Predicting Affective Episodes in Bipolar Disorder Using Statistical Process Control Analysis of GPS-Based Mobility...

Area of Science:

• Digital health
• Psychiatry
• Data science

Background:

• Bipolar disorders (BD) cause severe affective episodes impacting quality of life.
• Early prediction of BD episodes is challenging.
• Mobile sensing and digital phenotypes, like geolocation, offer new detection possibilities.

Purpose of the Study:

• To assess if spatial exploratory behavior, via GPS, predicts depressive and manic episodes in BD.
• To evaluate unique places visited and mobility metrics for predicting prodromal states and ongoing episodes.
• To apply statistical process control (SPC) for deviation detection.

Main Methods:

• Utilized the BipoSense dataset with high-resolution GPS data.
• Applied Density-Based Spatial Clustering to extract mobility indicators (unique places, location changes, time per location).
• Implemented exponentially weighted moving average (EWMA)-based SPC to detect deviations from individual baselines.

Main Results:

• Median time spent at locations best indicated depressive and (hypo)manic episodes.
• Number of unique clusters visited did not significantly correlate with phase transitions.
• EWMA-SPC detected behavioral deviations, but no single indicator consistently met high sensitivity and specificity thresholds for prediction.
• Optimized SPC improved performance, yet individual indicators lacked robust predictive accuracy for prodromal or acute episodes.

Conclusions:

• While unique place data alone is insufficient, EWMA-SPC applied to GPS data shows promise for digital phenotypes.
• Current predictive accuracy for upcoming episodes is not robust, but the framework is promising for individualized monitoring.
• Further research is essential to refine digital biomarkers and validate their clinical utility in managing BD phases.