[R] OT Futility Analysis

[Stephen A Roberts]

I would take the line that if they hadn't pre-specified any stopping rules, the only reason to stop is safety or new external data. I would be very suspicious of requests from the steering committee to stop for futility - they should be blinded so why are they thinking futility unless results have leaked? I would argue that they are obliged to finish the trial once they start.

This is an example of the need to sort out these things in advance - look up the stuff from the UK DAMOCLES project. The recent book edited by DeMets et al (Data Monitoring in Clinical Trials: A Case Studies Approach) is a good read on these sorts of issues and I think there is a more statistical book from the same group of authors.

 As far as software is concerned, futility analysis and conditional power are simply standard analyses with made up data and more-or-less justifiable assumptions.

Steve.


> -----Original Message-----
> From: r-help-bounces at stat.math.ethz.ch [mailto:r-help-
> bounces at stat.math.ethz.ch] On Behalf Of Spencer Graves
> Sent: 22 February 2006 03:45
> To: Kevin E. Thorpe
> Cc: R Help Mailing List
> Subject: Re: [R] OT Futility Analysis
> 
> 	  What does this particular Steering Committee think a "futility
> analysis" is?  Do they have any particular reference(s)?  What do you
> find in your own literature review?
> 
> 	  If it were my problem, I think I'd start with questions like that.
> Your comments suggested to me a confounding of technical and political
> problems.  The politics suggests the language you need to use in your
> response.  Beyond that, I've never heard before of a "futility
> analysis", but I think I could do one by just trying to be clear about
> the options the Steering Committee might consider plausible and then
> comparing them with appropriate simulations -- summarized as confidence
> intervals, as you suggest.
> 
> 	  And I hope that someone else will enlighten us both if there are
> better options available.
> 
> 	  Best Wishes,
> 	  spencer graves
> p.s.  For any attorneys who may read these comments, the suggestions are
> obviously warranteed up to the amount you paid for it, which is nothing.
>   If you follow them and they turn out to be inappropriate, you will pay
> the price.  I encourage you to share the problems with me, so I can
> learn from the experience.  However, the limits of my liability are as
> already stated.
> 
> Kevin E. Thorpe wrote:
> 
> > I beg your pardon if this is too off topic.  I am posting here
> > since I hope to find an R solution to my problem.  Please indulge
> > me while I give a little background about what I'm trying to do.
> >
> > I'm on a DSMB for a clinical trial.  The Steering Committee for the
> > trial has asked us to perform a futility analysis on their primary
> > outcome which is a time-to-event endpoint.  The trial was not designed
> > with group sequential methods, nor was any futility analysis spelled
> > out in the protocol.  Another thing which may be relevant is that
> > due to circumstances beyond the investigators' control, the trial
> > will stop recruitment prematurely unless there is some compelling
> > reason for them to find a way to continue the trial.  Lastly, the
> > trial has accrued not quite half of the planned sample size.
> >
> > Admittedly, I don't have a vast amount of experience implementing
> > stopping rules.  In other protocols I have seen where futility
> > analyses have been planned but a group sequential design has not
> > otherwise been employed, conditional power has been used for the
> > futility rule.  So naturally, that was my first thought (although
> > I may well be wrong) in this case.  I have done RSiteSearch() with
> > the following terms (three different searches):
> >
> > 	futility analysis
> > 	conditional power
> > 	stochastic curtailment
> >
> > Nothing that looked relevant to my problem jumped out at me.
> >
> > I have read, somewhat recently, that there are problems with conditional
> > power, although I don't remember the details at the moment.  This
> > has prompted me to consider other approaches to the problem.
> >
> > One simple thing that has occurred to me, although I don't know
> > what the implications are is to simply look at a confidence
> > interval around the hazard ratio for the treatment effect.  In
> > the event that the CI includes 1 and excludes any clinically
> > important difference, I would take that as an indication of
> > futility.
> >
> > I would appreciate your comments on this and to learn of any more
> > formal methods, particularly of implementations in R.
> >
> > Thank you for reading.
> >
> > Kevin
> >
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-
> guide.html