Representation Learning for Electronic Health Records: A Survey

Abstract

With the wide application of Electronic Health Record (EHR) in hospitals in past few decades, researches that employ artificial intelligence (AI) and machine learning methods based on EHR data have been explosive. With such EHR data, one can engage in many tasks such as risk prediction, treatment recommendation, information imputation, etc. The performance of classification or prediction highly depends on the quality of data representation, i.e., representing original records into numerical vectors to facilitate further learning. However, there is little research that focuses on the representation learning techniques for EHR data at present, which makes it hard to understanding the development trend of EHR learning in a global map. In this paper, we bridge this gap by systematically investigating the related research efforts that apply the representation learning on EHR data. We analyze and conclude the techniques used in the typical representation learning approaches as well as the limitations and advantages of them. The survey would provide a comprehensive reference for further analysis and application in EHR research.