Grant: Operating Grant

As more electronic health information is collected, there are increasing demands for disclosing this information for secondary purposes, such as health services research and public health. Privacy legislation allows the disclosure of health information without patient consent under specific circumstances, a key one being if the data is de-identified. The most common way to disclose data has been to de-identify it by, for example, generalizing the variables, suppressing records or values, and adding noise. However, these methods reduce the precision and overall quality of the data, limiting the kind of analysis that can be performed. It is possible for data custodians to disclose encrypted data only and allow the analysts to perform computations on the encrypted data. Existing secure multi-party computation protocols using homomorphic cryptosystems can be developed to allow data analysis without sharing any raw data or the disclosure of personal health information. The advantage of secure computation over traditional de-identification methods (such as generalization and suppression) is that the risk of re-identification is zero, and the results will not suffer from the lower precision introduced by generalization and suppression. However, secure computation can be slower than running the same analyses on de-identified data. Furthermore, many statistical analysis methods on encrypted data have not been developed yet. The purpose of the proposed project is to develop the secure computation protocols for common data analysis methods: descriptive statistics and general linear models. The performance of the protocol will be evaluated on typical data sets through an extensive simulation analysis. We will also develop a software platform that would implement this functionality and integrate it with SAS. The integration would allow end-users to work from a familiar environment, and hence simplifying its adoption.