Niels Jodts presented a demonstration of the patient trajectory analysis (PTRA) tool on data from multiple sclerosis patients. The work was a close collaboration with imec AIA (Exascience) and Biomed of UHasselt as part of the COOCK SaveData project.
PTRA is analytics software to provide temporal clustering of (long-term, sparse) longitudinal event data. An example would be recording the history of patient visits to a hospital physician or the history of medications. From this data, PTRA calculates statistically relevant pathways (trajectories) and a physician can quickly identify which different care pathways are possible for the patient.
PTRA is an unsupervised AI tool and is used to test existing hypotheses and, more importantly, to generate new ones. Once a domain expert analyzes the different paths, they can notice unexpected paths and thus discover that for example administering a certain drug for diabetes increases the risk of depression.
Once the pathways are calculated, they can help predict the pathway. Seeing which path a patient is on also predicts statistically relevant future elements. This can then form the basis for adjusting behavior.
PTRA is specifically designed to be scalable and calculate paths quickly and accurately on large data sets. For example, it takes 2 hours on 1 server to calculate paths on data from 273,000 patients and 6 million patient visits.
(from: savedata.be)
