Subtyping first-episode psychosis based on longitudinal symptom trajectories using machine learning

Area of Science:

• Psychiatry and Clinical Psychology
• Computational Neuroscience
• Medical Data Science

Background:

• The clinical course following a first episode of psychosis (FEP) is highly variable.
• Personalized treatment strategies require understanding and predicting longitudinal symptom trajectories in FEP patients.

Purpose of the Study:

• To identify distinct patient subgroups based on longitudinal positive and negative symptoms after FEP.
• To predict the cluster membership of FEP patients using baseline clinical data.

Main Methods:

• K-means clustering was applied to longitudinal symptom data from 411 FEP patients.
• Ridge logistic regression was used to predict cluster membership based on baseline characteristics.

Main Results:

• Three distinct clusters emerged: Low Symptoms (LS), Low Positive/Persistent Negative (LPPN), and Persistent Positive and Negative Symptoms (PPNS).
• Cluster membership was predicted with an AUC of 0.74.
• Predictors for LS included lower apathy and affective flattening; predictors for PPNS included hallucination severity and positive thought disorder.

Conclusions:

• Distinct symptom trajectories exist following FEP, characterized by different symptom profiles and antipsychotic dose requirements.
• Baseline clinical features can predict these trajectories, paving the way for personalized interventions in FEP.
• Understanding these heterogeneous pathways is crucial for tailoring treatments to individual patient needs.