Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Survival analysis multiple events
From
Steve Samuels <[email protected]>
To
[email protected]
Subject
Re: st: Survival analysis multiple events
Date
Wed, 26 May 2010 11:08:29 -0400
There is no need to do _anything_ about the 28 day period if you use
the "conditional risk set model (time from the previous event)" of
section 3.2.4. Just use the dates that you now have. You will need
to add a time0 = 0 to start each interval and a stratum variable to
indicate the number of prior failures.
You could just as well use time0 = 28. The only difference would be a
shift in the baseline survival curves S(t). In the first formulation,
S(t) = 1 for t<=28 provided there are no failures under treatment.
The Cox model results will not change at all. If you don't
understand why, consult one of the survival references.
The measurement error in using dates of visits for the dates of events
should average +3.5 days or so. If your period of followup for each
failure is long in comparison, you probably won't introduce much bias
into the Cox parameters. To get more accurate baseline curves, apply
the average correction (-3 or -4) to the date of visit. This will
center the measurement error closer to zero.
Steve
On Wed, May 26, 2010 at 10:23 AM, Ben Andagalu <[email protected]> wrote:
> My intention was to use the Cox regression model taking into account
> potential clustering of events at the individual level.
> The "not at risk" is due to the treatment given after experiencing the
> event. the failures were detected at weekly follow-ups and unplanned
> visits
> I had actually read section 3.2.4 at
> www.stata.com/support/faqs/stat/stmfail.html and other literature
> prior to posting, and my major problem remains to be how to add those
> 28 days. I started using Stata not so long ago, so pardon my
> ignorance.
> Thanks
> Ben Andagalu, MD
>
> On Wed, May 26, 2010 at 1:31 PM, Steve Samuels <[email protected]> wrote:
>>
>> The 28 day problem aside, what analyses did you have in mind?
>> Someone can be "not at risk" for different reasons, and the specific
>> setting could affect the choice of analysis.
>>
>> Have you read www.stata.com/support/faqs/stat/stmfail.html ? Your data
>> are almost set up for the "conditional risk set model (time from the
>> previous event)" , discussed in section 3.2.4. This model resets the
>> clock after each failure. You can do Cox regression, and baseline
>> survival curves computed for the time period after each failure will
>> remain at 1.0 for at least 28 days.
>>
>> This model would not be appropriate if you believe that risk of
>> failure is much more strongly related to time since enrollment than to
>> time from previous failure. If you need another model, -snapspan-
>> will convert your data to a format that -stset- can use, You might
>> consider withdrawing people from observation in the 28 day period
>> after each failure. You do this by adding 28 days to the start date of
>> the interval. The resulting survival curves and inference will refer
>> to people who are conceptually "at risk" for the entire time after
>> enrollment, even if this was not true of the people under study. The
>> section in the Stata Survival manual on -stset- has many examples
>>
>> There could be another issue. Is the "date of follow-up" visit the
>> date a failure occurred or did the failure occur at an unknown date
>> prior to the visit? If the latter, you have either grouped data, if
>> the follow-up visits are at fixed times from enrollment, or
>> interval-censored data, if the follow-up intervals are not the same
>> for all people.
>>
>>
>> Good references are:
>> 1) Stephen Jenkins's text Survival Analysis using Stata:
>> http://www.iser.essex.ac.uk/survival-analysis
>> 2) An Introduction To Survival Analysis Using Stata (Paperback)
>> by Mario A. Cleves William W. Gould , and Roberto G. Gutierrez
>>
>> Steve
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> On Tue, May 25, 2010 at 3:23 PM, <[email protected]> wrote:
>> > How do i go about setting up dataset to analyse multiple failures per individual with a 28 day 'not at risk' period after each failure? individuals were followed up weekly after randomization - The variables i have are: date of randomization, date of follow-up visit, visit id, failure event , patient id, and the predictor variables. The data is in the 'long' format. Thanks. Ben Andagalu.
>> >
>> >
>> > *
>> > * For searches and help try:
>> > * http://www.stata.com/help.cgi?search
>> > * http://www.stata.com/support/statalist/faq
>> > * http://www.ats.ucla.edu/stat/stata/
>> >
>>
>>
>>
>> --
>> Steven Samuels
>> [email protected]
>> 18 Cantine's Island
>> Saugerties NY 12477
>> USA
>> Voice: 845-246-0774
>> Fax: 206-202-4783
>>
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/statalist/faq
>> * http://www.ats.ucla.edu/stat/stata/
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
--
Steven Samuels
[email protected]
18 Cantine's Island
Saugerties NY 12477
USA
Voice: 845-246-0774
Fax: 206-202-4783
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/