Bias When Using an Interval Endpoint in Time-to-Event Analysis in a Longitudinal Study

• Activity Number: 38
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): David Couper*+
• Companies: The University of North Carolina at Chapel Hill
• Keywords: interval-censoring ; longitudinal study ; time-to-event analysis

Abstract

For some outcomes in longitudinal studies, the outcome is determined at least in part by measurements taken at study visits. The event of interest may have occurred between the visits but ascertainment occurs at a visit. An example is onset of diabetes, where one determinant of diabetic status is the fasting glucose level in blood drawn at the time of a study visit. Time-to-event methods are often used to investigate associations between putative exposures and incident events. It is known that defining the event time as the endpoint of the interval in which it occurred may yield biased estimates, so that an interval-censoring approach to the analysis is preferred. However, some studies involve complex sampling designs, for which appropriate interval-censoring methods are not available in standard software. We present the results of simulations investigating the effect of the inter-visit interval length on the degree of bias and the use of auxiliary information to reduce potential biases.