Comprehensive Summary
This study, presented by Wu et al., examines the usage of digital biomarkers from wearable devices to build predictive models for participants with bipolar disorder (BD). The 24 participants diagnosed with BD self-rated the Beck Depression Inventory (BDI) and Young Mania Rating Scale (YMRS) on a weekly basis. Each participant wore the wrist-watch-like sensor Garmin Vivosmart 4, which continuously measured and recorded the motor activity, sleep length, and heart rate 24 hours a day. With this data, the researchers then trained the depressive and manic models using Python, scikit-learn, and SHAP packages, and they adopted logistic regression, decision tree, k-nearest neighbors, random forest, adaptive boosting, and Extreme Gradient Boosting (XGBoost) to predict mood symptoms. After evaluating model performance, the prediction model for depressive symptoms achieved 83% accuracy, an area under the receiver operating characteristic curve (AUROC) of 0.89 and an F1-score of 0.65 on testing data, while the prediction model for manic symptoms achieved 91% accuracy, an AUROC of 0.88, and an F1-score of 0.25 on testing data. Therefore, the researchers concluded that digital biomarkers can be used to predict depressive and manic symptoms to help early detection of mood symptoms, as well as the treatment and recurrence of BD.
Outcomes and Implications
With the functional impairment and high disease burden of BD, this research is clinically important because it demonstrates the potential of digital biomarkers to predict early depressive and manic symptoms, allowing clinicians to use wearable devices as a means of passive, continuous real-world monitoring and data collection. Furthermore, the successful trial of including individualized features suggests this model can improve accuracy in diverse clinical populations by addressing heterogeneous physiological baselines. To conclude, the authors note that such algorithms that facilitate the early detection of depressive changes enables the timely introduction of adequate psychoeducation and clinical assessment, thus reducing the risk of recurrence. However, larger and more diverse samples or additional digital biomarkers are needed to enhance the predictive capabilities of the models before broad clinical implementation.