Dissecting first-episode psychosis heterogeneity with clustering analyses: A systematic review

Area of Science:

• Psychiatry
• Computational Neuroscience
• Genetics

Background:

• First episode psychosis (FEP) presents heterogeneously, necessitating precise diagnostic and prognostic tools.
• Unsupervised machine learning (ML) clustering offers a method to stratify FEP patient groups for improved outcomes.
• This review examines unsupervised ML clustering applications in FEP for clinical decision-making.

Purpose of the Study:

• To systematically review the application of unsupervised ML clustering in patients with FEP.
• To assess the potential of unsupervised ML clustering for clinical decision-making in FEP.
• To identify current limitations of unsupervised ML clustering in FEP research.

Main Methods:

• Systematic literature search conducted on PubMed, Embase, Psycinfo, and Scopus.
• Search performed from inception through December 31, 2025.
• Study selection followed PRISMA guidelines.

Main Results:

• 48 studies utilized unsupervised ML clustering for FEP patient stratification.
• Clusters were identified across cognitive, functional, immune, genetic, clinical, imaging, and neurophysiological domains.
• Immune/genetic and imaging/neurophysiological studies suggested a two-cluster model, with one indicating higher inflammation or brain damage.
• No consensus on cluster number or characteristics emerged across cognitive, functional, and clinical domains due to methodological variations.
• A consistent finding was the identification of a subgroup with significant impairment at psychosis onset.

Conclusions:

• Unsupervised ML clustering shows promise for enhancing FEP diagnosis and treatment personalization.
• Lack of reproducibility across studies is a significant limitation.
• Future research should integrate multi-omics and longitudinal data in larger studies for improved classification.