Currently working on a survival analysis of some lung cancer patients. I am building a Cox PH model and was working on validating the PH assumption for my covariates. To do this am using two techniques. Looking at the -log(-log(s)) plot of the survival function and also looking at the significance of the variable*log(survival time) interaction in my Cox model.

The issue I am getting is that for some of my covariates, I am getting conflicting data on the results. For example, the variable*log(survival time) is significant at p<0.05; however, the plot of the -log(-log(s)) shows curves that look very parallel.

Any thoughts on how to reconcile these two seemingly conflicting pieces of data? I am using SAS 9.4 btw.

Thank you!