Re: Estimating heritability with censored data

Hi,
Use a discrete time surivival analysis in ASREML as in my papers and the
referenced work:

B. R. Southey, S. L. Rodriguez-Zas, and K. A. Leymaster
"Discrete time survival analysis of lamb mortality in a terminal sire composite
population" J Anim Sci 2003 81: 1399-1405.

B. R. Southey, S. L. Rodriguez-Zas, and K. A. Leymaster
"Competing risks analysis of lamb mortality in a terminal sire composite
population" J Anim Sci 2004 82: 2892-2899.

Also Gary Snowder's work might be helpful :
G. D. Snowder, L. D. Van Vleck, L. V. Cundiff, and G. L. Bennett
"Bovine respiratory disease in feedlot cattle: Environmental, genetic, and
economic factors" J Anim Sci 2006 84: 1999-2008.

G. D. Snowder, L. D. Van Vleck, L. V. Cundiff, and G. L. Bennett
"Influence of breed, heterozygosity, and disease incidence on estimates of
variance components of respiratory disease in preweaned beef calves" J Anim Sci
2005 83: 1247-1261.

However, I addressed the genetic analysis of mortality not the recurrent events
you may have here. But it can be expanded to recurrent events.

The major problems that you will have are:
1) The data structure because you may not have sufficient recurrent events.
2) What is the relationship between sick and mortality? You might not be able to
treat these as independent.

>Two suggestions have been:
>
> 1. multiple threshold model with more than two categories
> 2. animal model with repeated measures for each individual
>
> but using the whole data set - live and dead animals. But I am not
> sure this properly accounts for the missing data when an animal is
> dead.

That is very correct, neither of these suggestions are sufficient alone. You can
fit a multiple threshold model in Matvec
(http://statistics.unl.edu/faculty/steve/software/matvec/ )
The problem with the second suggestion is that your events are conditional on
the previous time point and you really can not repeatly measure dead animals.

> Another suggestion is to define the trait as number of weeks "alive
> and healthy" so that all animals have a complete dataset.

This is not normally distributed and does not address dead animals have fewer
weeks 'alive and healthy' animals.

The problem with the two trait analysis approach that Arthur suggested is that
the sick and dead traits are correlated by definition. Definitely, these have at
least a part-whole relationship for the data given below:
   a) if the animal is alive, it is not dead
   b) if an animal is dead, it is not alive and it is not sick.
   c) if an animal is sick, it is alive but not dead.

If you have questions, just ask.

Regards
Bruce

---- Original message ----
>Date: Fri, 28 Jul 2006 16:43:47 +1000
>From: Sandra Eady <Sandra.Eady_at_CSIRO.AU>
>Subject: Estimating heritability with censored data
>To: ASREML-L_at_AGRIC.NSW.GOV.AU
>
> I would like to estimate the heritability of resistance to bacterial
> infection in grower rabbits.
>
>
>
> I have data on the weekly incidence of bacterial infection in grower
> rabbits and mortality data (with post-mortem findings of bacterial
> infection) that looks like this:
>
>
>
> Rabbit wk5 wk6 wk7 wk8 wk9 wk10
>
> 1 0 0 0 0
> 1 1 rabbit healthy for first 4 weeks
> then sick but alive for last two
>
> 2 1 dead dead dead dead
> dead rabbit sick wk5 and dead by wk6
>
> 3 0 0 0 0
> 0 0 rabbit healthy and alive every week
>
> 4 0 0 0 dead
> dead dead rabbit apparently healthy until wk8 when
> it was dead
>
>
>
> Is there a way of analysing this data in ASREML to account for the
> dead animals?
>
>
>
> Two suggestions have been:
>
> 1. multiple threshold model with more than two categories
> 2. animal model with repeated measures for each individual
>
> but using the whole data set - live and dead animals. But I am not
> sure this properly accounts for the missing data when an animal is
> dead.
>
>
>
> Another suggestion is to define the trait as number of weeks "alive
> and healthy" so that all animals have a complete dataset.
>
>
>
> Any suggestions welcome!
>
>
>
> Thanks,
>
>
>
> Sandra
>
>
>
>
>
>
>
>
>
>
>
> ***********************************
>
> Dr Sandra Eady
>
> Stream Leader Integrated On-Farm Systems
>
> CSIRO Livestock Industries
>
> Private Mailbag 1
>
> Armidale NSW 2350
>
> Australia
>
> Phone +61 2 6776 1394
>
> Fax +61 2 6776 1371
>
> *********************************
>
>
Received on Wed Jul 28 2006 - 08:39:16 EST