Dear David
I have sent an email to you too, if you answer this massage please ignore the email. Thnak you for your always kindly answers.
Please see the attaches file with a forest plot.
In this plot there are some between-study variation, but the I2 is zero in all. My interpretation is that the Cochrane Q is too smal and smaller than Degree of freedom, resulting in negative estimate and thus set as zero. I chose the fixed effects model to present the analysis as there were no differences (similar RR) and to keep study sizes as they are. However, I have been asked to present the confidence interval of I2 ! But how can we do that when it is negative and zero ? Is there any way to display the CI? And if there is some variation in CI (which I do not believe that) does this justify to present data as random effects model ?
I analysed the same data in Cochrane meta-analysis package, the results are the same, all I2 are zero.
I have used these commands both in fixed and random models but the results still show zero I2
admetan n t2 N T2, rr fixed forestplot (xlabel(.1 .5 1 5 20 50 100)) lcols (N T n t) label(namevar=Study) by (type2) keepall keeporder nooverall
metan9 n t2 N T2, rr fixed forestplot (xlabel(.1 .5 1 5 20 50 100)) lcols (N T n t) label(namevar=Study) by (type2) keepall keeporder nooverall
Best regards
Jawdat Abdulla MD PhD FESC
Section of Cardiology/Department of Medicine
Glostrup University Hospital
Copenhagen
Denmark
I have sent an email to you too, if you answer this massage please ignore the email. Thnak you for your always kindly answers.
Please see the attaches file with a forest plot.
In this plot there are some between-study variation, but the I2 is zero in all. My interpretation is that the Cochrane Q is too smal and smaller than Degree of freedom, resulting in negative estimate and thus set as zero. I chose the fixed effects model to present the analysis as there were no differences (similar RR) and to keep study sizes as they are. However, I have been asked to present the confidence interval of I2 ! But how can we do that when it is negative and zero ? Is there any way to display the CI? And if there is some variation in CI (which I do not believe that) does this justify to present data as random effects model ?
I analysed the same data in Cochrane meta-analysis package, the results are the same, all I2 are zero.
I have used these commands both in fixed and random models but the results still show zero I2
admetan n t2 N T2, rr fixed forestplot (xlabel(.1 .5 1 5 20 50 100)) lcols (N T n t) label(namevar=Study) by (type2) keepall keeporder nooverall
metan9 n t2 N T2, rr fixed forestplot (xlabel(.1 .5 1 5 20 50 100)) lcols (N T n t) label(namevar=Study) by (type2) keepall keeporder nooverall
Best regards
Jawdat Abdulla MD PhD FESC
Section of Cardiology/Department of Medicine
Glostrup University Hospital
Copenhagen
Denmark
Comment