It is known from previous works (Segão et al.,1998) that the indicator variable of distant metastasis is one of the most important prognostic factors in the lung cancer. By its own nature, the registry of this covariate is not always correct, and in some cases it happens that understaging of registry errors is observed (Segão et al.,1998 and Gong et al.,1990).
Under the assumption that, the true value of the indicator variable of distant metastasis is related to the survival time according to Cox proportional hazards model, several approaches were considered to estimate the relative risk (Segão, 1999 and Segão and Turkman, 2000a and 2000b). The approaches considered were based on an exponential model with misclassification, the idea of sensitivity in regression (Lin et al., 1998), the method of regression calibration (Carrol et al.,1995) and the robust estimator developed by Bednarski (1993). The proposed estimators were applied to a lung cancer data and a simulation study was made in order to evaluate their behaviour.

keywords: ll Censoring, classification error, Cox proportional hazards model, hazard function, metastasis, relative risk, survival time.