[R] Hazard curves
therneau at mayo.edu
Tue Sep 30 15:47:32 CEST 2008
-- begin included message -----
I am looking at a continuous variable, age. I am looking at time to
12-month remission and can calculate the HR and 95% confidence interval are
coxfita = coxph(Surv(rem.Remtime,rem.Rcens)~nearma$all.age,data=nearma)
However, because I am looking at age as a continuous variable I cannot draw
a Kaplan-Meier curve. Instead I need to draw a plot of hazard against age.
How do I do this? I don't think
plot(nearma$all.age,coxfita$linear.predictors) is quite right.
Thank you for your help,
--- End included
There is not a simple plot: hazard is a function of both time and age, so will
be a 2-d surface.
A common plan is to plot the predicted survival for a small set of ages.
ageset <- c( 40, 55, 70)
curves <- survfit(coxfita, newdata= data.frame(all.age = ageset))
This will give a plot with 3 predicted survival curves, for ages 40, 55 and 70.
The plot of hazard vs age at any particular time point is not very
interesting: it will be h(t, age) = k exp(coefficient * age). This does not
reflect the data in any way, just your decision that age would be a linear term
in the model. If you want to explore the effect of age, then fit the
generalized additive Cox model
addfit <- coxph(Surv(rem.Remtime,rem.Rcens)~ pspline(all.age), nearma)
temp <- predict(addfit, type='terms', se=TRUE)
temp$fit - 2* temp$se.fit,
temp$fit + 2* temp$se.fit)),
log='y', xlab="Age", ylab="Estimated Relative Risk")
I suspect this is what you actually want.
More information about the R-help