Survival analysis with uncertain endpoints

In some survival analysis applications, the endpoint of interest has a degree of uncertainty associated with it. These events are typically classified by the investigator or by an endpoint committee as true or false according to some decision rule, and the analysis proceeds using only the true endpoints. This procedure has two drawbacks: The cut point for the decision rule is somewhat arbitrary, and the information contained in the level of certainty is lost. This paper introduces a modification of the Cox regression model that allows all potential endpoints to be included in the analysis along with the level of certainty of each. Simulation results show this procedure to considerably increase the power of the standard procedure in a wide range of situations.