A Hybrid Stepwise Approach for De-identifying Person Names in Clinical Documents
Óscar Ferrández, Brett R. South, Shuying Shen, Stéphane M. Meystre · 2012
As Electronic Health Records are growing exponentially along with large quantities of unstructured clinical information that could be used for research purposes, protecting patient privacy becomes a challenge that needs to be met. In this paper, we present a novel hybrid system designed to improve the current strategies used for person names de-identification. To overcome this task, our system comprises several components designed to accomplish two separate goals: 1) achieve the highest recall (no patient data can be exposed); and 2) create methods to filter out false positives. As a result, our system reached 92.6 % F2measure when de-identifying person names in Veteran’s Health Administration clinical notes, and considerably outperformed other existing “out-of-the-box ” de-identification or named entity recognition systems. 1