Warfarin Dose Estimation on High-dimensional and Incomplete Data

Warfarin is a widely used oral anticoagulant worldwide. However, due to the complex relationship between individual factors, it challenging estimate optimal warfarin dose give full play its ideal ef?cacy. Currently, there are plenty of studies using machine learning or deep techniques help with selection. But few them can resolve missing values and high-dimensional data naturally, that two main concerns when analyzing clinical real world data. In this work, we propose regard each patient’s record as set observed represent in an embedding space, enables our method learn from incomplete date directly avoid negative impact feature set. Then, novel neural network proposed combine embedded vectors non-linearly, capable capturing their correlations locating informative ones for prediction. After comparing baseline models on open source International Pharmacogenetics Consortium, experimental results demonstrate outperform others by signi?cant margin. further model performance different dosing subgroups, conclude has high application value clinical, especially patients high-dose medium-dose subgroups.