| Topic： |
Seemingly consistent regression calibration estimator in survival analysis with exposure measurement error |
| Speaker： |
王清雲 教授 (Fred Hutchinson Cancer Research Center, USA) |
| Time： |
2015/05/29 (Fri.) AM 10:40 - 11:30 (Tea Time: AM 10:20 - 10:40, Room 821) |
| Place： |
GENⅢ 837 |
| Abstract： |
Observational epidemiological studies often confront the problem of estimating exposure-disease relationships when the exposure is not measured exactly. Regression calibration is a common approach to correct for bias due to covariate measurement error. It is well known that the regression calibration estimator is consistent under linear regression, but often there is bias under nonlinear regression or in censored outcome regression. In the paper, we investigate exposure measurement error in excess relative risk regression, which is a widely used model in radiation exposure effect research. The regression calibration is an approximation under this situation. However, our simulation results demonstrate that the finite sample bias of the estimator is very minimal. The bias of the regression calibration decreases when the sample size increases. The finite sample performance of the regression calibration estimator under this model is very much like a consistent estimator. This is completely different what the regression calibration would perform under either Cox regression or logistic regression. |
