Mohammad Ehsanul Karim
2007-Nov-29 01:11 UTC
[R] Survreg(), Surv() and interval-censored data
Can anybody give me a neat example of interval censored data analysis codes in
R?
Given that suvreg(Surv(c(1,1,NA,3),c(2,NA,2,3),type="interval2")~1)
works why does
survreg(Surv(data[,1],data[,2],type="interval2")~1)
not work where
data is :
T.1 T.2 Status
1 0.0000000 0.62873036 1
2 0.0000000 2.07039068 1
3 0.0000000 1.40297214 1
4 0.0000000 0.09112871 1
5 0.0000000 0.75115280 1
6 0.0000000 2.36476839 1
7 0.0000000 2.47483937 1
8 0.0000000 1.16913097 1
9 0.0000000 0.21498251 1
10 0.0000000 1.89557501 1
11 0.9286723 1.09883334 1
12 0.3428920 1.16436088 1
13 0.5924055 2.20572832 1
14 0.7985962 1.73894982 1
15 1.7493239 999.00000000 0
16 0.8653470 999.00000000 0
17 0.6917203 999.00000000 0
18 1.1452119 999.00000000 0
19 2.2587688 999.00000000 0
20 0.7854710 999.00000000 0
21 1.5986105 999.00000000 0
22 2.2189803 999.00000000 0
23 0.7905505 999.00000000 0
24 1.3111472 999.00000000 0
25 2.3394220 999.00000000 0
26 1.3667469 999.00000000 0
27 0.3094596 999.00000000 0
28 1.6855216 999.00000000 0
29 1.7934033 999.00000000 0
Thanks in advance.
Ehsan
Re: [R] Survreg(), Surv() and interval-censored data
This message: [ Message body ] [ More options ]
Related messages: [ Next message ] [ Previous message ] [ In reply to ] [ Re:
[R] Survreg(), Surv() and interval-censored data ]
From: Thomas Lumley <tlumley_at_u.washington.edu>
Date: Thu 16 Feb 2006 - 07:58:26 EST
On Mon, 13 Feb 2006, Stephen Richards wrote: > Can survreg() handle interval-censored data like the documentation
> says? I ask because the command:
>
> survreg(Surv(start, stop, event) ~ 1, data = heart)
>
> fails with the error message
>
> Invalid survival type
>
> yet the documentation for Surv() states:
>
> "Presently, the only methods allowing interval censored data are
> the parametric models computed by 'survreg'"
>
> Any pointers as to what I'm missing?
>
You are specifying left-truncated, right-censored data, not interval-censored.
You need the type= argument, eg > Surv(c(1,1,NA,3),c(2,NA,2,3),type="interval2")
[1] [1, 2] 1+ 2- 3
specifies interval censoring at [1,2], right-censoring at 1, left-censoring at
2, and an observed event at 3.
-thomas
The question (slightly rephrased) was
Why does survreg(Surv(T.1, T.2, type='interval2') ~1, data=x) fail,
when x is
T.1 T.2 Status
1 0.0000000 0.62873036 1
2 0.0000000 2.07039068 1
3 0.0000000 1.40297214 1
4 0.0000000 0.09112871 1
5 0.0000000 0.75115280 1
6 0.0000000 2.36476839 1
...
----------------
The default model for survreg is a Weibull, for which a survival time of
exactly zero is not a valid value. Your example triggers this error message.
This is a bug in survreg. In the 10+ years the code has been available, no
one has ever specified this particular data: a (0,x] interval for a Weibull.
Congratulations! Fixing this will require some thought, however.
To work around the problem, change 0 to NA, which marks the observation as
left-censored.
Terry Therneau
(author of survreg)