An Unexpected Connection from Our Personalized Medicine Approach to Bipolar Depression Forecasting

Abstract

As one of the most complicated and recurrent depressive disorders, bipolar depression holds the highest morbidity and high mortality risk, but effective early detection and appropriately targeted treatments are still missing. This requires a new innovative approach, one capable of forecasting of mood states, in particular manic one. In our recent work, we combined several data sources to extract the most relevant variables, describe its intrinsic dynamics by network-flow analysis, and apply several supervised machine learning models to predict mania in BDD. By applying several methods of extracting and selecting the features from those aggregated data, and consequently performed supervised machine learning we arrived at real personalized medicine approach to BDD forecasting. Here we are interpreting previously unpublished data on sleeprelated variables and its possible relation with irritability that was the most promising variable from daily self-report data. By putting this connection in the perspective of other recent neuroimaging and biochemical findings we are elucidating on another most important factor, namely the reason why some antidepressants shown to disrupt sleep dynamics can exacerbate the tipping point to mania, via the already mentioned link between sleep-related variables and irritability that our research demonstrated to be of most valuable predictable power.