2019/08/19 by Liang Zhao, Zhiyuan Ma, Zhao, Liang +9 · 1 citation
Computer Science · Medicine · #Anomaly Detection Techniques and Applications #Machine Learning in Healthcare #Data-Driven Disease Surveillance
paper · pdf · doi:10.48550/arxiv.1908.07147
Electronic health record is an important source for clinical researches and\napplications, and errors inevitably occur in the data, which could lead to\nsevere damages to both patients and hospital services. One of such error is the\nmismatches between diagnoses and prescriptions, which we address as 'medication\nanomaly' in the paper, and clinicians used to manually identify and correct\nthem. With the development of machine learning techniques, researchers are able\nto train specific model for the task, but the process still requires expert\nknowledge to construct proper features, and few semantic relations are\nconsidered. In this paper, we propose a simple, yet effective detection method\nthat tackles the problem by detecting the semantic inconsistency between\ndiagnoses and prescriptions. Unlike traditional outlier or anomaly detection,\nthe scheme uses continuous bag of words to construct the semantic connection\nbetween specific central words and their surrounding context. The detection of\nmedication anomaly is transformed into identifying the least possible central\nword based on given context. To help distinguish the anomaly from normal\ncontext, we also incorporate a ranking accumulation strategy. The experiments\nwere conducted on two real hospital electronic medical records, and the topN\naccuracy of the proposed method increased by 3.91 to 10.91% and 0.68 to 2.13%\non the datasets, respectively, which is highly competitive to other traditional\nmachine learning-based approaches.\n