Fig. 1.

Distributional survival analysis data generation mechanisms. (a) In the proposed event-conditional censoring model (CenTime), $t$ is drawn from the death time distribution and $c$ is uniformly sampled up to $t$. (b) In the classical model, $t$ and $c$ represent randomly drawn death and censoring times from the corresponding distributions. If $c<t$, the patient is censored and the observation is the censoring time. Otherwise, the patient is uncensored and observation is the death time.