Oncology Drug Regimen Prediction using Knowledge Graph Completion with Real-World Clinical Database

Abstract

Combining different drugs for cancer treatment is intended to improve therapeutic efficacy and manage adverse events by using drugs with different mechanisms of action. In this study, we constructed a knowledge graph based on drug combinations in actual clinical practice and examined the possibility of proposing new regimens through knowledge graph completion. Medical claims database provided by Medical Data Vision Co. Ltd. (MDV) was used as a data source and constructed a knowledge graph using the nodes of Drug, Regimen and Disease. The relations between nodes were predicted using knowledge graph embedding models through link prediction. As a result, the Drug-Regimen relations to be linked were predicted to be among the top combinations, suggesting that new drug regimens could be predicted using knowledge graph.