Priority-Based k-Anonymity Accomplished by Weighted Generalisation Structures

Abstract

Biobanks are gaining in importance by storing large collections of patient’s clinical data (e.g. disease history, laboratory parameters, diagnosis, life style) together with biological materials such as tissue samples, blood or other body fluids. When releasing these patient-specific data for medical studies privacy protection has to be guaranteed for ethical and legal reasons. k-anonymity may be used to ensure privacy by generalising and suppressing attributes in order to release sufficient data twins that mask patients’ identities. However, data transformation techniques like generalisation may produce anonymised data unusable for medical studies because some attributes become too coarse-grained. We propose a priority-driven anonymisation technique that allows to specify the degree of acceptable information loss for each attribute separately. We use generalisation and suppression of attributes together with a weighting-scheme for quantifying generalisation steps. Our approach handles both numerical and categorical attributes and provides a data anonymisation based on priorities and weights. The anonymisation algorithm described in this paper has been implemented and tested on a carcinoma data set. We discuss some general privacy protecting methods for medical data and show some medical-relevant use cases that benefit from our anonymisation technique.