machine learning and digital phenotyping via smartphones and wearables, show promise for detecting relapse in psychotic disorders but have significant limitations. The sensitivity of AI models ranged from 0.25 to 0.77 and specificity from 0.06 to 0.88, with area under the curve between 0.63 and 0.78. Models were heterogeneous and most findings were not replicated. The review concludes that while personalized approaches with individual-level modeling are promising, larger studies and methods such as large language models are needed before AI can be used in real-world clinical practice.