BookmarkSubscribeRSS Feed
Buck1480
Calcite | Level 5

All:

I need to conduct a Kaplan-Meier survival analysis using a staggered entry approach for wildlife survival data (see Pollock et al. 1989). Specifically, the issue that I have is that PROC LIFETEST only permits survival estimate calculation when all individuals enter the study at the same time. However, in the world of wildlife management, we frequently encounter issues of additional individuals being censored or added throughout the study period (i.e., staggered entry). Is there a way in PROC LIFETEST that you can calculate a staggered entry survival? I appreciate the comments!

Thanks,

Andy

6 REPLIES 6
Reeza
Super User

Are you sure about that?

I thought it didn't matter because you essentially look at the duration and if it's censored or not.

Buck1480
Calcite | Level 5

Reeza,

Yes. According to the Analysis of Wildlife Radio-Tracking Data book by Garrott and White (1990), they state, "PROC LIFETEST of SAS can be used to compute the Kaplan-Meier estimates for these data if all animals enter the experiment at the same time." This is not the case for wildlife telemetry datasets so the SAS code must be adjusted to account for staggered entry. Likewise, Pollock et al. (1989) published a paper on adjusting the Kaplan-Meier survival calculation for a staggered entry approach and mentioned at the conclusion of the paper that PROC LIFETEST could work if all animals entered the study at the same time. Again, this is not the case for wildlife telemetry studies since animals are added or removed throughout the entire study process. Hopefully this helps with your question.


I'm still looking for a way to run the analysis in SAS. Hopefully one does exist.

Best regards,

Andy

SteveDenham
Jade | Level 19

I see three scenarios here, and there is a slightly different approach to using LIFETEST in these.  The first works off of's comment.  Index days observed rather than calendar days.  So suppose Bear 1 is tagged on June 10, and Bear 2 tagged on July 14.  For follow ups, count days from initial tagging.  So on August 1, Bear 1 would have 52 days and Bear 2 had 18.

Approach 2 is to assume that all individuals are present in the habitat when the first animal is tagged.  You just haven't gotten around to them yet.  I can see this as appropriate if there are no additional treatments applied--probably a lot better for herbivores than predators, as well.

The third approach is to assume both left and right censoring.  Now LIFETEST is out, and LIFEREG is really the only tool I have used before.  Theory may enable you to select a reasonable distribution to fit, but I would try a variety and compare results.

Steve Denham

Buck1480
Calcite | Level 5

SteveDenham,

Thanks for the reply! Based on your explanation for Approach 2, wouldn't the second (or third, so on and so far) animal tagged on another date already be considered "at risk." The idea of the staggered entry approach is to treat each individual separately on their entry into the study. Therefore, if animal 1 is tagged on January 1, 2011 and animal 2 is tagged on March 15, 2011, the start day for each animal should be the day they are marked rather than using the first day the first animal is tagged. Does that make sense?

To me, it makes sense to use your first example since each animal is added into the study along the way. In my case, I may have an animal enter the study on January 1, 2011 and another enter the study on March 15, 2011; therefore, I think it would be most appropriate (based on the idea of staggered entry) to calculate the number of days "at risk" from the date of capture until a selected end date.

As animals fall out of the study due to death or radio-trasnmitter failure, do they get coded a 0 for censoring? For example (using the first approach), would you set your dataset up in the following manner:

ID = Animal ID

Days2Event = Days to death or end of study for 1 year

Status = censored (0) or not

ID    Days2Event     Status

1             35                0

2             364              1

3             190              0

Lastly, if the first date of capture for animal 1 is 12/14/2010 and I wanted to calculate annual survival (using approach 1 above where I set an end date (e.g., December 31, 2011), how would you treat the number of days between 12/14/2010-12/31/2011 as this is obviously longer than a year. This is where approach 2 seems more appropriate because you could set the annual survival based on the first day of capture. Just curious.

Thanks,

Andy

Reeza
Super User

The annual survival would be a point in time estimate would it not?

I'm *GUESSING* that the issue with standard survival methods for this scenario actually occur due to seasonality which would need to be accounting for, so the proportional hazard assumption for coxph would never be met though that isn't an assumption in KM curves.

I guess the question is are you trying to measure 1 year survival or survival at a particular point in time i.e. end of the 2011.

If end of 2011 that's a different problem that how many animals survive one year, which makes this different than the typical scenario I see, which is clinical trials.

What I would do, is ignore the staggered start date and create yearly data sets.  If the animal is tagged at 12/14/2010 then as of 12/31/2010 that would be a censored observation for that year.

Then repeat process for every year, so for 2011 the animal entered at Jan 1/2011 (left censored) and either had an event or was alive at the end of the year (right censored).  In that scenario you have to use LifeReg.  If you're only interested in one year analysis that isn't an issue.

What's your actual research question in the end?

Buck1480
Calcite | Level 5

Reeza,

Thanks for the reply! I'm trying to address both annual (year-to-year) survival and seasonal (fall, winter, spring, summer) survival. I have two research questions: 1) what is the annual survival of a select avian species?; and 2) what is the seasonal survival of the select avian species? I have 2.5 years of survival data that I would like to use. We started sampling (or tagging) the avian species on 12/14/2010 and ended on March 15, 2011. We then re-started tagging during June-July 2011. Then again in January 1, 2012-March 15, 2012. Then once again from January 1, 2013-March 15, 2013. Individuals were tagged throughout all sampling periods. Likewise, individuals were monitored throughout the entire year during 2011 and 2012; however, monitoring ceased after July 31, 2013. Therefore, I planned to only calculate annual survival estimates for 2011 and 2012 since I had a full year of data for each time period. Thereafter, I would break each year into separate seasons (to see if a seasonal difference in survival existed). The seasonal periods would then include the 2013 data up to July 31, 2013. Does that make sense? Thanks again for the help!

Best regards,

Andy

SAS Innovate 2025: Register Now

Registration is now open for SAS Innovate 2025 , our biggest and most exciting global event of the year! Join us in Orlando, FL, May 6-9.
Sign up by Dec. 31 to get the 2024 rate of just $495.
Register now!

What is Bayesian Analysis?

Learn the difference between classical and Bayesian statistical approaches and see a few PROC examples to perform Bayesian analysis in this video.

Find more tutorials on the SAS Users YouTube channel.

SAS Training: Just a Click Away

 Ready to level-up your skills? Choose your own adventure.

Browse our catalog!

Discussion stats
  • 6 replies
  • 2462 views
  • 0 likes
  • 3 in conversation