Analysis of Breast Cancer Survival Data with missing information on stage of disease and cause of death

摘要

Aim of this paper is to study whether social class is related to breast cancer survival, in a cohort of 4709 breast cancer patients diagnosed in Sweden in 1993 and followed until the end of 2001, while adjusting for possible demographics and tumor related confounders. The data are provided by the Swedish Cancer Registry and are matched to the death registry by using the unique Swedish Personal Registration Number. The statistical problem is that the most recent cases have not reported in the registry, as far as it concerns with the underlying cause of death, and standard cause specific survival analysis will turn to exclude those patients, then affecting our ability to detect any statistical difference in the effect of our covariate of interest. Furthermore, a related problem is that for some cases some important covariates (tumor stage) are missing, due the fact that the regional cancer registries have not provided the requested information. In this application simple missing data imputations have been incorporated into a standard survival data analysis problem, based on the estimation of the Kaplan-Meier estimator and Cox proportional hazards regression model. As the type of failure is truncated by time, imputing the cause of death will increase the follow-up time, therefore allowing to best study the survival distribution. Moreover, when also a confounder is missing completely at random, it is possible to detect the effect of the main exposure variable with more accuracy.