Personalized Prescription for Comorbidity

摘要

Personalized medicine (PM) aiming at tailoring medical treatment to individual patient is critical in guiding precision prescription. An important challenge for PM is comorbidity due to the complex interrelation of diseases, medications and individual characteristics of the patient. To address this, we study the problem of PM for comorbidity and propose a neural network framework Deep Personalized Prescription for Comorbidity (PPC). PPC exploits multi-source information from massive electronic medical records (EMRs), such as demographic information and laboratory indicators, to support personalized prescription. Patient-level, disease-level and drug-level representations are simultaneously learned and fused with a trilinear method to achieve personalized prescription for comorbidity. Experiments on a publicly real world EMRs dataset demonstrate PPC outperforms state-of-the-art works.