I applied Cox PH model to my data and the results I got was quite reasonable. But I am a bit confused because it seems that conclusions for one particular predictor does not coincide between the Cox model and Kaplan-Meier group-specific survival curve estimation.
In R i get :
coef exp(coef) se(coef) z p
chanell1 -0.12085 0.886 0.04532 -2.666 7.7e-03
(i have not included coefficient estimates for other predictors in model)
It seems that there is significant difference between hazards of the to chanells (1,2).
But when looking at the corresponding KM survival curve (see attachment), it looks that they are almost equal. They even cross each other at some moment. So I think Cox model should not give significant coefficient for chanell1.
What could be the explanation of this?