Adversarial Learning of Privacy-Preserving Text Representations for De-Identification of Medical Records

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5829–5839 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 5829 Adversarial Learning of Privacy-Preserving Text Representations for De-Identification of Medical Records Max Friedrich1, Arne K¨ohn2, Gregor Wiedemann1, Chris Biemann1 1Language Technology Group, Universit¨at Hamburg {2mfriedr,gwiedemann,biemann}@informatik.uni-hamburg.de 2Department of Language Science and Technology, Saarland University koehn@coli.uni-saarland.de Abstract De-identification is the task of detecting pro- tected health information (PHI) in medical text. It is a critical step in sanitizing electronic health records (EHRs) to be shared for re- search. Automatic de-identification classifiers can significantly speed up the sanitization pro- cess. However, obtaining a large and diverse dataset to train such a classifier that works well across many types of medical text poses a chal- lenge as privacy laws prohibit the sharing of raw medical records. We introduce a method to create privacy-preserving shareable repre- sentations of medical text (i.e. they contain no PHI) that does not require expensive man- ual pseudonymization. These representations can be shared between organizations to create unified datasets for training de-identification models. Our representation allows training a simple LSTM-CRF de-identification model to an F1 score of 97.4%, which is comparable to a strong baseline that exposes private infor- mation in its representation. A robust, widely available de-identification classifier based on our representation could potentially enable studies for which de-identification would oth- erwise be too costly. 1 Introduction Electronic health records (EHRs) are are valuable resource that could potentially be used in large- scale medical research (Botsis et al., 2010; Birk- head et al., 2015; Cowie et al., 2017). In addition to structured medical data, EHRs contain free-text patient notes that are a rich source of information (Jensen et al., 2012). However, due to privacy and data protection laws, medical records can only be shared and used for research if they are sani- tized to not include information potentially identi- fying patients. The PHI that may not be shared in- cludes potentially identifying information such as names, geographic identifiers, dates, and account numbers; the American Health Insurance Porta- bility Accountability Act1 (HIPAA, 1996) defines 18 categories of PHI. De-identification is the task of finding and labeling PHI in medical text as a step toward sanitization. As the information to be removed is very sensitive, sanitization always requires final human verification. Automatic de- identification labeling can however significantly speed up the process, as shown for other annota- tion tasks in e.g. Yimam (2015). Trying to create an automatic classifier for de- identification leads to a “chicken and egg prob- lem” (Uzuner et al., 2007): without a compre- hensive training set, an automatic de-identification classifier cannot be developed, but without ac- cess to automatic de-identification, it is difficult to share large corpora of medical text in a privacy- preserving way for research (including for train- ing the classifier itself). The standard method of data protection compliant sharing of training data for a de-identification classifier requires humans to pseudonymize protected information with substi- tutes in …

Content truncated to 500 words