An Analysis of Attention over Clinical Notes for Predictive Tasks

摘要: The shift to electronic medical records (EMRs) has engendered research into machine learning and natural language technologies analyze patient records, predict from these clinical outcomes of interest. Two observations motivate our aims here. First, unstructured notes contained within EMR often contain key information, hence should be exploited by models. Second, while strong predictive performance is important, interpretability models perhaps equally so for applications in this domain. Together, points suggest that neural may benefit incorporation attention over notes, which one hope will both yield gains afford transparency predictions. In work we perform experiments explore question using two corpora four different tasks, that: (i) inclusion mechanisms critical encoder modules operate fields order competitive performance, but, (ii) unfortunately, boost it decidedly less clear whether they provide meaningful support