Median survival not reached

[Tore Sand]

Hello fellow statalisters,

I would like to ask the community for some help regarding a survival analysis problem I'm facing;

I’m doing a survival analysis about tumor patients, this specific tumor type has generally a favorable prognosis, i.e. patients live 12-15 years after diagnosis. At the same time, follow-up in the available records is limited due to patients moving away etc., so median follow-up time is around 8 years.

This means that, with the event (death) not being encountered so often in the dataset, I can’t calculate median overall survival time for my cohort (and the respective disease subgroups).

STATA has the rmean option under the stci command. Based on what I've read online it’s primarily used in cases where the proportionality assumption is not fulfilled. Is it statistically sound to use restricted mean survival times when median times have not been reached?

Thanks in advance!
Tore

[Clyde Schechter]

Most survival distributions are right skew, often heavily so. If you haven't reached the median yet, you are not even close to reaching the mean. Restricted mean survival times restricted to a short fraction of the expected lifespan can still be useful measures of the short-term death process, but should not be interpreted as informative about the overall long-term nature of the process. So you can use them, but just understand their limitations and don't draw unwarranted conclusions from them.

Other things you might consider in this circumstance is reporting the 25th percentiles. Or, given that you have 8 years of follow-up you might report 5 and 7 year survival probabilities.

If you are contrasting the (short-term) survival outcomes of different groups within your data, you might still use the usual apparatus like Kaplan-Meier or Cox regression if the proportional hazards assumption holds over the time period of your data. It doesn't matter, for this purpose, whether you have reached median survival or not. Although, again, you are only getting information about the most immediate aspects of the death process.

[Tore Sand]

Thank you Dr. Schechter for the informative reply!

I will go the route of reporting 5/8-year survival probabilities. I am also using a Cox model for subgroup comparison, but wanted the median survival times for my Table 1.

To be honest, the restricted mean survival times I get do look pretty plausible, knowing the natural history of the tumor.

Thanks again for the feedback!

Tore