Safe, effective and explainable drug recommendation based on medical data integration
Abstract: Medicine recommendation denotes the task of predicting drug combinations for patients’ therapies in case of complex diseases such as cancer or diabetes. These patients often follow a treatment consisting of multiple drugs simultaneously. Previous research works have already made predictions of next drug combinations based on the integration of the patients’ health records with an adverse drug–drug interaction (DDI) knowledge graph in order to minimize drug side effects. However, they missed to consider additional valuable information coming from synergistic drug–drug interaction knowledge graphs. In this paper, we integrate Electronic Health Record graph data incorporating patient, disease, therapy and drug information with either a synergistic or an adverse DDI knowledge graph to recommend safe and explainable medications. To the best of our knowledge, we are the first ones to compare three different medical data integration strategies based on the analysis stage (i.e. early, intermediate and late) at which integration takes place. By identifying those drugs that either complement each other or behave adversely, we are able to improve the efficacy of the patient’s therapy and/or minimize the toxicity and side effects. Moreover, we develop models to support doctors in comprehensively screening candidate drugs and their possible substitutes by providing also robust explanations alongside with the recommended medicine. We run experiments with three real-life medical data sets. In terms of suggesting safe and effective drugs, our proposed method suggests 34 times more synergistic drugs than the baseline algorithm for the cancer data set and reduces the unwanted side effects to patients’ medication by 2350 times more than the baseline algorithm for the MIMIC III data set. Our results demonstrate that we can assist doctors to prescribe effective, safe and explainable medication for the patients’ treatment.
Loading