Risk adjustment for survival

Hi all - I am using jmp for my analysis.

I compared and analyzed two different patient populations (Virus infection Type 1 and Virus infection Type 2) with one specific treatment in regards to outcome.

Kaplan Meier among those groups did not show any difference - which is good - so can make the statement that the unadjusted survival among both groups is the same.

However, it seems like one group (eg Virus infection Type 1) is being "sicker" by going into the disease process, by either being older (as a continuous variable) or having more , or other comorbidities than the population Virus infection Type 2. I identified those differences in the pre-existing population by univariate analysis (logistic regression and chi-square depending on).

I want to adjust a priori before I do Kaplan Meier analysis to see if this population (Virus Type infection 1) if I adjusted for those risk factors this patient has actually a relative survival benefit ? Or in other words - I want to risk adjust for eg advanced age, and come to the conclusion - as has been shown that the survival is the same ?!

Thanks a lot, very timely and important analysis. Marc
I think the adjustment method in kaplan-meier analysis is 'stratified' kaplan-meier. I don't know if this is 'a-priori' adjustment, but is probably the way it should go. In sas i think this is in the 'strata' statement, and the test variable (virus infection type?) goes in group, as i recall. Parametric methods are usually better at handling these adjustments than kaplan-meier.


Less is more. Stay pure. Stay poor.
It sounds like you may want to use weights instead of controlling for the potential confounder(s) in the model. Of note, if you do this, you wont know their direct effect on the model terms, so keep that in mind.

Second, if you aren't finding a difference, you ABSOLUTELY CANNOT say there isn't a difference. What is happening is you are failing to show a difference. If you ran a priori power analyses you can state given those parameters and the posited effect size you did not see a difference. But you can't after you looked at the data say there is no effect because you didn't find an effect, you are changing your purpose and intentions. If you want to report such a result, next time before starting the study write that into your protocol and state your clinical margin of difference and conduct a proper non-inferiority study.