I'm writing a retrospective study presenting an overall survival analysis comparing two groups of patients. I implement a Cox regression model with some covariates and then present the results in various tables. In addition, I provide a table with 1, 3, 5 and 10-year overall survivals. My data regard patients treated until 2014.

One of the reviewers asked to clarify how I dealt with patients for whom there are no available data for 5-year survival (those treated from 2010) - "How were these cases examined with regard to the survival analysis?"

I am not sure about what he is asking and I hadn't thought about such aspect. How can I manage the issue and answer the reviewer?


I have to wonder a bit whether the reviewer understands survival analysis.

A great advantage of Kaplan-Meier analysis or Cox regression is that these approaches address the issue raised by the reviewer. The drop in a Kaplan-Meier survival curve at the time of an event, or the contribution of an event to calculating hazard in Cox regression, depends on the total number of cases still at risk at the time of the event. A case followed up for only 4 years without an "event" provides no direct information about risks at 5 years, but it does contribute to information on 1- and 3-year survival. By improving the estimates of survival at shorter times, these censored (non-event) cases tend to diminish the uncertainty in survival estimates at later times.

The simplest answer would be that the survival estimates from the Cox regression incorporate information from all cases to some extent, but that the estimates of later-time survival necessarily become less precise as fewer cases are at risk at those times.

Plotting survival curves for the two groups, with indications of when cases were censored, can help . Indicating the numbers of cases still at risk at critical times also helps, and is often required by high-quality journals.


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.