This Article Full Text Full Text (PDF) All Versions of this Article: M2702v1 15/5/601    most recent Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Google Scholar Articles by Friedlin, F. J. Articles by McDonald, C. J. PubMed PubMed Citation Articles by Friedlin, F. J. Articles by McDonald, C. J.

A Software Tool for Removing Patient Identifying Information from Clinical Documents

Regenstrief Institute, Indianapolis, IN

^_* Correspondence: Jeff Friedlin, DO, Regenstrief Institute, Inc., Medical Informatics, Health Information and Translational Sciences (HITS) Building, 410 West 10th Street, Suite 2000, Indianapolis, IN 46202 (Email: jfriedlin{at}regenstrief.org).

Received for publication: 12/20/07; accepted for publication: 05/30/08.

We created a software tool that accurately removes all patient identifying information from various kinds of clinical data documents, including laboratory and narrative reports. We created the Medical De-identification System (MeDS), a software tool that de-identifies clinical documents, and performed 2 evaluations. Our first evaluation used 2,400 Health Level Seven (HL7) messages from 10 different HL7 message producers. After modifying the software based on the results of this first evaluation, we performed a second evaluation using 7,190 pathology report HL7 messages. We compared the results of MeDS de-identification process to a gold standard of human review to find identifying strings. For both evaluations, we calculated the number of successful scrubs, missed identifiers, and over-scrubs committed by MeDS and evaluated the readability and interpretability of the scrubbed messages. We categorized all missed identifiers into 3 groups: (1) complete HIPAA-specified identifiers, (2) HIPAA-specified identifier fragments, (3) non-HIPAA–specified identifiers (such as provider names and addresses). In the results of the first-pass evaluation, MeDS scrubbed 11,273 (99.06%) of the 11,380 HIPAA-specified identifiers and 38,095 (98.26%) of the 38,768 non-HIPAA–specified identifiers. In our second evaluation (status postmodification to the software), MeDS scrubbed 79,993 (99.47%) of the 80,418 HIPAA-specified identifiers and 12,689 (96.93%) of the 13,091 non-HIPAA–specified identifiers. Approximately 95% of scrubbed messages were both readable and interpretable. We conclude that MeDS successfully de-identified a wide range of medical documents from numerous sources and creates scrubbed reports that retain their interpretability, thereby maintaining their usefulness for research.