Anonymity test attacks and vulnerability indicators for the “Patient characteristics” disclosure in medical articles

In the field of Privacy-Preserving Data Publishing (PPDP), a privacy violation attack based on a bias in the ratio of sensitive attribute values of disclosed information is called a homogeneity attack, and l-diversity has been proposed as an indicator of this vulnerability. In medical articles, especially in clinical trial, the ratio of attribute values is disclosed as “patient characteristics” which include statistical information such as the number of hypertension patients and age distribution of the patient group subject to clinical research. The patient characteristics could also be vulnerable to homogeneity attack but have not been studied. In this paper, we propose three new attack methods similar to the homogeneity attack that violate the anonymity of patient characteristics. We also propose three new indicators similar to l-diversity to evaluate anonymity against such attacks. Experimental results show that our new attacks can point out that actual patient characteristics leaks patient information that should be kept confidential. And the results also show that the new proposed indicators can measure the vulnerability to such attacks.