Trace Augmentation with Missing EHRs for Sepsis Treatments

Real-world Electronic Health Records (EHRs) are often plagued by limited data availability with missing values and small amount of traces. The limited data availability can extremely limit many downstream machine learning techniques that generally require complete data as input and prefer a great amount of data for training neural networks. Our study aims to provide robust and safe trace augmentation that can automatically induce high-quality and shareable data across research groups and institutions.