Diagnosis Prediction Using Knowledge Graphs

Abstract

Consultations between doctors and patients form the basis of the interaction between both parties, and lay the groundwork for administering appropriate treatment. Advances in machine learning, information, and communication technologies have enabled healthcare practitioners to enhance the manner in which data are captured and analysed during these information-rich meetings. The true potential of clinical data can only be realised if clinical data sources are synthesised in an appropriate data-representation and modelling approach. One such approach is the so-called knowledge graph (KG). The aim in this paper is to model consultation-related data in a KG and thereafter employ graph machine-learning techniques to identify missing links between entities in the graph through link prediction, thereby providing additional decision support to doctors. A case study data set comprising a list of patients, their respective conditions, and their medications forms the basis of the algorithmic analysis that is carried out.