C-D3-01: Developing and Implementation of Secure Linear Regression on Distributed Databases

Abstract

Background/Aims: Improving access to multiple sources of electronic health information is crucial for population-based health care research. But it also raises serious issues relating to data confidentiality and security. We propose to develop methods and tools (e.g., SAS macros) to enable multivariate analyses to be conducted using multiple datasets that are stored separately without exposing any patient level information. Phase I of this project focuses on linear regression. Methods: A two-step process is carried out by SAS macros. The macros will be installed on both the central and site computers that hold the analytic datasets. After the central computer initiates the process, each site computer executes the SAS program locally and submits intermediate summarized statistical results to the central computer. The central computer then combines the intermediate results, and computes the parameter estimates. A second round of program execution delivers the variance/covariance estimates of the parameter estimates. Results and Conclusions: Simulated separated data sets were used to test the macros. The results were identical to the outputs from the SAS PROC REG procedure when the data sets were combined. The macro enables remote execution of multivariate linear regression analysis within the context of a multi- institutional distributed research network. The successful implementation of the secure statistical analysis within a distributed database model will overcome one of the largest hurdles in fully developing such a model, namely, the current need to combine data for multivariate analyses. Obviating this need will save the time and resources currently spent developing data-sharing agreements and will greatly minimize the patient confidentiality and proprietary concerns