Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: RE: Estimation of differences of logistic regression models in three groups


From   "Kieran McCaul" <kamccaul@meddent.uwa.edu.au>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: RE: Estimation of differences of logistic regression models in three groups
Date   Sun, 22 Feb 2009 18:24:53 +0900

If you type:

testparm _IlapXtim_1_2 _IlapXtim_1_3

you will get a Wald's test of the interaction terms.  This is a overall test of the three period effects.

Alternately, you could fit
xi: logistic komplik i.lapscopi i.period
est store A
xi: logistic komplik i.lapscopi*i.period
lrtest A

which will give you a likelihood ratio test of the same thing.

Note, you are not testing for trend here.  The null hypothesis is no difference in the complication rate in the three periods.

If you are looking for trend, you could try:

xi: logistic komplik period i.lapscopi
est store A
xi: logistic komplik period*i.lapscopi
lrtest A

The effect of period is now the risk of complications associated with a one-unit change in period.
In the second model, the period effect is the trend for open appendectomy and the period + period*lapcopi effect will be the trend for laparoscopy.

Note also, I don't know how you have coded period. If the three periods are equal time intervals, then the model above will be OK.  But if the three periods are not equal time periods, then you will need to recode the levels of period to reflect this.

______________________________________________
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: kamccaul@meddent.uwa.edu.au
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: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of roland andersson
Sent: Saturday, 21 February 2009 7:08 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: RE: Estimation of differences of logistic regression models in three groups

I have started by myselt but need feedback.

by the commands

lincom  _Ilapscopii_1+ _IlapXtim_1_2	,	hr
lincom  _Ilapscopii_1+ _IlapXtim_1_3	,	hr

I understand that this will give the estimates for the risk of
laparoscopy for timeperiod 2 and 3. The estimate for timeperiod 1 is
the baseline estimate from the model. The riskdifference is
significantly higher in the third timeperiod. Is this correct? How can
I make a test for the trend over time?

perpostkir~k  Odds Ratio   P>z	[95% Conf.	Interval]
_Ilapscopi~1	1.591794	0.104	.9093168	2.786497
(1)                    1.249688   0.373	.7650747	2.041264
(1)                    2.485883   0.000	1.759211	3.512719

Roland

2009/2/21 roland andersson <rolandersson@gmail.com>:
> Thank you Kieran
>
> It should have been obvious for me to use interaction terms.
> I am not used to the lincom command. Can you guide how to use it with
> this result:
>
>
> komplik         Odds Ratio      P>z            [95% Conf.       Interval]
> _Ilapscopi              1.591794                0.104   .9093168        2.786497
> _Iperiod2               1.029331                0.858   .7497497        1.413169
> _Iperiod3               1.334           0.070   .9767403        1.821933
> _IlapXperiod2   .7850812                0.519   .3761646        1.638518
> _IlapXperiod3   1.561686                0.178   .816029 2.988697
>
> From this result I understand that laparoscopy has a higher rate of
> complications overall.
>
> There is also a trend towards an increasing frequency of complications
> for both surgical methods. How can I test for the trend over the 3
> time period?
>
> There is also maybe a trend for increasing complication rate for
> laparoscopy. I would like to test for trend also here.
>
> It is anyhow important that the time trend is towards an increasing
> risk and not decreasing, which should be expected.
>
> I would like to estimate the risk of complication for laparoscopy
> compared to open surgery for each timeperiod. Is laparoscopy
> associated with a higher risk in all time periods?
>
> I would appreciate your help.
>
> Roland Andersson
> surgeon, ass prof
> Jönköping, Sweden
>
>
> 2009/2/21 Kieran McCaul <kamccaul@meddent.uwa.edu.au>:
>> Fit an interaction between the surgery variable and the time period variable.
>> You can use -lincom- to estimate the surgery effect in each time period and use -testparm- to test significance of the interaction terms.  This will test if the surgery effect is different across time periods.
>>
>> Do the same thing for region.
>>
>> ______________________________________________
>> 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: kamccaul@meddent.uwa.edu.au
>> 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: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of roland andersson
>> Sent: Friday, 20 February 2009 9:47 PM
>> To: statalist@hsphsun2.harvard.edu
>> Subject: st: Estimation of differences of logistic regression models in three groups
>>
>> I am analysing the frequency of surgical complicationafter
>> laparoscopic compared to open appendectomy in alrge national sample of
>> patients.
>>
>> I use logistic regression to adjust for differences in comorbidity,
>> age and sex between the two surgical methods. I obtain a higher risk
>> of complications after laparoscopic appendectomy ( OR 1.97).
>>
>> I know want to analyse if this risk-difference between laparoscopic
>> and open appendectomy has decreased between three timeperiod, and
>> between three geographical regions.
>>
>> Shall I put the timeperiods and regions in the same model and use some
>> postestimation method to do the inference? Or shall I do separate
>> calculations for each timeperiod and geographic region? If so how can
>> I estimate the difference in risk between the timepriods and make the
>> statistical inference?
>>
>> Here is a simplified model with the important variables:
>>
>> xi:logistic perpostkirkomplik lapscopi i.appdgn  i.agekat sex i.period
>> i.region comrbidityvariables, or
>>
>> I assume this can be done with some postestimation methods but I need
>> your helpt o understand them.
>>
>> Roland Andersson
>> Jönköping, sweden
>>
>> *
>> *   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/
>>
>

*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index