Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

st: Re: "Meta-analysis" of Proportions


From   Steve Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: Re: "Meta-analysis" of Proportions
Date   Sun, 21 Apr 2013 09:37:16 -0400

Welcome to Statalist, Kathryn. 

You'd do much better if you:
1. Read the FAQ
2. Put a subject into your emails
3. Spelled Stata correctly -not "STATA" (see the FAQ)
4. Use Stata, not spreadsheet, terms: you have "variables" that you are inputting, not "columns"

Unlike a meta-analysis of a causal effect, there is no reason to
believe that proportions from the different studies will be identical, or
even similar.  So it is a bad idea to pool studies (add numerators &
denominators) to get an overall rate. The studies almost certainly have
different sample sizes and represent different populations. In the
example below, the largest study has the lowest rate and dominates the
pooled estimate.  Although you don't mention it, you also need confidence
intervals.

Here is model code:
*********CODE BEGINS************
/* Save this as "prop01.do" */
capture log close
set more off
log using prop01, replace
clear
input study events ntot
1 60 300
2 60 150
3 35 100
4 20 60
end

gen prop = events/ntot

/* Convert to a full data set */
gen no_events = ntot - events
save d1, replace
preserve

keep study events
gen event = 1
rename events count
tempfile t1
save `t1'

restore
keep study no_events
gen event = 0
rename no_events count
append using `t1'
sort study event
list

expand count  //expand to individual records
save d2, replace
/* Get separate CIs */
bys study: ci event, binomial agresti

/* Overall Proportion */

/* 1. Pool studies: bad idea
Weights studies by sample size
Incorrect idependence assumption for CI
*/
mean event
ci event, binomial agresti //same

/* 2. Arithmetic mean of study proportions: 
probably best idea*/
use d1, clear
ci prop


/*3. Random Effects Model: assumes normality of study
effects, which must be checked by plots. Don't use
it unless you understand it and do the checks. Here, it
is to the arithmetic mean of the 4 proportions
and I would use just that and its CI
 */
use d2, clear

xtmelogit event  || study:
nlcom invlogit(_b[_cons])
*****************CODE ENDS********************

On Apr 19, 2013, at 12:14 PM, kathryn dennick wrote:

Hi,

I am a registered user of STATA and I have a query that I would really like to post in statalist if that's possible at all, please? 

I am currently undertaking a systematic review of cross sectional studies investigating the prevalence of distress in diabetes. The data I have to synthesise is a simple proportion/percentage (those with distress as a proportion of the total diabetes sample). I have to to synthesise this data in a meta-analysis, and I recently attended a course where I learnt about using the metan code from the online STATA journal to conduct meta-analysis of RD, RR and OR data. I'm obviously not able to use this to undertake the meta-analysis as I don't have this type of data. I am know that in entering the data that I have as 2 columns (i.e. as those with depression and then the total sample size), STATA will not read this as a proportion as I need it to. My query is whether you are aware of any STATA command that exists which would enable me to synthesise the data I have and derive an overall proportion/percentage across the studies being reviewed. 
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


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