Monte Carlo Correction for Misclassification with Imperfect Internal Validation Data

• Activity Number: 373
• Type: Invited
• Date/Time: Wednesday, August 10, 2005 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Timothy L. Lash*+
• Companies: Boston University
• Keywords: Misclassification ; Internal validation ; Monte Carlo

Abstract

Established methods allow correction of effect estimates to account for misclassification of variables in the analysis. With internal validation data, corrections yield a revised estimate and interval that account for sampling error and the binomial error in the bias parameters . Similarly, Monte Carlo methods using external validation data yield a revised estimate and interval that account for sampling error and the uncertainty in the extrapolation of the bias parameters to a second population. The authors extend these Monte Carlo methods to studies with internal validation data that are measured with greater uncertainty than the binomial error. For example, inter-rater reliability data use no true gold standard, so are not truly validation data. In this paper, the authors illustrate the method with an application to a study of breast cancer patients for which data collection was by medical record review. Both inter-rater reliability data and an adjudicated gold standard were available. The authors compare the conventional estimate of effect and its interval with the estimates and intervals corrected with the inter-rater reliability data and with the gold standard.