• woman checking fitness and health tracking

BLS: Bi-polar Longitudinal Study

  1. Goal:  Predict adverse events - mania/psychosis - prior to their occurrence by detecting changes in the behavior of at-risk patients.
  2. Data collection designs:
  • Smartphone sensor data collected continuously and unobtrusively/passively: GPS, accelerometer, call/text logs, screen on/off logs, WiFi, bluetooth, app usage, and gyroscope.
  • Analysis of some continuous covariates (hazard modeling) and others after discretization (change point detection) due to circadian behavioral patterns, so discretizing to the level of the day leads to most meaningful/interpretable measures.
  •  Unsupervised detection of change point prior to the occurrence of an adverse event under hypothesis that behavioral changes precede adverse events.
Project Person