Learning to Classify Medical Discharge Summaries According to ICD-9.

CONTEXT: We present a post-hoc approach to improve the recall of ICD classification. METHOD: The proposed method can use any classifier as a backbone and aims to calibrate the number of codes returned per document. We test our approach on a new stratified split of the MIMIC-III dataset. RESULTS: When returning 18 codes on average per document we obtain a recall that is 20% better than a classic classification approach.

Patient Electronic Health Record as Temporal Graphs for Health Monitoring.

Machine learning methods are becoming increasingly popular to anticipate critical risks in patients under surveillance reducing the burden on caregivers. In this paper, we propose an original modeling that benefits of recent developments in Graph Convolutional Networks: a patient's journey is seen as a graph, where each node is an event and temporal proximities are represented by weighted directed edges. We evaluated this model to predict death at 24 hours on a real dataset and successfully compared our results with the state of the art.