Design and Analytic Considerations for Single-armed Studies Subject to Outcome Misclassification

• Activity Number: 110
• Type: Contributed
• Date/Time: Monday, August 12, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section*
• Author(s): Robert Lyles*+ and Hung-Mo Lin and John Williamson
• Affiliation(s): Rollins School of Public Health of Emory University and Pennsylvania State University and Centers for Disease Control and Prevention
• Keywords: design ; efficiency ; misclassification ; regression phenomenon ; selection bias

Abstract

Due to logistics or cost considerations, it is common for clinical studies to rely upon a potentially misclassified binary outcome variable for assessing an intervention effect. We consider nonrandomized studies, which are sometimes necessary for ethical reasons, and we focus on the situation in which subjects are selected to receive an intervention contingent upon a positive screening test. Both initial misclassification at screening and a regression effect impacting the error-prone follow-up outcome measure contribute to bias in the typical treatment effect estimate. We propose a study design involving the collection of internal validation data, assuming the availability of a more demanding gold standard outcome measure. In particular, we consider likelihood-based analysis and describe efficiency considerations relevant to two different treatment effect definitions. We identify four possible types of validation study observations and discuss finding the optimal allocation into these four types in order to minimize the variance of the estimated treatment effect. The optimal allocation appears highly dependent upon whether a ratio or a difference measure is adopted.