Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Re: case control design & analysis


From   Mike Lacy <Michael.Lacy@colostate.edu>
To   statalist@hsphsun2.harvard.edu
Subject   st: Re: case control design & analysis
Date   Thu, 03 Jul 2003 10:07:38 -0600

 Roger Webb wrote:


Date: Wed, 2 Jul 2003 09:51:57 +0100
From: "roger webb" <roger.webb@man.ac.uk>
Subject: st: re. case control design & analysis

Dear Statalist

I have a case control study in which cases were inpatient suicides
and eligible controls were alive and in inpatient care on the index
date (i.e. the date of death of the corresponding case). There is no
matching for age, sex, diagnosis, or any other explanatory variable.

I'd greatly appreciate advice on the following two issues:

1) Is this a matched design, or does the single criterion for
selection of controls (i.e. alive on index date) simply constitue
sampling from the risk set (as in Clayton & Hills, 1993)? Therefore
should I analyse the data using conditional logistic regression or
just use the simple 'logit' / 'logistic' commands?

Yes, it is a matched design, with the matching factor being "in the risk set at time X." I would analyze the data with both conditional and unconditional logit/logistic, and see if there is any difference in the results in terms of parameter estimates. If not, you could then just use the unconditional analysis. I have seen this strategy commonly pursued in published work, i.e. "We tried a conditional analysis and got the same results as the unconditional so we just relied on conditional." Perhaps someone can supply a citation justifying this approach.

Now, as for the followup question, as raised by another respondent to your question: What happens if you use a conditional analysis on a set of data for which there is no need to do so? This issue was discussed a few years ago (2-5) on the Epidemilogy list from U. Montreal. What I thought was the definitive answer was a comment by the author of a paper with a simple proof indicating that a matched analysis will *always* decrease estimation efficiency unless the matching factor is actually a confounder. The archive of the Epi list is at http://www.listes.umontreal.ca/wws/arc/epidemio-l.
The paper, who author was main a contributor to that discussion (and therefore possible a good search key in the Epi list archives) was:
Choi, B. G. K. 1984 "Unnecessary Stratification Pairing Can Never Increase Efficiency-- A Mathematical Proof." International Journal of Epidemiology 13 (1): 116-117.

When I looked this up in the Web of Science/Citation Index, I did see some demurring literature citing Choi, so one would want to follow up in that literature before taking him as the last word.

Regards,

Mike Lacy
Dept. of Sociology
Colorado State University


*
* 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/




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