Abstract
It is expected that medical information stored in hospitals are collected and published in order to improve the quality of medical services. Because medical information contains privacy relevant information, it is necessary to anonymize it to prevent revealing sensitive patient information. There is well-known anonymization method, named k–anonymization, which ensures that the probability of user identification is at most 1/k. However, existing methods of k–anonymization are considered in cases of anonymizing a dataset which stored in a provider. In practice, there are many cases of anonymizing a dataset which is combined from datasets stored in some providers. If an existing k–anonymization method performs on the combined dataset, there is a risk that the probability of user identification might be more than 1/k by comparing an original dataset of a provider and the anonymized dataset. Therefore, this paper proposes a new indicator, named k–site–anonymity, which represents the probability of identifying a user with consideration from a view of an attacker in a provider. Also, this paper proposes a new anonymization method which satisfies k–site–anonymity and evaluates it on a real rezept dataset.
