Bias-Corrected Estimates of the Longitudinal Slope Association with Baseline When the Outcomes Are Measured with Error

• Activity Number: 475
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Zhanghua Chen*+ and Anny H Xiang
• Companies: University of Southern California and Kaiser Permanente
• Keywords: measurement error ; baseline prediction ; longitudinal slope

Abstract

Mixed effects models are often used to evaluate the association between baseline and longitudinal change in repeated measures over time. Random measurement error in the observed outcome can bias the estimate of the true association. We developed bias correction formulae for the parameter estimate of the interaction between follow-up time and baseline, representing the extent to which the difference of slopes of repeated measures varies by baseline. Bias-corrected estimates for models with and without baseline as the dependent variable were derived using OLS estimates of two-stage models reconstructed from the mixed effects models. The correction formulae were functions of the naïve estimates, follow-up time, and the ratio of measurement error variance to observed total variance. Validity of the formulae was evaluated through simulation studies. Inference tests of the bias-corrected estimates were conducted by bootstrapping. Using a real dataset, application of the naïve analysis resulted in an un-interpretable, significantly negative association between the baseline and the slope of repeated measures, whereas the bias-corrected estimates demonstrated no significant association.