Section: New Results

Discriminant Chronicles for Care Pathway Analysis

Participants : Yann Dauxais, Thomas Guyet, David Gross-Amblard [Druid] , André Happe [Brest University Hospital] .

A care pathway is a sequence of events (drugs deliveries, hospitalisation, etc) extracted from medical databases (see section 4.3 for details). In some studies, each patient is labeled by a class (e.g. died or not died). This information can be taken into account for the discriminant analysis of care pathways. This year, our objective was to extract discriminant patterns from a dataset of care pathways that can discriminate patients on their labels. To this end we introduced the new task of discriminant chronicle mining. Conceptually, a chronicle is a graph whose vertices are events and edges represent quantitative time constraints between events. We also proposed $DCM$, an algorithm dedicated to discriminant chronicles mining. This algorithm is based on rule learning methods to extract the temporal constraints. Computational performances and discriminant power of extracted chronicles are evaluated on artificial and real data.

The paper describing this work has been accepted in the french national conference on data mining (EGC 2017) [4] and is nominated for the best paper award.