I am studying the effect of expression of a specific gene signature on patient survival, dividing patients into Low, Mid, and High expression of Signature X. The survival curves for these patients all look very different from each other, and I'm trying to get a statistical value to support this statement (patients with Low, Mid and High expression have different survival trends).

I used the coxph function in the survival package in R and obtained this:

    > coxph(Surv ~ Sig.X.Group, data = data)
                    coef      exp(coef)   se(coef)    z      p
    Sig.X.Group     1.0310    2.8039      0.0795      13     <2e-16

If I had divided patients into Low and High expression of Sig.X I understand that: "High" expression of Sig.X is associated with a 2.8 increase in risk of death compared to "Low" expression (or something along those lines).

But what is the English explanation to an HR of 2.8 when I run the coxph function on 3 groups rather than 2?

EDIT: Fixed. The problem was my Sig.X.Group was coded as a number. Fixed the issue when changed Sig.X.Group to a factor, which gave me two individual HRs:

                  coef       exp(coef)   se(coef)    z        Pr(>|z|)
    Sig.X.Group2  0.5821074  1.789806    0.1696     3.4318    0.0005993
    Sig.X.Group3  1.9186438  6.811714    0.1550     12.3753   0.0000000
  • 2
    $\begingroup$ If you have 3 groups but only one coefficient reported, it's possible that you have accidentally left Sig.X.Group coded as a numeric variable. E.g., groups called "1","2", and "3" will tend to be dealt with as numeric values instead of factor levels unless specified otherwise. Make sure that Sig.X.Group is coded as a factor before you call coxph. Also, summary(coxph()) provides more useful information. $\endgroup$ – EdM Jan 30 '18 at 21:48
  • $\begingroup$ @EdM Yes, that is in fact what was happening. I made my Sig.X.Group into a factor and now I get two HR's. Thanks a lot for pointing that out! I guess each HR now represents the likelihood of a death event when comparing Sig.X.Group = 1 vs. when it is equal to 2, or 3, which makes sense to me conceptually. $\endgroup$ – user3579613 Jan 30 '18 at 22:26

Browse other questions tagged or ask your own question.