# RE: st: RE: FWD: poisson regression on aggrgregated data

 From "Kieran McCaul" <[email protected]> To <[email protected]> Subject RE: st: RE: FWD: poisson regression on aggrgregated data Date Tue, 10 Feb 2009 06:28:51 +0900

```OK, I see.

You have the year that treatment started: 1997,1998,1999,...,2008
which you have coded as:
year_start_cat  = 1 if year_start=1997 and
year_start_cat  = 2 if year_start=1998,1999,...,2008

You have time since the treatment started (time_since_start):0_1, 1_2,
2_3,...,Max
coded as:
time_since_start_cat = 1 if time_since start = 0_1 .
time_since_start_cat = 2 if time_since start = 1_2
time_since_start_cat = 3 if time_since start = 2_3,...,Max

Then you modelled the incidence rates using these covariates and now you
want to predict the rates in each of the calendar years that treatment
started (1997,1998,1999,...,2008) and in each of the years since
treatment started (0_1, 1_2, 2_3,...,Max).

Your model is assuming that the rate in the treatment years 1998 to 2008
is the same.  Similarly, it is assuming that after two years of
follow-up have elapsed, the rate is the same; constant with time since
treatment started.  In fact, it is assuming that this constant rate is
the same for all treatment years, 1998 to 2008.

If these assumptions are correct, then you could use the model to
predict the rates, but you would need to provide some evidence that they
are correct.

If you have sufficient data, you could use -strate- to generate the
rates for year_start and time_since start to see if your assumptions are
essentially correct.

strate year_start time_since start

This would give you the annual incidence rates since treatment started
and you could graph these for each calendar year.  I did something
similar to this in a study of second primary melanoma rates:
McCaul et al. The incidence of second primary invasive melanoma in
Queensland, 1982-2003. Cancer Causes Control 2008; 19(5):451-8.

Alternately, you could go back to your original data and use
Kaplan-Meier to generate hazard curves by calendar year of treatment.
These should be constant three or more years after treatment and they
should be similar for 1998 through to 2008.

______________________________________________
Kieran McCaul MPH PhD
WA Centre for Health & Ageing (M573)
University of Western Australia
Level 6, Ainslie House
48 Murray St
Perth 6000
Phone: (08) 9224-2140
Fax: (08) 9224 8009
email: [email protected]
http://myprofile.cos.com/mccaul
http://www.researcherid.com/rid/B-8751-2008
______________________________________________
The fact that no one understands you doesn't make you an artist.

-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of [Redacted]
Sent: Monday, 9 February 2009 9:30 PM
To: [email protected]
Subject: RE: st: RE: FWD: poisson regression on aggrgregated data

Dear Kieran

No I am not modeling mortality rates, I am modeling the incident of
experiencing an event by patients using these 2 covariates:

year_start (year of starting treatment): 1997,1998,1999,...,2008
time_since_start:0_1, 1_2, 2_3,...,Max

My original data was not in aggregated form, and patients experienced
the event (1) or not (0). Then I splitted the follow up time in one year
periods and coded year_start,1997 as 1 and 1998+ as 2 and time_since
start ,0_1 as 1, 1_2 as 2 and 2+ as 3. then aggregated data using these
coded covariates:

strate year_start_cat  time_since_start_cat

saved the output, and used poisson on _D (count of events in each
group)-created by strate-, above 2 coded covariates (year_startcat,
time_since_startcat) and the exposure that would be created aggregated
follow up time _Y ,and then predicted number of events and incident
rates in each group using predict.

Now I would like to use the coefficients of this fitted poisson model
and have the predicted incident rates in disaggregated data (for
year_start and time_since_start before grouping them)

for example the incident rate for  year_start=2003 (whitch has been
coded as year_startcat=2) and year_since_start=3_4 (which has been coded
as year__since_startcat=3)

Would it be correct if I use the same predicted incident rate from
poisson on grouped data (1998+ and 2+) and use it for above example or
each case in this group or in this case more events than really is will
be predicted?

Many thanks,
Shell

> Subject: st: RE: FWD: poisson regression on aggrgregated data
> Date: Mon, 9 Feb 2009 04:09:31 +0900
> From: [email protected]
> To: [email protected]
>
> .
>
> It's not clear from this description what you are modelling. Is it
> mortality rates?
> Second, when you say that you have used aggregated data, is this data
> that you have aggregated or was the data already in aggregated form?
>
>
>
> ______________________________________________
> Kieran McCaul MPH PhD
> WA Centre for Health & Ageing (M573)
> University of Western Australia
> Level 6, Ainslie House
> 48 Murray St
> Perth 6000
> Phone: (08) 9224-2140
> Fax: (08) 9224 8009
> email: [email protected]
> http://myprofile.cos.com/mccaul
> http://www.researcherid.com/rid/B-8751-2008
> ______________________________________________
> The fact that no one understands you doesn't make you an artist.
>
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of Martin
Weiss
> Sent: Sunday, 8 February 2009 7:55 PM
> To: [email protected]
> Subject: st: FWD: poisson regression on aggrgregated data
>
> <>
>
> Dear [Redacted],
>
> short answer: I do not know! BTW, it is not clear to me why you want
to
> tap
> only my wisdom when you can have the entire list. So I am forwarding
> this
> topic to the list for you...
>
>
> HTH
> Martin
> _______________________
>
>
> Dear Martin
>
> I would like to ask the following question.
>
> I have used poisson regression on aggrgregated data with 2 covariates
> (in a
> cohort of HIV positive patients)
>
> 1.year of starting 3+ antiretroviral treatment (1997, 1998+)
> 2.time since starting 3+antiretrovirals (0-1, 1-2, 2+)
>
> and I have predicted incident rates in this aggregated data using
> predict
> option after the poisson model.
>
> Now my question is if I want to have predicted incident rates for
> patients
> starting 3+ arvs at 2000 and time since starting 3+ arvs , 3_4 years
> is it correct to use the same predicted incident rate obtained for
> aggregated data and in this example use predicted incident rate for
> 1998+
> and 2+?
>
> It would be greatly appreciated if you could answer my question,
> let
> me know if you would like me to provide more details.
>
> Thank you very much in advance,
>
> Kind regards,
> Shell
>
> *
> * 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/
_________________________________________________________________
News, entertainment and everything you care about at Live.com. Get it
now!
http://www.live.com/getstarted.aspx
*
*   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/
```