Power Analysis for Cox Proportional Hazards Model

[Antonio Pedroso-de-Lima]

In a recently submitted paper the referees have asked us to perform a
power analysis as some of the prognostic factors were marginally
significant. We were not able to find references nor software
available to do that. Please note that in our model we have
categorical (three levels) as well as quantitative covariates. I was
wondering if someone could point me out some references on this
subject.

Thanks,

Antonio

[BXC (Bendix Carstensen)]

If the reviewers are asking you to do a post-hoc power calculation, the
answer shoud be: No way!
Once data is collected, all the information is in the data, i.e. in the
estimates and confidence intervals.

A central reference to refuting demands of post-hoc power calculations
is:

@Article{Hoening.2001,
author = {JM Hoening and DM Heisey},
title = {The abuse of power: The pervasive fallacy of
power calculations for data analysis},
journal = {The American Statistician},
year = {2001},
volume = {55},
number = {1},
pages = {19--24},
OPTnote = {},
OPTannote = {}
}

Best,
Bendix
______________________________________________

Bendix Carstensen
Senior Statistician
Steno Diabetes Center
Niels Steensens Vej 2-4
DK-2820 Gentofte
Denmark
+45 44 43 87 38 (direct)
+45 30 75 87 38 (mobile)
+45 44 43 73 13 (fax)
b...@steno.dk http://www.biostat.ku.dk/~bxc

[John Uebersax]

I see your point, Bendix, but I could still imagine scenarios where it
is fair to ask for a power analysis after the fact -- but to be used
for reasons other than actual estimation of statistical power.

For example, a regulatory agency might, post-hoc, want to know if the
investigators had any business performing the analysis they did.

In any case, if the question is just, "what is a good reference for
power analysis of Cox regression given multiple predictors, some
categorical and some continuous?" I'd be curious to know the answer to
that myself.

--
John Uebersax PhD

> > Antonio- Hide quoted text -
>
> - Show quoted text -

[BXC (Bendix Carstensen)]

John,
as always when it comes to power calculations:

Once you are precise about your question, you have already specified a
simulation scheme. Then just let your computer do the job overnight,
and next morning you have the answer.

Formulae for power calculation are only still around because 43.7% of
all
active statisticians were broght up before computeres were powerful
enough
for this, and 94.8% of all statisticians have been taught by by people
from
this subset!

I normally use R for this sort of thing, but SAS is also quite well
suited for
data generation, and if you remember to include the number of the
simulated
sample in the total dataset the "by" is SAS will do all the analyses for
you
in one go!

Best
Bendix

[dkw]

On Sep 4, 9:31 am, "BXC (Bendix Carstensen)" <b...@steno.dk> wrote:
> John,
> as always when it comes to power calculations:
>
> Once you are precise about your question, you have already specified a
> simulation scheme. Then just let your computer do the job overnight,
> and next morning you have the answer.
>
> Formulae for power calculation are only still around because 43.7% of

> all active statisticians were brought up before computers were powerful

> enough for this, and 94.8% of all statisticians have been taught by by people
> from this subset!
>
> I normally use R for this sort of thing, but SAS is also quite well
> suited for data generation, and if you remember to include the number of the
> simulated sample in the total dataset the "by" is SAS will do all the analyses for
> you in one go!

No argument, but sometimes it helps to satisfy reviewers to have a
citation:

<http://links.jstor.org/sici?
sici=0006-341X(198609)42%3A3%3C507%3AEOSSAP%3E2.0.CO%3B2-D>
"Evaluation of Sample Size and Power for Analyses of Survival with
Allowance for Nonuniform Patient Entry, Losses to Follow-Up,
Noncompliance, and Stratification"
John M. Lachin, Mary A. Foulkes. Biometrics, Vol. 42, No. 3 (Sep.,
1986), pp. 507-519

or if a citation were needed for simulation methods, one could cite
Harrell's "Regression Modeling Strategies":

<http://lib.stat.cmu.edu/S/Harrell/help/Hmisc/html/cpower.html>
<http://lib.stat.cmu.edu/S/Harrell/help/Hmisc/html/spower.html>

<http://books.google.com/books?id=kfHrF-
bVcvQC&pg=PA499&lpg=PA499&dq=lachin+power+%22survival+analysis
%22&source=web&ots=32N-
S2bix_&sig=i45fKheD0cz9qpbWN9AgmEGa1sk#PPA501,M1>

--
David Winsemius

[dkw]

On Sep 5, 10:28 am, dkw <dwinsem...@comcast.net> wrote:
> On Sep 4, 9:31 am, "BXC (Bendix Carstensen)" <b...@steno.dk> wrote:
>
> > Once you are precise about your question, you have already specified a
> > simulation scheme. Then just let your computer do the job overnight,
> > and next morning you have the answer.
>
> > Formulae for power calculation are only still around because 43.7% of
> > all active statisticians were brought up before computers were powerful
> > enough for this, and 94.8% of all statisticians have been taught by by people
> > from this subset!
>
> > I normally use R for this sort of thing, but SAS is also quite well
> > suited for data generation, and if you remember to include the number of the
> > simulated sample in the total dataset the "by" is SAS will do all the analyses for
> > you in one go!

> No argument, but sometimes it helps to satisfy reviewers to have a
> citation:
>

Broken Link Sigh: If only GoogleGroups would accept the convention of
recognizing links within "<.>" the world would be an easier place:

http://tinyurl.com/yuyxnn

broken link deleted

> "Evaluation of Sample Size and Power for Analyses of Survival with
> Allowance for Nonuniform Patient Entry, Losses to Follow-Up,
> Noncompliance, and Stratification"
> John M. Lachin, Mary A. Foulkes. Biometrics, Vol. 42, No. 3 (Sep.,
> 1986), pp. 507-519
>
> or if a citation were needed for simulation methods, one could cite
> Harrell's "Regression Modeling Strategies":

Sigh #2:
http://tinyurl.com/2zk5xy

> --
> David Winsemius

[BXC (Bendix Carstensen)]

John,
we will only leave this era if the reference for the reviewers is
a link to the code doing the simulation. The various formulae and
tables are almost alway quite a bit off the sitauation you really want
to address.
Bendix

> -----Original Message-----
> From: MedS...@googlegroups.com
> [mailto:MedS...@googlegroups.com] On Behalf Of dkw
> Sent: Wednesday, September 05, 2007 4:29 PM
> To: MedStats
> Subject: {MEDSTATS} Re: Power Analysis for Cox Proportional
> Hazards Model
>
>
>
>

[dkw]

On Sep 5, 10:47 am, "BXC (Bendix Carstensen)" <b...@steno.dk> wrote:
> John,

> we will only leave this era if the reference for the reviewers is
> a link to the code doing the simulation. The various formulae and
> tables are almost alway quite a bit off the sitauation you really
> want > to address.

Again, I still do not disagree and would enthusiastically encourage
such a proposal, but let me point out that three out of my four links
were to code or explications of code.

Forward, ... toward simulations of the unknown.

--
David Winsemius

[Peri]

Hey can any one guide me how to unsubscribe from this group.

 

Thanks

 

--
Peri Goutam
Business Analyst
HSBC Group Services Center,
Mobile: +91-9903382563

[dkw]

On Sep 6, 12:26 am, Peri <goutam.p...@gmail.com> wrote:
> Hey can any one guide me how to unsubscribe from this group.
>

At your My Groups page find the link to "manage my memberships" and
then each group in the list has a pulldown menu (to the right)that
includes "unsubscribe".

[Bruce Weaver]

On Sep 4, 8:59 am, John Uebersax <jsueber...@gmail.com> wrote:
> I see your point, Bendix, but I could still imagine scenarios where it
> is fair to ask for a power analysis after the fact -- but to be used
> for reasons other than actual estimation of statistical power.
>
> For example, a regulatory agency might, post-hoc, want to know if the
> investigators had any business performing the analysis they did.
>
> In any case, if the question is just, "what is a good reference for
> power analysis of Cox regression given multiple predictors, some
> categorical and some continuous?" I'd be curious to know the answer to
> that myself.
>
> --
> John Uebersax PhD

John, PASS has a routine for Cox regression:

http://www.ncss.com/coxreg.html

I bet you'll find some references in the documentation for it.

--
Bruce Weaver
bwe...@lakeheadu.ca
www.angelfire.com/wv/bwhomedir
"When all else fails, RTFM."

[Val Gebski]

Surley the reg authorities would simply look at the protocol assumptions rather than doing a (dodgy) post-study power analysis?

Val Gebski

________________________________

From: MedS...@googlegroups.com on behalf of Bruce Weaver
Sent: Tue 11/09/2007 12:09 AM
To: MedStats

http://www.ncss.com/coxreg.html

#####################################################################################
This e-mail message has been scanned for Viruses and Content and cleared
by MailMarshal
#####################################################################################

####################################################################################################################

IMPORTANT NOTICE: This e-mail and any attachment to it are intended only to be read or used by the named addressee.
It is confidential and may contain legally privileged information. No confidentiality or privilege is waived or lost
by any mistaken transmission to you. The CTC is not responsible for any unauthorised alterations to this e-mail or
attachment to it. Views expressed in this message are those of the individual sender, and are not necessarily the
views of the CTC. If you receive this e-mail in error, please immediately delete it and notify the sender. You must
not disclose, copy or use any part of this e-mail if you are not the intended recipient.

#####################################################################################################################