Assessing Uncertainty When Using Linked Administrative Records
Linking subjects in a planned medical study to records in administrative databases, such as electronic health records and Medicare claims data, can enable researchers to evaluate long-term outcomes, as well as outcomes not measured in the planned study, without expensive de novo primary data collections. This chapter reviews methods that attempt to account for uncertainty arising from imperfect record linkage. It begins with a high-level description of key sources of uncertainty. The chapter discusses approaches to handle uncertainty via statistical modeling. It focuses on problems involving record linkage of two files, although related ideas arise with more than two files and with de-duplication in one file. Finally, the chapter concludes with a description of some open problems and practical implications.