Multiple Imputation to Correct for Covariate Measurement Error Based on Summary Statistics from External Calibration Data

• Activity Number: 177
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ying Guo*+ and Rod Little
• Companies: Merck & Co., Inc. and University of Michigan
• Keywords: calibration data ; measurement error ; multiple imputation ; regression calibration

Abstract

Covariate measurement error is very common in empirical studies, and currently information about measurement error provided from calibration samples is insufficient to provide valid adjusted inferences. We consider the problem of estimating the regression of outcomes Y on covariates X and Z, where Y and Z are observed, X is unobserved, but a proxy variable W that measures X with error is observed. Data on the joint distribution of X and W (but not Y and Z) are recorded in a calibration experiment. The data from this experiment are not available to the analyst, but summary statistics for the joint distribution of X and W are provided. We describe a new multiple imputation (MI) method that provides multiple imputations of the missing values of X in the regression sample, so that the regression of Y on X and Z and associated standard errors are estimated correctly using multiple imputation (MI) combining rules, under normal assumptions. The proposed method is shown by simulation to provide better inferences than existing methods, namely the naïve method and regression calibration, particularly for correction for bias and achieving nominal confidence levels.