Just looking for some clarity regarding time-to-event (death) data when there is a large proportion of right censored data (lots of survival). Is this a serious problem for a proportional hazards model? The data are basically left skewed because of the censored data where you might otherwise expect right-skewed time-to-event data.



In terms of estimation, both the Cox model and parametric models that belong to the proportional hazards (PH) family will not be affected by a high proportion of censored observations.

The reason is that the likelihood (or partial likelihood for Cox) is a cumulative product where only individuals who experience the event provide a contribution. Censored observations only provide a passive contribution in the definition of the individuals at risk for each event time.

In terms of interpretation, you will have to specify that the HR you are presenting summarizes a given comparison within a specific window of time where only p% of events were observed. In practice, let's say that in one year you have more events and redo the analyses, the HR will likely change (even under PH being valid).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.