Knowledge-based incremental induction of clinical algorithms

Abstract

The current approaches for the induction of medical procedural knowledge suffer from several drawbacks: the structures produced may not be explicit medical structures, they are only based on statistical measures that do not necessarily respect medical criteria which can be essential to guarantee medical correct structures, or they are not prepared to deal with the incremental arrival of new data.

In this thesis we propose a methodology to automatically induce medically correct clinical algorithms (CAs) from hospital databases. These CAs are represented according to the SDA knowledge model. The methodology considers relevant background knowledge and it is able to work in an incremental way.

The methodology has been tested in the domains of hypertension, diabetes mellitus and the comborbidity of both diseases. As a result, we propose a repository of background knowledge for these pathologies and provide the SDA diagrams obtained. Later analyses show that the results are medically correct and comprehensible when validated with health care professionals.