buhr at biostat.wisc.edu
2010-Feb-23 21:25 UTC
[Rd] survival package: Surv handles invalid intervals (start time > stop (PR#14221)
In the latest version of the survival package (2.35-8), the Surv
function handles invalid intervals (where start time is greater than
stop time) by issuing a warning that NAs have been created and then
setting the left endpoint of the offending intervals to NA. However, the
code that sets the endpoint to NA subsets incorrectly, and in some
circumstances an arbitrary selection of intervals will have an endpoint
set to NA.
For example, for the interval/event data:
interval event
(NA, 10] 1
(1, 5] 1
(6, 4] 1
the appropriate Surv call **should** result in a warning and the left
endpoint of the third, invalid interval being set to NA, as here:
> Surv(c(NA,1,6),c(10,5,4),c(1,1,1))
[1] (NA,10 ] ( 1, 5 ] (NA, 4 ]
Warning message:
In Surv(c(NA, 1, 6), c(10, 5, 4), c(1, 1, 1)) :
Stop time must be> start time, NA created>
However, the Surv call **actually** results in:
> Surv(c(NA,1,6), c(10,5,4), c(1,1,1))
[1] (NA,10 ] (NA, 5 ] ( 6, 4 ]
Warning message:
In Surv(c(NA, 1, 6), c(10, 5, 4), c(1, 1, 1)) :
Stop time must be> start time, NA created>
Note that the endpoint of the valid, second interval has been set to NA
in place of the invalid, third interval.
A similar problem exists for type="interval2" type data. The
**expected** behavior is:
> Surv(c(NA,1,6), c(10,5,4), type="interval2")
[1] 10- [ 1, 5] [NA, 4]
Warning message:
In Surv(c(NA, 1, 6), c(10, 5, 4), type = "interval2") :
Invalid interval: start> stop, NA created>
but the **actual** behavior is:
> Surv(c(NA,1,6), c(10,5,4), type="interval2")
[1] 10- [NA, 5] [ 6, 4]
Warning message:
In Surv(c(NA, 1, 6), c(10, 5, 4), type = "interval2") :
Invalid interval: start> stop, NA created>
The attached patch fixes the problem.
--
Kevin Buhr<buhr at biostat.wisc.edu> Phone: +1 608 265 4587
Assistant Scientist Fax: +1 608 263 0415
Statistical Data Analysis Center
Room 211, WARF Office Building, 610 Walnut St., Madison, WI 53726-2397