Estimating Survival Curves Under Proportional Hazards Model with Covariate Measurement Errors (Q4702178)

The authors propose a corrected cumulative baseline hazard estimator for the measurement error Cox model that is simple and easy to use. For a general additive measurement error model where the magnitude (standard deviation) of the measurement error \(\sigma^{2}\) is not assumed to go to zero, if a \(\beta\) estimator is consistent with an asymptotic normal distribution then the proposed estimator is also consistent and converges to a Gaussian process. One such consistent \(\beta\) estimator is the corrected maximum partial likelihood estimator (MPLE) proposed by \textit{T. Nakamura} [Biometrics 48, 829-838 (1992)]. The authors also derive corresponding modified estimators for the hazard functions and the survival functions of individuals with particular covariate values. Another purpose of this paper is to derive the confidence bands for both the cumulative hazard and survival functions when measurement error occurs, using a Monte Carlo technique developed by \textit{D.Y. Lin}, \textit{T.R. Fleming} and \textit{L.J. Wei} [Biometrika 81, 73-81 (1994)]. Some preliminary simulations performed by the authors indicate that the corrected confidence band indeed improves the naive one, even when \(\sigma^{2}\) is not that small.