RE: st: Re: st: Imputing with ice and using clogit‏

[Tasha Amin <tamin_wh@hotmail.com>]

Hi Daniel and everyone else:
-Mi estimate- does support -clogit-, but my issue has been after using sttocc, so I guess I should have clarified that. Below are more details of my individually matched survival analysis using controls picked via incidence density sampling:

Basically, this is what I need to do for my analysis where x1 is the main exposure of interest and  x2 is the binary variable that needs imputing
stset hcv_duration, id(id) failure(newcase==1)set seed 8021sttocc, number(2) match(ethnicity)clogit _case x1 x2, group(_set) or


Since I need to impute x2, I first imputed with the commands listed below under (A). I tried to analyze using the commands listed under (B), but instead of getting case-control sets within each imputation (sets within _mj=1 different from sets within _mj=2, e.g.) so that I can analyze each imputation separately, I am getting case-control sets across imputations (so, e.g. after sttocc, in _set 1, I get _case=0 from _mj=4 and _mj=5 while _case=1 is from _mj=3).

Not sure what I'm doing wrong. Any help would be great.

A) For imputing, mi set widemi stset hcv_duration, id(id) failure(newcase==1)sts generate NA=naice x2 x3 x4 x1 NA _d, m(5) cmd(x2:logit) seed (8031) saving(MI.dta, replace)



B) For analysis in MI.dta, drop if _mj==0set seed 8021sttocc, number(2) match(ethnicity)mim: clogit _case x1 x2, group(_set) or



Thanks,Tasha, PhD student, Epidemiology



> Date: Wed, 6 Jun 2012 09:48:19 +0200
> Subject: st: Re: st: Imputing with ice and using clogit‏
> From: klein.daniel.81@googlemail.com
> To: statalist@hsphsun2.harvard.edu
>
> Tasha,
>
> I have not (yet) tried to do a -clogit- with -mi- data, but you can
> clearly import -ice- data (Royston, available from SSC) into -mi- (see
> -help mi import-). Also, the help file for -mi estimate- (help
> mi_estimation##estimation_command) states, that -mi estimate- supports
> -clogit-, at least in Stata 12.
>
> Can you explain what exactly you mean by "[...]none of them can be
> properly combined for an ice-mi estimate clogit type combo"?
>
> Best
> Daniel
>
> --
> Hi all:
> I have survival data that requires multiple imputation for a variable
> with considerable missingness (>35%). I then need to generate
> case-control sets that are individually matched (via the sttocc and
> match commands), and then use conditional logistic regression while
> properly accounting for the within and between imputation variation
> using Rubin's rules.
>
>
> While Stata 11 and 12 allow the user-written ice command and mi
> estimate allows Cox, multinomial and regular logistic models, none of
> them can be properly combined for an ice-mi estimate clogit type
> combo. Has anyone had success doing a clogit after multiple
> imputation? Any help or suggestions would be appreciated.
>
>
> Thanks!Tasha Amin, PhD student, Epidemiology
> *
> *   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/