Evaluation of SIMEX Extrapolation Methods in Accelerated Failure Time Models With Covariate Measurement Error.

It is well known that ignoring measurement errors in covariates in the model leads to biased estimates. Various methods were proposed to address this issue. Simulation and extrapolation (SIMEX) method developed by (He et al. 2007) is a popular method due to its flexibility. It consists of two steps, simulation step and extrapolation step. Although it was investigated for applying to different settings, very little research was done on finding the best extrapolation function.. The objective of this study is to evaluate which extrapolation function gives best estimation in AFT models for data follows Weibull distribution. We use simulation studies to investigate the performance of three most popular extrapolation functions used for the SIMEX method. The results show linear extrapolation function is best for data with small measurement error, quadratic extrapolation function is best for the data with medium measurement error and nonlinear extrapolation function is best for the data with large measurement error. Then we applied these three extrapolation functions to a real data set patients with advanced lung cancer from the North Central Cancer Treatment Group to illustrate the usefulness of the research