Prediction of Clinical Outcomes in Psychotic Disorders Using Artificial Intelligence Methods: A Scoping Review

Area of Science:

• Psychiatry
• Computational Neuroscience
• Medical Informatics

Background:

• Psychotic disorders significantly impair physical, social, and psychological functioning.
• Accurate prediction of outcomes aids in identifying patient subgroups and optimizing treatment.

Purpose of the Study:

• To review the accuracy of artificial intelligence (AI) in predicting clinical outcomes for patients with psychotic disorders.
• To identify key predictors of these outcomes.

Main Methods:

• A scoping review following PRISMA guidelines.
• Searched seven electronic databases for relevant English articles published up to February 1, 2024.

Main Results:

• Thirty articles were included, primarily from the West (63%) and Asia (37%), published recently (83.3% in the last 5 years).
• Machine learning models (e.g., SVM, random forest, logistic/linear regression) utilized clinical, cognitive, and biological data (genetics, neuroimaging).
• Predictive accuracy varied (AUC 0.58-0.95), with no single AI approach consistently outperforming others. Predictors included demographics, social factors, illness features, treatment details, and neuroimaging abnormalities.

Conclusions:

• AI and machine learning show promise for predicting clinical outcomes in psychotic disorders.
• Further refinement of AI models is needed to understand the complex interplay of variables influencing outcomes.