Identifying psychosis spectrum disorder from experience sampling data using machine learning approaches

Area of Science:

• Digital mental health
• Computational psychiatry
• Psychosocial research

Background:

• Smartphones enable widespread use of the Experience Sampling Method (ESM) for collecting longitudinal data on daily experiences.
• ESM captures momentary mental states, offering insights into emotional fluctuations indicative of mental ill-health.
• Previous research has not fully explored the potential of aggregated ESM data in distinguishing clinical populations.

Purpose of the Study:

• To determine if aggregated ESM data can differentiate patients with psychosis spectrum disorder from controls using predictive modeling.
• To investigate higher-order patterns in daily life emotions captured through ESM.
• To assess the efficacy of machine learning techniques in analyzing ESM data for clinical applications.

Main Methods:

• Utilized ESM data from patients with psychosis spectrum disorder and control participants.
• Employed feature selection methods including variable importance, recursive feature elimination, and ReliefF.
• Trained and tested predictive models (Random Forests, SVM, Gaussian Processes, Logistic Regression, Neural Networks) using nested cross-validation and ROC analysis.
• Assessed model performance stability via Monte Carlo simulations.

Main Results:

• Aggregated ESM data, particularly patterns of acceleration in 'anxious' and 'insecure' emotions, significantly improved predictive power.
• Support Vector Machines with a radial kernel achieved the highest accuracy (82%) and sensitivity (82%) in distinguishing patients from controls.
• The study demonstrated the feasibility of using combined statistical and machine learning approaches on ESM data.

Conclusions:

• Patterns in emotional changes captured by ESM are discernible using advanced statistical and machine learning techniques.
• This proof-of-concept study highlights the potential of harnessing ESM data for identifying individuals with psychosis spectrum disorder.
• Synergistic application of machine learning and statistical modeling with ESM data offers a promising avenue for future mental health research and diagnostics.