[R] Survival analysis: same subject with multiple treatments and experience multiple events

Mike Marchywka marchywka at hotmail.com
Sat Apr 23 13:28:47 CEST 2011










----------------------------------------
> Date: Fri, 22 Apr 2011 10:00:31 -0400
> From: littleduck24 at gmail.com
> To: r-help at r-project.org
> Subject: [R] Survival analysis: same subject with multiple treatments and experience multiple events
>
> Hi there,
>
> I need some help to figure out what is the proper model in survival analysis
> for my data.
> Subjects were randomized to 3 treatments in trial 1, some of them experience
> the event during the trial;
> After period of time those subjects were randomized to 3 treatments again in
> trial 2, but different from what they got in 1st trial, some of them
> experience the event during the 2nd trial (I think the carryover effect can
> be ignored since the time between two trials is long enough.)
>
> What I am interested is whether the survival functions differ among
> treatments. How should I deal with the correlation between the observation
> since the same subject was treated with two different drugs in two trials?
> Should I add "TRIAL" , "whether the event happened before", or "number of
> times the event happened before" as covariate(s)?
>
> Any input will be appreciated. Thank you.
> Qian
>

No one else replied so I would just suggest a web search using the term "crossover design"

http://www.google.com/#q=cran+crossover+design+survival

and refer you to any FDA panel discussions regarding any drugs that have been debated with
similar trial designs as part of the debate.

http://www.google.com/#sclient=psy&hl=en&site=&source=hp&q=site:fda.gov+briefing+crossover

The point of the above is to get some idea what can happen as no battle plan survives first
contact with data. Usually the objective in these designs is to infer something about causality in some system
and you just use the statistics to avoid fooling yourself. Personally it seems to me that
understanding of disease and host dynamics is improving to the point where you can
do more with the "carryover effct" that you mention as well as parametric putative prognostic
factors but you can also see opinions vary.



 		 	   		  


More information about the R-help mailing list