Re: st: Re: Cox regression

[Steven Samuels <sjhsamuels@earthlink.net>]

Janet, You cannot fit a hazard model to a group with no observed  
failures.  However your main problem is that you have too few  
failures to trust the Cox results.  The p-values from the Cox mode  
require larger sample sizes.  Plot the survival curves with the  
asymptotic CI's (-sts graph-) and see how wide they are.  Your only  
recourse is to an "exact" test. At least one such exact long-rank  
test has been published, but I don't have the reference offhand.   If  
follow-up times are similar among the censored observations, or the  
distributions of censoring times in the three groups are similar, you  
can try pairwise Fisher's exact tests with failure as a binary outcome.

Steve

> I want to use a Cox proportional hazards model with 1
> independent variable corresponding to drug types.
> There are 3 drugs and for 1 of these there are no
> failures.
>
> . xi: stcox i.fadh
> i.fadh            _Ifadh_1-3          (naturally
> coded; _Ifadh_1 omitted)
>
>          failure _d:  death
>    analysis time _t:  dt
> Cox regression -- no ties
>
> No. of subjects =           60
> Number of obs   =        60
> No. of failures =           15
> Time at risk    =      1326.29
>                                                    LR
> chi2(2)      =      9.61
> Log likelihood  =   -47.766052
> Prob > chi2     =    0.0082
>
> ---------------------------------------------------------------------- 
> --------
>           _t | Haz. Ratio   Std. Err.      z    P>|z|
>    [95% Conf. Interval]
> ------------- 
> +----------------------------------------------------------------
>     _Ifadh_2 |   1.68e+09   8.79e+08    40.49   0.000
>    5.99e+08    4.68e+09
>     _Ifadh_3 |   4.33e+09          .        .       .
>           .           .
>
> My question is - how do I handle the situation where
> one of the groups has no failures, and how do I
> calculate a meaningful hazard ratio.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/