Measuring Risk and Information Preservation: Toward New Metrics for De-identification of Clinical Texts

Current metrics for de-identification are based on information extraction metrics, and do not address the real-world questions “how good are current systems”, and “how good do they need to be”. Metrics are needed that quantify both the risk of re-identification and information preservation. We review the challenges in de-identifying clinical texts and the current metrics for assessing clinical de-identification systems. We then introduce three areas to explore that can lead to metrics that quantify reidentification risk and information preservation.