[R] Kaplan Meier analysis: 95% CI wider in R than in SAS

[Terry Therneau]

On 04/14/2012 05:00 AM, r-help-request at r-project.org wrote:
> Am replicating in R an analysis I did earlier using SAS. See this as a test of whether I'm ready to start using R in my day-to-day work.
> ?
> Just finished replicating a Kaplan Meier analysis. Everything seems to work out fine except for one thing. The 95% CI around my estimate for the median is substantially larger in R than in SAS. For example, in SAS I have a median of 3.29 with a 95% CI of [1.15, 5.29]. In R, I get a median of 3.29 with a 95% CI of [1.35,?13.35].
> ?
> Can anyone tell me why I get this difference?
>

The confidence interval for the median is based on the confidence 
intervals for the curves.  There are several methods for computing 
confidence intervals for the curves: plain, log, log-log, or logit 
scale.  There are opinions on which is best, and it is a close race: 
except for the first of these.  The type "plain" intervals are awful, 
it's like putting me in one lane of a championship 100 meter dash.

Until about version 9 the only option in SAS was "plain", then for a 
time it was still the default.  By 9.2 they finally went to loglog.

Terry Therneau