Interpretation of Forest Plot

[Hatty Wong]

Good morning/afternoon everyone,

I am currently performing the statistical analysis part to my Thesis exploring the association of recreational use of cannabis and depression among adolescents. I have copied and pasted the codes I use to run my meta-analysis below but would like to confirm with experts here that 1) My interpretation is correct and 2) Whether I am missing any crucial elements when I ran this meta-analysis that I should include.

The Meta inferential statistics, the metafunnel and metatrim did not work
(I'm not expert in this, so I would greatly appreciate if someone may explain why (if applicable), they are required for meta-analysis and 2) why they didn't work for my set of data).


Thank you very much!

 generate logor=log(OR)

 generate selogor=(log(UpperCI)-log(LowerCI))/1.96

 metan logor selogor, fixedi second(random)

 metainf logor selogor

 metafunnel logor selogor eform xtitle("Odd Ratio-Log Scale")

 metatrim logor selogor

My data is presented below:

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte StudyNumber str21 StudyName double(OR LowerCI UpperCI) float(logor selogor _LCI _UCI _WT)
 1 "Geogiades"                3.779  2.38   5.99  1.3294594  .4709137   .4064855 2.2524333 1.9647503
 2 "Gage"                      1.34  1.18   1.51  .29266962 .12581389  .04607894  .5392603 27.525354
 3 "Fergusson"               3.0219   2.3   3.97  1.1058857 .27849844   .5600388 1.6517327  5.617519
 4 "Gilder"                       3  1.45    6.2  1.0986123  .7413192  -.3543467 2.5515714   .792829
 5 "Harder"                  .74205   .53   1.03 -.29833865  .3389985  -.9627635  .3660862  3.791356
 6 "Kleinman"                1.1517   .63   2.12   .1412391   .619108 -1.0721902 1.3546684   1.13673
 7 "Rey"                     2.9546  1.98   4.42  1.0833633  .4097157   .2803352 1.8863914 2.5955224
 8 "Boys"                      6.55  3.87   11.1   1.879465 .53759724   .8257939  2.933136  1.507565
 9 "Rasic"                     1.24  1.06    1.5  .21511137  .1771409 -.13207845  .5623012 13.885205
10 "Saluja"                  2.6568  2.39   2.95   .9771224 .10740398   .7666144 1.1876303 37.770206
11 "Rossow"                       1     1      1          0         0          0         0         0
12 "Fergusson and Horwood" 1.403357 .9888 1.9918   .3388672  .3572969 -.36142185 1.0391563  3.412963
end

[Clyde Schechter]

It isn't helpful to say that something didn't work. We can't guess what that means: there are many ways in which a command might have failed. To get help on that please show us, in addition to the commands, exactly what Stata responded with (and, if it isn't obvious, how that differs from what you were expecting.) To show those responses, just copy them from the Results window or your log file into a code block.

[Hatty Wong]

Thank you Dr. Schechter.

The output STATA provided was most definitely obvious. I guess I'm just curious as to whether I'm even supposed to have executed those codes to begin with because to be completely honest, I am following strictly from manuals but don't really have a good understanding as to why they should be performed in the first place.

For Meta-trim, based on a definition I found from literature:
^metatrim^ performs the Duval and Tweedie nonparametric "trim and fill" method of accounting for publication bias in meta-analysis. The method, a rank-based data augmentation technique, formalizes the use of funnel plots, estimates the number and outcomes of missing studies, and adjusts the meta-analysis to incorporate the theoretical missing studies. The authors claim that the method is effective and consistent with other adjustment techniques.

metainf logor selogor

Standard error/variance/confidence limit must be>0 or missing for all studies

metatrim logor selogor

Standard error/variance/confidence limit must be>0 or missing for all studies

metafunnel logor selogor, xtitle(Log odds ratio) ytitle(Standard error of log OR) egger

note: default data input format (theta, se_theta) assumed
standard error of theta must exceed 0 in each study

Lastly, based on the Forest plot produced, I would like to believe and confirm that ES from either fixed-effect/random-effect demonstrate that there is an effect of cannabis use on depression.

[Clyde Schechter]

In all three instances, the problem you are having is coming from StudyNumber 11 where the standard error is zero.

I don't quite understand that observation in any case: how can a study produce a confidence interval for the odds ratio of 1 through 1: i.e. the odds ratio is 1.0 with no sampling error possible. That doesn't make sense. And if you look at the detailed output from the -metan- command you will see that that study is excluded. The other programs, rather than excluding it and hoping you'll notice, are telling you up front that there is invalid data and refusing to proceed. Correcting that observation (or eliminating it if you can't correct it) should solve your problems here.