I have the survival dataset of a population with a special disease. I´d like to compare this population with the general population to see whether this population has a decreased life-expectancy overall. What I had in mind was to create a control for each patient in the dataset and enter the age, sex and cohort specific life-expectancy from the national statistics databank and just run a Kaplan-Meier analysis.

However, I´m unsure as to how I should deal with the censoring issue. Should I just censor the control if the life-expectancy for the x-aged, y-sexed, z-cohort exceeds todays date, i.e.: a 50 year old male in 2000 was expected to live 28 years in the general population? My take is that he should enter with 11 years and a censoring status.

Or is there some other more mathematically savvy way of doing this taking into account the uncertainty with the projected life-expectancy for the population?

  • $\begingroup$ How many observation for each combination of age/sex/cohort you have? I assume it's not too much, right? $\endgroup$ – Anton Korobeynikov Jan 15 '12 at 7:39

You are looking for relative survival. You need more than life-expectancy for this analysis, the entire life-table is needed. This is usually not an issue in the US - these are available. The main idea is that you would compute the expected number of deaths in your population (given the followup time) assuming that the control population's lifetable is followed, and then you compare it to the observed number of deaths. There are multiple ways to do it, the relsurv package in R implements some of them.

| cite | improve this answer | |
  • 1
    $\begingroup$ Thx.. relsurv seems to be what I´m looking for. But the methodology seems to be very similar to what I sketched above,except for the cox analysis in relsurv instead of kaplan meier analysis. $\endgroup$ – Misha Jan 31 '12 at 19:56

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.