Distantly supervised end-to-end medical entity extraction from electronic health records with human-level quality

Medical entity extraction (EE) is a standard procedure used as a first stage\nin medical texts processing. Usually Medical EE is a two-step process: named\nentity recognition (NER) and named entity normalization (NEN). We propose a\nnovel method of doing medical EE from electronic health records (EHR) as a\nsingle-step multi-label classification task by fine-tuning a transformer model\npretrained on a large EHR dataset. Our model is trained end-to-end in an\ndistantly supervised manner using targets automatically extracted from medical\nknowledge base. We show that our model learns to generalize for entities that\nare present frequently enough, achieving human-level classification quality for\nmost frequent entities. Our work demonstrates that medical entity extraction\ncan be done end-to-end without human supervision and with human quality given\nthe availability of a large enough amount of unlabeled EHR and a medical\nknowledge base.\n