This study applies social network analysis to predict phase transitions of patients with bipolar affective disorder (manic depressive).
During a manic period, which includes a high degree of social activities, some patients tend to drop medication. When mood swings from manic to depressive these patients are endangered committing suicide. Timely prediction of mood swings and adjustment of medication could potentially safe patient's lifes. We analyse phase-specific patterns of social network activities (communication, in-person-meetings, moving to specific locations, social network dynamics) and derive phase transition predictors (early warners). For data collection by mobile smartphones we developed SIMBA, a social activity monitoring system, which is based on MIT's funf sensing framework (Aharony et al. 2011). The resulting overlapping networks based on communication, personal contact and location (two-mode) are analysed and SNA methods are used for clustering and evaluation. Predictors are derived by a Gaussian mixture model.
An expectation-maximization algorithm is used for parameter estimation. The resulting phase transition predictors are used as signals for attending physicians and psychiatrists.