Practical exercise 5

 

In this exercise we will use Cox regression to study the effect of some covariates for the melanoma data. As we are interested in the mortality from malignant melanoma, we treat death form other causes as censored observations in the analysis.

 

As described in practical exercise 1, you may read the data into R by the command:

melanoma=read.table("http://www.uio.no/studier/emner/matnat/math/STK4080/h08/computing/melanoma.txt", header=T)

 

a) You may make a Cox regression analysis for the covariate sex by the commands

cox.mel<- coxph(Surv(lifetime,status==1)~factor(sex), data=melanoma)

summary(cox.mel)

 

Perform the commands and interpret the results. (Remember to attach the survival library.)

Remark: It is not really necessary to use factor for a binary categorical covariate like sex, but it is good practice to do so.

 

b) Make a Cox regression analysis for the categorical covariate ulceration and interpret the results.

 

c) Make a Cox regression analysis for the categorical covariate grouped tumor thickness (grthick) and interpret the results.

 

d) We may use tumor thickness as a numeric covariate (i.e. not a factor) either by using the recorded tumor thickness (thickn) or by using the logarithm of the tumor thickness (logthick). Explain why the results of question c makes it more reasonable to use the logartithm of the tumor thickness than the tumor thickness itself. Fit a Cox model with the logarithm of the tumor thickness as the only covariate, and interpret the results.

 

e) Fit a multivariate Cox regression with the three covariates sex, ulceration and logarithm of tumor thickness. Interpret the results and compare with those of questions a, b and d.