Cox Regression Survival Question

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

Posts: 1
Joined: Wed Apr 02, 2014 10:03 pm

Cox Regression Survival Question

Postby ProSubzero » Wed Apr 02, 2014 10:16 pm

I have practical experience working with SPSS analyzing data derived in our lab; however, I don't have the knowledge to always explain the specifics of how a test is performed and results presented. I'm pretty sure what I did was right, but I can't explain in any meaningful why it looks the way it does.

So, below in A it is just a survival analysis depending on whether the patients have a mutation or not. As expected, the mutated (red/positive) samples have a worse outcome. However, age is a major factor with outcome of course, so B is the survival curve and is again based on mutation status with age included in the cox regression. I have age and mutational status set as the covariants. And under options I have display model information at each step and probably for stepwise entry 0.05 and removal 0.10 and maximum iterations as 20. Those are the defaults. As expected the results show that age is significant and that mutational status remains.

The specific question I have:

In the cox regression curves, the steps down are mirrored between the two. It makes it look as if deaths are happening at the exact same time in both groups. That is of course not true. In the KM, that isn't seen since that is reflection when the event happens. In the Cox graph, it is no longer a true reflection of the time. It seems to me that while adjusting for age, the stats basically gates the time. It looks at something like 50 days, adjusts for age and calculates. Then the next 50 ect ect. Looking around online, when people do the Cox, this mirrored step wise result happens for all the analysis I've seen. So I'm pretty sure it is right, I just can't figure out how to search for this specific question and get something that is citable for an explanation. Again, this might be a very simple straight forward answer for people more familiar with the algorithms.


I'd GREATLY appreciate an explanation that a lab geek and statistical novice can understand.

Who is online

Users browsing this forum: No registered users and 1 guest