Nice explanations, but I think that for testing the proportional hazard assumption through the LML, the variable should be included as a stratification variable instead of as a categorical covariate. Otherwise the LML curves will always be parallel.
Very helpful. Thank you. Log log plot looks perfectly parallel. Isn't it the predictor has to go to the strata instead for the log minus log plot? I guess, different parallels will be observed then. I read that having your predictor in covariates box, and plotting log minus log will always give you perfect parallel, since it is done with the the proportional hazard analysis.
thanks for this. do always multiple T_ by the time varying variable being considered? also, i am measuring time to death as age, but also age as a variable. what should i use as time to death? time since the study started or age itself? thank you
Hello, Does it mean, if we didnt find interaction between time and covariate, we just stick to use equation derived from conventional cox regression (without time to covariate in the model)? Thank you.
You don't describe how to add the interacting covariates to the cox regression. That is, how do you get "SMOKER_01*T_COV_"? When I add my covariates, it just lists both separately (no interaction).
Jason Axford This interaction term has to be defined in the "Cox model" box (time 12:32 of the video). You press MAJ and click on the 2 variables on the left panel (so T_COV_ and the variable you are interested in estimating interaction with time) and then click on a*b in the middle of the box, just next to the "covariates" panel. The interaction term will now appear in the "covariates" box. Alternatively, you can create the time-dependent covariate as an interaction term. So you define T_COV_ as SMOKER_01*T_ in the "Compute Time-Dependent covariate" box.
Thank you. I have a time-interacting variable (cox ph violated) that seems to increase in hazard as time increases. This variable is a factor and describes three bacterial infections in an organism (and one uninfected control I'm using as a baseline). I'm uncertain if I should introduce a time-interaction in SPSS or if I should use another model such as Accelerated Failure Time. However the time variable (time to event) is not normal. Maybe I should just stick to Kaplan Meier.
I am sorry but you make a mistake when you graphically analyze the curves parallelism. In this way you will always find parallel graph. I totally agree with JGiltay
Nice explanations, but I think that for testing the proportional hazard assumption through the LML, the variable should be included as a stratification variable instead of as a categorical covariate. Otherwise the LML curves will always be parallel.
Very helpful. Thank you. Log log plot looks perfectly parallel. Isn't it the predictor has to go to the strata instead for the log minus log plot? I guess, different parallels will be observed then. I read that having your predictor in covariates box, and plotting log minus log will always give you perfect parallel, since it is done with the the proportional hazard analysis.
corrrect
thanks for this. do always multiple T_ by the time varying variable being considered? also, i am measuring time to death as age, but also age as a variable. what should i use as time to death? time since the study started or age itself? thank you
How do you know when to use ln(T)?
Very helpful, informative, and coherent. Thank you!
Hello,
Does it mean, if we didnt find interaction between time and covariate, we just stick to use equation derived from conventional cox regression (without time to covariate in the model)? Thank you.
Fantastic, you saved my day!!!
You don't describe how to add the interacting covariates to the cox regression. That is, how do you get "SMOKER_01*T_COV_"? When I add my covariates, it just lists both separately (no interaction).
Jason Axford
This interaction term has to be defined in the "Cox model" box (time 12:32 of the video). You press MAJ and click on the 2 variables on the left panel (so T_COV_ and the variable you are interested in estimating interaction with time) and then click on a*b in the middle of the box, just next to the "covariates" panel.
The interaction term will now appear in the "covariates" box.
Alternatively, you can create the time-dependent covariate as an interaction term. So you define T_COV_ as SMOKER_01*T_
in the "Compute Time-Dependent covariate" box.
Thank you. I have a time-interacting variable (cox ph violated) that seems to increase in hazard as time increases. This variable is a factor and describes three bacterial infections in an organism (and one uninfected control I'm using as a baseline). I'm uncertain if I should introduce a time-interaction in SPSS or if I should use another model such as Accelerated Failure Time. However the time variable (time to event) is not normal. Maybe I should just stick to Kaplan Meier.
Very good Nicolas :)
I am sorry but you make a mistake when you graphically analyze the curves parallelism. In this way you will always find parallel graph. I totally agree with JGiltay
A