Cox regression- testing for effects

Question

[berry]

Hi all,

I performed hierarchical clustering on my expression data, and I have 3 clusters. I also categorized available treatment types into three groups. Now I want to analyze whether one of my clusters benefit significantly from one of the three treatment types in the presence of gender, age, KPS, and MGMT status.

I ran my coxph model using R coxph function adding interaction terms as below:

coxph(Surv(OS months, eventofdeath) ~ genders + age + kps + mgmt + clusters + therapy + clusters:therapy, data = mydata)

This is my first time performing survival analysis. I am not sure if what I am doing is enough to test my model for statistical interactions and/or for effect modification. I appreciate any input from you.

[German.M.Demidov]

try all the models and choose the most significant one =) (don't do this)

feature selection is an art still and there are no general recommendations for this

[Kevin Blighe]

There are a lot of covariates in your model - is their respective inclusion justified? Have you followed any tutorial before doing this analysis? You mention that this is your first time performing survival analysis, but doing this requires at least some basic understanding of regression models (?)