Home > Teaching > Tutorials > Survival Analysis in R
The function survfit() may be used to find pointwise confidence intervals for the survival function, which is where the probability of the survival curve being outside of these bounds at a given point is 95% (or whatever level is chosen). Confidence bands over an interval (tL, tU) are a bit different. If 95% confidence bands over (tL, tU) are created, then the (approximate) probability the true survival curve goes outside the bands at any point is 0.05. This being said, it should be clear that these bands will be further from the estimate of S(t) than the pointwise bounds.
Also, just so there are no surprises, conf.bands() takes in arguments, does some computations, and then prompts the user to find the appropriate coefficient using a table (hopefully this may be updated in the near future to automatically find the coefficient). After the coefficient is input, the function will output the confidence bands. Note that without a table with these coefficients (like in the appendices of Klein and Moeschberger), conf.bands() is useless.
surv.object :: a survival object from the function Surv().
conf.type :: This is the transformation to use to construct the confidence bands. The options include 'plain' (the default), 'log-log', and 'asin-sqrt' (arcsine-squareroot).
type :: Either 'ep' (the default) or 'nair'.
tL :: The latest time for the confidence band. If not specified, it will be the earliest observed event.
tU :: The latest time for the confidence band. If not specified, it will be the latest observed event.