An evaluation of kernel smoothing to protect the confidentiality of individual locations

With advances in spatial data management technologies, accurate geographic information about individual patients increasingly has become available. Researchers should protect the privacy of patients, which includes their locational information, in public health data analyses. Protecting privacy involves a trade-off between information loss and disclosure risk. Estimation of a kernel density surface commonly has been used to mask confidential point locations. However, the literature lacks an extensive discussion of reverse transformations from a kernel density estimation surface to points, and evaluations of recovered points compared to their original point counterparts. This paper presents a method to recover relatively precise point locations from a kernel density estimation surface using geometric centres of clusters, and evaluates recovered points in terms of protecting locational privacy and maintaining locational accuracy. An application illustrates this method utilizing late-stage colorectal cancer points in the Pensacola metropolitan statistical area, Florida that examines various kernel density estimation surfaces with different bandwidths and cell sizes.