Skip to content

cox model for haplotypes

2 messages · SUBIRANA CACHINERO, ISAAC, Thomas Lumley

#
On Fri, 24 Mar 2006, SUBIRANA CACHINERO, ISAAC wrote:

            
A relatively straightforward approach is to use haplo.em from the 
haplo.stats package to impute the haplotypes. It will give you all the 
possible haplotype pairs (diplotypes) for each individual, with 
probabilities.  You can then use all these possible diplotypes with the 
probabilities as weights in coxph() and use cluster() to get the right 
standard errors.

The haplo.ccs package does this for logistic regression, so you can follow 
how it works for more details.

This approach will underestimate the standard errors of rare haplotypes, 
especially in small samples, and is not fully efficient. The ideal 
solution is Danyu Lin's semiparametric maximum partial likelihood 
estimator, but I don't know of any implementations (the URL describe in 
the paper lists the software as 'under development').

 	-thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle