In his post after Maarten's reply, Richard wrote:
Also Q2: We have followed the cohort to every destination. Our
participants leave and return on multiple occasions and these movements have
been documented..
I don't know if Maarten and other Statalisters agree, but I suppose Richard
is experiencing problems related to interval truncation (or gaps): patients
withdraw from the study and then drop in again, causing gaps in his follow
up.
The (really helpful) reference pointed out in the previous Maarten's reply
Cleves MA, Gould WW, Gutierrez RG. An introduction to Survival Analysis
using Stata. Revised edition College Station: Stata Press, 2006.
deals with this issue openly (pag. 46).
Kind Regards,
Carlo
-----Messaggio originale-----
Da: [email protected]
[mailto:[email protected]] Per conto di Richard Gibson
Inviato: mercoled� 27 febbraio 2008 14.49
A: [email protected]
Oggetto: Re: st: Survival analysis gurus
Thanks Maarten, your response and the reference are very helpful!
I am still thinking (my bedtime and your coffee time), but off the cuff;:
re Q2: the time varying suggestion for the treatment is most interesting
-seems like a good idea - thanks. and.. Fortunately, this is my job and not
my dissertation (that is some other story).
Also Q2: We have followed the cohort to every destination. Our
participants leave and return on multiple occasions and these movements have
been documented..
I suppose you are experiencing problems related to
interval truncation (or gaps): people withdraw from the study and then drop
in again, causing a gap in your follow up.
The (really helpful) reference pointed out in the previous Maarten's reply
Cleves MA, Gould WW, Gutierrez RG. An introduction to Survival Analysis
using Stata. Revised edition College Station: Stata Press, 2006.
deals with this issue openly (pag. 46).
Re Q3: Sure, I will have to take the clustering into account, but sought
the overall framework of time first,
Also Q3," I would add age as a time varying control variable." In that
case, age would be perfectly correlated with time.... I am not sure that
applying age as a time varying covariate is such a good idea in this
situation...I think the choices are baseline age, or centred age (much the
same) or age as time....
Richard
>>> Maarten buis <[email protected]> 27/02/2008 23:16:09 >>>
question 1:
Cox regression is pretty typical in these kinds of studies. In Stata
this is implemented as -stcox-.
question 2:
If you continue to follow respondents that move than that is not a
problem but actually a bonus, especially if they move from institions
that are a control to an institution that are a treatment or vice
versa. This allows you to add your treatment dummy as a so called time
varying covariate.
If you do not follow them than that is a problem. Survival analysis is
designed for dealing with respondents that leave the study without
experiencing the event (called right censoring), but it assumes that
this the probability of being right censored is unrelated to
experiencing the event. I have known a PhD student studying the effect
of friendship networks on health events in elderly people who was
forced to abondon that research because of the same problem you just
described (they did not follow the respondents when they moved).
question 3:
I would add age as a time varying control variable.
One question you have not asked is how to deal with the fact that your
respondents are nested within institutions. -stcox- allows for this
using the -shared- option.
When you are doing such an analysis you will want to have a good book
on your desk. A book I have found very useful is "An introduction to
survival analysis using Stata" by Mario Cleves, William Gould, and
Roberto Gutierrez:
http://www.stata-press.com/books/saus.html
Hope this helps,
Maarten
--- Richard Gibson < [email protected] > wrote:
> Survival analysis is not my usual territory and so I would appreciate
> some advice.
>
> I have data on around 5,700 residents of aged care facilities
> enrolled into a study on the effect of an intervention delivered to
> the facilities for the benefit of residents. There were around 90
> facilities randomised to control or intervention. We wish to assess
> the effect of the intervention in preventing certain events. The
> analysis plan included a survival analysis. We can resolve time to
> days (or hours/minutes with some assumptions) over a period of 18
> months of observation.
>
> Part of my problem is that about 35% (I am guessing - the data are at
> work - but it was a lot) of the residents either changed facility
> (often moving to higher care) or had time out from the facility (eg
> hospitalisation) or both. The other part is that those residents who
> changed facility, especially from low to high care were much more
> likely to experience the events of interest; presumably they were
> moved having been assessed as being at higher risk. If I were to say
> that those who changed facility were no longer under observation then
> I loose about 60% (again a guestimate - but reasonable) of events!
>
> Complicating matters is that residents who moved may have moved from
> an intervention facility to a control facility or vice versa. Those
> moving from intervention to control could carry some of the impact of
> the intervention to the new facility - eg fitness or medication
> regimen, while those moving from a control facility to an
> intervention facility may obtain new benefit from the change (caveat:
> providing the intervention is working). Residents could move in and
> out of the same facility several times or change locations completely
> several times. There is also a wide age range to account for with
> older residents being at greater risk of experiencing any event.
>
> My questions are:
>
> 1. What would be the best way to estimate the impact or not of the
> intervention?
>
> 2. How do/can I treat resident movements in a survival analysis?
>
> 3. As age is a risk factor, should age be the measure of time?
>
> With respect to question 1, I am thinking that a logistic regression
> with the outcome being the event of interest and having several
> covariates describing movements could be one appropriate analysis -
> but I am not sure that would be best. ...
>
> I am not sure how to answer question 2. I have done some reading
> around but nothing has lept out at me yet. One idea is to model
> final facility adjusting for origin (intervention or control and
> relative time), number of movements in and out of hospital and say
> number of facility changes. Another idea is to model initial
> facility, time in a control facility, time in an intervention
> facility, number of hospitalisations (ignoring time out). So many
> options, but they may be naive.
>
> Advice and suggestions will be greatly appreciated.
>
> Richard
>
>
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434
+31 20 5986715
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
___________________________________________________________
Rise to the challenge for Sport Relief with Yahoo! For Good
http://uk.promotions.yahoo.com/forgood/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
