Methods to Use Audit Data in Observational Studies

• Activity Number: 321
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Committee on Applied Statisticians
• Author(s): Bryan Shepherd*+ and Pamela Shaw
• Companies: Vanderbilt University and University of Pennsylvania
• Keywords: HIV ; measurement error ; missing data ; epidemiology ; audit

Abstract

A data audit involves verifying whether data sent to a coordinating center match those found in source documents. Data audits are common in clinical trials and are slowly being recognized as important for studies with observational data, where data errors are typically much more frequent. As the data coordinating center for a large multicenter HIV cohort, we have performed data audits, discovered errors, and then wondered what to do next. Should we request that the sites re-enter all of the error prone data? Or should we ignore the errors, or perhaps simply mention them as a potential limitation with our observational data? We propose a compromise: use information from the data audit to improve estimates using the entire database. We develop and apply statistical methods that are related to methods of measurement error or missing data. I will describe the statistical approaches we have incorporated, discuss some of their assumptions/properties, and demonstrate their use with data from the Caribbean, Central, and South American Network for HIV Research.