Forecasting depression in bipolar disorder.

Area of Science:

• Psychiatry
• Computational Psychiatry
• Time Series Analysis

Background:

• Bipolar disorder involves recurrent mood episodes, impacting function and increasing suicide risk.
• Self-rated mood data collection via SMS offers a novel approach for monitoring bipolar disorder.
• Understanding mood fluctuations is crucial for effective bipolar disorder management.

Purpose of the Study:

• To forecast next-week depression ratings in bipolar disorder patients using self-rated mood data.
• To evaluate the effectiveness of different forecasting methods for bipolar disorder mood data.
• To assess the variability and predictability of depression time series in bipolar disorder.

Main Methods:

• Utilized SMS text messaging for collecting weekly self-rated mood data from bipolar disorder patients.
• Employed exponential smoothing and Gaussian process regression for time series forecasting.
• Compared forecasting accuracy against unconditional mean and persistence baselines.

Main Results:

• Significant heterogeneity observed in depression time series across patients, including variations in mean and correlation structure.
• Almost half of the time series were better predicted by the unconditional mean than by persistence.
• Neither exponential smoothing nor Gaussian process regression improved forecasting accuracy over a persistence baseline.

Conclusions:

• The high heterogeneity of bipolar disorder depression time series limits the accuracy of automated mood forecasting across diverse patient populations.
• The collected SMS mood data represents a valuable resource for future research in bipolar disorder.
• Further research is needed to develop clinically useful tools for automated mood forecasting in bipolar disorder.