total zero event studies in meta-analysis: continuity correction

[Dika Luijendijk]

I am performing a meta-analysis with 23 studies. Fifteen studies have zero events in both treatment arms. These studies are omitted when using the metan command to calculate a pooled OR, but the common recommendation is to include them. I could not find the required Stata code. Also, Stata does not accept that I impute 0.5 in all zero cells by hand ("Non integer cell counts found"). Cheng et al have suggested that there is a method to do perform this continuity correction for total zero events studies in Stata (BMJ Open 2016). Can someone tell me how? I would be very grateful.

[Marcos Almeida]

Hello Dika,

Welcome to the Stata Forum.

Under the user-written - metan - program, the option - cc(0.5) - will provide the continuity correction.

You may wish to type - help metan - and see the help file so as to get further information on the core commands.

Best,

Marcos

[Dika Luijendijk]

Dear Marcos, Thank you for the suggestion. If I use cc(0.5), studies with zero in one arm are included. I checked help metan but cannot find additional instructions with respect to studies with zero events in both arms. Best, Diks

[Marcos Almeida]

Dear Dika,

It seems a quite specific situation, only "when treatment effects are unlikely", according to the authors of the cited article.

Just to provide "food for thought", and, besides, to follow the FAQ on how to provide references, here I share the link to the article written by Cheng at al. Impact of including or excluding both-armed zero-event studies on using standard meta-analysis methods for rare event outcome: a simulation study. BMJ Open 2016;6:e010983 doi:10.1136/bmjopen-2015-010983 :

http://bmjopen.bmj.com/content/6/8/e010983.full

Hopefully you will get further advice!

Best,

Marcos

[Dika Luijendijk]

Yesterday, I mailed Alessandro Battaggia with my question because he reported the inclusion of total zero events studies in a publication (https://www.ncbi.nlm.nih.gov/pubmed/26237758). He knew the Stata code, and with his permission I gladly post his response here:

Dear Dika,

I'm greateful for your attention.
With Stata's <metan> pakage it is not possible to perform a meta-analysis including trials with zero+zero events.
So, you can alternatively do that operation introducing a manually continuity correction as described in Sweeting 2004 (see 3.2 and Table 6 of accluded work) as "reciprocal of the opposite treatment arm size correction".

suppose in your dataset:

a= number of events in treatment arm
b= number of no events in treatment arm
c= number of events in control arm
d= number of no events in control arm
ab= a+b
cd= c+d

so, your do-file can be launched in this manner:

gen R=cd/ab
gen taTCC=1/(R+1)
gen taCCC=R/(R+1)
replace d=d+taCCC if a==0
replace c=c+taCCC if a==0
replace b=b+taTCC if a==0
replace a=a+taTCC if a==0
replace d=d+taCCC if c==0
replace b=b+taTCC if c==0
replace a=a+taTCC if c==0
replace c=c+taCCC if c==0
metan a b c d, fixed or noint lcols(trial)

Note that the < noint > option must be used because in this manner non integer numbers can be generated in cell of 2X2 table

Best wishes

Alessandro Battaggia

[Marcos Almeida]

Hello Dika,

Thank you so much for sharing Battagia's code!

Thanks, also, to Alessandro Battagia, for providing the code and giving permission to share it here.

Limitations and pitfalls of this particular method are presented in Cheng's article you mentioned in #1 and I shared in #4.

I gather your message provided a nice closure to this thread.

Thank you again.

Best,

Marcos