On Graph Construction for Classification of Clinical Trials Protocols Using Graph Neural Networks

A recent trend in health-related machine learning proposes the use of Graph Neural Networks (GNN’s) to model biomedical data. This is justified due complexity healthcare data and modelling power graph abstractions. Thus, GNN’s emerge as natural choice learn from increasing amounts While formulating problem, however, there are usually multiple design choices decisions that can affect final performance. In this work, we focus on Clinical Trial (CT) protocols consisting hierarchical documents, containing free text well medical codes terms, a classifier predict each CT protocol termination risk “low” or “high”. We show while using solve classification task very successful, way constructed also importance one benefit making priori useful information more explicit. consider an independent pose problem classification, consistent performance improvements be achieved by considering them super-nodes unified connecting according some metadata, like similar condition intervention, finally approaching node rather than classification. validate hypothesis experimentally large-scale manually labeled database. provides insights flexibility graph-based modeling for domain.