Summary: Data from fitness trackers can detect mood swings in people with bipolar disorder with 89.1 percent accuracy for mania and 80.1 percent accuracy for depression. The researchers used passively collected, non-invasive data and machine learning algorithms to identify mood swings, demonstrating the potential for real-time monitoring.
These findings could improve clinical care by alerting healthcare professionals to anxiety attacks between appointments, allowing for rapid intervention. The approach focuses on personalized algorithms that can provide comprehensive support to patients without the need for specialized equipment or invasive data sharing.
Key facts:
- Fitness tracker data showed an accuracy of 89.1% for mania and 80.1% for depression.
- The study used passively collected, non-invasive data for real-world clinical applications.
- Algorithms can alert doctors to mood changes and thus improve treatment for bipolar disorder.
Source: Brigham and Women’s Hospital
Researchers at Brigham and Women’s Hospital, a founding member of the Mass General Brigham Healthcare System, examined whether data collected by a fitness tracker could be used to accurately detect mood changes in people with bipolar disorder.
Their results, published in Acta Psychiatrica Scandinavica, indicate that using data from fitness trackers, it is possible to detect with great accuracy the periods of time when bipolar disorder patients experience depression or mania.
“Most people carry personal digital devices, such as smartphones and smartwatches, that record everyday data that may be relevant to psychotherapy.
“Our goal was to use this data to identify when study participants diagnosed with bipolar disorder were experiencing anxiety attacks,” said Dr. Jessica Lipshitz, corresponding author and researcher in the Department of Psychiatry at Brigham and Women’s Hospital. “In the future, we hope that machine learning algorithms like ours can help patient care teams respond more quickly to new or ongoing episodes and reduce their negative impact.”
Bipolar disorder (BD) is a chronic psychiatric condition characterized by rapid mood swings, including depression, mania, and hypomania, followed by periods of remission. Identification and treatment of new and persistent mood changes are essential to limit the impact of BD on patients’ lives.
Previous research has shown that personal digital devices can accurately detect mood changes. However, previous studies have not used methods suitable for widespread use in clinical settings.
As an implementation scientist, Lipschitz and his colleagues focused on using methods that were widely applicable in clinical practice. Specifically, they used commercially available personal digital devices, limited data filtering, and data collection in a completely passive and non-invasive manner.
Using a new type of machine learning algorithm, they were able to detect clinically significant symptoms of depression with 80.1% accuracy and clinically significant symptoms of mania with 89.1% accuracy.
The researchers note that “overall, the results move the field one step closer to personalized algorithms that are appropriate for the entire patient population, rather than for those with high levels of medication adherence, access to specialized equipment, or a desire for invasive data sharing.”
Their next step is to apply these predictive algorithms to routine care, where they could be used to improve treatment for bipolar disorder by notifying doctors when their patients experience depressive or manic episodes between scheduled appointments. The researchers are also working to extend this work to major depressive disorder.
Authorship: In addition to Lipschitz, Chelsea K. Pike and Katherine E. Burdick are the authors of the book on Massachusetts General Brigham. Other authors include Sidian Lin and Soroush Saghafian.
Disclosures: Burdick chairs the steering committee and serves as scientific director of the non-profit Breakthrough Discoveries Foundation’s Integrated Network for People with Bipolar Disorder (BD2), and receives grants and honoraria in that capacity. He has also received honoraria in the past 12 months as a member of Merck’s Scientific Advisory Board. However, he declares no competing financial interests.
Lipschitz is a consultant with Solara Health Inc. but declares no financial conflicts of interest. The other authors declare no financial or other conflicts of interest.
Funding: This research was supported by a Brain and Behavior Research Foundation Young Investigator Grant (#28537; to JML), the Harvard Brain Science Initiative Bipolar Disorder Seed Grant Program, and a Pathways Research Award from Alkermes, Inc. Data collection for the longitudinal study was supported in part by the Bipolar Disorder Seed Grant Program (to KEB).
Lipschitz’s time was supported in part by grant MH120324 from the National Institute of Mental Health (NIMH). Saghafian’s time was supported by a grant from the Harvard Middle East Initiative’s Kuwait Science Program, which focuses on improving public health through machine learning and AI-based mobile health interventions.
The funding agencies had no role in the study design, data collection, data analysis and interpretation, nor in the writing of this manuscript.
About this bipolar disorder and neurotech research news
Author: Cassandra Falone
Source: Brigham and Women’s Hospital
Contact: Cassandra Falone – Brigham and Women’s Hospital
Image: The image is credited to StackZone Neuro
Original Research: Closed access.
“Digital phenotyping in bipolar disorder: Using longitudinal Fitbit data and personalized machine learning to predict mood symptomatology” by Jessica Lipschitz et al. Acta Psychiatrica Scandinavica
Abstract
Digital phenotyping in bipolar disorder: Using longitudinal Fitbit data and personalized machine learning to predict mood disorders
Background
Effective treatment of bipolar disorder (BD) requires rapid response to mood swings. Early studies suggest that predictions based on passive sensor data from personal digital devices can accurately detect mood swings (for example, between routine medical appointments), but no methods have been developed that are suitable for widespread use.
This study examined whether a novel, personalized machine learning approach, trained entirely on passive Fitbit data and with limited data filtering, could accurately detect mood symptoms in patients with bipolar disorder.
Methods
We analyzed data from 54 adults with bipolar disorder (BD) who wore Fitbits and completed biweekly self-reports for nine months. We applied machine learning (ML) models to the Fitbit data, collected over a two-week observation period, to detect the onset of depressive and hypomanic symptoms. These symptoms were defined as a two-week interval with scores above established clinical cutoffs on the Patient Health Questionnaire-8 (PHQ-8) and the Altman Self-Rating Mania Scale (ASRM), respectively.
Results
As predicted, the Binary Mixed Model (BiMM) forest of different machine learning algorithms achieved the highest AUC-ROC (area under the receiver operating curve) percentage during the validation process. In the test set, the AUC-ROC was 86.0% for depression and 85.2% for (hypo)mania.
Using improved thresholds calculated with Youden’s J statistic, the predictive accuracy was 80.1% for depression (sensitivity 71.2% and specificity 85.6%) and 89.1% for (hypo)mania (sensitivity 80.0% and specificity 90.1%).
Conclusion
We have achieved remarkable results using methods designed for widespread use to detect mood symptoms in patients with bipolar disorder. These results strengthen the evidence that Fitbit data can produce accurate predictions of mood symptoms. Furthermore, to our knowledge, this is the first application of the BiMM forest to predict mood symptoms.
Overall, the findings bring the field one step closer to personalized algorithms that are suitable for the entire patient population, rather than for those with high levels of medication adherence, access to specialized equipment, or a desire for invasive data sharing.
