JMIR Publications

ABSTRACT

Background:

Bipolar disorder is a prevalent mental disease imposing a high societal burden. Accurate forecasting of symptom scores can be used to improve disease monitoring, enable early intervention and eventually help prevent costly hospitalizations. While several studies have examined the use of smartphone data to detect mood, only few studies deal with forecasting mood one or more days ahead of time.

Objective:

The objective of this work is to examine the feasibility of forecasting daily subjective mood based on daily self-assessments collected from bipolar disorder patients via a smartphone-based system in a randomized clinical trial.

Methods:

We apply hierarchical Bayesian regression models, a multi-task learning method, to account for individual differences and forecast mood up to seven days ahead based on 15,975 smartphone self-assessments from 84 bipolar disorder patients participating in a randomized clinical trial. We report the results of two time-series cross-validation one day ahead prediction experiments corresponding to two different real-world scenarios and compare the outcomes to commonly used baselines methods. We then apply the best model to evaluate a seven-day forecast.

Results:

The best performing model used a history of 4 days of self-assessments to predict future mood scores with historical mood being the most important predictor variable. The proposed hierarchical Bayesian regression model outperformed pooled and separate models in a one-day forecast time series cross-validation experiment and achieved predicted R^2=0.51 and RMSE=0.32 for mood scores on a scale of -3 to 3. When increasing the forecast horizon, forecast errors also increase and the forecast regresses towards the mean of the data distribution.

Conclusions:

We found that our proposed method can forecast mood several days ahead with low error compared to common baseline methods. The applicability of a mood forecast in clinical treatment of bipolar disorder is also discussed.