Abstract:
|
For time-to-event outcomes, a rich literature exists on the bias introduced by covariate measurement error in regression models and the methods of analysis to address this bias. One popular, easy to implement method is SIMEX, which adjusts for additive covariate measurement error. Less attention has been given to understanding the impact of or addressing errors in a failure time outcome. For linear models, it is well known that non-differential errors in the outcome do not bias regression estimates. However, the same cannot be said for non-linear outcomes. We introduce an extension of SIMEX to correct for bias in the Cox model. Detailed numerical studies are presented to examine the performance of SIMEX under varying distributions and levels of the outcome error. In these numerical studies, SIMEX outperforms the naive method of ignoring the error, with smaller MSE in all cases, and kept the bias under 10-15% for moderate error. Even for dramatically skewed error, SIMEX was an improvement, generally halving the bias. We consider other analysis options to address error in the survival outcome and conclude with general recommendations.
|