eform option in metan command

[Soodabeh Navadeh]

HI.
I am doing a meta-analysis in which I need to generate pooled prevalence rates. I transformed my data to the logarithmic scale to have normal distribution and back-transform using the "eform" option in the "metan" command. The problem is the horizontal lines that show the confidence intervals in the forest plot are not proportional to the length of CI. For example, in the below graph, the length of CI in study#21 is ~8. It is ~4 in the study#22 but the horizontal line for the study#22 is taller than study#21!!


I would be thankful if somebody help me.

Soodabeh Navadeh MD, MPH
Research Specialist
University of California, San Francisco

[Hannah Jongsma]

Hi Soodabeh,

I think this is to do with the scale of the x-axis - it appears to be on a log scale? I have come across this problem too, and can't get my x-axis to be on a linear scale.
The relevant bit of syntax is as follows: xlabel (10, 20, 40,60,80) force nolog. This allows you to specify the points on the x-axis, and 'force' forces it to actually use those points.
However, even using the nolog and the force option, still doesn't get my x-axis on a linear scale.

The full syntax I've used is as follows: metan logir se, random eform lcols(Author Year Country Notes) title ("Incidence of all psychotic disorders") texts(130) nulloff nowt xlabel (10, 20, 40,60,80) force nolog xtitle("Rate per 100,000 person-years", size(vsmall)) xsca(titlegap(2))

And the resulting graph can be found in the attachment. Happy to discuss further!

If anybody else has got any clues on how to get the x-axis to be linear, that would be much appreciated.

Hannah

[Dan Bratton]

Hi Hannah, Soodabeh,

Instead of specifying the log IR and its SE in the metan command, have you tried instead specifying the IR itself and its confidence limits (not on the log scale). For example, try "metan IR cli uci" where lci and uci are the lower and upper limits of the CI of the IR. This works in the examples I have tried.

Hope that helps

Dan

[Abdul-Kareem Abdul-Rahman]

Hi all,

Apologies for a digression here.

At a more basic stage, please can you share how (in terms of full syntax/command) you generated forest plots that report rates in 100, 1000 or other multiples of person-years (as opposed to simple fractions)?. I seem to be stuck at this point.

(I am conducting a systematic review and meta-analysis of mortality rates. When I use rates (as simple fractions of 'count of events'/'person years') and use these rates in a metan command

(e.g. metan log_mortality_rate standard_error_of_mortality_rate, eform random sortby(log_mortality_rate) lcols(author year person_year) texts(200) nulloff xlabel(5, 10)),

the forest plot does not report rates in multiples of person-years (1000 person-years) - understandably.

Otherwise, if I force the 1000 into the nominator or denominator in the MR calculation per se, I would get rates in 1000person-years, but this will cause the confidence intervals in the forest plots to be too narrow, and non-varying)

Thanks a lot.

Kareem