JMIR Publications

ABSTRACT

Background:

Wearable devices allow continuous and objective collection of digital biosignals in day-to-day life. In bipolar disorder (BD), digital biosignals such as electrodermal activity (EDA) and heart rate (HR) can fluctuate depending on mood state and symptom severity, but their association with illness course remains unclear.

Objective:

This study investigated whether naturalistic EDA and HR profiles differ by BD illness course variables (subtype, predominant polarity, years since onset, lifetime episodes) in euthymic patients, evaluating their potential as illness-course biomarkers.

Methods:

For this exploratory post hoc analysis, we used data from 80 euthymic BD spectrum patients from the INTREPIBD–TIMEBASE cohort. They wore an Empatica E4 wristband for 48 hours in naturalistic settings. We evaluated the effects of four illness course variables (BD subtype, predominant polarity, years since illness onset, and number of lifetime episodes) on three biosignals collected with the E4 (EDA mean, EDA peaks per minute, and HR) with generalized additive models (GAMs). The interaction of the clinical variables with the biosignals’ circadian pattern was also tested in a second set of GAMs. All models were adjusted for age, sex, current antipsychotic use and a random effect for subjects.

Results:

61.2% of the participants were females and the mean age was 56.9 years old. After adjusting for covariates, only years since illness onset was associated with EDA peaks per minute, with the group with more than 22 years of BD showing fewer EDA peaks than the rest (β = -0.80, SE = 0.34, p = 0.02). This difference did not remain significant after multiple comparison corrections. BD subtype, predominant polarity, and number of lifetime episodes did not predict any EDA components or HR in the main results. None of the clinical variables showed a significant interaction with EDA or HR circadian pattern. In the sensitivity analysis, years since BD onset and predominant polarity predicted EDA peaks, even after adjusting for multiple comparisons.

Conclusions:

In euthymic BD, only illness duration showed modest EDA phasic effects after covariate adjustment, supporting digital biosignals' predominant state-dependent role. Illness-course biomarker utility appears limited, requiring longitudinal validation in larger cohorts.