RE: st: Data Setup for Survival Analysis With Unequal Interval

["Jesper Lindhardsen" <JESLIN01@geh.regionh.dk>]

Hi Peter,

Equal time-intervals are not neccesary (and would be unrealistic restrictive) when using Stata's survival analysis tools (st commands). The first example of -stset- with multiple records in the manual shows this. As the -help stset- info is very sparse on the use of 'timevar' (i.e. -stset timevar, options-), I would refer to the manual, which is excellent on this subject BTW.
In short, timevar is the last point in time (in your case day) that the record/row is valid. So, your last row with only the date of death makes no sense - the timevar variable in the preceding row should be equal to this date, if I understand you correctly. 

BW Jesper


Peter wrote:

Thanks for you quick reply.  I did use stset command with id option
for multiple records, but I am not sure that Stata technically handle
this kind of data setup because the examples I had read have equal
time-interval.

You understood correctly.  Date of death is one of the variables in
our data.  So you got it right again that the last record (row) of
each patient have no value of those covariates except death_status
(==1).  My best guess is that this row (of each patient) may affect on
the estimators (could be under or over-estimates).  I try to think how
to handle this row methodologically.

Peter

On Tue, May 10, 2011 at 10:26 PM, Jesper Lindhardsen
<JESLIN01@geh.regionh.dk> wrote:
>
> Hi Peter
>
> Yes, you can use multiple records per individual i.e. stset with more than one record per person. However, from my experience it really pays off to consult the manual (on stset) and also thoroughly check if the variables constructed by -stset- are consistent with what you expect.
>
> Regarding 1) -stset- doesn't care if the time between the visits are off, but perhaps your interpretation of data does. You would have to conside what timespan these variables are relevant to. Before the visit, after the visis etc. Perhaps it makes most sense to crude categories, for example, high, low & normal bloodpressure.
> Perhaps
>
> 2) I'm not sure I get this - it sound more like you have been provided a date of death ?? (I would presume the bloodpressure is quite low when the patient's passed away BTW ;-) )
>
> BW Jesper
>
>
> Jesper Lindhardsen
> MD, PhD candidate
> Department of Cardiovascular Research
> Copenhagen University Hospital, Gentofte
> Denmark
>
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of TA Stat
> Sent: 10 May 2011 16:59
> To: statalist@hsphsun2.harvard.edu
> Subject: st: Data Setup for Survival Analysis With Unequal Interval
>
> Dear All
>
> I would like to know the best way to run survival analysis (Cox PH
> Model) in Stata (I am not sure I did it correctly) in a long format.
> I also have a few points to discuss.
>
> 1.  The data is in a long format.  One row represents one hospital
> visit.  Each visit we measured blood pressure and other lab tests
> which will be time-varying covariates in Cox PH model.  However, the
> time interval between hospital visits is not the same for each
> patient.  Can Stata handle this?  I can calculate the numbers of days
> or months from the first visit.  But I am not sure if I have to setup
> the data in such a format that each visit is one or two months apart
> for all patients.
>
> 2.  For the last visit, if the patient passed away, that patient won't
> have a value of any time-varying covariates, would this effect the
> analysis?  I am thinking about using missing data calculation to fill
> in those values, but it doesn't make sense to me because the day that
> patient passed away, those value such as systolic blood pressure
> should be lower or higher than their normal level.
>
> Thanks
> Peter
>

*
*   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/