Re: [S] Comparing two survival curves with very high survival rates out to 10 years

Frank E Harrell Jr (fharrell@virginia.edu)
Fri, 26 Jun 1998 11:21:12 -0400


Eric, the spower function in the Hmisc library will simulate power for this
situation using a log-rank test, which will be conservative in your case as
you probably expect proportional hazards to fail. You can give spower
other functions to base the power on, if you write a little function that
quickly computes the test statistic given a current sample of simulated data.
Then give spower a 'treatment effect over time function', symbolically.

See the document by Alzola & Harrell on our web page under Statistical
Computing Tools for examples of spower, with graphical output showing the
hazard ratio function, etc. See Chapter 5. Note that this document is undergoing
a major revision which will be posted in less than 2 weeks (Chapter 5 will not
change).

---------------------------------------------------------------------------
Frank E Harrell Jr
Professor of Biostatistics and Statistics
Director, Division of Biostatistics and Epidemiology
Dept of Health Evaluation Sciences
University of Virginia School of Medicine
http://www.med.virginia.edu/medicine/clinical/hes/biostat.htm

-----Original Message-----
From: eric.gibson@pharma.novartis.com <eric.gibson@pharma.novartis.com>
To: s-news@wubios.wustl.edu <s-news@wubios.wustl.edu>
Date: Friday, June 26, 1998 11:09 AM
Subject: [S] Comparing two survival curves with very high survival rates out to 10 years

>
>dear S-News,
>
>Suppose there is a medical device, call it xyq.
>There is to be an undetermined number of patients
>enrolled in the study which will last roughly 8 years.
>Half the patients will get drug A, half will get drug B, roughly.
>It is desired to show that device xyq last longer in patients
>on drug A than in patients on drug B.
>So there will be a survival curve for xyq in patients on A
>and a survival curve for xyq in patients on B.
>
>The problem is that based on past experience, the survival curves for both
>A and B will
>hover around 95 percent from the start of the study through 5 years. Only
>after 5 years do
>the survival curves begin to lower and separate noticably.
>This is compounded by the fact that between 5 and 8 years we expect
>30% or more of the patients to be lost to follow up.
>
>Any ideas for sample size calculations that don't result in enrolling
>100,000 patients?
>Any ideas for comparing the two survival curves in a case like this?
>Any ideas how to do either in SPlus? Literature references?
>
>A substantial part of the study is already handled in Splus,
>so it would be ideal to carry out the rest in Splus as well.
>
>Kind regards,
>Eric Gibson
>Novartis Pharmaceuticals
>
>
>-----------------------------------------------------------------------
>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
>

-----------------------------------------------------------------------
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