Summary
The episode works through four questions that students commonly find difficult in time-to-event analysis: the choice of time zero, the reading of a Kaplan-Meier curve, the effect of competing events on one minus the Kaplan-Meier estimate, and the translation of a hazard ratio into risks. It uses the reanalysis of a Canadian study of Academy Award winners and a hypothetical cardiac rehabilitation cohort to show how immortal time can produce an apparent effect where none exists. Sarah computes a Kaplan-Meier estimate for a small hypothetical cohort with tied event and censoring times, and the hosts work a harder life-table example in which treating deaths as censoring overstates the risk of dementia. The hosts close by debating whether the hazard ratio should be the headline result of a time-to-event study, and they agree that it should be reported alongside survival curves and absolute risks at times fixed in the protocol.