R help - Jun 2008 - Survival Analysis with two different events

Hello all,

I am hoping to use survival analysis to examine whether parasite attack
increases nest death in a species of social wasp. I therefore have data for

1. Whether the nest "died" in the 6 week census period
("Status", where
1=died, 0=survived)
2. The day number of death/last recorded day it was observed alive.
3. Whether the nest was attacked by the parasite (0/1 as with 1.)
4. The day number of attack/ last recorded day the nest was observed without
a parasite.

i.e. example dataset:

status   death   para   paraday
0           42       0       42
1           32       0       42
1           25       1       13
0           42       1       25 ...

I've looked over r-help, as well as in Crawley etc., but I have yet to find
a solution. Can anyone point me in the right direction or literature?

Many thanks,

Edd Almond
-- 
View this message in context:
http://www.nabble.com/Survival-Analysis-with-two-different-events-tp18183526p18183526.html
Sent from the R help mailing list archive at Nabble.com.

sickboyedd <sickboyedd <at> gmail.com> writes:
> 
> 
> Hello all,
> 
> I am hoping to use survival analysis to examine whether parasite attack
> increases nest death in a species of social wasp. I therefore have data for
> 
> 1. Whether the nest "died" in the 6 week census period
("Status", where
> 1=died, 0=survived)
> 2. The day number of death/last recorded day it was observed alive.
> 3. Whether the nest was attacked by the parasite (0/1 as with 1.)
> 4. The day number of attack/ last recorded day the nest was observed
without
> a parasite.
> 
> i.e. example dataset:
> 
> status   death   para   paraday
> 0           42       0       42
> 1           32       0       42
> 1           25       1       13
> 0           42       1       25 ...
> 
> I've looked over r-help, as well as in Crawley etc., but I have yet to
find
> a solution. Can anyone point me in the right direction or literature?
> 
   You might want to send this to r-sig-ecology if you need further
discussion.  In the meantime, the very simplest
thing (conditioning on whether the nest was
attacked or not) would be

library(survival)
c1 = coxph(Surv(death,status)~para,data=mydata)

(you should definitely read up a bit on survival analysis,
Cox proportional hazards, etc..  I think there's a chapter
in the book by Scheiner and Gurevitch, geared towards
ecologists).

Dealing with parasite attack in a more fine-grained
way (i.e. assessing mortality before and after parasitism)
would be a little trickier, but I wouldn't worry about
it until after you've understood the first stage of
the analysis.

  Ben Bolker

sickboyedd <sickboyedd <at> gmail.com> writes:
> 
> 
> Hello all,
> 
> I am hoping to use survival analysis to examine whether parasite attack
> increases nest death in a species of social wasp. I therefore have data for
> 
> 1. Whether the nest "died" in the 6 week census period
("Status", where
> 1=died, 0=survived)
> 2. The day number of death/last recorded day it was observed alive.
> 3. Whether the nest was attacked by the parasite (0/1 as with 1.)
> 4. The day number of attack/ last recorded day the nest was observed
without
> a parasite.
> 
> i.e. example dataset:
> 
> status   death   para   paraday
> 0           42       0       42
> 1           32       0       42
> 1           25       1       13
> 0           42       1       25 ...
> 
> I've looked over r-help, as well as in Crawley etc., but I have yet to
find
> a solution. Can anyone point me in the right direction or literature?
> 
 The classic solution in biomedical work is a time-dependent covariate.  Create
 a new data set like this:
  time1   time2  status  parasite
   0        42       0     0
   0        32       1     0
   0        13       0     0
   13       25       1     1
   
   ...
   
    The key is lines 3 and 4, which show the colony parasite free from day 0 to 
13, and with parasite from day 13 to 25.  Then one uses a Cox model with
    	fit <- coxph(Surv(time1, time2, status) ~ parasite)
    	summary(fit)
    	
It estimates the increase in death rate with parasite versus no parasite.  These
models were originally developed for treatment regimens that change over time.
A given colony (subject) can have as many lines of data as you want, subject to 
the fact that the time intervals can't overlap (which would correspond to
two
copies of the same person alive at the same time).  The status variable for a 
multi-line dataset =1 if THIS interval ends with an event.  Look at the survival
analysis chapter of Venables & Ripley, Modern Applied Statistics with S, for
further insight. (or many other books)

	Terry Therneau