What are the optimum quasi-identifiers to re-identify medical records?

Abstract

Recently, medical records are shared to online for a purpose of medical research and expert opinion. There is a problem with sharing the medical records. If someone knows the subject of the record by using various methods, it can result in an invasion of the patient's privacy. To solve the problem, it is important to carefully address the tradeoff between data sharing and privacy. For this reason, de-identification techniques are applicable to address the problem. However, de-identified data has a risk of re-identification. There are two problems with using de-identification techniques. First, de-identification techniques may damage data utility although it may decrease a risk of re-identification. Second, de-identified data can be re-identified from inference using background knowledge. The objective of this paper is to analyze the probability of re-identification according to inferable quasi-identifiers. We analyzed factors, inferable quasi-identifiers, which can be inferred from background knowledge. Then, we estimated the probability of re-identification from taking advantage of the factors. As a result, we determined the effect of the re-identification according to the type and the range of inferable quasi-identifiers. This paper contributes to a decision on de-identification target and level for protecting patient's privacy through a comparative analysis of the probability of re-identification according to the type and the range of inference.