Evaluation and Limitations of Disclosure-Treated Health Data Using Random Substitution and Sub-sampling

• Activity Number: 332
• Type: Topic Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): Joshua Michael Borton*+ and Peter K. Kwok and Avinash C. Singh
• Companies: NORC at the The University of Chicago and NORC at the The University of Chicago and NORC at the The University of Chicago
• Keywords: Statistical disclosure limitation ; inside intrusion ; de-identification ; stochastic treatment ; disclosure risk measure ; data utility

Abstract

In order for medical data, such as the Medicare Claims Data, to be shared for research purposes, the data provider must be certain that the risk of re-identification of such records is sufficiently low. Statistical disclosure limitation (SDL) is the process of reducing disclosure risk. In order for the data to be useful to researchers, the chosen SDL method should have as little impact as possible on inferences made from the data. It is the constant tension between data confidentiality and data quality that is at the heart of all decisions regarding SDL. We present a method by which both data utility and data confidentiality are evaluated throughout the process. This information is used to determine the amount of random substitution and subsampling that should be used as part of our SDL process. Application of the method to the claims data is discussed.