Deriving symptom networks from digital phenotyping data in serious mental illness

Area of Science:

• Psychiatry and Computational Neuroscience
• Digital Phenotyping
• Network Analysis

Background:

• Serious mental illnesses involve complex, multidimensional symptoms.
• Generative models can illuminate the intricate relationships within symptom networks.

Purpose of the Study:

• To employ generative models for identifying unique, moment-to-moment interactions among schizophrenia symptoms.
• To analyze temporal dynamics of symptom experience in schizophrenia.

Main Methods:

• Utilized a digital phenotyping app to collect data on mood, anxiety, psychosis, sleep, social function, and cognition over 90 days from 47 schizophrenia patients.
• Employed generative models to calculate transition probabilities between symptom states across 3-day intervals.
• Analyzed data from patients retrospectively grouped by clinical measurements.

Main Results:

• Identified high transition probabilities between anxiety-inducing mood and subsequent symptoms (0.357), psychosis-inducing mood and symptoms (0.276), and anxiety-inducing poor sleep and symptoms (0.268).
• Validated findings against a pilot study cohort, showing no significant differences.
• Discovered unique symptom networks specific to clinical subgroups.

Conclusions:

• Generative models analyzing digital phenotyping data demonstrate that specific schizophrenia symptoms can escalate others over time.
• Symptom networks provide a feasible framework for clinically interpretable models of psychosis-spectrum illness.
• Results support temporal dynamics research, preventative clinical care, and enhanced patient understanding.