From Mood Episodes to Digital Signatures: Passive and Active Phenotyping of Bipolar Disorder Over Time

Abstract

Background:

Digital phenotyping has emerged as a promising approach to capture real-time behavioral and physiological data in individuals with bipolar disorder (BD). By integrating passive and active data streams, this approach may enable the identification of dynamic patterns associated with mood instability. However, the conceptual integration of these data into clinically meaningful digital signatures remains insufficiently defined and lacks standardized operational frameworks.

Methods:

This narrative review synthesizes current evidence on digital phenotyping in BD and proposes a conceptual framework integrating passive sensing (e.g. smartphones, wearables, mobility and communication data, physiological signals) and active assessments (e.g. ecological momentary assessment, self-reported mood, cognitive tasks). The framework outlines how multimodal digital biomarkers can be analyzed using computational approaches, including machine learning and longitudinal modeling, to derive individualized digital signatures.

Results:

The proposed framework describes how continuous behavioral and physiological data can be transformed into multimodal digital biomarkers reflecting sleep–wake rhythms, motor activity, mobility patterns, social interaction dynamics, and autonomic physiology. Through multimodal data integration and personalized baselines, computational models can identify temporal deviations associated with mood changes. These individualized digital signatures capture the dynamic processes underlying mood regulation and may provide early warning signals of relapse, as well as markers of treatment response.

Conclusions:

Digital signatures derived from integrated digital phenotyping data represent a promising step toward precision psychiatry in BD. However, this concept remains an emerging framework requiring further empirical validation and methodological standardization. This approach highlights the potential for early detection of mood instability, prediction of mood episodes, and personalized clinical decision-making. Future research should focus on validation in longitudinal clinical cohorts, standardization of methodologies, and ethical considerations related to data privacy and implementation.