Multifractal dynamics of activity data in Bipolar Disorder: Towards automated early warning of manic relapse

Predicting the onset of a manic episode so that timely medical intervention can occur is an important issue in the long-term management of people diagnosed with Bipolar I Disorder. In this proof-of-concept in vestigation of a single case, activity time series acquired from a wrist-worn actigraph were used to detect quantitative precursors of a manic episode that resulted in the hospitalization of a young man diagnosed with the disorder at the end of week 14. Using transformed activity data, a multifractal analysis showed that the data were significantly multifractal as indicated by the lack of overlap between Hurst functions computed from the data and surrogate series. When a continuous Shannon entropy measure derived from a log normal distribution was fit to the multifractal spectra for each of the 14 weeks of data, a decrease in entropy of practical significance was observed after week 12. If this newly discovered indicator of relapse had been used to initiate preventative treatment, it is possible that hospitalisation might have been avoided. As this analysis is based on a single case study and just one episode , larger samples will need to be prospectively monitored into manic and depressive relapses to generalize the proposed methodology and permit its widespread usage for automated episode monitoring and prevention.