Machine Learning Methods and Schizophrenia Spectrum Disorders: A Scoping Review (Preprint)

UNSTRUCTURED Schizophrenia spectrum disorders (SSD) pose significant challenges in terms of functional impairment, disability, and economic burden. The application of machine learning (ML) in mental health, particularly in SSD research, has grown rapidly in recent years. ML techniques offer the potential to analyze complex datasets, identify subtle patterns, and make predictions that could lead to more personalized and effective interventions for SSD. This scoping review aims to characterize how ML has been applied to SSD research, with a focus on determining clinical and functional outcome trajectories. We conducted a comprehensive scoping review following the methodology adapted from Arksey and O'Malley, adhering to the PRISMA extension for scoping reviews and following the five stages for scoping reviews. A health sciences librarian assisted in designing a search strategy that was implemented across seven bibliographic databases. Two independent reviewers screened articles at the title/abstract and full-text levels. Data extraction focused on study characteristics, machine learning techniques used, and key findings related to SSD outcomes. In total, 88 studies were included in the review which illustrated a diverse range of ML applications in SSD research, categorized into five primary outcome domains: clinical, treatment, functional, behavioral, and provider outcomes. Clinical outcome studies utilized various data types, including electronic medical records, clinical assessment scales, neuroimaging, and mobile sensing data, employing methods and algorithms such as support vector machines, random forests, and neural networks to predict symptom burden, remission, and relapse. Treatment outcome research focused on predicting response to specific interventions, treatment resistance, and adverse effects, primarily using supervised learning techniques. Functional outcome studies employed advanced ML methods, including deep learning, to analyze social cognition and functioning. Behavioral outcome research applied ML to predict risks of self-harm, aggression, and offending. Provider outcome studies mainly examined prescribing patterns using supervised learning approaches. Across all outcome domains, there was a trend towards integrating multiple data sources and applying increasingly sophisticated ML techniques to address the complex nature of SSD. There is a growing potential of ML in SSD research across various outcome domains. The integration of diverse data types and advanced ML techniques offers promising avenues for improving diagnosis, prognosis, and treatment planning in SSD. However, methodological challenges, ethical considerations, and the need for robust validation remain important issues to address. Future research should focus on refining ML models, addressing data quality and privacy concerns, and translating these computational approaches into clinical practice to enhance personalized care for individuals with SSD.