Evaluation of regression calibration and SIMEX methods in logistic regression when one of the predictors is subject to additive measurement error.

Abstract

Background: This paper presents an evaluation of two methods of measurement error adjustment based on recently-developed computer routines (RCAL and SIMEX) under logistic regression models, when one of the two predictors is subject to additive measurement error or Berkson error.

Methods: Computer simulations were used to generate data under a variety of conditions and the methods compared in terms of bias, mean squared error and confidence interval coverage of the regression estimates.

Results: Based on our investigations, RCAL was shown to perform very well in all situations considered, except in the presence of Berkson error when the predictor variables were highly correlated.

Conclusions: Since measurement error can lead to misleading inference, it is important to adjust for measurement error in the application of logistic regression. Until better measurement error adjustment methods become available, we recommend RCAL on the basis of our simulation results.