HitzalMed: Anonymisation of Clinical Text in Spanish

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 7038–7043 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 7038 HITZALMED: Anonymisation of Clinical Text in Spanish Salvador Lima, Naiara Perez, Laura Garc´ıa-Sardi˜na, Montse Cuadros SNLT group at Vicomtech Foundation, Basque Research and Technology Alliance (BRTA) Donostia/San-Sebasti´an, 20009, Spain {slima, nperez, lgarcias, mcuadros}@vicomtech.org Abstract HITZALMED is a web-framed tool that performs automatic detection of sensitive information in clinical texts using machine learning algorithms reported to be competitive for the task. Moreover, once sensitive information is detected, different anonymisation techniques are implemented that are configurable by the user –for instance, substitution, where sensitive items are replaced by same category text in an effort to generate a new document that looks as natural as the original one. The tool is able to get data from different document formats and outputs downloadable anonymised data. This paper presents the anonymisation and substitution technology and the demonstrator which is publicly available at https://snlt.vicomtech.org/hitzalmed. Keywords: clinical data, sensitive data detection, anonymisation 1. Introduction Data has become an invaluable resource for both research and commercial purposes. When it comes to data contain- ing personal information, such as health records, there is an ethical and legal responsibility towards respecting the individuals’ privacy. This has led to the introduction of specific laws that address this issue, such as the European Union’s General Data Protection Regulation (GDPR) direc- tive or the United States’ Health Insurance Portability and Accountability Act (HIPAA). A possible solution to the privacy problem is anonymisa- tion. Anonymisation can be defined as the process of re- moving all information from a document that could po- tentially point to a given person, such as names, phone numbers or e-mail addresses. The resulting documents al- low us to use real data that cannot be linked to a real per- son. Medlock (2006) describes three different approaches to anonymisation: removal (‘replacing a reference with a blank place-holder’), categorisation (‘replacing a reference with a label in some way representing its type or category’) and pseudonymisation (‘replacing a reference with a variant of the same type’). Manual anonymisation is a tiresome and expensive process. For this reason, considerable efforts are being made to au- tomatise the task. This is straightforward in the case of structured text or tabular data, but the task becomes con- siderably more challenging when dealing with unstructured natural language. This article presents HITZALMED1, a web-framed tool that assists with the anonymisation of clinical free text in Spanish. HITZALMED uses a hybrid approach that com- bines Machine Learning (ML) techniques to detect Pro- tected Health Information (PHI) and a more traditional rule- based system for their de-identification. The main features of HITZALMED are presented below: • It supports multiple document formats • It features an automatic PHI recogniser and classifier that has proven to be competitive in the MEDDOCAN: 1https://snlt.vicomtech.org/hitzalmed Medical Document Anonymization (Marimon et al., 2019) shared task, having achieved F1-scores higher than 0.95 • The classifier distinguishes 21 fine-grain …

Content truncated to 500 words