Cox Model with Covariate Measurement Error and Unknown Changepoint

Sarit Agami, David M. Zucker, Donna Spiegelman

Research output: Working paper/preprintPreprint


The standard Cox model in survival analysis assumes that the covariate effect is constant across the entire covariate domain. However, in many applications, there is interest in considering the possibility that the covariate of main interest is subject to a threshold effect: a change in the slope at a certain point within the covariate domain. Often, the value of this threshold is unknown and need to be estimated. In addition, often, the covariate of interest is not measured exactly, but rather is subject to some degree of measurement error. In this paper, we discuss estimation of the model parameters under an independent additive error model where the covariate of interesting is measured with error and the potential threshold value in this covariate is unknown. As in earlier work which discussed the case of konwn threshold, we study the performance of several bias correction methods: two versions of regression calibration (RC1 and RC2), two versions of the fitting a model for the induced relative risk (RR1 and RR2), maximum pseudo-partial likelihood estimator (MPPLE) and simulation-extrapolation (SIMEX). These correction methods are compared with the naive estimator. We develop the relevant theory, present a simulation study comparing the several correction methods, and illustrate the use of the bias correction methods in data from the Nurses Health Study (NHS) concerning the relationship between chronic air pollution exposure to particulate matter of diameter 10 $\mu$m or less (PM$_{10}$). The simulation results suggest that the best overall choice of bias correction method is either the RR2 method or the MPPLE method.
Original languageEnglish
StateSubmitted - 24 Aug 2018

Bibliographical note

26 pages. arXiv admin note: text overlap with arXiv:1808.07662


  • stat.AP


Dive into the research topics of 'Cox Model with Covariate Measurement Error and Unknown Changepoint'. Together they form a unique fingerprint.

Cite this