Evaluation of Re-identification Risk for Anonymized Clinical Documents

Clinical data sharing for data transparency has potential to strengthen academic research, the practice of medicine and the integrity of clinical trial systems. This topic is becoming popular nowadays amongst leading pharmaceutical companies, and it raises concerns around the potential leak of personal health information of patients participating in clinical trials. The European Medicines Agency (EMA) has published phase 1 of their policy 0070 around requirements for anonymization of clinical documents for all studies submitted to EMA for market authorization, signaling a commitment towards data sharing, and increasing the urgency for addressing risk of potential reidentification. This paper will be discussing different scenarios in anonymization of clinical documents, implications of re-identification risk and the need for automation. INTRODUCTION It is not only EMA that is moving towards clinical data sharing but other associations/big pharmaceutical companies are trying to make data sharing mandatory. Pharmaceutical companies have started submitting anonymized clinical documents to the EMA with redaction of any text meeting the criteria of personal information or Commercially Confidential Information (CCI). Clinical documents in this paper is meant for all the reports that needs to be submitted under phase 1 policy of Policy 0070. In July 2013, Pharmaceutical Researchers and Manufacturers of America (PhRMA) & European Federation of Pharmaceutical Industries and Associations (EFPIA) member companies demonstrated commitment to share complete Clinical Study Reports (CSRs) along with Individual Patient Data (IPD) with qualified scientific and medical researchers, as necessary to conduct legitimate research. Recently in June 2017, International Committee of Medical Journal Editors (ICMJE) have introduced a requirement for data sharing statements for clinical trials before publishing any articles in journals abiding to ICMJE. However, in this paper, we shall be mostly interested in discussing about risk assessment methods for submission of clinical documents under EMA Policy 0070. Data redaction may lead to compromised data utility; hence the EMA is interested to approach to alternative techniques for data anonymization, including replacement, generalization, date offsetting, etc., with a more clearly established method for calculating quantitative risk of re-identification. This needs lots of efforts and resources to complete it manually. Industry needs an automated process of anonymizing clinical documents like study reports and Clinical Narratives. Anonymization methodologies must include a way of measuring re-identification risk and have a repeatable process to follow. Submissions done at EMA under phase 1 of policy 0070 have qualitative risk assessments. This paper will be discussing different ways of risk assessment methods suggested in literature for anonymized clinical documents.