sally.x.wetten at gsk.com
2007-May-04 10:24 UTC
[R] Predicted Cox survival curves - factor coding problems...
I am trying to use the survfit() function with the newdata argument to
produce predicted survivor curves for a particular covariate profile.
The main purpose of the plot will be to visualise the effect of snp1,
coded 0 and 1. In my Cox model I have stratified by one variable, edu, and
so I know I will automatically get a separate curve for each strata. My
problem is how to deal with the variable sitenew, which is a 3 level
factor (coded 0,1,2).
coxsnps$edu <- as.factor(coxsnps$edu)
coxsnps$sitenew <- as.factor(coxsnps$sitenew)
test1 <- coxph(Surv(time,censor) ~ snp1 + sex + sitenew + geno + eth_self
+ strata(edu), data=coxsnps)
coef exp(coef) se(coef) z p
snp1 -0.4324 0.649 0.1073 -4.031 5.5e-05
sex 0.0604 1.062 0.0915 0.660 5.1e-01
sitenew1 0.5128 1.670 0.1381 3.712 2.1e-04
sitenew2 0.2688 1.308 0.1313 2.047 4.1e-02
geno 0.9702 2.639 0.0663 14.630 0.0e+00
eth_self 0.0871 1.091 0.1088 0.801 4.2e-01
profile1 <- data.frame(snp1=factor(c(0,1), levels=c("0",
"1")), sex=0,
sitenew=0, geno=0, eth_self=0)
plot(survfit(test1, newdata=profile1))
Putting sitenew=0 in the dataframe produces the error
Error in x2 %*% coef : non-conformable arguments
implying the rows and columns of the matrices don't match up. This makes
sense as R creates dummy variables for sitenew. I naively tried using
sitenew1=0, sitenew2=0 in the dataframe instead, which produced the same
error message.
Is there a way to produce curves by edu and snp1 but just one level of
sitenew?
If not, what would be the procedure for producing curves for all of them
(for reference)?
Many thanks,
Sally
-----------------------------------------------------------
This e-mail was sent by GlaxoSmithKline Services Unlimited
(registered in England and Wales No. 1047315), which is a
member of the GlaxoSmithKline group of companies. The
registered address of GlaxoSmithKline Services Unlimited
is 980 Great West Road, Brentford, Middlesex TW8 9GS.
-----------------------------------------------------------
[[alternative HTML version deleted]]
Reasonably Related Threads
- Predicted Cox survival curves - factor coding problems..
- Merging data frames, or one column/vector with a data frame filling out empty rows with NA's
- array dimension changes with assignment
- snp-chip table
- a question about 'read.table' with or without 'read.table'.(urgent)
