stcurve or sts graph for adjusted Kaplan Meier

[javier rodriguez]

Hi all,

I've been asked by a reviewer to plot an adjusted Kaplan Meier graph. I understood that with sts graph and adjustfor, the adjustment was to zero values of all the variables.

So I tried to use stcurve, survival but I'm having a different graph with more steps. Besides, it is called "Cox proportional hazards regression" so I don't really understand if with stcurve, I am really getting an adjusted Kaplan Meier.

What do you think?
Thank you so much in advance!

Javier

[Raoul Reulen]

I have never come across "an adjusted KM curve". If you use the adjustfor option Stata will derive the survival function from a Cox regression in which all variables are set to the reference category. If you have provided hazard ratios from a Cox regression I dont see what it would add to have an adjusted curve. You could set the covariates to certain levels so you get a curve for an "average" person - for example someone age 50, male, etc....

[javier rodriguez]

Hi Raoul,

Thank you for your reply! Well the reviewer says that KM plots without controlling the effects of confounders in observational studies can be misleading. So I don't really understand what he/she wants.

I was thinking to show the cumulative incidence of death (adjusted) but apparently I can't do that after stcox but only after stcrreg. Do you know how to plot the cumulative incidence after stcox?

Or mayve, I will plot the survivor function with stcurve, survival.

I don't think they want to see an adjusted KM for an average person but rather in the whole population but it's really unclear. I've always provided classic KM...and I agree the adjusted curve after a cox regression doesn't add anything except to provide a graph...

Thank you,
best,

Javier

[Paul Dickman]

You've already received extensive answers in another thread you started. The reviewer is correct that "KM plots without controlling the effects of confounders in observational studies can be misleading" but how you respond to that depends on your study design, research question, and what other results you are showing. For example, if your other results are sufficient to support your conclusion then an appropriate response might be to sim,ply remove the K-M plot. If you do want/need to show a graph of survival curves then marginal (population-averaged) curves would probably be preferable but nobody here can advise you on exactly what to do without knowing all the details. You mentioned in another thread that you a re a med student; I suggest you get advice from your advisor or the senior investigator on the project. Plots of cause-specific cumulative incidence may be motivated, but they present a different measure with a different interpretation to the K-M curves. That is, they answer a different question.

[javier rodriguez]

Hi Paul,

Yes, I will talk to my supervisor.
Thank you all for your replies. It's already helpful.

Best,
Javier