The Effect of Bipolar Disorder Illness Course on Physiological Digital Signals: Observational Study

Although bipolar disorder (BD) lacks objective biomarkers of illness progression, wearable devices enable continuous, real-world collection of physiological digital biosignals, offering a promising approach to address this gap. Measures such as electrodermal activity (EDA) and heart rate variability (HRV) fluctuate with mood state and symptom severity. Previous studies have reported reduced EDA during depressive episodes and increased HRV following symptom resolution after acute mood episodes. Furthermore, greater illness burden in BD has been associated with lower HRV, suggesting that digital biosignals may capture not only state-related changes but also aspects of illness course.

This study investigated whether naturalistic EDA and heart rate (HR) profiles differ according to illness course variables in BD-including diagnosis, predominant polarity, years since onset, and lifetime number of episodes-in euthymic patients, to evaluate their potential as illness course biomarkers.

This cross-sectional post hoc analysis used data from 80 euthymic patients on the BD spectrum from the INTREPIBD-TIMEBASE cohort. Participants were recruited at Hospital Clínic de Barcelona (Spain) between 2021 and 2024 and wore an Empatica E4 wristband for 48 hours in naturalistic settings. Inclusion criteria comprised a current or past DSM-5 (Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition) diagnosis of BD confirmed by structured interview and sustained euthymia. We examined the effects of 4 illness course variables (diagnosis, predominant polarity, years since onset, and number of lifetime episodes) on 3 biosignals (mean EDA, EDA peaks per minute, and HR) using generalized additive models. Interactions between illness course variables and circadian patterns were also assessed. All models were adjusted for age, sex, and current antipsychotic use and included a participant-specific random effect.

Of the 80 participants, 49 (61.3%) were female; participants' mean age was 46.9 (SD 14.1) years. After adjustment for covariates and multiple comparisons, no significant associations were observed between illness course variables and EDA or HR. In sensitivity analyses excluding patients with schizoaffective disorder (n=75), years since illness onset and predominant polarity were associated with EDA peaks per minute after multiple comparison correction. Specifically, patients with more than 22 years since illness onset had fewer EDA peaks compared with other groups (β=-1.06, 95% CI -1.71 to -0.41; P=.001). No significant interactions were found between illness course variables and circadian patterns of EDA or HR.

To our knowledge, this is the first study to examine the association between illness course variables and EDA and HR recorded in naturalistic settings in BD. In contrast to prior findings linking illness burden with HRV, our results suggest that years since illness onset may show modest associations with phasic EDA activity. These findings indicate that digital biosignals may be more sensitive to state-related changes than to illness course characteristics, limiting their use as longitudinal biomarkers of disease progression.