Discovering Quantitative Temporal Functional Dependencies on Clinical Data

Approximate functional dependencies, even with suitable temporal extensions, have been recently proposed as a methodological tool for mining clinical data. It allows healthcare stakeholders to derive new knowledge from overwhelming amount of healthcare and clinical data. Some examples of the kind of knowledge derivable from data through dependencies may be "month by month, patients with the same symptoms get the same type of therapy" or "within 15 days, patients with the same diagnosis and the same therapy receive the same daily amount of drug". The main limitation of such kind of dependencies is that they cannot deal with quantitative data, when some tolerance can be allowed for numerical values. In particular, such limitation arises in clinical data warehouses, where analysis and mining have to consider one or more measures (related to quantitative data as lab test results, vital signs as blood pressures, temperature and so on), with respect to many dimensional (alphanumeric) attributes (as patient, hospital, physician, diagnosis) and to some time dimensions (as the day since hospitalization, the calendar date, and so on). According to this scenario, we introduce here a new kind of approximate temporal functional dependency, named multi approximate temporal functional dependency (MATFD), which consider dependencies between dimensions and quantitative measures from temporal clinical data. Such new dependencies may provide new knowledge as "within 15 days, patients with the same diagnosis and the same therapy receive a daily amount of drug within a fixed range". Moreover, we provide an original algorithm to mine such kind of dependencies and to derive some core dependencies, both for the discovered temporal window and for the involved dimensional attributes. Finally, we discuss some first results we obtained by pre-processing and mining ICU data from MIMIC III database.