TABLE OF CONTENTS The discussion so far has given rather little consideration as to how censored observations may have come about. Indeed, because they seem to pose no practical difficulty for analysis, it is far too easy to give them inadequate consideration. Clearly censored observations coincident with a data-independent end of a trial pose no special source of concern, but this may not be the case for losses to follow-up occurring during the process of the trial. Losses may occur through patient noncompliance or inadequacies in the implementation of the trial. These raise all the issues and concerns relating to missing data, discussed in previous chapters. In other circumstances, losses may be disguised as failures, the result of some parallel survival process that cannot be suppressed during the period of the trial. Death from other causes than the focal cause is the typical example. On the assumption that this other process is conditionally independent of the process of interest, then these failures from other causes can be treated as providing censored observations from the process of interest.
Table 8.8 compares the Cox model estimates and standard errors for the effects of age, surgery and mismatch score on survival following heart-transplant for an all-causes outcome and for an outcome restricted to death from rejection (all other outcomes contributing censored observations). The data come from the much analysed Stanford Heart Transplant programme (Crowley and Hu, 1997). The increased estimate and significance for the effect of quality of transplant match, when...
