Smartphone keyboard dynamics predict affect in suicidal ideation.

Area of Science:

• Digital health
• Computational psychiatry
• Mental health assessment

Background:

• Digital phenotyping enables unobtrusive, real-time mental health monitoring.
• Integrating diverse data modalities (e.g., self-reports, sensor data) presents challenges due to high dimensionality and differing sampling rates.
• Accurate mental health assessment requires robust methods for combining heterogeneous data.

Approach:

• Developed an integrated pipeline to harmonize digital phenotyping data by aligning sampling frequencies.
• Employed temporal independent component analysis (ICA) for high-dimensional self-report data.
• Utilized linear mixed-effects models to fuse processed modalities and identify associations.

Key Points:

• Temporal ICA on self-report data identified components linked to well-being, anhedonia, irritability, and social dysfunction.
• Analysis of BiAffect keyboard dynamics revealed that reduced phone movement during typing was significantly associated with increased anhedonia (β = -0.12, p = 0.00030).
• The pipeline effectively integrates dense, longitudinal digital phenotyping data for mental health research.

Conclusions:

• The proposed integrated pipeline addresses key challenges in combining digital phenotyping data for mental health assessment.
• This approach facilitates the discovery of meaningful associations between digital behaviors and clinical symptoms.
• The method is broadly applicable to various dense, longitudinal digital phenotyping datasets for mental health research.