Meta-analysis of prevalence data using metaprop command

[Ajay Kohli]

Hi There,

I am seeking some help regarding the meta-analysis of the prevalence data using the metaprop command. Below is the data set I have the command I used and the output I got.


Data-set

Study	case	population
Xyz et al (2015)	49	539
Xez et al (2016)	12	167
Abc et al (2011)	134	803
Bcs et al (2013)	197	1205
Dah et al (2009)	69	1168
Hwc et al (2016)	6	144
Pns et al (2016)	4	152

Command used

metaprop case population, random


Output

Study | ES [95% Conf. Interval] % Weight
---------------------+---------------------------------------------------
1 | 0.09 0.07 0.12 14.44
2 | 0.07 0.04 0.12 13.41
3 | 0.17 0.14 0.19 14.35
4 | 0.16 0.14 0.19 14.62
5 | 0.06 0.05 0.07 14.93
6 | 0.04 0.02 0.09 13.90
7 | 0.03 0.01 0.07 14.37
---------------------+---------------------------------------------------
Random pooled ES | 0.09 0.05 0.13 100.00
---------------------+---------------------------------------------------

Heterogeneity chi^2 = 134.71 (d.f. = 6) p = 0.00
I^2 (variation in ES attributable to heterogeneity) = 95.55%
Estimate of between-study variance Tau^2 = 0.00

Test of ES=0 : z= 4.14 p = 0.00

I need help in getting the prevalence of cases in % (instead of a decimal as it is right now) and please confirm whether I am using correct command and method to get the pooled prevalence.

Also, I want to names of the studies to be displayed instead of the numbers.


KR
Ajay

[Leonardo Guizzetti]

I have converted your example data into a usable dataset for Stata, using -dataex- as you are asked to do in the FAQ.

* Example generated by -dataex-. To install: ssc install dataex
clear
input str32 study int(case pop)
"Xyz et al (2015)"  49  539
"Xez et al (2016)"  12  167
"Abc et al (2011)" 134  803
"Bcs et al (2013)" 197 1205
"Dah et al (2009)"  69 1168
"Hwc et al (2016)"   6  144
"Pns et al (2016)"   4  152
end

I think your questions of effect size scaling and study labeling are pretty clearly addressed if you take the time to read the help file for -metaprop- (a user-contributed command). See the output of -help metaprop-. Hint: -label()- and -power()- are of interest.

A random-effects model seems sensible to me as a default with this many studies. However, , some of your studies have prevalence rates close to the extreme low end and you may wish to consider a suitable transformation. A discussion of these issues can be found, for example, here (Barendregt, J. J., Doi, S. A., Lee, Y. Y., Norman, R. E., & Vos, T. (2013). Meta-analysis of prevalence. J Epidemiol Community Health, 67(11), 974-978). A PDF version can be found via Google Scholar.