Thread: Comparing survival against standard population

1. Comparing survival against standard population

Hello. I have the survival data (survival time + time of death) of a population after a certain treatment, as well as a standard population built from year-, age- and sex-specific mortality rates in the general population.

The Kaplan-Meier curve for the treatment-population falls sharply initially, and flattens out after 2-3 years. The standard popoulation survival curve has a somewhat opposite shape, naturally falling more and more as the average age increases.

Initially, I thought of using a log-rank test to compare the survival, but as with the cox proportional hazards model, I've learned that the survival curves should be parallel with each other, as is not true in this case. I could have used a later point in time (when the treatment pop. survival flattens out) as the starting point for my comparison, but I struggle to find an appropriate starting point which is not just based on me looking at the graph.

If anyone here has ideas on this matter, I would be very grateful!

2. Re: Comparing survival against standard population

It would be helpful if you could provide more detail about your research question. What exactly are you comparing your treated population to?

In general, when you want to compare mortality rates for a treated population, you use the Standardized Mortality Rate (SMR) which compares the average vs. expected # of deaths for the two groups. I haven't seen much use of population data within a survival analysis, because generally the population data is recorded on a yearly basis (# of deaths per year, for example), while treated sample data has a more accurate timeframe (person A died 52 days before person B, for example). Would using the SMRs be appropriate for your question?

3. Re: Comparing survival against standard population

Thanks for the answer, and sorry for not making clear enough what my objective is, I should mention that I do not have a great deal of statistical knowledge, but I try to read up on anything relevant.

My treatment population consists of severely ill patients receiving a certain treatment. I do not have a reference population to compare this population with, so instead, I was aiming to test the hypothesis that the mortality in my treatment population does not differ significantly from the general population. Therefore, I built a representative population based on age-/sex-/year-specific death rates.

So I have a treatment population with individual survival times in years and a binary variable (1=death, 0=survive) and a standard population with annual death rates for each matched individual. I was wondering which method to use to test this hypothesis, if possible at all. Could one use the SMR to test the hypothesis in any way?

I hope this was more understandable! :-)

 Tweet