Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: confidence intervals of predicted risk after stcox


From   Steve Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: confidence intervals of predicted risk after stcox
Date   Tue, 30 Oct 2012 16:09:05 -0400


Mario wrote to me privately. When this happens I refer the writer to the
FAQ paragraph that starts "Do not request private replies...". 

> My problems are:
> 
> 1. If i gen survival probability and not risk, the upper and lower
> limits of ci seem inverted:
> 
They should be inverted.  You previously computed CIs for the failure ("risk") curve, now
you are computing the  CIs for the survival curve. A high risk for failure implies a low
risk for survival and vice-versa. This is mathematically so because F = 1 -S, as you 
previously calculated. Thus the upper and lower limits for the two curves are
switched. 

> Gen survlb15 = base15^exp(lb)
> gen survub15 = base15^exp(ub)
> gen surv15 = base15^exp(xb)
> sum survlb15 surv15 survub15
> 
>    variable |       obs        mean    std. Dev.       Min        max
> -------------+--------------------------------------------------------
>    survlb15 |        48    .7716674    .2110915    .381534   .9887756
>      surv15 |        48    .5276728     .343838   .0012602   .9593767
>    survub15 |        48    .2910488    .3057171   8.11e-21   .8586743
> 
> 
> 2. To compare the survival result at 15 months obtained in subjects
> with age40 = 20 and drug = 3, i used the stata command survci
> (-survci- sj11-4: St0217_1 ? Initially written by yulia marchenko of
> statacorp  and thereafter adjourned by matthew cefalu (the stata
> journal (2011), number  1, pp. 64?81 pointwise confidence intervals
> for the covariate-adjusted survivor function in the cox model).
> 
> The code i used is:
> Survci, at(age40=20 drug=3) outfile(age20_drug3).
> 
> The results of survci at 15 months are:
> _t     _d        _surv       _se          _lb                _ub
> 15     1        .85324285   .07846486   .6101214      .95029744
> 
> those of my code are the same as regard surv15 but very different as
> regard the confidence intervals (lower and upper)
> 
> surv15             survlb15              survub15
> .8532429          .9722255                4089079
> 
There are two mistakes here.

1. The minor one is that the -survci- gives CIs for someone 
with   with age = 20 and drug = 3, whereas the *15 values
you show are averages over the entire sample. There's no
reason to expect similarity.

2. The major mistake is that your original CIs reflect only variability of
the estimated betas, whereas -survci-  properly takes into account the fact
that the baseline survival probability is also an estimate.


Steve 

> -----------------------------------------
> 
> date: Fri, 26 oct 2012 17:15:04 -0400
> from: Steve samuels <sjsamuels@gmail.Com>
> subject: Re: St: Confidence intervals of predicted risk after stcox
> 
> mario, when i run your code, and sum the new variables, i get:
> 
> Variable |       obs        mean    std. Dev.       Min        max
> - -------------+--------------------------------------------------------
>        lb15 |        48    .2283326    .2110915   .0112244   .6184661
>      risk15 |        48    .4723272     .343838   .0406233   .9987398
>        ub15 |        48    .7089512    .3057171   .1413257          1
> 
> i see nothing strange about them. What is it that does not make sense to you?
> 
> 
> Steve
> 
> on oct 26, 2012, at 6:45 am, petretta@unina.It wrote:
> 
> Hi all,
> 
> i use stata/ic 12.1 for windows (32-bit).
> 
> I run a cox model to calculate for each patient the predicted probability
> of event at 15 months, using predict xb, predict basesurv and adjiusting
> the baseline risk
> 
> webuse cancer, clear
> gen age40=age-40
> stcox age40 i.Drug
> predict xb, xb
> predict basesurv, basesurv
> sum basesurv if _t<16
> scalar base15 = r(min)
> gen risk15 = 1 - base15^exp(xb)
> 
> i ask if it is possible to have also the 95% confidence intervals of these
> estimate.
> 
> I try:
> 
> Predict se_xb, stdp
> gen lb = xb - invnormal(0.975)*se_xb
> gen ub = xb + invnormal(0.975)*se_xb
> gen lb15 = 1 - base15^exp(lb)
> gen ub15 = 1 - base15^exp(ub)
> 
> but the results seems do not have sense.
> 
> Thanks for your consideration.
> 
> Mario petretta
> dpt. of internal medicine, cardiology and heart surgery
> naples university federico ii -italy
> 
> 



Mario, when I run your code, and sum the new variables, I get:

Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
       lb15 |        48    .2283326    .2110915   .0112244   .6184661
     risk15 |        48    .4723272     .343838   .0406233   .9987398
       ub15 |        48    .7089512    .3057171   .1413257          1

I see nothing strange about them. What is it that does not make sense to you?




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


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index