This study advances the field of mood episode detection by demonstrating that passive Fitbit data, analyzed with cutting-edge machine learning methods, can reliably predict mood symptomatology in patients with bipolar disorder.
Effective management of bipolar disorder (BD) requires early identification and treatment of mood episodes, which can range from depressive lows to manic or hypomanic highs.
Traditional methods rely on patients reporting symptoms during medical appointments, which can lead to delays in recognizing mood changes. Recent studies suggest that passive data collected from everyday digital devices, such as Fitbit trackers, may help detect mood episodes.
This study, conducted by researchers at Harvard Medical School, aimed to test a new personalized machine learning (ML) approach using only Fitbit data. Importantly, the method was designed to be broadly applicable, avoiding reliance on extensive filtering or highly invasive data collection.
The study involved 54 adults diagnosed with bipolar disorder who wore Fitbit devices continuously for nine months. During this time, participants completed biweekly self-report questionnaires to monitor their mood.
Mood symptomatology was assessed using clinical tools. Depression was identified using the Patient Health Questionnaire-8 (PHQ-8), with scores above a standard clinical cutoff indicating depressive symptoms.
Manic or hypomanic symptoms were measured using the Altman Self-Rating Scale for Mania (ASRM), with scores above a defined threshold marking these episodes. These mood states were analyzed over two-week windows.
Fitbit devices provide passive data such as heart rate, sleep patterns, and physical activity levels. These data were aggregated into two-week intervals for analysis, corresponding to the mood assessment periods.
A personalized machine learning approach was used, meaning that models were tailored to individual participants rather than applying a single algorithm.
Binary Mixed Model Forest (BiMM Forest), a sophisticated machine learning algorithm, was applied alongside other ML models to classify mood states. Binary Mixed Model Forest (BiMM Forest) is a machine learning technique that combines the logic of mixed models with random forests to analyze complex data.
In mixed models, the data has both fixed components (common to all individuals) and random components (specific to each individual), allowing for personalization.
By integrating this approach with random forests—which use multiple “decision trees” to improve accuracy and robustness—BiMM is especially effective at handling individual variation and making personalized predictions. It is useful in cases such as monitoring symptoms in patients, where responses can vary significantly between individuals.
Model performance was assessed using the receiver operating characteristic area under the curve (ROC-AUC), which measures the model’s ability to distinguish between mood states.
Predictive accuracy was determined using Youden’s J statistic, which optimizes the balance between sensitivity (correctly identifying mood episodes) and specificity (avoiding false positives).
The results show that among several ML models tested, the BiMM Forest algorithm performed best. On the test data:
1- For depression, the ROC-AUC was 86.0%, indicating high accuracy in distinguishing depressive episodes.
2- For (hypo)mania, the ROC-AUC was 85.2%, showing equally strong performance.
Using optimized thresholds, the model achieved an accuracy of 80.1% (sensitivity of 71.2% and specificity of 85.6%) for depression, and an even higher accuracy for (hypo)mania, at 89.1% (sensitivity of 80.0% and specificity of 90.1%).
This study represents the first use of the BiMM Forest algorithm for mood prediction in bipolar disorder. The findings demonstrate that accurate mood predictions can be made using only passive Fitbit data, without relying on invasive or highly specialized monitoring tools.
The personalized nature of the ML approach ensures that it can adapt to individual differences, making it suitable for a wide range of patients. This includes those who may not wear specialized medical devices or share sensitive data.
By relying solely on passive data from widely available devices such as Fitbits, the method is accessible, cost-effective, and easy to use. Accurately detecting mood episodes between routine appointments could enable faster intervention, potentially improving outcomes for patients with bipolar disorder.
This study advances the field of mood episode detection by demonstrating that passive Fitbit data, analyzed with state-of-the-art machine learning methods, can reliably predict mood symptomatology in patients with bipolar disorder.
By achieving strong performance in identifying depressive and (hypo)manic states, the findings pave the way for the development of practical, personalized tools for early intervention. The use of the BiMM forest algorithm introduces a new approach, showing promise for wider adoption in mental health care.
READ MORE:
Digital phenotyping in bipolar disorder: Using longitudinal Fitbit data and personalized machine learning to predict mood symptomatology
Jessica M. Lipschitz, Sidian Lin, Soroush Saghafian, Chelsea K. Pike, Katherine E. Burdick
Acta Psychiatrica Scandinavica. 13 October 2024
Abstract:
Effective treatment of bipolar disorder (BD) requires prompt response to mood episodes. Preliminary studies suggest that predictions based on passive sensor data from personal digital devices can accurately detect mood episodes (e.g., between routine care appointments), but studies to date do not use methods designed for broad application. This study whether evaluated a novel, personalized machine learning approach, trained entirely on passive Fitbit data, with limited data filtering could accurately detect mood symptomatology in BD patients. We analyzed data from 54 adults with BD, who wore Fitbits and completed bi-weekly self-report measures for 9 months. We applied machine learning (ML) models to Fitbit data aggregated over two-week observation windows to detect occurrences of depressive and (hypo)manic symptomatology, which were defined as two-week windows with scores above established clinical cutoffs for the Patient Health Questionnaire- 8 (PHQ-8) and Altman Self-Rating Mania Scale (ASRM) respectively. As hypothesized, among several ML algorithms, Binary Mixed Model (BiMM) forest achieved the highest area under the receiver operating curve (ROC-AUC) in the validation process. In the testing set, the ROC-AUC was 86.0% for depression and 85.2% for (hypo)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%). We achieved sound performance in detecting mood symptomatology in BD patients using methods designed for broad application. Findings expand upon evidence that Fitbit data can produce accurate mood symptomatology predictions. Additionally, to the best of our knowledge, this represents the first application of BiMM forest for mood symptomatology prediction. 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.
Comments