Wednesday, 22th February, 2017, 12:00 - 13.00
The widespread use of electronic health record (EHR) datasets has facilitated the massive collection of patient health information, thereby enabling researchers to conduct large-scale studies of comorbidities. The term comorbidity can be defined as the co-occurrence of two or more diseases within the same individual. The factor of time has, typically, not been taken into account in most of the relevant works. However, by incorporating the time dimension into a comorbidity study more complex disease patterns and their temporal characteristics can be revealed. In this work, a large-scale temporal comorbidity study is performed on a local (Catalonian) health database. The disease-history vectors of individual patients are compared between each other in order to extract common disease trajectories (i.e. shared by at least 2 patients). By using statistical-significance tests on the common disease trajectories of length=2, significant pairwise disease associations are identified and their temporal directionality is assessed. Subsequently, a novel unsupervised clustering algorithm, based on the Dynamic Time Warping (DTW) technique, is applied on all extracted common disease trajectories (length>=2), in order to group them according to the temporal patterns that they share. It will be shown that DTW can successfully cluster the disease-trajectory signals under investigation, which consist of various time scales and durations, although they do not exhibit any obvious temporal alignment. In this manner, important key clusters can be identified with trajectories that share the same time-dependent characteristics. A time-dependent comorbidity analysis is expected to facilitate the early diagnosis of a disease and prevent any adverse outcomes, by permitting the prediction of the disease progression along time.
Speaker: Alexia Giannoula, Integrative Biomedical Informatics, GRIB (IMIM/UPF)
Room Aula 473.10 (PRBB, 4th floor)