Research at Brigham and Women’s Hospital: Advancing Bipolar Disorder Care with Wearable Technology

March 30, 2025
Jessica Lipschitz, PhD
Passively collected data from wearable devices may improve the early identification of mood episodes in individuals with bipolar disorder.

Bipolar disorder (BD) is a chronic and difficult-to-control psychiatric disorder characterized by extreme mood swings, including depression, mania, and hypomania followed by periods of remission. Identification and treatment of new and unremitting mood episodes between routine care appointments, is essential for limiting the impact of BD on patients’ lives.

The increasing use of personal digital devices, such as smartphones and smartwatches, offers opportunities for collecting clinically significant measurements outside of a clinical setting. It is now feasible to seamlessly collect biobehavioral information on a variety of indicators of mood episodes, such as sleep patterns, daily activity, and heart rate. This type of passive sensor data from personal digital devices could allow detection of mood episodes between routine care appointments. While prior research shows that data from personal digital devices can accurately detect mood episodes in BD, these studies have employed methods that are unlikely to be acceptable to patients in real-world care settings (e.g., invasive data streams like GPS and voice) or may bias results toward only highly compliant patients (via use of significant data filtering).

In a recent study, Dr. Jessica Lipschitz, PhD, and colleagues, at the BWH Digital Behavioral Health and Informatics Research Program investigated use of mainstream, commercially-available personal digital devices to identify periods of clinically significant mood symptoms in patients with bipolar disorder. They recruited a sample of patients with bipolar disorder to engage in a 40-week observational study, during which participants were asked to wear a study-provided Fitbit to collect biobehavioral metrics like sleep, movement and heart rate. The study team also tracked fluctuations in participants’ mood symptoms using weekly self-report measures (the Patient Health Quesitonnaire-8 (PHQ-8) for depressive symptoms and the Altman Mania Rating Scale (AMRS) for manic symptoms).

The study team evaluated whether a novel, personalized machine learning approach, trained entirely on passive Fitbit data, with limited data filtering could accurately detect mood symptomatology in BD patients.

Key findings:

  • Using random forest imputation, mood episode status was detectable over the full monitoring period in 90% of participants (in comparison to prior studies which typically filter out at least half of participants owing to noncompliance).
  • Among several machine learning algorithms tested in the validation process, Binary Mixed Model (BiMM) forest achieved the highest area under the receiver operating curve (ROC-AUC).
  • In the test dataset, the ROC-AUC was 86.0% for depression and 85.2% for mania. Using optimized thresholds calculated with Youden’s J statistic, predictive accuracy was 80.1% for depression (sensitivity of 71.2% and specificity of 85.6%) and 89.1% for (hypo)mania (sensitivity of 80.0% and specificity of 90.1%).

Findings suggest that mood episodes can be identified in the full population of patients with bipolar disorder using only passively-collected Fitbit data. Overall, results move the field a step toward personalized algorithms suitable for the full population of patients, rather than only those with high compliance, access to specialized devices, or willingness to share invasive data.

Next Steps

Dr. Lipschitz’s team is currently working to expand this research to patients with unipolar mood disorders like major depressive disorder. So far findings in this population show even greater predictive accuracy. Ongoing studies also include qualitative research focused on identifying appropriate strategies for implementing these methods in routine care. The team is actively pursuing funding to study the clinical impact of applying these methods in the context of routine psychiatric care to monitor patients’ symptoms between scheduled appointments.

Read More

Lipschitz JM, Lin S, Saghafian S, Pike CK, Burdick KE. Digital phenotyping in bipolar disorder: Using longitudinal Fitbit data and personalized machine learning to predict mood symptomatology. Acta Psychiatr Scand. 2025 Mar;151(3):434-447. 

Share this with your network: