[S] time dependent covariate in coxph

[James Song <jsong@bcm.tmc.edu>]

Dear S users,

This is a cox regression model fitting question. I have the ages of a
group of people. I want to compare their ages to the baseline population
hazard (mu(t)). mu(t) is taken from the published lifetable, it includes
hazard from 0 year old to 101 year old. The model I am trying to fit is a
cox model: hazard(t)=hazard0(t)exp(b1+b2 log(mu(t))). I included mu(t) as
a time dependent covarite. However, the coxph returns an error message of
X matrix being singular. Here is part of the input data.

cen: censoring status,
stop: death age of the subject
start: stop - 1
logus1: log(mu(stop)), log baseline hazard for age=stop. for example, the
first subject died at age 64. The baseline hazard from population for age
64 is 0.024699, so log(mu(64))=-0.3765395.

Am I using the correct command below?  Thanks for your advice!

James Song 


> temp1
    start cen treat id    logus1 stop 
  1    63   1    
1  1 -3.765395   64
  2    74   1    
1  1 -2.992841   75
  3    78   1    
1  1 -2.703324   79
  4    76   1    
1  2 -2.848123   77
  5    79   1    
1  2 -2.632055   80
  6    24   1    
1  3 -5.230226   25
  7    56   1    
1  4 -4.319501   57
  8     0   1    
1  5 -1.947467    1
  9     0   1    
1  5 -1.947467    1
 10    14   1     1  5
-5.892593   15
 11    63   1     1  5
-3.765395   64
 12    65   1     1  5
-3.635211   66
 13     1   1    
1  6 -3.302148    2
 14    81   1     1  6
-2.489816   82
 15    22   1     1  7
-5.195699   23
 16    23   1     1  7
-5.204998   24
 17    73   1     1  7
-3.063109   74
 18    74   1     1  7
-2.992841   75
...

> coxph(Surv(start,stop,cen)~logus1,temp1)
Call:
coxph(formula = Surv(start, stop, cen) ~ logus1, data = "">

       coef exp(coef) se(coef) 
z  p 
logus1   NA       
NA        0 NA NA

Likelihood ratio test=0  on 0 df, p=NA  n= 184 
Warning messages:
1: X matrix deemed to be singular; variable 1 in: coxph(Surv(start, stop,
cen) ~ logus1, temp1)
2: One or more nonpositive parameters in: pchisq(logtest, df)
> 



-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu.  To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message:  unsubscribe s-news